MR Regression

Public Economics
Gareth D. Myles
October 2001
Contents
Introduction
xiii
Foundations
1 Introduction
1.1 Public economics . . . . . . . . . . . .
1.2 Motivation . . . . . . . . . . . . . . .
1.3 Eciency versus equity . . . . . . . .
1.4 Information . . . . . . . . . . . . . . .
1.5 Methodology . . . . . . . . . . . . . .
1.6 Preview . . . . . . . . . . . . . . . . .
1.7 Public sector income and expenditure
1.8 Notation . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
3
5
6
7
8
9
11
16
2 General Equilibrium and Welfare Economics

2.1 Introduction . . . . . . . . . . . . . . . . . . .
2.2 The Arrow-Debreu economy . . . . . . . . . .
2.2.1 The institutional framework . . . . . .
2.2.2 Commodities . . . . . . . . . . . . . .
2.2.3 Consumers . . . . . . . . . . . . . . .
2.2.4 Producers . . . . . . . . . . . . . . . .
2.2.5 Equilibrium . . . . . . . . . . . . . . .
2.2.6 Walras law . . . . . . . . . . . . . . .
2.2.7 Normalizations . . . . . . . . . . . . .
2.2.8 Relaxation of strict convexity . . . . .
2.2.9 Existence of equilibrium . . . . . . . .
2.2.10 Analysis of policy . . . . . . . . . . . .
2.2.11 Core of the economy . . . . . . . . . .
2.2.12 Net trades . . . . . . . . . . . . . . . .
2.3 Welfare properties of competitive equilibrium
2.3.1 Basic definitions . . . . . . . . . . . .
2.3.2 The First Theorem . . . . . . . . . . .
2.3.3 The Second Theorem . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
19
19
20
21
21
22
25
27
28
29
30
32
33
35
37
38
38
39
41
iii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
iv
CONTENTS
2.4
2.5
2.6
A critical appraisal . . . . . . . . . . . . . . . . .
2.4.1 Lump-sum transfers and taxes . . . . . .
2.4.2 The institutional assumptions . . . . . . .
2.4.3 Pareto optimality . . . . . . . . . . . . . .
Interpersonal comparisons and social welfare . . .
2.5.1 Interpersonal comparability . . . . . . . .
2.5.2 Social choice . . . . . . . . . . . . . . . .
2.5.3 Bergson-Samuelson social welfare function
Summary . . . . . . . . . . . . . . . . . . . . . .
3 Topics in Measurement
3.1 Introduction . . . . . . . . . . . . . . . .
3.2 The measurement of income . . . . . . .
3.3 Equivalence scales . . . . . . . . . . . .
3.3.1 Minimum needs . . . . . . . . . .
3.3.2 Engel and Rothbarth . . . . . . .
3.3.3 Prais and Houthakker . . . . . .
3.3.4 Barten . . . . . . . . . . . . . . .
3.3.5 General case . . . . . . . . . . .
3.3.6 In practice . . . . . . . . . . . .
3.4 The measurement of inequality . . . . .
3.4.1 Basic definitions . . . . . . . . .
3.4.2 Statistical measures . . . . . . .
3.4.3 Statistical measures and welfare
3.4.4 Generalised Lorenz Curves . . .
3.4.5 Welfare-theoretic indices . . . . .
3.4.6 Axiomatic inequality measures .
3.4.7 Summary . . . . . . . . . . . . .
3.5 The measurement of poverty . . . . . .
3.5.1 Relative or absolute? . . . . . . .
3.5.2 The poverty line . . . . . . . . .
3.5.3 Standard measures . . . . . . . .
3.5.4 Axiomatic approach . . . . . . .
3.5.5 Variable poverty line . . . . . . .
3.6 Conclusions . . . . . . . . . . . . . . . .
II
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
45
45
49
50
53
53
55
58
59
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
61
61
62
63
64
64
67
67
69
71
72
73
74
78
82
83
86
88
88
89
90
90
91
95
95
Analysis in the competitive model
4 Commodity Taxation
4.1 Introduction . . . . .
4.2 Methodology . . . .
4.3 The Ramsey rule . .
4.3.1 The economy
4.3.2 Derivation . .
4.3.3 Interpretation
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
97
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
99
99
100
101
101
102
104
CONTENTS
4.3.4 Implications . . . . . . . . . . . .
4.3.5 Inverse elasticities rule . . . . . .
4.4 Extension to many-households . . . . .
4.4.1 The optimal tax rule . . . . . . .
4.4.2 Reduction to the Ramsey rule . .
4.4.3 A cautionary note . . . . . . . .
4.5 Numerical results . . . . . . . . . . . . .
4.5.1 Applications . . . . . . . . . . .
4.5.2 Further results . . . . . . . . . .
4.6 Generalising the production technology
4.7 Untaxed goods . . . . . . . . . . . . . .
4.8 Uniform taxes . . . . . . . . . . . . . . .
4.9 Production eciency . . . . . . . . . . .
4.10 Summary . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
106
107
108
108
111
112
113
113
116
117
121
124
126
129
5 Income Taxation
5.1 Introduction . . . . . . . . . . . . . . . .
5.2 The Mirrlees economy . . . . . . . . . .
5.2.1 The optimisation . . . . . . . . .
5.2.2 Basic structure . . . . . . . . . .
5.2.3 The structure of utility . . . . .
5.2.4 Self-selection . . . . . . . . . . .
5.3 Characterisation of optimal tax function
5.3.1 The general problem . . . . . . .
5.3.2 Linear taxation . . . . . . . . . .
5.3.3 Nonlinear taxation . . . . . . . .
5.3.4 Summary . . . . . . . . . . . . .
5.4 Numerical results . . . . . . . . . . . . .
5.4.1 Simulations . . . . . . . . . . . .
5.4.2 Choice of specification . . . . . .
5.5 Extensions and omissions . . . . . . . .
5.5.1 Income and commodity taxes . .
5.5.2 Omissions . . . . . . . . . . . . .
5.6 Summary . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
131
131
132
132
133
134
138
140
140
141
144
153
153
154
156
159
160
162
163
6 Policy Reform
6.1 Introduction . . . . . . . . . . . . . . . . . . . . .
6.2 The reform problem . . . . . . . . . . . . . . . .
6.2.1 Productive feasibility . . . . . . . . . . . .
6.2.2 Improving reforms . . . . . . . . . . . . .
6.3 Characterising possibilities . . . . . . . . . . . . .
6.3.1 Solution via Theorems of the Alternative
6.3.2 Geometric analysis . . . . . . . . . . . . .
6.3.3 Informational requirements . . . . . . . .
6.4 Productive eciency and maximising reforms . .
6.4.1 The possibility of ineciency . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
165
165
165
166
169
171
171
172
176
177
178
vi
CONTENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
180
181
182
183
184
186
187
188
188
191
7 Risk
7.1 Introduction . . . . . . . . . . . . . . . . .
7.2 General equilibrium with risk . . . . . . .
7.2.1 Risk in the Arrow-Debreu economy
7.2.2 Eciency . . . . . . . . . . . . . .
7.2.3 Including time . . . . . . . . . . .
7.2.4 Complete markets . . . . . . . . .
7.2.5 Comments . . . . . . . . . . . . . .
7.3 Private and social attitudes to risk . . . .
7.3.1 Private attitudes . . . . . . . . . .
7.3.2 Social attitudes . . . . . . . . . . .
7.3.3 Summary . . . . . . . . . . . . . .
7.4 Household choice and taxation . . . . . .
7.4.1 Wealth taxation . . . . . . . . . .
7.4.2 Income tax . . . . . . . . . . . . .
7.4.3 Imperfect loss osets . . . . . . . .
7.4.4 Generalisations . . . . . . . . . . .
7.5 Labour supply and occupational choice . .
7.5.1 Labour supply . . . . . . . . . . .
7.5.2 Occupational choice . . . . . . . .
7.6 Optimal taxation . . . . . . . . . . . . . .
7.6.1 Private risk . . . . . . . . . . . . .
7.6.2 Aggregate risk . . . . . . . . . . .
7.7 Conclusions . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
193
193
194
194
196
197
198
199
200
200
202
208
208
209
212
214
216
217
217
218
221
221
224
226
8 Corporate Taxation
8.1 Introduction . . . . . . . . . . . . . . . . . . . .
8.2 Taxation of the firm . . . . . . . . . . . . . . .
8.2.1 Input and output taxes . . . . . . . . .
8.2.2 Profit tax . . . . . . . . . . . . . . . . .
8.2.3 Personal taxes . . . . . . . . . . . . . .
8.2.4 Why tax the corporation? . . . . . . . .
8.3 Tax incidence . . . . . . . . . . . . . . . . . . .
8.3.1 Tax incidence in the Harberger economy
8.3.2 Extensions . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
227
227
228
228
229
230
230
231
231
237
6.5
6.6
6.7
6.8
6.9
6.4.2 Interpretation . . . . . .
Lump-sum taxes . . . . . . . .
The inverse optimum . . . . . .
6.6.1 Theory . . . . . . . . .
6.6.2 Applications . . . . . .
Marginal social cost of taxation
6.7.1 Derivation . . . . . . . .
6.7.2 Empirical application .
Political Constraints . . . . . .
Conclusions . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
CONTENTS
8.4
III
vii
8.3.3 Taxation and finance . . . . . . . . . . . . .

8.3.4 Systems of corporate and personal taxation
8.3.5 Finance and investment with certainty . . .
8.3.6 Uncertainty . . . . . . . . . . . . . . . . . .
Conclusions . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Relaxing the Assumptions
9 Public Goods
9.1 Introduction . . . . . . . . . . . . . . . . . . . .
9.2 Definitions . . . . . . . . . . . . . . . . . . . . .
9.2.1 Pure public goods . . . . . . . . . . . .
9.2.2 Impure public goods and congestion . .
9.3 Optimal provision . . . . . . . . . . . . . . . .
9.3.1 Pure public good . . . . . . . . . . . . .
9.3.2 Free disposal . . . . . . . . . . . . . . .
9.3.3 With congestion . . . . . . . . . . . . .
9.3.4 Public input . . . . . . . . . . . . . . .
9.4 Personalised prices and the Lindahl equilibrium
9.4.1 Simple model . . . . . . . . . . . . . . .
9.4.2 A general treatment . . . . . . . . . . .
9.4.3 Core equivalence . . . . . . . . . . . . .
9.4.4 Cost-share equilibria . . . . . . . . . . .
9.4.5 Comments . . . . . . . . . . . . . . . . .
9.5 Private provision of public good . . . . . . . . .
9.5.1 Equilibrium . . . . . . . . . . . . . . . .
9.5.2 Pareto improvements . . . . . . . . . . .
9.5.3 Quantity of provision . . . . . . . . . .
9.5.4 The number of households . . . . . . . .
9.5.5 Invariance results . . . . . . . . . . . . .
9.5.6 Alternative formulations . . . . . . . . .
9.5.7 Summary . . . . . . . . . . . . . . . . .
9.6 Finance by taxation . . . . . . . . . . . . . . .
9.6.1 Identical consumers . . . . . . . . . . .
9.6.2 Dierentiated households . . . . . . . .
9.6.3 Summary . . . . . . . . . . . . . . . . .
9.6.4 Definitions . . . . . . . . . . . . . . . .
9.6.5 Dominant strategies . . . . . . . . . . .
9.6.6 Nash equilibrium . . . . . . . . . . . . .
9.6.7 Bayesian equilibria . . . . . . . . . . . .
9.6.8 Conclusions . . . . . . . . . . . . . . . .
9.7 Experimental evidence and market data . . . .
9.7.1 Experimental evidence . . . . . . . . . .
9.7.2 Market data . . . . . . . . . . . . . . . .
9.8 Conclusions . . . . . . . . . . . . . . . . . . . .
240
241
242
249
254
255
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
257
257
258
258
259
259
260
262
262
263
264
265
267
270
271
272
272
273
275
276
277
279
280
282
282
282
286
287
288
289
294
296
296
297
297
300
301
viii
10 Externalities
10.1 Introduction . . . . . . . . . . . . . .
10.2 Externalities . . . . . . . . . . . . .
10.2.1 Definitions . . . . . . . . . .
10.2.2 Representation . . . . . . . .
10.3 Market ineciency . . . . . . . . . .
10.3.1 Equilibrium and ineciency .
10.3.2 Pareto irrelevant externalities
10.4 The Coase theorem . . . . . . . . . .
10.4.1 The Theorem . . . . . . . . .
10.4.2 Markets for externalities . . .
10.4.3 Non-existence of markets . .
10.4.4 Bargaining . . . . . . . . . .
10.4.5 Conclusions . . . . . . . . . .
10.5 Welfare-improving changes . . . . . .
10.5.1 Local results . . . . . . . . .
10.5.2 Global results . . . . . . . . .
10.6 Corrective taxation . . . . . . . . . .
10.6.1 Non-uniform taxation . . . .
10.6.2 Uniform taxation . . . . . . .
10.6.3 Summary . . . . . . . . . . .
10.7 Tradeable licenses . . . . . . . . . .
10.7.1 Certain costs and benefits . .
10.7.2 Prices versus quantities . . .
10.7.3 Nonlinear pricing . . . . . . .
10.8 Internalisation . . . . . . . . . . . .
10.9 Conclusions . . . . . . . . . . . . . .
CONTENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
303
303
304
304
305
306
306
308
309
309
311
313
316
320
320
321
323
324
325
327
331
331
332
333
335
335
336
11 Imperfect Competition
11.1 Introduction . . . . . . . . . . . . . . . . . . . . .
11.2 Imperfect competition and general equilibrium .
11.2.1 Objective and subjective demand . . . . .
11.2.2 Price normalisations . . . . . . . . . . . .
11.2.3 The economy . . . . . . . . . . . . . . . .
11.3 Imperfect competition and welfare . . . . . . . .
11.3.1 Failure of Pareto optimality . . . . . . . .
11.3.2 Measures of welfare loss . . . . . . . . . .
11.4 Commodity taxation . . . . . . . . . . . . . . . .
11.4.1 Tax incidence . . . . . . . . . . . . . . . .
11.4.2 Optimal taxes . . . . . . . . . . . . . . . .
11.5 Ad valorem and specific taxes . . . . . . . . . . .
11.5.1 Tax incidence . . . . . . . . . . . . . . . .
11.6 Optimal combinations . . . . . . . . . . . . . . .
11.7 Production eciency and the taxation of labour
11.7.1 Production eciency . . . . . . . . . . . .
11.7.2 Taxation of labour . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
337
337
338
338
339
340
343
343
344
346
347
351
357
357
359
362
363
366
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
CONTENTS
ix
11.8 Other forms of regulation . . . . . . . . . . . . . . . . . . . . . . 367

11.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
12 Tax
12.1
12.2
12.3
12.4
12.5
Evasion
Introduction . . . . . . . . . . . . . . . . . . . . . . . .
The extent of tax evasion . . . . . . . . . . . . . . . .
Evasion as a decision with risk . . . . . . . . . . . . .
Optimal auditing and punishment . . . . . . . . . . .
Tax evasion and labour supply . . . . . . . . . . . . .
12.5.1 Labour supply . . . . . . . . . . . . . . . . . .
12.5.2 Allocation of hours . . . . . . . . . . . . . . . .
12.6 Public goods . . . . . . . . . . . . . . . . . . . . . . .
12.6.1 The valuation of public funds . . . . . . . . . .
12.6.2 Congestible public goods . . . . . . . . . . . .
12.7 Empirical evidence . . . . . . . . . . . . . . . . . . . .
12.8 Honesty and social norms . . . . . . . . . . . . . . . .
12.9 Optimal auditing with an independent revenue service
12.10Tax evasion by firms . . . . . . . . . . . . . . . . . . .
12.10.1 Competitive firms . . . . . . . . . . . . . . . .
12.10.2 Imperfect competition . . . . . . . . . . . . . .
12.11Optimal taxation with evasion . . . . . . . . . . . . .
12.11.1 Commodity taxation . . . . . . . . . . . . . . .
12.11.2 Income taxation . . . . . . . . . . . . . . . . .
12.12Summary . . . . . . . . . . . . . . . . . . . . . . . . .
IV
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Introducing Real Time
13 Overlapping Generations Economies

13.1 Introduction . . . . . . . . . . . . . . . . . . . . .
13.2 Overlapping generation exchange economies . . .
13.2.1 The Samuelson economy . . . . . . . . . .
13.2.2 Overlapping generations as Arrow-Debreu
13.2.3 Existence of equilibrium . . . . . . . . . .
13.2.4 Money, dynamics and indeterminacy . . .
13.2.5 Summary . . . . . . . . . . . . . . . . . .
13.3 An aggregate production economy . . . . . . . .
13.3.1 Consumers . . . . . . . . . . . . . . . . .
13.3.2 Production . . . . . . . . . . . . . . . . .
13.3.3 Equilibrium . . . . . . . . . . . . . . . . .
13.3.4 Steady State Solution . . . . . . . . . . .
13.3.5 Golden rules . . . . . . . . . . . . . . . .
13.3.6 Pareto optimality . . . . . . . . . . . . . .
13.4 Conclusions . . . . . . . . . . . . . . . . . . . . .
371
371
371
373
376
377
378
378
380
381
382
385
387
389
392
392
394
395
395
397
397
399
. . . . . .
. . . . . .
. . . . . .
economies
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
401
401
402
403
405
407
412
418
419
419
420
421
424
425
428
429
x
14 Social Security
14.1 Introduction . . . . . . . . . . . . . .
14.2 Fully funded and pay-as-you-go . . .
14.3 An optimal program . . . . . . . . .
14.4 Some extensions . . . . . . . . . . .
14.4.1 Labour supply and retirement
14.4.2 Eect on savings and capital
14.4.3 Ricardian equivalence . . . .
14.4.4 Demographics . . . . . . . . .
14.5 Determination and justification . . .
14.5.1 Voting equilibria . . . . . . .
14.5.2 Altruism . . . . . . . . . . . .
14.5.3 Myopia . . . . . . . . . . . .
14.5.4 Uncertainty . . . . . . . . . .
14.6 Lifetime uncertainty . . . . . . . . .
14.6.1 Symmetric information . . .
14.6.2 Adverse selection . . . . . . .
14.6.3 Moral hazard . . . . . . . . .
14.7 Conclusions . . . . . . . . . . . . . .
CONTENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
431
431
432
433
436
436
441
443
445
447
447
449
452
454
459
460
462
466
467
15 Debt and Taxes

15.1 Introduction . . . . . . . . . . . . . . . . . . .
15.2 The eect of debt . . . . . . . . . . . . . . . .
15.2.1 External and internal debt . . . . . .
15.2.2 Eects of debt . . . . . . . . . . . . .
15.3 Optimisation of debt and lump-sum taxes . .
15.3.1 Lump-sum taxation and debt . . . . .
15.3.2 Optimal lump-sum taxation and debt
15.3.3 Summary . . . . . . . . . . . . . . . .
15.4 Debt neutrality . . . . . . . . . . . . . . . . .
15.4.1 Two special cases . . . . . . . . . . . .
15.4.2 Intergenerational altruism . . . . . . .
15.4.3 Generalisations . . . . . . . . . . . . .
15.5 Income and interest taxation . . . . . . . . .
15.6 Conclusions . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
469
469
470
470
471
477
478
480
482
483
484
485
490
492
496
16 References
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
499
Preface
This is lightly edited version of the first edition that has beenconverted to Tex
from Word. Since this has necessitated re-typing all of the maths, errors may
have been introduced. You are welcome to use it provided you report any typing
errors you discover to G.D.Myles@ex.ac.uk so that I can put things right.
I would also welcome any comments you may have which could improve the
second edition. I will be adding exercises and additional chapters on voting,
rent seeking, political economy, clubs and local public goods, and international
issues in taxation.
xi
Introduction
Do I really need an introduction here?
xiii
Part I
Foundations
Chapter 1
Introduction
1.1
Public economics
In the broadest interpretation, public economics is the study of economic policy,

with particular emphasis upon taxation. The subject therefore encompasses topics as diverse as responses to market failure due to the existence of externalities
and the determination of optimal social security policies. This characterization
reflects an extension of the scope of public economics from its initial emphasis
upon the collection and disbursement of government revenues to its present concern with all aspects of government economic intervention. The intention of this
book is to provide an introduction to the vast literature of public economics,
emphasizing the foundations upon which future research can be laid
Public economics has a long history as a discipline within economics and
many eminent economists have written on the subject. For example, Ricardo
(1817) discussed the eects of public debt, the incidence of taxation in imperfectly competitive markets was analyzed by Cournot (1838), Edgeworth (1925)
considered the eects of taxation on multi-product firms and Pareto (1909) set
out the foundations for making social decisions. The explanation for this interest in public economics is no doubt contained in the close connection of the
analysis with policy and application, which are the ultimate inspiration of most
economists. Exposing a theoretical construction to policy analysis also highlights its value and provides a test of its relevance. However, it is also true that
before a good policy can be designed an adequate theory must be developed.
One of the challenges of public economics is that much of the subject area is
still in its infancy with considerable work still to be done.
Although a number of partitions could be used to break down the subject
matter of public economics into convenient portions, the most instructive division is between that of determining the eects of alternative policies and that
of determining the optimal policy. This division represents the distinction between the exercise in positive economics involved in calculating the change in
equilibrium caused by the introduction of a policy and the normative exercise
3
CHAPTER 1. INTRODUCTION
of evaluating, in terms of welfare, the outcome of policy. To achieve the first

objective requires a theory that describes how economic agents choose their actions and how these actions are aected by changes in policy. The individual
agents must then be combined to form an economy and a theory of equilibrium
provided for this economy. The evaluation of policy, and the choice of optimal
policy, necessitates the specification of an objective for the policy maker that
is capable of providing a measure of the performance of each policy based on
the relevant features of the equilibrium resulting from the policy. This evaluation process represents an application of normative economics. The success of
public economics has largely followed from the systematic application of these
methods.
The theory that is described in the following chapters has developed mainly
since 1970 and has built upon developments in microeconomics, macroeconomics, general equilibrium theory and game theory. One of its characteristic
features is the use of duality techniques to allow problems to be phrased in the
manner most amenable to solution. These techniques permit optimization exercises to be phrased in terms of the natural choice variables. In this context,
the work of Diamond and Mirrlees (1971) was of fundamental importance in
introducing these methods into public economics. The use of duality theory has
also allowed many problems to be studied with great generality and has often
overcome the need to impose restrictive sets of assumptions. A second characteristic feature is the consistent use of general equilibrium theory to provide a
rigorous foundation for the policy analysis. A general equilibrium analysis of
policy captures both the direct eects of policy and the secondary eects. As
the latter may well outweigh the former, a convincing policy analysis cannot
be conducted except within a general equilibrium context. These underlying
methods of duality and general equilibrium provide the cohesion to what at
first glance may appear to be a number of disparate topics.
An emerging trend in the public economics literature has been the use of numerical methods. These have taken the form of both simulations of economies
in order to test their behavior and the evaluation of policy proposals using empirical data. The latter technique indicates a promising convergence between
theory and application and is clearly a direction in which the subject will continue to move. Although this book is primarily intended to be a text on the
theory of public economics, numerical results are given prominence due to their
obvious importance.
The dominant setting for the analysis of public economics is within the
mixed economy so that individual decisions are respected but the government
intervenes to aect these choices. The design of policy can then be interpreted
as the manipulation of individual choices by the choice of policy parameters so
as to arrive at an equilibrium preferred to that which would arise in the absence
of policy. This makes the results of the studies applicable to most developed
economies and concurs with the present ascendancy of such a form of economic
organization. To provide a benchmark from which to judge the outcome of the
economy under alternative policies the perfectly controlled command economy
with an omniscient planner is often employed. Naturally, this usage of the
1.2. MOTIVATION
command economy implies no claim that such perfect control is possible, or

even desirable.
1.2
Motivation
The motivation for the study of public economics follows naturally from the
observation that unregulated economic activity does not lead to a socially optimal outcome. At a very basic level, an economy could not function eectively
if there were no contract laws since this would inhibit satisfactory exchange. In
addition, although the anarchic equilibrium that would occur without contracts
may be in the core of the economy, it need not be particularly stable (Bush and
Mayer (1974)). It must therefore be accepted that no economy could operate
without law enforcement and that in order for organized economic activity to
take place, there must be a clearly defined and enforced set of contract laws.
These laws cannot be policed free of cost. There is also a need for the enforcement of more general criminal laws and for the provision of a means of defence
for the nation. These are also costly activities.
Consequently, even the minimal requirements of the enforcement of contract
and criminal laws and the provision of defence need the collection of revenue
to provide the required finance. This is the case whether these services are
provided by the state or by private sector organizations. The coordination of
the collection of revenue and the provision of services to ensure the attainment
of ecient functioning of economic activity therefore provides a natural role
for a central state in any economy that wishes to develop beyond the most
rudimentary level. In addition, this reasoning also illustrates that to achieve
even the most minimal level of eciency and organization of economic activity
some unavoidable revenue requirements are generated and require financing.
Having determined that the organization of economic activity must generate
a revenue requirement, one aspect of the role of public economics is to determine
how this revenue can be collected at the least cost to the economy. Although
the concept of least cost has several possible interpretations, both positive and
normative, under any interpretation the aim of the economic policy design would
be that of finding an ecient means of revenue collection. Such design would
involve the identification of feasible policy instruments from the set of possible
policies, the choice of policy instruments to be imposed from amongst those
that are feasible and the calculation of the optimal level of each instrument.
The issue of eciency in policy design is a continuing and central theme of
public economics.
Moving beyond the basic requirements for organized economic activity, it is
arguable that there are other situations where state intervention in the economy has the potential to increase welfare. Unlike the basic revenue requirements
however, there will always be a degree of contentiousness about further intervention motivated on these grounds. The situations where state intervention
may be warranted can be divided into two categories: those that involve market
failure and those that do not. With market failure, the argument for consid-
ering whether intervention would be beneficial is compelling. For example, if

economic activity generated externalities, so that there is divergence between
private and social costs and the competitive outcome is not ecient, it may
be felt necessary for the state to intervene to limit the ineciency that results.
This latter point can also be extended to other cases of market failure such as
those connected to the existence of public goods and of imperfect competition.
Where market failure does not occur, state intervention can be motivated
by the observation that although an equilibrium may be ecient it need not
be optimal according to the states welfare criterion. Such a situation may
arise if the equilibrium of the economy is characterized by widespread poverty
and an inequitable distribution of income. In such circumstances, the level of
economic welfare as viewed by the state may well be raised by a program of
income redistribution. Similar arguments can be applied to the provision of
state education, social security programs and compulsory pension schemes. It
should be stressed that such potential increases are with respect to normative
assessments of welfare, unlike the positive criteria lying behind the concept of
economic eciency.
In the cases of both market failure and welfare-motivated policies, policy
intervention concerns more than just the ecient collection of revenue. The
reasons for the failure of the economy to reach the optimal outcome have to be
understood and a policy that can counteract these has to be designed. It must
also be recognized that the actions of the state, and the feasible policies that it
can choose, are often restricted by the same features of the economy that make
the competitive outcome inecient. In each case, policy intervention can only
be justified by proving that the state can actually improve upon the market.
That it can always do so should not be taken for granted. Extending the scope
of the public economics to address such issues provides the breadth to public
economics.
1.3
Eciency versus equity
In conducting an economic policy the state will generally have two conflicting
aims. On the one hand, it will aim to implement the policy with the minimum
loss to society. The use of policy will cause a loss due to the resources used in
the implementation process and from the economic distortions that the policy
will cause. Minimizing these losses is the eciency aspect of policy design.
Conversely, the state may also feel that it is desirable to intervene in the economy
in order to attain a more equitable distribution of the economys resources. This
is often accompanied by a corresponding reduction in the degree of concern for
the aggregate level of economic activity. This motivation represents the equity
side of policy design.
Due to their distinct natures, it is inevitable that the aims of equity and
eciency regularly conflict. It is often the case that the ecient policy is highly
inequitable whilst the equitable policy would introduce into the economy significant distortions and disincentives. Given this fact, the design of optimal policy
1.4. INFORMATION
can be seen as the process of reaching the correct trade-o between equity and
eciency objectives. This optimum trade-o will depend upon the concern for
equity that is expressed in the objectives of the policy maker. In many analyses
of policy problems, the resolution of the trade-o between equity and eciency
is the major determinant of the resulting policy program, with aspects of the
policy being attributable to one or the other. This distinction is often a helpful
way in which to think about optimality problems and their solutions. It is worth
stressing that the conflict between equity and eciency does not always arise.
For example, in some instances of uncertainty, such as the provision of social
insurance discussed in chapter 8, the two aims of eciency and equity may not
be competing.
To illustrate this discussion, a simple example of the conflict between equity and eciency can be found in the optimal taxation of commodities. Under
assumptions that will be described later, it is ecient to tax goods with low
elasticities of demand, as shown by the well-known inverse elasticity rule, since
this introduces the least distortion into the pattern of demand. However, goods
with low elasticities of demand tend to be necessities that are consumed disproportionately by less well-o households. Taxing these goods highly would then
cause a proportionately greater reduction in the welfare of poor households. The
proposed tax program is therefore highly inequitable and equity criteria would
shift the taxes onto goods consumed by higher income groups. The DiamondMirrlees tax rule that is developed in Chapter 5 shows how this conflict between
eciency and equity is resolved.
In this context, it is worth adding one final note concerned with modelling
techniques. A standard simplification that will be employed on a number of
occasions in this book is to work with one-consumer economies or with economies
composed of a population of identical consumers. In such economies there can
be no distributional issues, so the resulting policy recommendations are based
only on considerations of eciency. This generally leads to results that can
be expressed much more clearly and precisely than would be possible if equity
considerations were present, which is of considerable assistance when an issue
is analyzed for the first time. In considering the practical value of such results,
the implications of introducing equity considerations must always be borne in
mind.
1.4
Information
The role of information is central to public economics. The availability of information to private agents determines the nature of the equilibrium without
policy intervention and the information set of the government determines feasible policy instruments. If information deficiencies, particularly asymmetric
information between agents in the economy, lead the market outcome to be inecient, the state can only improve the outcome if it is not subject to the same
informational limitations.
As will be made clear in Chapter 2, the first-best outcome could be sustained
if the state levied lump-sum taxes that were contingent upon all economically
relevant characteristics of the agents in the economy. Naturally, some of these
characteristics will be private information and therefore not directly observable
by the state. If the state cannot induce the agents to truthfully reveal these characteristics, then the lump-sum tax system that supports the first-best cannot
be implemented. Policy design then involves the optimal utilization of the available information. The outcome that is achieved will necessarily be second-best.
This simple example demonstrates the essential consequences of informational
restrictions and captures themes that will recur throughout the book.
Further examples are easily found. In the context of commodity taxation,
limited information prevents commodity taxes being dierentiated between consumers. It also results in the use of an income tax levied upon the observable
income of households, rather than an ability tax on their unobservable earning potential. The optimal provision of public goods is also prevented by the
fact that the government cannot observe consumers willingness to pay for such
goods. The outcome in each of these cases is described in the relevant chapters.
Although asymmetries of information are at the heart of most of the analysis
that follows, their nature will rarely be made explicit. Instead, the nature of
information will be implicit in the assumptions that describe the structure of
the economies employed and the restrictions that are placed upon feasible policy
instruments. In considering results derived below, it will always be advantageous
to reflect upon the nature of the informational restrictions involved and the
consequences of their relaxation or strengthening.
1.5
Methodology
The method of analysis adopted within this book is invariably to consider policy within the context of a general equilibrium representation of the economy.
In Parts I to III of the book, the underlying framework is the Arrow-Debreu
economy and its extensions. The overlapping generations economy employed
in Part IV is a particular infinite horizon version of the general formulation.
The general equilibrium perspective is maintained wherever it is feasible to do
so. This aim is not achieved in places such as when only partial equilibrium
treatments are available in the existing literature, or when a general equilibrium
treatment would only obscure the major issues.
A general equilibrium economy is undoubtedly the most appropriate framework to adopt since it is the only means by which all the repercussions of a
policy may be captured. In contrast, concentration upon partial equilibrium
can lead to important consequences of policy being overlooked, particularly if
there are significant adjustments in markets other than those forming the focus
of the analysis. However, partial equilibrium analysis is often useful as a means
of obtaining preliminary insights into a problem, but its limitations should never
be underestimated. Although most attention in the literature has been focused
upon competitive economies, with market imperfections are now being widely
used and policy analysis within such economies will also be considered.
1.6. PREVIEW
Within a general equilibrium context, changes in policy, or alternative policies, can be viewed as resulting in dierent equilibria for the economy. To treat
the question of optimality in policy choice, the equilibria for dierent policies
are contrasted via some welfare measure. The optimal policy is then defined as
the feasible policy yielding the greatest level of welfare. Since Pareto optimality
generally provides too incomplete a ranking of states to be of use as a guide
for policy, the welfare criterion that is used is typically a Bergson-Samuelson
social welfare function. This procedure naturally invokes questions about the
comparability of individual utilities and the formulation of welfare measures.
The analytical tools used in the book are generally fairly simple and the
mathematics rarely uses anything more dicult that the theory of constrained
maximization and comparative statics analysis. There are exceptions to this
rule. Separation arguments are used in Chapters 2 and 9. Chapters 6 and
15 employ the maximum principle, Chapter 7 employs a theorem of the alternative and Chapter 13 touches upon nonlinear dynamic systems. Dynamic
programming is used in 15. Since a self-contained treatment of general equilibrium analysis and welfare economics is given in Chapter 2, the economics
that is employed, but not otherwise introduced in the text, should be covered
by any advanced undergraduate or graduate course in microeconomics and involves mainly standard duality results in producer and consumer theory. An
excellent source of reference for this material is Varian (1992).
1.6
Preview
Following the discussion of the methodology that will be employed, it is clear

that a necessary starting point is a review of the competitive general equilibrium
economy and the standard results of welfare economics. These represent the
content of Chapter 2. The chapter introduces the agents involved in the economy
and characterizes economic equilibrium. It also introduces a useful perspective
from which to view Walras law and formalizes the notion of how policy changes
are modelled. Emphasis is placed upon the institutional assumptions underlying
the competitive equilibrium economy since much of the subject matter of public
economics can be motivated by discomfort with the nature of the equilibrium
of the competitive market or by the failure of one or more of the institutional
assumptions. The Two Theorems of Welfare Economics are proved and a critical
analysis of their scope and relevance is provided. This naturally leads into a
discussion of the measurement of social welfare and the contrast between the
implications for interpersonal comparability of ordinal and cardinal utility.
In practice, if economists are to be concerned about equity it is necessary
to find measures that capture aspects of the distribution of welfare. Chapter
3 considers three topics that have been the subject of considerable interest.
Alternative methods of constructing equivalence scales are reviewed as a prelude
to developing measures of distribution. The measurement of poverty and of
inequality are then used to illustrate the axiomatic approach to measurement,
which specifies the properties that an index should have and then constructs the
10
indices that have those properties, and the statistical approach which simply
uses readily accessible formulae. These three introductory chapters constitute
Part I of the book.
Part II is concerned with the analysis of five topics of fundamental importance for public economics which are conducted within the competitive economy.
Chapter 4 analyses the eects of corporate taxation and the issues that need
to be taken into account in tax design. Chapters 5 and 6 consider the characterization of optimal commodity and income taxes respectively. Both of these
chapters illustrate the resolution of the equity/eciency trade-o in the design
of policy and the consequences of government informational limitations. In addition to the theoretical analysis, the results of simulations and applications of
the methods to data are considered. The numerical results are useful since the
theoretical analysis leads only to characterizations of optimal taxes rather than
explicit solutions. Policy reform is the subject matter of Chapter 7 and the
link is drawn between the existence of improving reforms and non-optimality.
The question of production eciency along the reform path is also considered.
Chapter 8 on public economics in the presence of risk completes part II.
The focus of Part III is upon the consequences of relaxing certain of the
institutional assumptions on which the competitive economy is based. Chapter
9 introduces public goods into the economy and contrasts the allocation that
is achieved in the private provision equilibrium with the optimal allocation.
Methods of financing public goods are also considered and this analysis has
close parallels with the chapter on commodity taxation. Private provision and
preference revelation are also addressed. The treatment of externalities in Chapter 10 relaxes another of the institutional assumptions. It is shown why market
failure can occur and reviews alternative policy schemes designed to improve
eciency. Imperfect competition and its consequences for commodity taxation
is the subject of Chapter 11. The measurement of welfare loss is discussed and
a general equilibrium economy with imperfect competition is developed. Emphasis is given to the incidence of taxation and the determination of optimal
taxes. A distinction is also drawn between the eects of specific and ad valorem
taxes. Part III is completed by Chapter 12 on tax evasion and administration.
Estimates of the size of the black economy are reviewed, the tax evasion decision
is modelled and the eect of evasion and administration costs on optimal taxes
are determined.
Part IV concentrates upon intertemporal issues in public economics. The
first Chapter, 13, describes the overlapping generations economy that is the
main analytical tool of this part. The relationship between the overlapping
generations economy and the Arrow-Debreu economy is emphasized, as are the
dierent eciency properties of the two economies. The concept of the Golden
rule is introduced for economies with production and capital accumulation. The
chapter also touches upon the dynamic adjustment process of the economy.
Chapter 14 analyses social security policy and relates this to the potential nonoptimality of the competitive equilibrium. Both the motivation for the existence
of social security programs and the determination of the level of benefits are addressed. The interaction between debt and taxation is the subject of Chapter
1.7. PUBLIC SECTOR INCOME AND EXPENDITURE
11
15. The eects of government debt and the issue of debt neutrality are considered. Ricardian equivalence is linked to the existence of gifts and bequests. The
choice between the taxation of income and expenditure is also analyzed.
1.7
Public sector income and expenditure
The public sector plays an important role in the mixed economies of the major
industrialized countries. To show quite how important, this section presents
some summary statistics concerning the size and structure of the public sector.
Whilst there are some well-recognized issues concerning the appropriate definition of the public sector, these do not aect the validity of the broad sketch
given here.
Table 1.1 shows the pattern of public sector total outlay as a percentage of
nominal GNP over the period 1978 - 1993 for seven of the major industrialized
countries from North America, Europe and Asia. For these countries, public
sector expenditure falls in the range of 30 - 55% of GNP with Japan and the
United States having the smallest public sectors and Italy and France the largest.
Even though the range is large, the public sector is significant in every case.
Expenditure in Italy shows sustained growth through the period, as it does in the
U.S. but to a lesser extent. Other than these countries, the pattern is generally
one of the public sector being a constant proportion of GNP. This relative
stability over the recent past is in sharp contrast to the period of expansion of
the public sector experienced by the industrialized countries from 1890 through
to 1970.
The major implication of Table 1.1 is that it clearly justifies the claim that
the public sector is significant in the economies of the industrialized countries
and the mixed economies of these countries are characterized by substantial
government involvement. They are far from being free-market with minimal
government intervention. The size of the public sector alone is justification for
the study of how it should best choose its means of revenue collection and its
allocation of expenditure. It is also worth noting that data on expenditure typically understates the full influence of the public sector upon the economy. For
instance, regulations such as employment laws or safety standards infringe upon
economic activity but without generating any measurable government expenditure or income.
Table 1.2 shows the proportion of Japanese government income derived from
various sources and the division of its expenditure. The chart for income shows
that direct taxation is the largest single component. Social security contributions and indirect taxation are the next largest and make fairly similar contributions to income. In terms of expenditure, social security spending is the largest
category followed by purchases of goods and services. Interest on public debt is
also a significant item of expenditure.
A similar breakdown of income and expenditure is reported for the United
Kingdom in Table 1.3. Contrasted to Japan, the U.K. shows greater reliance
upon indirect taxation, with indirect taxation generating slightly more revenue
12
Government expenditure
60
50
United States
Japan
Germany
France
Italy
United Kingdo
Canada
40
30
20
1970
1980
1990
2000
Year
Figure 1.1: General government total outlay as % of nominal GDP
Division of income
Direct tax
Social security
Indirect taxes
Property income
Figure 1.2: Japanese Government Income 1991
13
Division of expenditure
Goods and services

Interest on public debt
Subsidies
Social security
Other transfers
Figure 1.3: Japanese Government Expenditure 1991
than direct taxation. The relative size of social security contributions is also
much less than in Japan. The relative sizes of the expenditure items are very
similar, although the U.K. spends more on goods and services but less on subsidies. The social security item in Japan is equivalent in relative size to the
transfers in the U.K.
Tables 1.2 and 1.3 demonstrate the importance of direct and indirect taxation in the collection of revenue for the U.K. and for Japan. Taken together,
these generate 73% of revenue in the U.K. and 63% in Japan. The third item of
income, social security contributions, are 17% of income in the U.K. and 27% in
Japan. These figures support the prominence given to the design of commodity taxation in chapter 5, income taxation in chapter 6 and social security in
chapter 14.
An alternative perspective on the relative importance of the three major
categories of income is given in Table 1.4. This shows receipts as a percentage
of GDP for the U.S. and as an average for other OECD countries. For the U.S.,
consumption taxes are relatively less important than as shown for Japan and
the U.K. above and as against the average over OECD countries. However,
consumption tax receipts still equal over 4% of U.S. GDP. Social security taxes
raise twice the income of consumption taxes whilst income tax receipts represent
one tenth of GDP. In contrast, the OECD average shows rather more equality
between receipts from income and consumption taxes.
Table 1.5a shows the expenditure of the U.S. Federal Government broken
down into type and function, expressed as a percentage of total expenditure.
Similarly, Table 1.5b has the same breakdown for State and Local Government.
These tables reveal that the major items of expenditure for Federal Government
14
Division of income
Direct tax
Social security
Indirect taxes
Property income
Figure 1.4: UK Government Income
Division of expenditure
Goods and services

Interest on public debt
Subsidies
Transfers
Figure 1.5: UK Government Expenditure
15
Figure 1.6: Comparison of Tax Yields
Other
Commercial activities
Net interest paid
Labour training
Economic development
Postal service
Transportation
Natural resources
Agriculture
Energy
Recreational and cultural
Housing and community service
Veterans benefits
Income support, social security
Health and hospitals
Education
Civilian safety
National defence
Space
International affairs
Central executive
0
10
20
30
40
50
(a) Federal Government
Figure 1.7:
are income support and social security, and defence. In contrast, the major item
for State and Local Government is education followed by income support and
social security. Other than these, the most significant items are the net interest paid by the Federal Government and transportation and civilian safety
paid for by State and Local Government. The items can be placed into separate categories representing the breakdown of public sector objectives: defence
expenditure is one of the minimal requirements; income support is evidence
of concern for equity; and education represents provision of a public good to
counter market failure.
Although brief, this review of statistics on the size and structure of public
sector income and expenditure has illustrated the significant extent of public
sector intervention in the mixed economies of the industrialized countries. The
16
Other
Commercial activities
Net interest paid
Labour training
Economic development
Postal service
Transportation
Natural resources
Agriculture
Energy
Recreational and cultural
Housing and community service
Veterans benefits
Income support, social security
Health and hospitals
Education
Civilian safety
National defence
Space
International affairs
Central executive
-10
10
20
(b) State and Local
Figure 1.8: US State and Federal Expenditure

relative importance of alternative sources of revenue has been shown as has the
range of expenditure.
1.8
Notation
There are many symbols that have a constant meaning through the text. The
most important of these are listed below.
H number of households
m number of firms
n number of goods
pi pre-tax price (of good i)
qi post-tax price
ti tax rate
R government revenue requirement
G public good supply
xh consumption plan (of household h)
X h consumption set
`h level of labour supply (when distinguished from other commodities)
h initial endowment
I h lump-sum income
M h total income
h wealth
T h lump-sum tax paid by h
1.8. NOTATION
17
h marginal utility of income

h social marginal utility of income
U h () household utility function
V h () indirect utility function
E h () expenditure function
W () social welfare function
y j production plan (of firm j)
Y j production set
j profit level
hj share of h in j
Xi aggregate demand (for good i)
Yi aggregate supply
Zi excess demand
E [] expectations operator
Kt capital stock (in period t)
kt capital-labour ratio
rt interest rate
wt wage rate
In addition to these symbols, there are others which are more chapterspecific.
The is one further notational convention that must be noted. Throughout
the text round brackets proceeded by a symbol always represent functions so
that f (x) denotes the function f of x. In contrast, square brackets denote
algebraic operations so that f [x] is read f times x. The only exception to this
rule is the expectations operator .
Chapter 2
General Equilibrium and

Welfare Economics
2.1
Introduction
The competitive general equilibrium economy described in this chapter has been
developed over a considerable period of time and has been subjected to closer
study than any other form of economy. The earliest formal construction is
generally attributed to Walras (1874) and this was developed in the German
literature of the 1930s. These developments reached maturity in the seminal
contribution of Arrow and Debreu (1954) and in the monograph by Debreu
(1959); hence its common title of the Arrow-Debreu economy. The importance
of the Arrow-Debreu contribution was the demonstration that an equilibrium
existed for the economy under reasonable assumptions and the formal elaboration of the welfare properties of equilibrium. This chapter provides an informal
but self-contained introduction to the Arrow-Debreu general equilibrium economy. The discussion emphasizes those aspects of the economy that are most
relevant to the study of public economics. Technical aspects of the economy
that are not strictly necessary for later analysis will be noted but, where there
is no harm from doing so, will not be discussed in detail. Formal treatments
of the issues dealt with here can be found in Arrow and Hahn (1971), Debreu
(1959) and Hildenbrand and Kirman (1988).
The Arrow-Debreu economy is studied for two primary reasons. Firstly, it
provides the analytical foundation for the economies analyzed in later chapters
which are often simplifications or modifications of the general framework. The
focus of Part III of the book will be on the consequences of relaxing the assumptions on which the Arrow-Debreu economy is based so the results of this chapter
act as a benchmark from which to judge the eects of the relaxation. The second
reason is that welfare properties of the economy, which are commonly known as
the Two Theorems of Welfare Economics, are used as the basis for claims concerning the eciency, and thus desirability, of the competitive outcome. These
19
20CHAPTER 2. GENERAL EQUILIBRIUM AND WELFARE ECONOMICS

theorems represent the formulization of Smiths (1776) notion of the eciency
of the invisible hand of competition and play a central role in welfare economics.
The First Theorem states that a competitive equilibrium is optimal, in a sense
to be made precise below, and the Second Theorem that any optimum can be
decentralized as a competitive equilibrium. An understanding of the economy
is therefore a prerequisite for appreciating the content and limitations of these
claims.
The Theorems provide the motivation for two alternative viewpoints upon
economic policy. One viewpoint would be to take the Theorems as evidence
that policy should always be attempting to move the economy as close to the
competitive ideal as possible. In this interpretation, the Theorems are seen as
prescriptive of what should be achieved. The alternative view is that the Two
Theorems provide a description of what could be achieved if the economy were
competitive and a demonstration of why it cannot, and possibly should not, be
achieved in practice. The cannot refers to assumptions of the economy that will
not be met in practice and the should not to distributional aspects of competitive equilibrium. The mainstream of public economics combines parts of both
of these caricatured viewpoints. Many policy analyses take for granted the presumption that competitive behavior should be advocated and encouraged when
it leads to economic eciency. However, its failings are also readily admitted
and policies designed to correct for them. Indeed, much of public economics
implicitly takes as its starting point the rejection of the practical value of the
Second Theorem.
The chapter begins by introducing the assumptions that describe the underlying framework of the economy and this is followed by discussions of Walras
law, normalizations and the eect of policy; all being of particular relevance for
the understanding of the techniques employed in public economics. A formal
demonstration of the Two Theorems and a critical discussion of their content
and implications is given. The criticism focuses upon the possibility of designing and employing optimal lump-sum taxes, the assumptions underlying the
structure of the economy and on the limitations of the Pareto criterion. The
discussion of the Two Theorems leads naturally into a number of other basic
topics in welfare economics. The Pareto criterion is considered in detail and the
role of interpersonal comparisons of utility and the possibility of making such
comparisons are described. This is followed by a discussion of the formulation
of social welfare functions and the relation of these to the permissible extent of
interpersonal comparisons.
2.2
The Arrow-Debreu economy
This section introduces the constituent parts of the Arrow-Debreu economy,

defines an equilibrium and sketches a proof of existence. The role of price
normalizations and Walras law are emphasized for their importance in public
economics.
2.2. THE ARROW-DEBREU ECONOMY
2.2.1
21
The institutional framework
The institutional framework consists of those assumptions that describe the basic structure of the economy. Included within this structure are the nature of the
agents that constitute the active participants in the economy and the description of the trading environment. Such assumptions need to be distinguished
from the technical assumptions (such as convexity and continuity) made upon
preferences and technology in order to prove theorems about the economy.
There are two agents in the standard Arrow-Debreu economy: consumers
(or households) and producers (or firms). A third agent, the government, will be
included at a later stage. The households own initial endowments of goods and
have shares in firms which yield dividend payments. They engage in trade to
maximize their satisfaction or utility. Producers use inputs to produce outputs
subject to the technological knowledge they have available. Their aim is to
maximize profit, which is the dierence between revenue and costs. All profits
earned are distributed to shareholders.
It is assumed that all trade takes place at a given date. The possibility
of making contracts at the trading date for delivery at future dates allows the
introduction of time but the accounts for such contracts must be settled when
the contract is formed. Uncertainty can also be introduced by allowing contracts
to be written with delivery contingent upon the state of nature that arises, but
this will not be done formally until risk is considered in Chapter 8. More
importantly, no trade is permitted to take place except at equilibrium prices.
The essential requirement that must be satisfied for the economy to be competitive is that all agents treat prices as parametric. That is, in determining
their optimal action, the agents do not believe that their decision can aect the
prices observed. Formally, this belief can only be consistent with reality when
agents are infinitesimally small relative to the market (Aumann (1964)). Since
the economies that are considered below (with the exception of the overlapping generations economy of Part IV) have a finite number of consumers, Aumanns conditions cannot be satisfied and competitive behavior is imposed as
an assumption. The competitive assumption implies that there is no monopoly
power and hence no market distortions through price-setting. Households and
firms are all taken to act as independent units and to interact only via the price
system. This ensures that there are no external eects and no public goods.
These assumptions provide the basic framework of the Arrow-Debreu economy. They should not be viewed as immutable concepts but rather as a starting
point for the analysis. The eventual aim should be to choose the correct set of
assumptions that capture the economic reality of the situation of interest. The
analysis of Part III shows how this can be done by introducing external eects,
public goods and imperfect competition.
2.2.2
Commodities
The concept of a commodity is central to the economy and the flexibility in

the definition of commodities gives the economy its breadth of interpretation.

Commodities are simply defined as the set of goods that are available during the
operation of the economy. The definition brings with it the implication that at
the trading date all present and future commodities must be known to firms and
households. This does not imply that all commodities are available for delivery
at the trading date because some contracts may only be written for delivery at
some future date. In this way, the economy can cope with the introduction of
new products and is consistent with the observation that the set of available
products tends to change over time.
It is assumed that there is a finite number n of commodities for which trades
can take place. These commodities are indexed i = 1, ..., n. Each commodity
is distinguished by its location and its time of availability. Thus bread available for delivery today is a dierent commodity to bread available for delivery
tomorrow and bread available for delivery today in Coventry is a dierent commodity to bread available in Exeter. The list of commodities is intended to be
exhaustive and, as already noted, includes all presently available commodities
and all commodities that will be available for delivery at some specified future
date. A quantity of some, or possibly all, of the commodities is held as an initial
stock by each household; this is the households initial endowment. The initial
stock of a commodity is augmented by the productive activities of firms if it is
an output and diminished if it is an input.
To each commodity is associated a price. For good i the price is denoted
pi . This price can be given two essentially identical interpretations. The first
interpretation is that pi is the number of units of numeraire that have to be
surrendered in exchange for one unit of commodity i, where the numeraire is
that good denoted as having unit price. The alternative interpretation is that
pi is the price in terms of some unit of account (ie. money). The distinction
between these definitions is that money may be a purely artificial construct,
whereas the numeraire is one of the commodities. For the functioning of the
economy, the interpretation does not actually matter. The structure of the
household and firm decision problems that are described below make it clear
that it is only relative prices, ppji , that determine choices. Since these ratios are
independent of the interpretation of prices, the interpretation is irrelevant.
2.2.3
Consumers
Consumers are one type of economic agent in the economy. Each consumer
brings to the economy an initial endowment of goods and also holds shareholdings in the firms. The consumers use the income from the sale of the endowment
and from dividend payments to purchase their preferred choice of commodities.
These commodities are then consumed.
The number of consumers is fixed and is given by H. This can either be
interpreted as the number of individual consumers or as the broadest partition of
the set of consumers into distinct consumption units. In terms of the economies
used in public economics, the second interpretation is often adopted and, when it
is, a consumption unit can be viewed as a household rather than as an individual.
This is acceptable provided the household acts as if following a single objective.
23
In this book the terms household and consumer are interchangeable in Parts
I to III (with household invariably adopted) but are carefully distinguished in
Part IV.
Each of the H households, h = 1, ..., H has a consumption set, X h , that
describes feasible consumption plans. This should be distinguished from the
budget set, defined below, which describes aordable consumption plans. As
an example of the kind of restriction that is embodied in the consumption set,
twenty-four hours of work per day and no food would not be regarded as a
feasible consumption plan and the consumption set would therefore not include
this plan. The consumption set is assumed to be convex.
Each household, h, also has a utility function that represents its preferences.
The adoption of utility functions to represent preferences is not restrictive since,
as shown by Debreu (1954a), the conditions necessary for a functional representation of preferences to exist are very weak. In fact, the required assumptions
are that preferences are reflexive, transitive, complete and continuous. However, the comparability of utilities between households is more troublesome and
is discussed later in the chapter. The utility function is assumed to be strictly
quasi concave, so the set
h

h h
x : U h xh U h x
, x Xh ,
(2.1)
is strictly convex for any x

h X h . This assumption is equivalent to assuming
the preference order is strictly convex.
The utility function of household h is written
U h xh1 , ..., xhn ,

(2.2)
where xhi is the consumption of good i by household h. If good i is supplied
by the household, as would be various forms of labour services, then xhi < 0.
Household h also has an initial allocation, or endowment, of the n goods given
by the vector
h h1 , ..., hn .
(2.3)
Included in this endowment is the stock of labour services that the household
can supply. The household is assumed to liquidate this endowment and to use
the resulting income to purchase desired commodities.
The shareholdings of household h in the m firms in the economy are denoted
h1 , ..., hj , ..., hm ,
(2.4)
where each hj 0. Hence if firm j makes a profit of amount j , household h

receives a dividend of size
hj j ,
(2.5)
from firm j. As all profits are distributed, it must also be true that the firm is
fully owned by households. Across all households, the individual shares must
therefore sum to 1, hence
H
X
hj = 1, all j = 1, ..., m.
(2.6)
h=1
In the framework of the basic competitive economy there is no market for these
shareholdings so they remain fixed at their initial values.
With these definitions it is now possible to describe the economic behavior
of the household. Each household h chooses a consumption vector
xh xh1 , ..., xhn ,
(2.7)
to maximize their utility function U h () subject to the budget constraint

n
X
i=1
pi xhi
n
X
i=1
pi hi +
m
X
hj j ,
(2.8)
j=1
and subject to the consumption vector being feasible in the sense that it belongs
to the consumption set of h. The budget constraint simply requires that the
value of expenditure is not more than the value of the endowment plus dividends
received.
Under the maintained assumption of strict quasi concavity of the utility
function, the solution of the households maximization problem will result in a
demand function for each good, i, from household h . The demand of household
h for good i is written in the form
xhi = xhi p, h , h ,
(2.9)
where p, h , h , are the variables that the household takes as parametric. As
already noted, if xhi > 0, good i is regarded as being consumed and if xhi < 0,
good i is regarded as being supplied. The corresponding outcome when the
assumption of strict quasi concavity is relaxed will be discussed in Section 2.9.
The important concept for the existence of equilibrium is the level of aggregate demand from the consumption sector which is calculated by summing the
individual demands of households. Carrying out the summation, the aggregate
demand for good i is given by
Xi =
H
X
xhi = Xi (p, , , ) .
(2.10)
h=1
As individual demands may be negative, it is clearly possible for the aggregate

demand for a good to be negative.
In this discussion of the household it has been implicit that a solution to
the utility maximization problem exists. Conditions that guarantee that this is
permissible will be given in the discussion of the existence of equilibrium.
25
Figure 2.1: A Typical Production Set
2.2.4
Producers
The producers in the economy are the firms that take inputs and turn them
into outputs. Inputs may come from the initial endowments of households or
they may be intermediate goods that are produced by other firms. Each firm
is characterized by the technology that it has available and aims to maximize
profits by their choice of a production plan.
The description of the individual firms begins with the production technology. Each firm, j, is described by its production set, Y j , which represents the
technology of the firm. This set describes feasible input-output combinations
that are known to the firm. In other words, it is a list of all the alternative combinations of inputs and outputs of which the firm has knowledge. An example
of a typical production set for a firm operating in an economy with two goods
is illustrated in Figure 2.1.
Figure 2.1 has adopted the standard convention in general equilibrium theory
of measuring inputs as negative numbers and outputs as positive. The reasoning
behind this convention is that the use of a unit of a good as an input represent a
subtraction from the stock of that good available for consumption. In addition,
when aggregation takes place across firms, the inputs of one firm will cancel
with the output of another so that the aggregate represents net changes in the
stock.
The important features of Figure 2.1 that capture commonly made assumptions on the structure of production sets are that it is a strictly convex set
and the origin and the negative orthant are included in the set. In addition,
no strictly positive vector is in the production set. The inclusion of the origin
captures the possibility of inactivity on the part of the firm; at some prices the

firm may choose not to produce at all. Including the negative orthant can be
interpreted as allowing the firms to freely dispose of inputs. Finally, a strictly
positive vector would represent the production of outputs with no use of inputs.
This cannot be permitted.
Consider the firm shown in Figure 2.1 choosing the production plan described
by the vector y j = (2, 3). When faced with the price vector p = (2, 2), the
firms level of profit, which is given by the inner product of the price vector and
the production vector, is
j = py j = (2, 2) . (2, 3) = 2.
(2.11)
The representation of profit as an inner product is mathematically convenient

and illustrates the value of the sign convention. In addition, dividing the inner
product into positive and negative components shows that the positive part
can be given the interpretation of sales revenue and the negative part becomes
production costs.
More generally, in an n-good economy, each firm will choose a production
plan y j , where
y j y1j , ..., ynj ,

(2.12)
to maximize profit subject to y j being in the production set Y j . It is assumed
that the production set is strictly convex, this will be relaxed in Section 2.9.
The firm thus solves the maximization problem
py j subject to y j Y j .
max
j
{y }
(2.13)
The maximization in (2.12) determines the firms supply of each good, which
will be negative if the good is an input and positive if an output, as a function
of the price vector. Firm js supply function for good i is
yij = yij (p) .
(2.14)
As noted in the discussion of household demand, it is the level of aggregate

supply rather than individual firms supply that is important for equilibrium
existence questions. Aggregate supply is formed from the supply decisions of
the individual firms by summing across the firms. This gives aggregate supply
as
m
X
Yi =
yij (p) = Yi (p) .
(2.15)
j=1
Note that if good i is an input for some firms and an output for others, the
sign convention results in these being cancelled out so that Yi (p) represents net
supply from the productive sector. The vector of aggregate supplies Y (p) =
(Y1 (p), ..., Yn (p)) will have both positive and negative elements.
2.2.5
27
Equilibrium
The equilibrium of the economy occurs when demands and supplies are in balance. In such a state, each agent is able to carry out its planned action and
has no reason for wanting to modify its plan. The equilibrium state is important because it is presumed to be the position that the economy will achieve.
However, why the economy should actually reach the equilibrium is not entirely
clear. It is sometimes argued that equilibrium is reached as the outcome of a
dynamic adjustment process but such dynamics fit very uncomfortably with the
static nature of the model. This issue is still subject to debate. The procedure
adopted here is to follow the tradition of focusing upon equilibrium both for the
above reason and because of the lack of any satisfactory alternative.
To permit a precise definition of equilibrium, first observe that the level of
profit of each firm can be written as a function of the price vector by using the
supply function defined in (2.14) to write
j = py j = py j (p) = j (p) .
(2.16)
Hence, using (2.16) and the fact that w and are constant to eliminate them
as arguments of the functions, aggregate demand (2.10) can be written
Xi = Xi (p, (p)) = Xi (p) .
(2.17)
Equation (2.17) expresses aggregate demand as a function of the price vector

alone. Next, define the excess demand for good i, , as the dierence between
demand and supply, or
Zi (p) = Xi (p) Yi (p)
H
X
hi .
(2.18)
h=1
In reading (2.18) it should be recalled that the total supply of each good is the
sum of the initial endowment and the additional net output of the firms. If
Yi (p) < 0 then the available quantity is less than the endowment due to some
of good i being used in the production process.
A natural definition of equilibrium is that demand must equal supply or
supply is greater than demand and the price is zero. The second part of the
definition reflects, for example, the possibility that the economy may be endowed
with a quantity of a good for which no household or firm has any use. Phrasing
this in terms of excess demand, equilibrium occurs when excess demand is zero
or negative for all goods, with the price of a good being zero if its excess demand
is negative. Stated formally, a set of equilibrium prices satisfies
Zi (p) 0, i = 1, ..., n and if Zi (p) < 0, pi = 0.
(2.19)
The use of excess demand is just a convenient alternative way of saying that
either demand equals supply or there is excess supply and the good is free.
The equilibrium is assumed to be found by adjustment of the price vector
until (2.19) is satisfied. Any price vector that satisfies (2.19) is termed an

equilibrium price vector or a set of equilibrium prices. The existence question
is concerned with whether there is a solution to (2.19). However, even though
it can be shown that an equilibrium price vector will always exist under fairly
weak conditions, the question remains as to how such an equilibrium is reached
since there is no notion of price formation in the description of the economy.
2.2.6
Walras law
Walras law provides a result that carries significant implications for the analysis
of the general equilibrium economy. Two dierent statements of the law will be
given, with the second having an important role in public economics.
Taking excess demand to be less than or equal to zero for the n goods
provides n equations in (2.19) to be solved simultaneously. However, the content
of Walras law is that these n equations are not independent and that only n 1
actually need to be solved. To show this result, first note that as each household
is satisfying their individual budget constraint and the firms are distributing
their entire profit to shareholders, the value of each agents demand is equal to,
or less than, the value of their supply. Summing over all agents it must be true
that the aggregate value of demand cannot be greater than the aggregate value
of supply. Starting with the individual budget constraint
n
X
i=1
pi xhi
m
X
hj j +
j=1
n
X
pi hi ,
(2.20)
i=1
summing over all households gives

n
X
i=1
pi Xi
H
m X
X
hj j +
j=1 h=1
H X
n
X
pi hi ,
(2.21)
h=1 i=1
where the first term follows from the definition of aggregate demand. Now
recalling (2.6) and (2.16), (2.21) can be written
n
X
i=1
pi Xi
n X
m
X
i=1 j=1
pi yij +
H X
n
X
pi hi .
(2.22)
h=1 i=1
Using the definition of aggregate supply in (2.15), (2.22) can be expressed as

n
X
i=1
pi Xi (p)
m
X
j=1
pi Yi (p) +
n
X
i=1
pi
H
X
hi .
(2.23)
h=1
or
n
X
i=1
pi Zi (p) 0.
(2.24)
29
Equation (2.24) is the standard form of Walras law. In words, Walras

law can be stated succinctly: the aggregate value of excess demand is nonpositive. It should be noted that Walras law holds for all price vectors, not
just equilibrium prices. In the important case in which all households are nonsatiated, so that they spend their entire income, there will be equality in (2.24)
and the value of excess demand will be precisely zero.
Assuming non-satiation and returning to the equilibrium conditions (2.19),
the equality form of Walras law implies that if n 1 markets have zero excess
demand so must the nth. Hence there are only n 1 independent equations in
(2.19) since the value of any n 1 imply the value of the nth. This may seem
to imply that there are only have n 1 equations to determine the n prices
so that an equilibrium price vector will generally not exist but, as it is only
relative prices that determine trade patterns, it is only necessary to solve for
n 1 relative prices.
Walras law does have another implication that has been exploited in public economics. The statement above can be modified to the following. If n
markets are in equilibrium and all agents but one are satisfying their budget
constraint, the remaining agent must also be satisfying its budget constraint.
The consequence of this statement in an economy with a government is that
if the n markets are in equilibrium and households are meeting their budget
constraints, the government must also be meeting its budget constraint. When
describing such an economy it is therefore optional to include the government
budget constraint as an equation and to consider equilibrium on n1 markets or
to specify equilibrium on n markets and leave the government budget constraint
to be implicitly satisfied. Both of these approaches will be employed below.
2.2.7
Normalizations
It has already been noted that only relative prices matter in determining demands and supplies; this is clear from studying the household and firm maximization problems. This observation implies that there is a degree of freedom
in the measurement of prices since the scale of prices does not matter.
In proving the existence of equilibrium this freedom is invariably removed
by restricting prices to a suitable compact set that is capable of capturing all
feasible price ratios. The use of compactness provides a helpful restriction on
the set that has to be searched to find an equilibrium price vector. The most
commonly
used compact sets are the simplex, so that P
the prices must satisfy
Pn
n
2
i=1 pi = 1, or the unit sphere where the restriction
i=1 [pi ] = 1 must be
met.
When the Arrow-Debreu economy is used in public economics it is also necessary to choose a normalization rule. Without such a rule, it is not possible
to assign any real meaning to rates of taxation since some basis is needed from
which to measure these. The normalization rule that is normally adopted is to
select a good as the numeraire and to set its price at unity and, often in addition
to this, to fix its rate of tax at zero. Other prices and taxes are then measured
relative to this. This procedure is used repeatedly below. Further discussion of
Figure 2.2: Non-Uniqueness of Optimal Choice

normalizations is given in Chapters 5 and 11.
2.2.8
Relaxation of strict convexity
The assumption of strict convexity that has been made for preferences and
production sets results in the optimal choice for the household or firm being
unique at any given set of prices. As prices vary, so does the unique choice, but in
a way that can be represented by a point-valued function. It is this observation
that permitted the use of demand and supply functions in the development
above. This section briefly explores the consequences of relaxing the assumption
of strict convexity.
If the utility function is assumed instead to only be quasi concave rather than
strictly quasi concave, which is equivalent to the underlying preferences being
convex, this allows the possibility that the indierence curves may posses flat
sections. For example, a household who regards two goods as perfect substitutes
has indierence curves which are straight lines. Similarly, a firms production
set is convex rather than strictly convex, its boundary may be flat at some
points. The obvious example of this possibility is the constant returns to scale
technology for which the boundary is flat along any ray from the origin.
Whenever preferences or production sets have flat sections there will exist
price ratios at which the optimal choice will not be unique. This is illustrated
in Figures 2.2a and 2.2b. When such non-uniqueness arises the relationship
between prices and the optimal choice must be described by a set-valued function
or, as they are more commonly known in economics, a correspondence.
It should be apparent from Figure 2.2 that provided convexity is assumed,
the set of optimal choices will be convex. Furthermore, the definition of continuity for point-valued functions can be generalized to a definition for set-valued
functions. A correspondence (x) , : x y, y compact, is said to be up-
31
Figure 2.3: Semi-Continuity
per semi
at the point x0 if xq x0 , y q (xq ) , y q y 0 implies
0continuous

0
y x . It is lower semi continuous at the point x0 if xq x0 , y 0 x0
implies there is y q such that y q (xq ) , y q y 0 , . A correspondence that
is both upper and lower semi continuous at x0 is said to be continuous. The
distinction between upper and lower semi continuity is shown in Figure 2.3. The
bold lines and the area between them represent the graph of the correspondence.
In Figure 2.3a it can be seen how upper semi continuity allows explosion of the
graph at x0 while lower semi continuity in 2.3b permits implosion.
The relevance of semi continuity is that if preferences are convex, and some
further technical assumptions are satisfied, the resulting demand correspondence
is upper semi continuous. The aggregate demand correspondence is then upper
semi continuous as the sum of upper semi continuous functions. Similarly, if
the production set is convex then the supply correspondence, and the aggregate
supply correspondence, are upper semi continuous. Together, these result in
upper semi continuity of the excess demand correspondence. The interpretation
of Walras law is then that the value of any point in the image set of a price
vector must have non-positive value. This extension allows the important case
of constant returns to scale to be accommodated in the analytical framework.
To minimize the level of technical knowledge required to read this text, there
are very few places where correspondences are employed below. Instead, it is
typically assumed that sucient convexity is present to permit the employment
of point-valued functions. It should be borne in mind, however, that almost
all the results given can be extended to apply to correspondences and hence to
incorporate constant returns to scale.
2.2.9
Existence of equilibrium
The existence of an equilibrium for the competitive economy has been the subject of a long literature which is elegantly summarized in Debreu (1980). A
sucient set of conditions for equilibrium to exist are:
a) For each household preferences are continuous and convex, there is no
point of satiation in the consumption set and their endowment is interior to the
consumption set.
b) Each firm may choose to be inactive.
c) The aggregate production set is closed and convex, contains the negative
orthant and satisfies irreversibility.
It should be noted that the conditions in (a) are sucient to imply the existence of a utility function that represents the preferences. The role of these
assumptions is to guarantee upper semi continuity of the excess demand correspondence. Without such continuity, it would clearly not be possible to prove
the existence of an equilibrium as a supply-demand diagram for a single market
makes clear. The assumption of irreversibility states that if a production plan
y 6= 0 is feasible then the plan y is not. This assumption is made to guarantee
that the set of feasible production plans for the economy is bounded.
A sketch of the existence argument for the two-good case can easily be given
when it is assumed that excess demand is a continuous function of the price
vector. To do this it is useful to restrict prices so that they sum to unity, hence
p1 + p2 = 1, which demonstrates the use of the simplex in price normalization.
Any price vector must therefore end on the line joining (1, 0) to (0, 1).
Now consider the price vector (1, 0) in Figure 2.4. Along the horizontal axis
of the figure are measured the price of good 1 and the quantity of good 1. The
price and quantity of good 2 are measured on the vertical axis. Assume that
preferences satisfy non-satiation so that the equality form of Walras law can
be used and that the excess demand for each good becomes unbounded as its
price falls to zero. An investigation of the conditions that guarantee the second
assumption is given in Arrow and Hahn (1971). With non-satiation, equality in
the statement of Walras law in (2.23) implies that p and Z must be orthogonal
so that the angle between them is 90o . For p close to (1, 0) excess demand
must then lie in quadrant A since the price of good 2 is close to zero and excess
demand for it must be positive. Denote this excess demand vector by Z(1e, e).
Consider next a price vector p = (e, 1 e) close to (0, 1). By the same
argument excess demand must lie in quadrant C since excess demand for good
1 must be positive. Denote this excess demand vector by Z(e, 1 e).
The argument is completed by considering the eect of continuously changing
prices from p = (1 e, e) to p = (e, 1 e), which is a rotation of the price vector
upwards. It has already been assumed that the excess demand begins in A and
ends in C and excess demand has also been assumed to be a continuous function
of prices and to satisfy Walras law. Putting these facts together, the excess
demand vector must cross from quadrant A to C at some point in the rotation
of p. As excess demand is continuous and cannot enter quadrant B, it follows
that there must be at least one price vector for which excess demand is the zero
33
Figure 2.4: The Existence of Equilibrium

vector. It is otherwise not possible for excess demand to move continuously
from A to C.
The proofs of equilibrium in the n-good case given by Gale (1955) and Debreu
(1959) are essentially extensions of this brief sketch of the argument which
overlooks a number of technical diculties. In addition a formal proof of the
existence result requires the use of a fixed point theorem; a fact that was not
apparent in the argument given.
2.2.10
Analysis of policy
It is helpful at this point to consider the general method of analyzing policy

in the equilibrium economy set out above. A policy is considered to aect the
maximization decisions of the economic agents and, by changing the solution
to the maximization, to aect their behavior. The change in behavior is then
manifested by changes in demands and supplies and hence of the equilibrium.
To proceed formally, consider some set of L policy instruments whose values
are denoted by the vector = ( 1 , ..., L ). These could be a list of tax rates
or supplies of public goods or any other such policies. It will then follow that
demands and supplies will depend upon the values of these policy instruments,
so that excess demand can be written
Z = Z (p, ) .
(2.25)
Equilibrium prices will therefore depend on , as will the equilibrium quantities.

Thus
p = p () , X = X () .
(2.26)
Two alternative policies and 0 will generate, via the equilibrium prices, two
arrays of consumption levels
Figure 2.5: Continuity of Equilibrium

1
x () , ..., xH () , x1 0 , ..., xH 0 .
(2.27)
These arrays can then be compared according to the chosen indicator of

economic performance, be it a welfare measure or a statistic such as total output.
Naturally the policy achieving the higher value on this indicator will be preferred
and out of any set of potential policies, that achieving the highest value will be
the optimal policy.
This procedure invites two further questions. Firstly, assuming that individual households and firms decisions are continuously dependent upon the
policy parameters, will it be the case that the equilibrium values will also be
continuously dependent upon the policy? It is evident that the use and analysis
of policy will be dicult if a small change in policy leads to a discrete jump
in the equilibrium values. Fortunately, the answer to this question is that the
equilibria are almost always continuous with respect to the policy parameters.
The formal definition of almost always can be found in Mas-Collel (1980a), but
advanced methods are needed to provide the formulization.
Diagrammatically, the claim amounts to stating that equilibria such as a
in Figure 2.5, which is such that any small policy change that moves Z(p, )
downwards switches the equilibrium to b, are exceptional and that b, c and d
are the typical cases.
The exceptional nature of a can be judged by noting that slight modifications
of Z(p, ) will either remove the equilibrium at a entirely or lead to two equilibria
like c and d that are continuous in policy parameters. Therefore very few excess
demand functions out of the set of possible excess demands can generate such
equilibria and for practical purposes they can be ignored.
The second question is that of uniqueness of equilibrium: given any set of
policy parameters can we be certain that there is a single equilibrium? Unfortunately the answer to this is invariably no. Very strong restrictions are
35
required to ensure uniqueness, far stronger than can be justified by standard

restrictions on preferences and technology. Such results as have been found involve restrictions on the Jacobian matrix of the excess demand function and are
summarized in Arrow and Hahn (1971). The typical response to this diculty
is to simply assume uniqueness or at least not to worry about it. Although this
is unsatisfactory, there is little alternative.
2.2.11
Core of the economy
The concept of the core of the economy is derived from considering economic
activity as a cooperative game. This provides insights into the nature of the
competitive economy. To introduce the concept, consider an exchange economy
(so there is no production) consisting of H households. Each household has
an initial endowment of the n goods. Rather than conduct bilateral exchanges,
it is now assumed that the households form coalitions and allocate the total
endowment of each coalition amongst the members of that coalition. A coalition
can therefore be composed of between 1 and H households. At any time, a
household can belong to only one coalition.
H
Given some allocation for the economy x1 , ...,
x
h , coalition S can improve
upon this
for h S such that
Pif there is some allocation x
P allocation
h = hS h ;
(i) hS x
(ii) x
h is preferred to xh by all h.
h
Condition (i) asserts that the allocation x
is feasible for the coalition in
the sense that they can meet the allocation
from their
initial endowment. If
such a coalition exists, then the allocation x1 , ..., xH would not be accepted
by the members of the coalition. Instead, they would rather form the coalition
and benefit from the improved allocation. This process of potential coalition
formulation would continue indefinitely unless their exist some allocations which
cannot be improved upon. The core of the economy is defined as the set of
allocations which cannot be improved upon by any coalition.
The first issue that arises is whether there are any allocations that are in
the core. That the core is non-empty is proved in the following theorem
h
Theorem 1 If x
is
the equilibrium allocation for an Arrow-Debreu economy
h
withendowments
and p the corresponding equilibrium price vector, then
x
h is in the core of the economy.
Proof. Assume that the claim is not true

h so that the exists some
h coalition
S that can improve upon the allocation x
with allocation x
, h S.
P
P
This implies, from (i) that hS x
h = hS h . However, since x
h were
the optimal choices for the households in the economy at prices p, condition
(ii)
P
implies px
h > p h for all h S. Summing this over h S gives p hS x
h =
P
p hS h . This contradicts feasibility and proves the theorem. ||
This theorem shows that allocations achieved a competitive equilibria are
in the core of the economy. That there generally exist allocations in the core
Figure 2.6: Core of the Economy

that are not competitive equilibria can be seen in an Edgeworth box for a twohousehold, two-good economy. The core of such an economy is that part of the
contract curve which is in the trading set and the competitive equilibria are
generally a subset of this. This is shown in Figure 2.6.
Introducing additional households into a two-household economy makes possible the formation of new coalitions and it is clear that some coalitions will be
able to improve upon allocations that are in the core of the two-household economy. For instance, if the economy of Figure 2.6 is enlarged by the addition of
a second type 1 household and a second type 2 household, the coalition of two
type 1s and one type 2 can improve upon the allocation labelled x. The limit
of this process is described in the Debreu-Scarf core convergence theorem.
Theorem 2 (Debreu/Scarf ) Consider an exchange economy with m types of
household and r households of each type. Assume the preferences of each type
of household satisfy insatiability (given any consumption plan xh there exists x
h
h
strictly preferred to x ), strong convexity and continuity and that the endowment
of each household is strictly positive. Then, if an allocation is in the core for all
values of r, it is a competitive equilibrium allocation.
Proof. See Debreu and Scarf (1963). ||
Theorem 2.2 establishes that as the size of the economy increases by replication, the core of the economy shrinks to the set of competitive equilibria. This
observation provides an additional motivation for considering the competitive
equilibrium as the relevant outcome in a large economy. Although replication
by increasing the number of each type of household may seem rather artificial,
it is not actually necessary in order to derive the conclusion that the core converges to the set of competitive equilibria in the limit. As shown by Aumann
37
(1964), for a continuum economy the core and the set of competitive equilibria
are identical.
2.2.12
Net trades
When taxation is introduced into the competitive economy, it becomes necessary

to be more careful about the actual trades that are carried out by households.
The preceding analysis was conducted as if the household liquidated their entire endowment to obtain income and then purchased the consumption plan
they desired. This was the essence of the budget constraint in (2.7). Without
taxation, it is equivalent to think of the households as determining only net
trades in commodities, which are the quantities of each commodity they will
purchase over and above their initial endowment. For example, if a household
is endowed with 2 units of each of the two available goods and wishes to enjoy
the consumption plan (1, 3), their net trade pattern will be (1, 1) which will
be feasible provided the cost of the plan (1, 1) is less than or equal to zero.
All the arguments given above can be recast in terms of net trades simply by
subtracting the initial endowment from the demand functions determined in
(2.8).
Now consider commodity taxation in a competitive economy with no production. Such taxes can only be levied upon the trades that are observed by
the government. When levied they have the eect of driving a wedge between
the selling price of a commodity, p, and the purchase price, q, with the dierence q p being the commodity tax paid. If a household h were to liquidate
their endowment it would be valued at p h but would cost q h to re-purchase.
If all taxes were positive, the value of the endowment would be less than its
cost. It follows from this observation that it pays the household to minimize
the number of purchases of commodities that are subject to positive taxes. For
such commodities it is only the net trades that will be observed and taxed. In
contrast, the household will attempt to maximize the supply of those commodities with negative taxes: the entire endowment will be sold at price p and some
bought back at q < p. This distinction between selling prices and purchase
prices introduces a kink into the budget constraint at the endowment point.
When firms are introduced, the situation becomes slightly more complex.
Firms will always buy and sell at the prices p. If a household either buys from a
firm or supplies a firm they will transact at the prices q. When q > p the dierence between the two is a tax on the household when buying from the firm and a
subsidy to the household when supplying the firm. In contrast, in a transaction
between households, the household supplying the good must receive price p and
the household purchasing must pay price q. The actual budget constraint of a
household then becomes dependent upon whether transactions are conducted
with firms or with other households. To eliminate this diculty, the economies
studied in the following chapters implicitly assume that households only trade
with firms so that they face prices q for all transactions. Although this is an
important restriction, it does make the analysis considerably more tractable.
2.3
Welfare properties of competitive equilibrium
Section 2 has described the structure of the competitive Arrow-Debreu economy

and has characterized its equilibria. That the economy can be shown to possess
an equilibrium under very weak assumptions can be judged as fairly remarkable.
What is even more surprising is that this equilibrium also has certain features
of eciency; out of a system in which individual households and firms are all
pursuing their independent objectives, a final state emerges that exhibits a
considerable degree of cohesion.
The belief that the competitive equilibrium has certain properties of optimality appears to have been widely held for some considerable time prior to an
actual demonstration of the fact. The first proof of the result is generally attributed to Barone (1935) and this line of analysis, involving the use of calculus
to characterize optima, reached its final form in Lange (1942). This approach
was, however, severely limited by its use of calculus and the assumptions of interior maxima and smoothness of functions. Employing convexity and separation
theory permits a more formal and more general proof of the Two Theorems.
The modern presentation began with the work of Arrow (1951a). Although of
great generality, the model of Arrow worked with an aggregate production possibility set rather than with individual profit-maximizing firms. Debreu (1951)
presented similar results but with individual firms. The Theorems were presented in what has become their final form in Debreu (1954b). An excellent
discussion of this line of work and its interpretation is contained in Koopmans
(1957), and Takayama (1985) is a good source for further discussion.
2.3.1
Basic definitions
The first step is to give a formal definition of a Pareto optimum. For this it is
necessary to define the concept of a feasible array of consumption vectors.
Definition 3 Feasibility
An array of consumption vectors {x1 , ..., xH } is feasible if xh X h , all h,
and there exists an array of production vectors {y 1 , ..., y m }, each y j Y j , such
that
x y + ,
(2.28)
where
x=
H
X
h=1
xh , y =
m
X
j=1
yj , =
H
X
h.
(2.29)
h=1
This definition states that a given allocation of consumption bundles to

households is feasible if it can be produced using the economys fixed initial
endowment and production technology.
Pareto Optimality can now be defined.
2.3. WELFARE PROPERTIES OF COMPETITIVE EQUILIBRIUM
39
h
Definition 4 Pareto Optimality (P.O.) A feasible consumption
array x
is
h
Pareto optimal if there does not exist a feasible array x
such that
with
h
h
Uh x
, h = 1, ..., H,
Uh x
h
h
Uh x
, for at least one h.
Uh x
(2.30)
(2.31)
h
Therefore x
is Pareto optimal if it is not possible to find an alternative
h
feasible array which gives every household at least as much utility as x
and
gives strictly more utility to at least one household.
The competitive equilibrium described in section 2 can be given the following
formal statement.
h j
Definition 5 Competitive Equilibrium (C.E.) An array p, x
, yb is a
competitive equilibrium if
x
h X h , px
h p h +
m
X
j=1
hj pybj , h = 1, ..., H,
ybj Y j , j = 1, ..., m,
(2.32)
(2.33)
and

P
h
ybj ,
h U h xh for all xh X h such that pxh p h + m
(i) U h x
j=1 j p
j
j
j
j
(ii) pyb py , all y Y ,
(iii) x
yb + .
Equations (2.32) and (2.33) require households demands to be in their consumption sets and firms choices to be in their production sets. (i) implies that
the equilibrium choices of households maximize utility, (ii) that firms maximize
profits and (iii) that the equilibrium is feasible.
2.3.2
The First Theorem
To develop the First Theorem it is necessary to state a preliminary result. Let

household h have locally non-satiated preferences and let x
h be their chosen
consumption plan. Lemma 2.1 relating to the costs of preferred individual
consumption plans can then be proved.
Lemma 6 Let x
h be a locally non-satiating choice for household h at prices p.
Then

h
(i) U h xh > U h x
bh ,
b pbxh > pbx
h
h
h
h
h
(ii) U x = U x
bh .
b pbx pbx
Proof. If (i) were false then clearly xh would have satisfied the households
budget constraint and would have been chosen in preference to x
bh . To prove (ii)
suppose pbxh < pbx
bh . Since x
bh is not a point of local satiation neither is xh . There

h

h
then exists x
eh at a distance of from xh with U h x
e > U h xh = U h x
b .
The distance may be chosen small enough so that pbx
eh < pbx
bh . This contradicts
the assumption that x
bh was the optimal choice. ||
With this background it is now possible to state and prove the First Theorem.
h j
Theorem 7 (The First Theorem of Welfare Economics)
Let
, yb be
h pb, x
a competitive equilibrium with no household satiated at x

b . Then x
bh , ybj
is a Pareto optimum.
h j
Proof. Suppose x
b , yb
is not a Pareto optimum. Then there exists
h j
x , y
with xh X h , y j Y j and
(i)x y +,
h
(ii)U h xh U h x
b all h,
h
h
h
h
(iii)U x > U x
b some h.
P
P
Given (ii) and (iii), (a) and (b) imply H
bxh > H
bx
bh . Under local
h=1 p
h=1 p
nonsatiation (iii) of C.E. gives pbx
b = pbyb+b
p so it follows that pbx = pbyb+b
p. Profit
maximization ((ii) of C.E.) implies pybj py j , all y j Y j and, in particular,
that pybj py j . Summing over j, pyb
py.Hence
pbx > pby + pb or pb [x y ].
From this inequality it follows that xh , y j is not feasible thus proving
the theorem by contradiction. ||
To appreciate the generality of this theorem, it is worth reconsidering the
assumptions that were required in its proof. When this is done it can be seen
that all that was used, in addition to profit and utility maximization, was the
non-satiation of preferences, which is a very mild restriction, the existence of the
competitive equilibrium and finiteness of the number of goods and households.
Therefore, being based on such weak assumptions, the result cannot easily be
dismissed. The consequences of relaxing the assumption that the number of
goods and households is finite are addressed by the overlapping generations
economy of Chapter 13 which shows how Pareto optimality can fail when both
are infinite.
For a two household exchange economy, the First Theorem can be demonstrated by using an Edgeworth box diagram. In Figure 2.7 the Pareto optima
are given by the tangencies of the indierence curves and the locus of tangencies
determines the contract curve. A competitive equilibrium is given by a price line
through the initial endowment point, , which is tangential to both indierence
curves at the same point. The common point of tangency results in household
choices that lead to the equilibrium levels of demand. Such an equilibrium is
indicated by point e. It is clear that there is no other point which is preferred
by both households to point e. The equilibrium is therefore Pareto optimal.
In addition, it should be noted that the set of Pareto optima for an exchange
economy, given by the contract curve, generally consists of an infinite set of
points.
The importance of the non-satiation assumption is shown in Figure 2.8 in
which household 1 has an area of satiation. Trading from , the equilibrium
is at point e but this is not Pareto optimal since the satiated household would
41
Figure 2.7: The First Theorem

be equally satisfied with less consumption which would allow an increase in the
utility of the non-satiated household.
2.3.3
The Second Theorem
The Second Theorem is concerned with the converse of the First and its focus can
be summarized as follows: given a Pareto optimum, can a competitive economy
be constructed for which the Pareto optimum is a competitive equilibrium? In
terms of an Edgeworth box, the same question can be formulated as asking
whether it is possible to decentralize all points on the contract curve. Using
the diagram, it can be seen that this is possible in an exchange economy if
the households indierence curves are convex. With convexity, the common
tangent at a Pareto optimum provides the equilibrium prices. To decentralize
the economy, a point on this budget line is chosen as the initial endowment
point.
This process is illustrated in Figure 2.9 where the Pareto optimum e is decentralized by selecting 0 as the endowment point. Note that if the endowments
of the households are initially given by and the equilibrium at e is to be
decentralized, some transfer of endowment or, equivalently, of income will be
necessary.
The Second Theorem is now stated formally and its proof given.
h j
Theorem 8 (The Second Theorem of Welfare Economics) Suppose that x
b , yb
is a Pareto optimum such that at least one household is not satiated. Then, with:
(a) convex preferences; (b) convex production sets; (c) the allocation x
bh interior
to the consumption set
of
h
,
for
all
h;
and
(d)
continuity
of
preferences,
there
h j
exists pb 6= 0 such that pb, x
b , yb
is a competitive equilibrium.
Figure 2.8: The Non-Satiation Assumption
Figure 2.9: The Second Theorem in an Exchange Economy
43
h
Proof. Assume that household 1 is not satiated at the allocation x
b and
h
h
h
h h

h
h h
h
h
h
b x :U x >U x
b
and C x
b x : U h xh U h x
b
.
define the sets C x
Given assumption (a), these sets are convex for all h. Now
define
the
convex
set
o
n
h
Pm
1 1 PH
Z by Z = C x
b + h=2 C h x
b and the set W by W = w : w = + j=1 y j , y j Y j .
Since the individual production sets are convex, W , the set of feasible
produc
tion plans for the economy, is also convex. In addition, as x
bh is a Pareto
optimum, it must not be feasible to produce a preferred consumption array,
thus no member of W is in Z.
The separating hyperplane theorem implies that there exists a vector pb such
that pbz pbwPfor all z in Z and w in W. By definition any
Pm point in Z can be
H
written z = h=1 xh and any point in W as w = + j=1 y j . Noting that
Pm
PH
feasibility implies, h=1 xh j=1 y j , pbz pbw is equivalent to
pb
"
H
X
h=1
H
X
h=1
x
b
pb
"
H
X
h=1
H
X
h=1
yb
0,
(2.34)
h
1 1
j
j
for all x1 C x
b , xh C h x
b , and
y 1 Y
1.Now as the inequality in (2.34)
1
1
1
must hold for all x such
b , it must also hold in the limit
that U 1 x > U x
for any x1 with U 1 x1 = U 1 x

b . Hence (2.34) holds for all xh C h x
bh ,
and y j Y j . Taking all but one of the xh , y j equal to x
bh , ybj , then (2.34) shows
that
h
pbxh pbx
bh , for all xh C h x
b , for all h,
pby j pbybj , for all y j Y j , for all j.
(2.35)
(2.36)
Inequality (2.36) shows that at the price system supporting the Pareto optimum
each firm is maximizing profit (consider (ii) of C.E.).
To show that households are maximizing utility requires a demonstration
h
that pbxh pbx
b .
bh implies there is no xh with pbxh = pbx
bh and U h xh > U h x
Now assume that there is some xh such that pbxh = pbx
bh and U h xh > U h x
bh .
h
h
h
From assumption
(c)
also exists x
e with pbx
e < pbx
b but (2.35) shows
h
there
that U h x
e < Uh x
bh . Now take a convex combination of x
eh and xh . As
preferences are continuous from assumption (d), this convex combination can
be made to satisfy
h
h
U h te
x + [1 t] xh > U h x
b ,
(2.37)
and
pb te
x + [1 t] xh < pbx
bh ,
(2.38)
by moving
arbitrarily close to xh . This contradicts pbxh pbx
bh , for all xh
Ch x
bh , all h. Hence the households must be maximizing utility and the price
vector decentralizes the Pareto optimum. ||
Figure 2.10: Proof of the Second Theorem
Two points about the proof are worth noting. Theorem 2.4 implicitly assumes that the households are given sucient income to purchase the optimal allocation. In fact, household h must receive an endowment that satisfies
bh . Secondly the reasoning of the proof is based on the fact that a price
pb h = pbx
vector that supports the sum of sets also supports the individual sets that form
that sum. This is illustrated in Figure 2.10.
Before proceeding further, it is worth emphasizing that the proof of the
Second Theorem required more assumptions than the proof of the First so there
may be situations in which the First Theorem is applicable but the Second is
not. For instance, an equilibrium may exist with some non-convexity in the
production sets of the individual firms but the separation theorem can then not
be applied which prevents the proof of the Second Theorem.
A moments reflection is sucient to realize the importance to economic
policy of the Second Theorem. In designing policy, it is almost certain that a
policy maker would wish to achieve a Pareto optimum, otherwise welfare could
be increased at no cost. The theorem demonstrates that the objective of the
policy maker can be achieved by making the economy competitive, selecting
the equilibrium that is to be decentralized and providing each household with
sucient income to aord their allocation. The only policy instrument employed
is a lump-sum redistribution of endowments to ensure that each household has
the required income. If this approach could be applied in practice, then economic
policy analysis reduces to the calculation and redistribution of the lump-sum
2.4. A CRITICAL APPRAISAL
45
taxes and the subject matter of public economics is completed.
2.4
A critical appraisal
There are several viewpoints from which the Two Theorems can be criticized.
Firstly, it is possible to question the practical value of the Second Theorem on
the basis of its requirement for lump sum transfers in implementing the decentralizations. Secondly, it is possible to question the merit of Pareto optimality
as a method of judging the acceptability of a given equilibrium. In addition
it is also possible to consider the structure of the economy: is it an accurate
representation of reality and what are the consequences for welfare economics
of relaxing its assumptions?
This section will consider each of these issues. The use of lump-sum transfers
will be taken up first since these are central to applications of the Second Theorem. This will be followed by an assessment of the structure of the economy
and the implications of this for the subject matter of the subsequent chapters.
The value of Pareto optimality as a criterion for social decision making is then
considered.
2.4.1
Lump-sum transfers and taxes
The strong implications of the Second Theorem for policy design have already
been noted. However, the practical value of the Second Theorem is dependent on
the possibility of making the lump-sum transfers of endowments that it requires.
Without recourse to such transfers, the decentralizations would not be possible.
The exchange economy illustrated in Figure 2.11 makes clear the role and
nature of lump-sum transfers. The initial endowment point is denoted and the
Pareto optimum at point e is to be decentralized. This requires the endowment
point to be moved to a point on the budget line through e. Assume that the
new endowment point 0 leaves the distribution of good 2 constant. This move
can be supported by the transfer of x
e11 units of good 1 from household 1 to
0
household 2. Trading from will lead to the competitive equilibrium at e as
required.
Such a transfer of part of an endowment from one household to another is the
most basic form of a lump-sum transfer. The lump-sum nature of the transfer is
due to the fact that neither household can alter the size of the transfer by changes
in their behavior; there is simply no scope for such changes in the economy
described. The transfer is an optimal transfer if the resulting equilibrium at e
maximizes the policy makers objective function.
Although the lump-sum transfer above involved quantities of goods it can
be reinterpreted to introduce lump-sum taxes. If at the equilibrium e the prices
of the two goods are p1 and p2 , then the value of the transfer in Figure 2.11 is
p1 x
e11 . The notional income of household 1 prior to taxation is
M 1 = p1 11 + p2 12 .
(2.39)
Figure 2.11: The Use of Lum-Sum Transfers

Now, rather than actually redistributing quantities of the goods, the government
could tax household 1 an amount T 1 , T 1 = p1 x
e11 , which reduces their income to
f1 = p1 11 + p2 12 p1 x
e11 ,
M
(2.40)
and give household 2 an amount of income T 2 , = T 1 . If it is intended that the

equilibrium be attained at e then T 1 is the optimal lump-sum tax on household
1 and T 1 is the optimal tax on 2, as this pair of taxes ensures the households
face the budget line through e.
The important point of this reinterpretation is that the tax scheme consisting
of the taxes{T 1 , T1 } is equivalent in its eect to the original transfer of
e11 . This equivalence demonstrates that it is possible to
endowment e
x11 , x
view the transfers needed to achieve the decentralizations as taking the form of
either real transfers of goods between households or as transfers of income in
the form of lump-sum taxes. This description of lump-sum taxes can be easily
generalized to an H household economy in which a chosen equilibrium would
be decentralized by a vector of lump-sum taxes
H
1
X
H
T , ..., T
T h = 0.
,
(2.41)
h=1
Since these taxes sum to zero, they represent a simple redistribution of resources.
Lump-sum taxes have a central role in public economics due to their eciency in achieving distributional objectives. It should be clear from the discussion above that the economys total endowment is not reduced by the application
of the lump-sum taxes. This point applies to lump-sum taxes in general. As
47
households cannot aect the level of the tax by changing their behavior, lumpsum taxes do not lead to any ineciency. There are no resources lost due to the
imposition of lump-sum taxes and redistribution is achieved with no eciency
cost. Having identified the nature and value of optimal lump-sum taxes, the
question of their applicability is now considered.
In practice, the endowment of most households is simply their future labour
supply. Given this, it would be impossible to conduct lump-sum transfers of endowments as a quantity of future labour cannot be transferred from one household to another without the reintroduction of slavery. It is therefore possible
to dismiss the idea of transferring quantities of goods, except in very particular
and inconsequential cases, and to focus upon the design of optimal lump-sum
taxes.
In order for a transfer, or tax, to be lump-sum the household involved must
not be able to aect the size of the transfer by changing their behavior. It is clear
that lump-sum taxes can be used, for example by taxing each household some
fixed amount a lump-sum tax is imposed. Setting aside minor details, this was
eectively the case of the U.K. poll tax. This example motivates the following
important observation. The eciency of lump-sum taxation rests partly on the
fact that their imposition is costless but this was far from the case with the
U.K. poll tax. In fact, the diculties of actually collecting and maintaining
information on the residential address of all households made the imposition
of a uniform lump-sum tax prohibitively expensive. Therefore, although the
structure of lump-sum taxes makes them appear deceptively simple to collect,
this may not be the case in practice since the tax base, people, is highly mobile
and evasive.
However, the costs of collection are only part of the issue. What is the primary concern here is the use of optimal lump-sum taxes. Optimality requires
the tax to be based on all relevant economic characteristics and households must
not be able to alter these characteristics in response to the taxes. It may be
possible to dierentiate lump-sum taxes according to sex, age or eye-colour for
instance, but these are unlikely to be the relevant characteristics on which to
base the tax. For the exchange economy examples, the characteristics were the
endowments and preferences of the households. More generally they may be the
expected future labour incomes of the households or the determinants of each
households human capital. Such characteristics are unlikely to be directly observable by the government and either it must either rely on households honestly
reporting their characteristics or the characteristics must be inferred from the
actions of households. In the latter case, there is invariably scope for changes
in market behavior which implies the taxes are no longer lump-sum. When reports are the sole source of information, unobserved characteristics cannot form
a basis for taxation unless the tax scheme is such that there is an individual
gain to truthful revelation.
As an example of the interaction between taxes and reporting, consider the
following. Let the system of lump-sum taxes be based on the characteristic IQ
level. If the level of tax was inversely related to IQ and if all households had
to complete IQ tests, then the tax system would not be manipulated since the

incentive would always be to maximize the score on the test. In contrast, if taxes
were positively related to IQ, a testing procedure could easily be manipulated
and the mean level of tested IQ would be expected to fall considerably. This
indicates the potential for misrevelation of characteristics.
These ideas have been developed formally by Mirrlees (1986) who presents
theorems on the (im)possibility of designing non-manipulable lump-sum tax
schemes. The central theorem considers a population of households who each
have the utility function
U h = U h c1 xh1 , ..., cn xhn .

(2.42)
The households vary, however, in the values of the constants c = (c1 , ..., cn )
that appear in the utility function. The preferences of each household are fully
described by their c vector. Since the cs are the only dierentiating characteristic
between households, any optimal set of lump sum taxes must be based on these
characteristics. Now assume that the government cannot observe the c vectors,
that households only truly report their characteristics when they do not lose by
doing so and that misrepresentation can only take place by a household claiming
that the values of the characteristics are above their true values.
With this formulation, a tax policy T = T (c) conditional on the characteristics can only be administered, in the sense that it generates truthful revelation
e h (c)
from the households, if the final utility allocation U h = U h (T (c)) = U
generated by the taxes is non-increasing in c. The following theorem shows
e h (c) will be increasing in ci with the optimal tax policy if good i is northat U
mal. Since some goods must be normal, the policy optimal policy cannot be
administered.
Theorem 9 (Mirrlees) If the utility of each household is given by (2.42) and
social welfare is defined as
Z
W = U h c1 xh1 , ..., cn xhn (c1 , ..., cn ) dc1 , ..., dcn ,

(2.43)
where () is the distribution function of the cs, then with the optimal tax
e h (c) is increaspolicy that maximizes (2.43) given the economys endowment, U
ing in ci if commodity i is always a normal good.
Proof. The optimum occurs when the marginal utility of income is identical
for all households. If this were not the case, transfers of income would raise welfare. Inverting this reasoning, the marginal income required to obtain another
unit of utility must be the same for all households. Corresponding to the utility
function (2.42) is the expenditure function for h which takes the form
pn h
p1
h
h
E =E
, ..., , U
.
(2.44)
c1
cn
Optimality then implies
EUh = ,
(2.45)
49
for some constant . Dierentiating (2.45) with respect to ci gives

eh
U
pi
EUh i 2 = 0,
(2.46)
ci
ci
or, since Eih = hi pc11 , ..., pcnn , U h where hi pc11 , ..., pcnn , U h is the Hicksian or
compensated demand for good ifrom h,
EUh U
EUh U
eh
pi hi
U
=
.
ci
ci U h
(2.47)
With a concave utility function EUh U > 0 and the normality of good i implies
h
i
U h
eh
U
> 0. Therefore the normality of good i implies c
> 0 at the optimum as
i
was to be proved. ||
This theorem shows why optimal lump-sum taxes may not be feasible. The
government cannot observe the relevant characteristics and relies on the households to reveal them. However, under the optimal tax policy it is not in the
households interest to truthfully reveal their characteristics. This theorem is
dependent upon the precise assumptions made but it is indicative of the general
results that emerge when policy relies on individual revelation of information.
The main points of the argument can now be summarised. To have any
content the Second Theorem relies on the use of optimal lump-sum transfers but
such transfers are unlikely to be available in practice or to satisfy all the criteria
required of them. The taxes may be costly to collect and the characteristics
upon which they should be levied may not be observable. When characteristics
are not observable, households may have incentives to make false revelations. It
is therefore best to treat the Second Theorem as being of considerable theoretical
interest but of very limited practical relevance.
2.4.2
The institutional assumptions
Before considering the concept of Pareto optimality, some comments are now
oered on the assumptions underlying the model and the consequences for the
Two Theorems of their relaxation. The assumptions can be conveniently placed
into two categories: those whose relaxation is possible without destroying the
validity of the theorems and the others whose relaxation does so. Time and uncertainty belong to the former category, although when they are introduced the
interpretation becomes less tenable. Public goods, externalities and monopoly
power, or imperfect competition generally, are firmly in the latter.
In the presence of time and uncertainty the existence of sucient markets
permits a re-interpretation of the economy and the proof of Two Theorems
proceeds as above. However, the passage of time introduces the questions of
investment, savings and growth which cannot be addressed satisfactorily in the
Arrow-Debreu economy. Fortunately a number of more appropriate structures
are available. One of these, the overlapping generations economy, will be studied
in Chapter 13. Uncertainty introduces aspects such as the correct amount of

risk-taking and portfolio choice. This is an area of public economics that is being
developed particularly with respect to the normative implications of uncertainty
and the important aspects of this literature are reviewed in Chapter 8.
Turning to externalities, public goods and monopoly power, since any one
of these will remove the Pareto optimality of equilibrium, the Two Theorems
are no longer applicable to any economy in which they are present. However,
the theorems do show some properties of the state that could be achieved if the
externality etc. could be eliminated. The failure of the Two Theorems in these
cases can be viewed as providing the motivation for the analysis given in Part
III.
2.4.3
Pareto optimality
Even when all the criticisms noted above are set aside, the Two Theorems
are still dependent upon the value of Pareto optimality as a welfare criterion.
For this reason, and also because Pareto optimality is regularly encountered
in public economics, it is worthwhile taking a closer look at Pareto optimality
and considering its implications and deficiencies as a means of choosing between
economic states. From this perspective it is possible to assess the usefulness of
Pareto optimality as a criterion for guiding public policy.
The Pareto criterion was introduced by Pareto (1909) and given its name
by Little (1950). The motivation for its use was that it provided a means
of comparing economic states without requiring interpersonal comparisons of
utility; an issue that will be returned to below. To begin the discussion, it is
helpful to start with a restatement of Pareto optimality.
Consider a set of states, S = (s1 , s2,... ) and a set H, indexed h = 1, ..., H, of
economic agents. Writing h for the strict preference relation of household h
and h for at least as good as, then state s1 is Pareto preferred to state s2 if
s1 h s2 for all h = 1, ..., H,
(2.48)
s1 h s2 for at least one.
(2.49)
and
A preference order is then induced on the set of states by (2.48) and (2.49).
This preference order is denoted by P , hence
s1 P s2 (2.48) and (2.49).
In addition, define P by
s1 P s2 (2.48)
(2.50)
(2.51)
State s1 is then defined as being Pareto optimal if there exists no other state
that is Pareto preferred to s1 . Formally, s1 is Pareto optimal if no s2 S
such that s2 P s1 . It is important to note that this general definition has
been in terms of two states and that no utility concepts were involved. This
demonstrates that Pareto optimality is a very broad concept and can be used
for far more than assessing allocations of utility.
51
Now consider s1 and s2 to represent dierent allocations of a fixed stock of

goods amongst the H agents. Then allocation s1 is Pareto optimal if there exists
no re-allocation that is Pareto preferred to s1 . In other words, starting from s1
it must be impossible to re-allocate the fixed stock of goods in manner which
makes one household better-o without making another worse-o. Alternatively,
s1 and s2 could represent two states with alternative political parties being in
power. Then s1 is Pareto preferred if no voter prefers party s2 and at least one
would strictly prefer party s1 .
With this background it is now possible to proceed to an assessment of Pareto
optimality. By considering even the simplest examples, it is not dicult to see
several deficiencies of the Pareto criterion. First consider the division of a cake
between two people where both people like cake and their preferences satisfy
non-satiation. In such circumstances every division of the cake is Pareto optimal,
from the most inequitable to the most equitable. This claim follows from noting
that, starting from any division, there are no changes in the distribution that
satisfy (2.48) and (2.49). In other words, any reallocation that provides more
to one household must take it away from the other. The latter point illustrates
that situations exist in which the set of Pareto optima
{si : si S and no sj S such that sj P si } ,
(2.52)
is infinite.
From this simple example it is possible to infer two deficiencies of Pareto optimality. Firstly, extreme allocations may be Pareto optimal and, consequently,
although an equilibrium may be Pareto optimal there is no reason why anyone
should advocate it as good in any other sense. Secondly, it is possible for there
to be a multiplicity of Pareto optimal allocations even for this very allocation
problem. When the Pareto criterion does not provide a unique optimal allocation it is not be of much assistance as a social rule for choosing an allocation.
The points made in the cake division example are also relevant when considering allocations in a two-household exchange economy. The contract curve in
Figure ?? gives the set of Pareto optima and there is generally an uncountable
number of these optima, so the ordering P does not select a unique maximal
element. In addition, the competitive equilibrium may be as the one illustrated
in the bottom left corner. This is Pareto optimal but highly inequitable.
Another failing of Pareto optimality is that the ordering P need not provide
a complete ranking of states in S. That is, there may be some pairs of states
which are incomparable under the Pareto criterion. Incomparability arises if, in
the move between two states s1 and s2 , some agent gains and another agent loses.
This is illustrated in Figure ?? where allocations s1 and s2 cannot be compared
although s1 and s3 and s2 and s3 can. Such gains and losses are invariably a
feature of policy choices and much of policy analysis consists of weighing-up the
gains and losses. In this respect, the Pareto criterion is inadequate as a basis
for policy choice.
To summarise these arguments, Pareto optimality does not embody any
concept of justice and highly inequitable allocations may be optimal under the
Figure 2.12: Pareto Optimal Allocations
Figure 2.13: Incompleteness of the Pareto Ranking
2.5. INTERPERSONAL COMPARISONS AND SOCIAL WELFARE
53
criterion. In many situations, the set of Pareto optimal allocations is infinite

and the criterion then provides little guidance for policy choice. Finally, the
Pareto criterion may not provide a complete ordering of states so that some
states will be incomparable under the criterion. The source of all these failing
is that the Pareto criterion avoids making interpersonal comparisons whereas it
is such comparisons that have to be made in most allocation decisions.
2.5
Interpersonal comparisons and social welfare
Pareto optimality was originally proposed as a means by which it was possible to analyse the consequences of alternative allocations without requiring
interpersonal comparisons of welfare to be made and it is from this that the
failures of Pareto optimality emerge. To make further progress it is necessary to
consider the scope for comparability between the utility levels of dierent individuals. This section reviews alternative degrees of comparability and considers
the problem of aggregating individual preferences into social preferences.
2.5.1
Interpersonal comparability
If suciently strong interpersonal comparisons are possible, statements such as:

the welfare loss of agent i in the new allocation is more than oset than the
gain of agent j would be permissible. Given the relative weights attached to
individuals i and j in social preferences it would then be possible to make a
judgement on whether the change was socially beneficial or not.
The first step in making such comparisons is the construction of a comparable
utility index to measure
is and js welfare.
h If, for example, i gains in the move
from allocation xh to allocation x
b
and j loses, with other households
i
U i against the loss U j U
j . Such
unaected, the aim is to weigh the gain U
comparisons run against the spirit of the Pareto criterion which, as already
stated, was originally intended to replace such utility comparisons since these
were felt to be making judgements that were scientifically untenable; a viewpoint
expounded most strongly by Robbins (1932).
The issues motivating Robbins denial of interpersonal comparability can
best be understood by reviewing the development of utility theory. Nineteenth
century economists assumed that utility was something measurable and that it
was naturally comparable between individuals. As such, it followed that the
welfare of society could be inferred in some straightforward manner from the
welfare levels of the individuals forming that society. Reasoning of this form
provides the scientific basis for utilitarianism.
The notion of measurable and comparable utility began to be dispelled in
the early Twentieth century after the demonstrations by Hicks (1939) and others that the entire theory of the household could be based on ordinal noncomparable utility. In other words, since it was the households preference or-

dering that determined consumption choices, utility was only relevant as a convenient functional representation of preferences, with no greater meaning attached
to it, and that given a utility function representing the preference ordering, any
increasing monotonic transformation of that utility function would also be an
equally valid representation of preferences. Ordinality and non-comparability
then became the accepted concepts. This acceptance left no scientific basis upon
which to justify the comparability of dierent households utility levels.
Ordinality means that if x1 h x2 then it is given a higher utility number
but the actual relationships between the numbers assigned to dierent xs, for
instance the size of the dierence between any two, have no content. This,
consequently, prevented welfare comparisons between people and created the
necessity of developing concepts, such as Pareto optimality, that were free of interpersonal comparisons. However the weaknesses of these criteria soon became
obvious and, following Scitovsky (1951), the trend since the 1960s has been to
explore the consequences of re-admitting interpersonal comparability into the
analysis. Some economists, for instance Mirrlees (1971) and Ng (1985), have
argued most strongly for this, although there is some divergence of opinion on
the acceptable extent of comparability.
Between non-comparability and full comparability there are several intermediate cases that vary as to precisely how comparable are the utility levels. For
instance, the claim that one household has a higher level of utility than another
requires rather less comparability than claiming it has 15% more utility. Dierent degrees of comparability carry obvious implications for the welfare measures
that can be based upon them.
The various degrees of comparability have been classified by Sen (1977) and
Roberts (1980b). The starting point for the classification is a utility function U h
for each household h and the set of admissible transformations of this utility

function. That is, = 1 , ..., H if h U h is an equally valid utility
function for h. Alternative degrees of measurability and comparability then
appear as restrictions on the set . The alternatives that have been considered
in the literature, in increasing order of comparability, can now be stated as
follows.
(i) Ordinality and non-comparability (ONC)
is a list of H independent strictly monotonically increasing transformations.
This is the set of transformations that arises when utility is viewed only as
the representation of preference in conditions of certainty. It implies that there
can be no interpersonal comparability.
(ii) Cardinality and non-comparability (CNC)
is a list of H independent strictly positive ane transformations:
h U h = ah + bh U h , bh > 0.
When there is uncertainty involved in the decision-making environment CNC
can be justified by appeal to the von Neumann and Morgernstern (1953) utility
representation theorem. Condition CNC demonstrates that cardinality, in itself,
does not carry any implications for interpersonal comparisons.
55
(iii) Ordinality and level-comparability (OLC)

is a list of
Hidentical strictly monotonically increasing transformations: h U h = U h , independent of h.
With OLC it is possible to rank the households in terms of their utility
levels, so that one household can be seen as having more or less utility than
another, but changes in utility levels cannot be compared across households.
This degree of comparability formalises what is probably intended by claims
that households have roughly similar preferences and can hence have their utility
levels compared.
(iv) Cardinal unit comparability (CUC)
is a list of H strictly
positive ane transformations which dier only
in their constants: h U h = ah + bU h , b > 0.

Although CUC does not allow comparability of welfare levels, since the constant terms may dier, it does allow the comparability of gains and losses in
utility. This follows from the restriction of the utilities to the same units of
measurement.
(v) Cardinal full comparability (CFC)

is a list of H identical strictly positive ane transformations: h U h =
a + bU h , b > 0.
CFC represents the maximum degree of comparability that is possible without the introduction of a natural zero for utility since it allows comparison of
both levels and gains and losses. If a zero is introduced, the set of transformations then becomes
(vi) Cardinal ratio scale (CRS)

is a list of H identical strictly positive linear transformations: h U h =
bU h , b > 0.
Amongst these alternative degrees of comparability, only ONC and CNC are
formally justified via representation theorems on preferences. Moving further
down the list requires an increasing degree of pure faith that the procedure is
justified. The next section relates the degree of comparability to permissible
welfare criteria.
2.5.2
Social choice
In moving from individual to social preferences the basic problem is one of aggregation. From the individual preference relations must be constructed a social
preference relation that satisfies certain attractive criteria. What aggregation
procedures are possible depends critically on the level of comparability assumed.
The starting point for the literature on social choice was Arrows (1950,
1951b) impossibility theorem. Arrow considered a set S of states and H individual preference orderings h , h = 1, ..., H, over S and attempted to construct
from these a social welfare function, F , that would determine the social ordering, , of S for any set of individual orderings, so = F ({h }). An ordering
is defined here as being complete, reflexive and transitive. Conditions that may
be imposed upon such a social welfare function include the following.

Unrestricted domain (U ): The domain of F includes all logically possibly
H-tuples of individual orderings of S.
Independence of irrelevant alternatives (I ): The restriction of to the pair
{s1 , s2 }, si S, i = 1, 2, is a function
only of the
restrictions of individual
preferences to that pair |{s1 ,s2 } = F h |{s1 ,s2 } .
Weak Pareto principle (P): For any pair {s1 , s2 } if s1 h s2 for all h, then
s1 s2 .
Non-dictatorship (D): There must be no individual h such that for all individual preference orders, for each ordered pair s1 , s2 , si S, i = 1, 2, s1 h s2
implies s1 s2 ..
From these conditions follows Arrows Impossibility Theorem.
Theorem 10 (Arrow) If H is finite and S has at least 3 elements then there
is no social welfare function satisfying conditions U, I, P and D.
Proof. See Sen (1986).
The implication of this theorem is that a social welfare function does not exist
that can aggregate the individual preference orderings without conflicting with
one, or more, of the conditions U, I, P or D. At the heart of this non-existence
result is the limited information contained in the individual preference orderings. This information is not sucient for the social ordering to be constructed
without violating one of the conditions. Following the first demonstrations of
this theorem, attempts were made to derive possibility results by weakening the
conditions U, I, P and D and relaxing the requirement of transitivity for the
social ordering. None of these relaxations provided a convincing path out of the
impossibility.
An alternative approach is to allow greater information about individual
preferences, in particular to consider the consequences of alternative forms of
interpersonal comparability. That admitting interpersonal comparability increases the scope for making consistent welfare judgements, and generates Possibility Theorems, has been demonstrated by Roberts (1980a,b) and Sen (1977).
The formal method is to consider a social welfare functional that determines the
social ordering over
for any given H-tuple of individual utility functions over
S
S, that is = F U h . The conditions U, I P and D can be translated into
equivalent restrictions on the social welfare functional. The degree of interpersonal comparability then determines the permissible transformations of the
individual utility functions. For example, with ONC the same ordering must
be generated whatever monotonic
are applied to the individual
n transformations

o
utilities or = F U h = F
h U h
for any set of monotonic transforn o
mations h . Since ONC is equivalent to the information content in Theorem
2.6, the translation of that theorem applies directly.
When further information is introduced, possibility theorems can be proved.
Taking ordinal level comparability (OLC) first, since this is sucient to permit
the ranking of individuals in terms of utility levels, it also permits social welfare
functionals that are defined as the welfare of the hth individual in the ranking
57
for any h. One special example of such a welfare function is the Rawlsian or
max-min given by

W = min U 1 x1 , ..., U H xH .
(2.53)
With unit comparability it is possible to asses the gains and losses between
individuals. Furthermore, once one set of the constants appearing in the utility
function are specified, it is meaningful to consider the sum of utilities as a welfare
criterion. Formally, CUC implies that the welfare functional
W =
H
X
ah U h , ah > 0,
(2.54)
h=1
is permissible. If the welfare functional is also assumed to satisfy the condition

of anonymity given byn o

e h is a re-ordering of the utility vector U h , then
Anonymity (A): if U
F
n o

eh
= F Uh ,
U
(2.55)
then the constants ah in (2.54) cannot vary with h so that the welfare functional
reduces to the standard utilitarian form
W =
H
X
h=1

U h xh .
(2.56)
Full comparability leads to the class of welfare functionals determined by
(2.57)
W = U (x) + g U (x) U (x) , U (x)
H
X U h xh

,
(2.58)
= U 1 x1 , ..., U H xH , U (x) =
H
h=1
where g() is homogeneous of degree 1. An example of such a welfare functional

is
(2.59)
W = U (x) + min U h (x) U (x) ,
which is Utilitarian for = 0 and Rawlsian for = 1. Finally, CRS leads to
W = W (U (x)) ,
(2.60)
where W () is a homothetic function. When anonymity is also imposed, the

social welfare functional becomes
H h h 1
X
U x
W =
.
(2.61)
1
h=1
2.5.3
Bergson-Samuelson social welfare function
The Bergson-Samuleson social welfare function was first introduced into economics by Bergson (1938) since which there has been considerable controversy
over its correct interpretation and its validity. The general Bergson-Samuelson
social welfare function is defined as a function of the state W = W (si ), where
si S, and is intended to represent the social preference ordering. If the social
welfare function is neutral towards non-utility features of the states it is evaluating, so that only the utility levels of the individuals matter for social welfare,
then the Bergson-Samuelson social welfare function takes the form

W = W U 1 x1 , ..., U H xH ,
(2.62)
There has been much discussion of the relationship between the social welfare
function in the sense of Arrow and that of Bergson-Samuelson. Samuelson
(1977) asserts that a Bergson-Samuelson social welfare function can be constructed even when the utility levels are ordinal and non-comparable. If such a
social welfare function is expected to hold for all possible preference orderings,
then Theorem 3.4 shows that it must violate one (or more) of U, I, P and D. One
potential route out of this impasse was to suggest that the Bergson-Samuelson
social welfare function needs to hold only for the given set of preference orders
that characterise the economy under consideration. In this case, condition U
is not applicable. This restriction to the given set of preferences represents the
single-profile approach to social choice where the social ordering only has to
satisfy the specified criteria for a single-profile of preference orders. In contrast, the Arrow theorem is characteristic of the multi-profile approach where
the ordering must be constructed for any possible set of preference orderings.
Unfortunately, the restriction to single-profiles does not avoid the impossibility
result. As shown by Kemp and Ng (1976) and Roberts (1980c) the impossibility
result remains intact when the requirements of the social ordering are translated
to the single-profile context.
The use of a Bergson-Samuleson social welfare function therefore implies the
adoption of comparability assumptions on individual utilities. Accepting this,
the maximisation of a Bergson-Samuelson social welfare functions is invariably
adopted as the objective of policy in public economics. In such exercises it is
usually assumed that the social welfare function satisfies the Pareto criterion
h
h
h
h
Uh x
Uh x
, all h implies W U h x
W Uh x
,
(2.63)
and, when the social welfare function is dierentiable, that

2W
W
0 and
0.
h
U
U h2
(2.64)
The assumption on the second derivative is often strengthened to that of strict

concavity or strict quasi-concavity. Unfortunately, as is discussed further in
Chapter 4, this does not always guarantee second-order conditions will be satisfied when the welfare function is used as the objective in a maximisation.
2.6. SUMMARY
59
Figure 2.14: Social Indierence Curves

For a given value of W () it is possible to draw its level sets, typically called
social indierence curves, as in Figure 2.14.
The convexity of these was the subject of discussion in the 1950s. Gorman
(1959b) demonstrated the important result that convexity in itself does not
require interpersonal comparability.
2.6
Summary
The aim of this chapter was to introduce the Arrow-Debreu economy and to
review a number of fundamental issues in welfare economics. This material
underlies much of the analysis in the following chapters, providing both the
analytical method and a baseline against which changes can be assessed.
The Two Theorems, which characterise the eciency properties of the competitive economy, were described and proved. Although the Theorems have
strong implications, it was argued that they both are limited in their value; the
First Theorem due to the weakness of Pareto optimality and the Second because
of its reliance upon lump-sum taxation. It was argued that criteria for judging
between economic states that did not employ interpersonal comparisons were
too weak to be of practical usefulness. If comparability is permitted, stronger
criteria are possible.
There are two major conclusions to be drawn from the discussion. Firstly,
the limitations of the Two Theorems provide a natural point of departure for
the study of public economics. Secondly, if decisions about economic policy are
to be taken on welfare grounds, it is necessary that interpersonal comparisons
must be made.
Chapter 3
Topics in Measurement
3.1
Introduction
The previous chapter has reviewed the theory and concepts involved in the
measurement of welfare and the motivations for adopting these approaches. The
welfare judgements that emerged from that analysis were based unavoidably
upon utility concepts and this basis is found by many to be objectionable. In
response to this, alternative means of assessing the economic outcome that rely,
in the main, on observable and measurable quantities are now discussed. At
first sight these appear to be rather more tangible concepts than those of utility
and social welfare. Most of what is said below applies equally to a range of
economic indices far broader than just those considered here.
The chapter begins with a discussion of the definition of income which is
the basic observable and measurable quantity from which indices of inequality and poverty are formed. Although there is little doubt about the meaning
of income in a world of certainty, it is rather less well specified when there is
uncertainty about the values of future variables. The distinctions between the
available definitions of income are clarified and their relative merits noted. Data
on household incomes, though, cannot be used directly in the construction of
economic indices since the households will dier in their size and composition
and, consequently, in their requirements. This observation motivates the consideration of equivalence scales, which are a means of adjusting observed household
incomes to take account of household composition eects and to provide a set
of comparable income measures.
This is followed by a review of the measurement of inequality. The link is
drawn between statistical measures and their implied social welfare functions.
This motivates the consideration of explicit welfare-based measures and axiom
systems. A similar approach is taken to the measurement of poverty. It is worth
noting at the outset that the major theme to emerge from the discussion is that
the construction of a successful index requires the welfare assumptions on which
it is based to be made explicit.
61
62
3.2
CHAPTER 3. TOPICS IN MEASUREMENT
The measurement of income
Many of the indices that measure economic well-being, including those of inequality and poverty, are formed from data on incomes on the grounds that
income is the means to achieve welfare. Given this important role, it is necessary that the income statistics should provide accurate data. This will be
achieved if there is precise collection of information, which is a topic that will
not be pursued here, and the correct definition of what is to be measured is
employed. It is the latter aspect that is the subject of this section.
In the static and deterministic economy of Chapter 2 a households income
can be unambiguously defined. If the household enters the economy with a zero
stock of goods and is given its endowment and list of shareholdings after entry,
then its income, for any given price vector, is the value of the endowment plus
dividend income. Defined in this way, income is the value of consumption that
can be undertaken whilst leaving the household with the same stock of goods,
which was zero, at the end of the economy as it had at the start of the economy.
The definitions of income below will also try to apply this reasoning, but only in
the deterministic setting can it be applied without diculty of interpretation.
Fisher (1930) proposed the view that income was formed by the set of pleasurable experiences that the consumer enjoyed over the period in which income
was to be measured. This was based on the view that neither the actual receipt
of money or the purchase of goods represented the end product of economic
activity; it was actual consumption that played this role. The diculty of this
approach is that it leads to income as an unmeasurable and purely personal
construct. As an approximation, Fisher suggested that the level of expenditure
in the period, less the disutility of any labour performed, should be the measure
of income since it is from expenditure that consumption results. Unfortunately,
even this approximation cannot be directly calculated due to its inclusion of the
disutility of labour.
An alternative definition provided by Hicks (1939) constitutes what is generally taken as the standard definition of income. In Hicks words income is
the maximum value which a man can consume during a week and still expect
to be as well-o at the end of the week as he was at the beginning (Hicks
1939, p. 172). This definition embodies both the consumption aspect emphasised by Fisher (1930) and reduces to the stock-of-goods measure in a static
and certain environment. The diculties of applying this definition follow from
the inclusion of the word expect and from the consequent forward-looking nature of the definition. The definition does not make it clear whose expectations
are referred to and how to resolve dierences in expectations, nor does it show
how the possibility of false expectations should be accommodated. A literal
application of the definition would not count windfall gains as income, since by
definition they are not expected, although they unarguably raise the potential
level of consumption. For these reasons, the Hicks definition of income does not
command universal acceptability.
In contrast, Simons (1938) adopts the position that a workable and calculable definition is preferable to one that is formally correct but non-operational.
3.3. EQUIVALENCE SCALES
63
Simons definition is backward-looking, measuring income in retrospect. In full,

the definition is that Personal income may be defined as the algebraic sum
of (1) the market value of rights exercised in consumption and (2) the change
in the value of the store of property rights between the beginning and end of
the period in question (Simons 1938 p. 50). This can easily be applied and
does not involve expectations so removing some of the drawbacks of the Hicks
formulation. As Simons required, it provides an operational definition.
These alternative definitions highlight the distinctions between ex ante and
ex post measures of income. In practice, tax assessments adopt the backwardlooking viewpoint and measure income as all relevant payments received over
the relevant period. The definition of income for the purposes of tax codes does
not precisely satisfy any of the definitions given above. For instance, capital
gains if not realised during the period in question are not usually measured as
income. Such observations have motivated the concept of the comprehensive
income tax, under which income from all sources would be treated equally. This
would accord with the definitions of income given above. In practice, though,
the ideal measure of income cannot be achieved so preventing the application
of a truly comprehensive income tax. The systems that do arise can then be
viewed as deriving from the resulting second-best problem. For the purposes of
the following chapters, the fact that much of the analysis is conducted within
certain environments eliminates the problems of definition. When uncertainty
is explicitly introduced, care will be taken to highlight the measure of income
adopted.
3.3
Equivalence scales
The income level of a household is often treated as a proxy for its level of
welfare since, at the very least, income is the means to achieve welfare. The fact
that households dier in make-up and composition implies that if the welfare
levels of two households are to be compared it will not generally be sucient,
taking environmental variables etc. as constant, to simply compare their income
levels. To make the issue concrete, a household of one adult with no children
needs less income to achieve a given level of welfare than a household with two
adults and two children. The question is then, of course, how much less income?
Equivalence scales are a means of adjusting measured incomes into comparable
quantities.
The typical causes of dierentiation between households are the number of
adults and the number and ages of their dependants; the relevant features of such
dierentiation are typically referred to as demographic variables. The general
problem in designing equivalence scales is to achieve the adjustment of observed
income to take account of demographic dierences. The discussion begins with
an approach based on identifying the minimum needs that a household requires
in order to survive at some chosen level of welfare, typically a level on or just
above that identified as implying poverty. This method represents more of
an expert judgement of equivalent incomes than an application of economic
64
theory. Following this, the methods of Engel and Rothbarth are considered,
both of which employ restrictive assumptions on the underlying preferences of
households. Relaxing these assumptions then leads into the Barten method and
its generalisation.
3.3.1
Minimum needs
The calculation of equivalence scales based upon minimum needs can be traced
back to at least Rowntree (1901). Such scales are calculated by determining the
cost of an identified bundle of goods and services that are seen as representing
the minimum needs for the household. The exact bundle will dier between
households of varying size. In the work of Rowntree the set of goods included
in the minimum needs were food, rent and rates, and household sundries.
There have been many studies since that of Rowntree employing this methodology and some illustrative numbers are given in Table 3.1 below. In each case a
household with zero children is assigned the index of 100 and other compositions
are measured relative to this. Hence with the Beveridge scale, a single child is
seen as requiring an increase in the income of a couple of 22% in order for the
same minimum level of consumption needs to be attained.
Single Person
Couple
+1 child
+2 children
+3 children
Rowntree (1901)
60
100
124
161
222
Beveridge (1942)
59
100
122
144
188
US Poverty Scale (1942)

78
100
123
152
208
Note: For the Beveridge scale, children are taken to be in the 5-9 age group
Table 3.1: Minimum Needs Equivalence Scales
Table 3.1 demonstrates that these equivalence scales assume that there are
returns to scale in household size so that, for example, a family of two adults
does not require twice the income of a single person. The major shortcomings of
this method of computing equivalence scales are that by focusing on the cost of
obtaining a minimum level of consumption they are inappropriate for applying
to incomes above the minimum level and, sine they do not take account of
optimisation by the households, they cannot be claimed to measure the true
economic cost of demographic dierences.
3.3.2
Engel and Rothbarth
The next two methods of computing equivalence scales both have a basis in the
theory of household optimisation. They follow from the following observation: if
the welfare level of a household can be judged by its consumption of some specific
commodity, then equivalent incomes are those which lead dierent households
to consume the same quantity of the commodity. The distinction between the
two scales is in the commodities they select as relevant for the comparison.
65
Figure 3.1: Construction of Engel Scale
Engel
The Engel (1895) approach to equivalence scales rests on the hypothesis that
the welfare of a household can be measured by the proportion of its income that
is spent on food. In particular, Engels law asserts that the food share of expenditure falls as income, and hence welfare, rise. If this is accepted, equivalence
scales can be constructed for demographically dierent households by calculating the income levels at which their expenditure share on food is equal. This
is illustrated in Figure 3.1 in which the expenditure share on food as a function of income is shown for two households with demographic characteristics d1
and d2 . Using the Engel method, incomes M 1 and M 2 are equivalent and the
2
equivalence scale is given by the ratio M
M1 .
There are two shortcomings to this approach. Firstly, although its content
may be empirically true, this alone does not provide a basis for making welfare
comparisons. Secondly, even if welfare conclusions can be inferred via Engels
law, it can be argued that the Engel method overestimates the cost of additional
children for a household. Nicholson (1976) argues that a child is largely a foodconsuming addition to a household. If, after the addition of a child, a household
is compensated suciently to restore the share of food in its expenditure to
its original level, this would represent over-compensation with respect to other
commodities. Deaton and Muellbauer (1986) provide a formalisation of this
argument.
The approach of Engel has been extended to the more general iso-prop
method in which the expenditure shares of a basket of goods, rather than simply food, becomes the basis for the construction of scales. Seneca and Taussig
(1971) employ a basket consisting of food, housing, clothing and transportation
and find that this provides less compensation than the Engel method at low in-
66
Figure 3.2: Rothbarth Equivalence Scale

comes but more compensation at high incomes. However, considering a basket
of goods does not overcome the theoretical shortcomings of the Engel method.
Rothbarth
The procedure of Rothbarth (1943) selects for attention a set of goods that are
consumed only by adults, termed adult goods, and such that the expenditure
upon them can be treated as a measure of welfare. If these goods also have
the property that changes in demographic characteristics only aect their demand via income eects, then the extra income required to keep their consumption constant when household composition changes can be used to construct
an equivalence scale. Typical examples of such goods that have been used in
practice are tobacco and alcohol.
For the chosen consumption level x, the procedure is illustrated in Figure 3.2
with three Engel curves representing dierent household compositions. On the
basis that they generate the same level of demand as family composition changes,
the Rothbarth procedure would assert that M 1 , M 2 and M 3 are equivalent
incomes and the equivalence scale would be constructed from their ratios.
There are a number of diculties with this approach. It rests upon the
hypotheses that consumption of adult goods accurately measure welfare and
that they are aected only via income eects; theoretically both hypotheses are
asserting separability between the utility from adult goods and that from goods
which are aected by demographic characteristics. These hypotheses could be
refuted by empirical evidence and, as noted by Cramer (1969), the typical adult
goods, alcohol and tobacco, are empirically unresponsive to income eects. In
addition, it is not clear that their demand is unaected via changes in demographic characteristics. The evidence reported in Deaton, Ruiz-Castillo and
Thomas (1989) supports these doubts about the methods validity. Further-
67
more, the ratios of M 1 to M 2 and M 3 will depend upon the level of demand
chosen for the comparison except in the special case in which the Engel curves
are straight lines through the origin, so that utility is homothetic, and the ratios may vary for dierent goods. This leads into a further aggregation problem
of forming some total ratio out of the ratios for each good. These observations
suggest that the formulation of equivalence scales should be more closely related
to consumer preferences.
3.3.3
Prais and Houthakker
The approach to equivalence scales based on the Engel curve has been extended
by Prais (1953) and Prais and Houthakker (1953) and investigated more recently by Muellbauer (1974) and McClements (1977). This method begins by
specifying the Engel curve for good i as
h
xhi
M
h
= xi
, i = 1, ..., n,
(3.1)
e
ai
e
a0
where e
ai , i = 1, ..., n, is a function of the vector of demographic characteristics
d = (d1 , ..., dm ) that describe the household. For example, d1 could the number
of children under 1 and d2 the number between 1 and 3. The income deflator is
then a function of the e
ai . In this specification of the Engel curve, real income and
the eective consumption level of each good are obtained by deflating by terms
determined by household composition. This provides two channels through
which household composition can aect demand.
From (3.1) the eect of a change in characteristic k of household composition
upon demand can be expressed in elasticity terms as
dk e
M h xhi dk e
dk xhi
ai
a0
=
.
e
ai dk
a0 dk
xhi dk
xhi M h e
Pn
Pn
xh
The budget identity j=1 pj djk = 0 and the condition that j=1
Pn pj xh ea
e
a0
= j=1 M hj deakj dkj . Substituting into (3.2) gives
1 imply deak0 d
k
(3.2)
h xh
pj xh
j M
j
M h xh
M h
j
n
n
dk xhi
ai X M h xhi X pj xhj dk e
aj
dk e
=
.
e
ai dk j=1 xhi M h j=1 M h e
aj dk
xhi dk
(3.3)
In principle, (3.3) provides a basis for empirically estimating the eects of household composition on demand elasticities. An example of an empirical application
of (3.3), which forms the basis for the equivalence scales used by the Department
of Social Security in the UK, can be found in McClements (1977).
3.3.4
Barten
Since the shortcomings of the methods discussed so far are due to their lack of
choice-theoretic foundation, the value of building equivalence scales from a basis
68
in utility theory is clear. Although each of the methods so far can be interpreted
in terms of the restrictions they place on preferences, see Muellbauer (1977) and
Coulter et al. (1992), this was not explicit in their original motivation. The first
attempt at making preference restrictions explicit can be attributed to Barten
(1964) who approached the derivation via the direct utility function. Extensions
of the method using duality can be found in Gorman (1978) and Muellbauer
(1974).
It has been noted that some goods may not be consumed by children, the
adult goods of Rothbarth, whilst children consume relatively large proportions
of goods such as food. This observation motivates letting a household of type
d be equivalent to ai (d) adults for the consumption of good i, i = 1, ..., n. The
households utility is then
h
xhn
x1
,
...,
hn ,
Uh = Uh
1 , ..., x
(3.4)
= Uh x
1
n
a (d)
a (d)
xh
i
. The behaviour of the household is determined by choosing
where x
hi = ai (d)
h
1 , ..., x
hn to maximise U h x
hn subject to the budget conthe quantities P
x
1 , ..., x
n
straint M h = i=1 pbi x
hi , where pb = (b
p1 , ..., pbn ) is the demographically adjusted
price vector with pbi = pi ai (d). The form of this price vector and budget constraint motivates the observation that lies at the heart of this approach When
you have a wife and a baby, a penny bun costs threepence (Gorman (1978) p.
9).
h
h
h
Solving the maximisation gives the demand
functions
x
=
x
p
b
,
M
or
i
i
h
i
h
1
n
h
i p1 a (d) , ..., pn a (d) , M . The latter of these shows how a
xi = a (d) x
change in composition has two eects upon demand. It aects demand directly
via the equivalence term ai (d) and indirectly via the equivalence terms aecting
the demographically adjusted prices. This specification should be contrasted to
(3.1). In elasticity form the eect of a change in demographic characteristic k
upon demand for good i is
n
ai dk X pj i b
dk aj
dk xhi
xhi
=
+
a
.
dk ai j=1 xhi pj
aj dk
xhi dk
(3.5)
Equation (3.5) provides a basis for empirically estimating the eects of household composition on demand elasticities and an application can be found in
Muellbauer (1977).
Contrasting (3.5) to (3.3) dierences emerge only in the final term. If the
Slutsky equation for the eect of a change in the price of good j on the demand
p
for good i is multiplied by xhj it becomes
i
pj xhi
pj hi
M h xhi pj xhj
=
,
xhi pj
xhi pj
xhi M h M h
(3.6)
where hi is the compensated demand for good i from household h. The specifications in (3.3) and (3.5) can then be seen to be identical if
h
i
pj
= 0 so that
69
the compensated demand for good i is independent of the price of good j. The
two methods are therefore equivalent when the compensated demands depend
are independent.
To construct the equivalence
scale, the expenditure function dual to the
utility function, U h x
hn , can be defined by
h1 , ..., x
h
h
subject to U h x
Uh ,
E h pb, U h = min pbx
Using the definition of demographically adjusted prices it follows that
E h pb, U h = E h p1 a1 (d) , ..., pn an (d) , U h ,
(3.7)
(3.8)
so the expenditure function captures all the demographic information via the
h
equivalence terms ai (d). For some give level of welfare, U , the equivalence
scale for two-household compositions db and de is given by

h
E h p1 a1 db , ..., pn an db , U

h .
(3.9)
E h p1 a1 de , ..., pn an de , U
If the specification in (3.4) is correct, (3.9) provides an exact equivalence scale

in the sense that it measures the true economic cost of demographics. Its form
also makes apparent how equivalence scales will generally be dependent upon
h
the price level and the base level of welfare, U , at which the comparison is
made. Equivalence scales that are independent of the base level of utility will
be considered in the next section.
For the two good case , this procedure is illustrated in Figure 3.3. The outward shift of the indierence curve is caused by the increase in the number of
family members requiring an increase in household consumption to keep household utility constant. The extent to which the budget line has to be shifted
outward determines the extra income required to compensate for the change in
demographic characteristics.
The Barten approach provides a model of household welfare formation that
is empirically implementable. The specification of utility chosen is a particularly
precise one but has been rejected as inappropriate in some econometric tests,
see Muellbauer (1977). In common with all the methods discussed so far, it
also treats the demographic variables as lying outside the control of the household. In some circumstances this may be appropriate; the consequences of this
assumption and the eect of relaxing it are considered in the next section.
3.3.5
General case
The preceding equivalence scales were constructed upon the implicit assumption
that the demographic variables were exogenous and were not an object of choice
by the household; in the terminology of Pollak and Wales (1979) they were
conditional scales. If the demographic vector, or some components of it, aect
70
Figure 3.3: The Barten Scale

utility directly rather than just through its interaction with consumption, an
unconditional equivalence scale is required. To see the distinction, let utility
take the form

U h = U xh , d , d .
(3.10)
An estimation procedure based upon observed market demand
will only be
able to infer information on the separable component U xh , d and will not
uncover the function (). Unfortunately, for welfare comparisons and for the
construction of equivalence scales, the function () is required.
The relevance of the distinction between conditional and unconditional equivalence scales is critically dependent upon whether the demographic vector enters
the utility function. If it does not, there is no distinction between the two and
the previous approaches are valid. When it does, Blundell and Lewbel (1991)
prove that demand data observed in a single price regime can provide no information on equivalence scales. There are three responses to this. The first
is simply to use the demand
data to construct cost functions relating to the

component of utility U xh , d and use these to form cost of living indices for
demographically dierent groups. These cannot be used to make welfare comparisons. An alternative procedure is to impose a convenient form upon the
function () and proceed as if this were the correct form. Whether the imposed form is correct will not be testable. Finally, it may be possible to employ
evidence other than demand data that sheds some light upon preferences over
demographic variables. None of these alternatives provides an exact solution to
the identification problem.

Given the utility function U h = U xh , d , d , let the corresponding ex-

penditure function be

E h p, d, U h = min pxh subject to U xh , d , d U h .
71
(3.11)
As in (3.9), the equivalence scale for two demographic compositions db and de is

defined by
b Uh
E h p, d,
.
(3.12)
e Uh
E h p, d,
A desirable property for an equivalence scale to possess is for it to be independent of the level of utility at which it is evaluated since this permits the
same equivalence scale to be used to adjust all levels of income. If the equivalence scale satisfies this property, it is said to be independent of base. For the
equivalence scale in (3.12) to be independent of the base level of utility, Lewbel
(1989, 1991) has shown that the cost function must be of the form
E h p, d, U h = F (p, d) G p, U h .
(3.13)
The suciency of this condition can be seen by forming the ratio in (3.12) to
obtain the equivalence scale

F p, db
,
(3.14)
F p, de
which is independent
cases
of the base level of utility. Two

special
arise from
(3.13). If G p, U h is independent of p, hence G p, U h = G U h , the cost
function represents homothetic preferences. When F () is independent of p the
underlying utility function takes the form
h
x
h
U =U
,
(3.15)
f (d)
which is a special case of the Barten model.

The main conclusion to be drawn from this analysis is that equivalence
scales are only independent of the base level of utility if the cost function is
separable between utility and demographic characteristics. Whether this is true
in practice, and if it were many applications of equivalence scales would receive
theoretical justification, is a matter of empirical testing. Evidence to date,
namely that of Blundell and Lewbel (1991), rejects the independent of base
assumption.
3.3.6
In practice
There have been many equivalence scales constructed (Buhmann et al (1989) list
34 and their study is not comprehensive) and so it is necessary to find a simple
means of summarising their content. Buhmann et al. (1989) suggest that an
72
acceptable approximation of existing equivalence scales is to write equivalised

incomes as
h
fh = M ,
M
(3.16)
[dh ]
where the demographic variable, dh , now measures only the number of family
members and the elasticity, , varies between 0 and 1. Four types of equivalence
scales are then identified. Expert statistical scales are designed for statistical
purposes such as counting the number above or below a given standard of living. Their typical value of is 0.75. Expert programme scales are designed
to assist with welfare programmes and have a typical value of of 0.55. Both
these expert scales are constructed in the manner of the Beveridge scale where
an outside judgement is made on needs. Consumption scales are based on observed expenditures and lead to an average value of 0.36. Finally, subjective
scales, constructed on the basis of questionnaire evaluations of income, have the
smallest of 0.25.
There is considerable variation in these scales which is linked to the means
of construction and the purpose for which the scale is required. The expert
statistical scales assign the largest increase in cost for increases in family size
and give little weight to potential economies of scale in consumption. Conversely,
the subjective scales find the greatest economies of scale but, being based on
consumers own perceptions of their welfare status, are not free from reporting
bias.
In assessing these methods of constructing equivalence scales, it should
be noted that the minimum needs and the Engel/Rothbarth approaches are
straightforward but as a consequence are not without their faults. The exact
measures capture precisely what the other two are attempting to approximate
but correspondingly are more dicult to calculate. It should also be noted
that by basing the exact measures upon utility analysis, they are subject to
the diculties concerning interpersonal comparability discussed in Chapter 2
and dependent on the appropriateness of the utility maximisation assumption.
Furthermore, there is also the diculty of imposing a single, well-defined, utility function such as (3.10) as representation of a households preferences. The
household utility function must represent the aggregation of the preferences of
the individuals constituting the household but, although on a smaller scale, such
aggregation is subject to the diculties already noted for social preferences in
Chapter 2. This observation raises obvious questions about the interpretation
of (3.10).
3.4
The measurement of inequality
Given a set of income levels for the households in an economy, or a specified subgroup of an economy, adjusted by an appropriate equivalence scale, an obvious
question for anyone who has any concern for distributive justice is to ask how
equally distributed is income in that economy or sub-group. Income inequality
may be of interest in its own right or it may be relevant because of its perceived
3.4. THE MEASUREMENT OF INEQUALITY
73
consequences. In either case, in order to meaningfully discuss inequality, it must

first be necessary to measure it.
There are many measures of inequality that can be employed and all have
features to commend them. Broadly speaking, measures of inequality can be
divided into three groups. The first group are those measures that can be
termed statistical and, as such, are typically standard statistical indices applied
to inequality. The welfare measures constitute the second group and take as
their starting point an explicit formulation of social welfare from which the
inequality measure is developed. This division does not imply that statistical
measures have no welfare implications; in fact, a fundamental observation in the
theory of inequality measurement was the recognition of their implicit welfare
assumptions. The final category of measures can be termed axiomatic. These
are derived by specifying properties that it is thought a satisfactory measure
should possess and deriving all measures that satisfy those conditions.
After providing basic definitions, the most relevant of the statistical measures are noted. This is done both for completeness and because a number of
these measures feature in the ensuing analysis. The connection between social
welfare and statistical measures is then clarified and this is followed by discussion of explicitly welfare-based indices. The axiomatic derivation of inequality
measures is then reviewed. As a final point, the discussion is given in terms
of the measurement of income inequality. However, the measures that are described can be applied to the measurement of the inequality of any vector or
distribution of observations be it data on wages, wealth, welfare or even noneconomic data such as weights. The only distinction between these applications
are the properties that it may be desirable for the index to possess.
3.4.1
Basic definitions
Measures of income inequality can either be defined in terms of discrete or

continuous distributions of income. Although the former is correct in an observational sense, the latter often allows simpler derivation of results and is a valid
approximation for large populations. Most attention will be paid below to the
discrete case but the continuous case will be used where appropriate.
With a discrete distribution of income, it is assumed that there are H households labelled h = 1, ..., H with the labelling chosen so that their incomes, M h ,
form an increasing sequence. Hence
M 1 M 2 M 3 ... M H .
(3.17)
The mean income, , is defined by

=
H
1 X h
M .
H
(3.18)
h=1
For a continuous distribution, the basic data is a density function for income.
Denoting the edensity by (M ) and assuming the support of this distribution to
74
RM
be 0, M , where M is the maximum level of income, so that 0 (M )dM = 1,
the mean level of income is
=
M (M )dM.
(3.19)
The basic task in inequality measurement is to assign a single number to the

distribution M 1 , ..., M H that describes the inequality of the distribution and
permits alternative distributions to be ranked. Formalising this, given the set
of incomes M = {M 1 , ..., M H }, an inequality measure is a function I(M ) such
f) > I(M
c) then the income distribution M
f is judged as having more
that if I(M
c
inequality than distribution M . Typically, a normalisation is adopted so that
I : <H
+ [0, 1] with a value of 0 representing complete equality and 1 maximum
inequality. For continuous
the inequality
distributions,
measure is defined as the

functional I with I : C 0, M [0, 1], where C 0, M represents the space of
continuous distribution functions on 0, M . These definitions imply that the

inequality measure has to provide a complete ranking of income distributions.
Other than defining its range and domain, the form of the function I(M ) has
not been restricted. This permits the maximum freedom in the choice of I(M )
and allows its form to be restricted only by considerations directly relevant to
the measurement of income inequality.
Given an income vector M = {M 1 , ..., M H }, an index that assigns the same
inequality to the vectors M and M for any > 0 is termed a relative index.
Equivalently, I(M ) = I(M ) so a relative index is homogeneous of degree 0
in M . This property is not entirely innocuous. Its acceptance leads to an
interpretation of inequality as a relative concept in which a scaling up of all
incomes leaves inequality unchanged. An alternative view would be that an
addition of an equal amount to all incomes should leave inequality unchanged so
that I(M ) = I(M + ) for all > M 1 . The implications of these assumptions
are investigated further below.
3.4.2
Statistical measures
The statistical measures of inequality that are presented represent only a selection from the many that exist. The selection is on the grounds of extent of
use in practice or theoretical relevance; Sen (1973) and Cowell (1977) discuss
further statistical measures.
The Lorenz Curve
The Lorenz curve has played an important role in the measurement of inequality since its introduction by Lorenz (1905) and constitutes a helpful graphical
device for presenting a summary of data on income distribution. Although not
an inequality index as defined above, Lorenz curves are considered here because of their use in illustrating inequality and the central role they play in the
motivation of other inequality indices.
75
Figure 3.4: Lorenz Curves as an Incomplete Ranking

The Lorenz curve is constructed by arranging the population in order of
increasing income and then graphing the proportion of income going to each
proportion of the population. If all households in the population had identical
incomes the Lorenz curve would then be the diagonal connecting the points
(0, 0) and (1, 1). If there is any degree of inequality, the ordering in which
the households are taken ensures that the Lorenz curve lies below the diagonal
since, for example, the poorest half of the population must have less than half
the proportion of income. For a continuous income distribution, the coordinates
(x, y) of the Lorenz curve are given by
Z m
Z
1 m
(M ) dM,
M (M ) dM
(3.20)
0
0
as varies from 0 to M .
Although the Lorenz curves does not constitute an index of inequality in
the sense of assigning a number to the distribution, it can still be employed to
unambiguously rank some income distributions. As illustrated in Figure 3.4, if
the Lorenz curve for distribution B lies entirely outside that for distribution A,
distribution B can be seen to be less equal. In particular, B could have been
derived from A by transfers from poor to rich. In this case, A can be said to
Lorenz dominate B. If the Lorenz curves representing distributions A and B
cross, an unambiguous ranking of the two distributions cannot be derived from
the Lorenz curve alone. The concept of Lorenz domination therefore provides
only a partial ordering of income distributions.
A fundamental concept in the theory of inequality measurement is the PigouDalton Principle of Transfers which was first formulated by Dalton (1920). The
basis of this principle is that any transfer from a poor household to a rich one
must increase inequality. The principle can be defined formally as follows.
Definition 11 (Pigou-Dalton Principle of Transfers) The inequality index must
decrease if there is a transfer of income from a richer household to a poorer
76
household which preserves the ranking of the two households in the income distribution and leaves total income remaining unchanged.
Any inequality measure that satisfies the Principle is said to be sensitive to
transfers. The Pigou-Dalton Principle is generally viewed as a condition that
any acceptable measure of inequality should meet and is therefore adopted in
most axiomatisations of inequality indices. Jenkins (1991) assesses the transfer
sensitivity of a number of statistical inequality measures. For the Lorenz curve,
a transfer of income from poor to rich moves the Lorenz curve further from the
diagonal. The Lorenz curve therefore satisfies the Pigou-Dalton Principle.
The Gini Coecient
The Gini coecient, attributed to Gini (1912), has been the subject of extensive
attention in the literature on inequality measurement, finding both proponents
and opponents. This interest has lead to a thorough understanding of the implications of its structure and to extensions of the basic index. The Gini coecient
has also seen considerable use in applied economics.
The basic Gini index, G, can be expressed in several alternative, but equivalent, ways. The first method of expressing the Gini is the most straightforward
mathematically but in itself carries little obvious meaning. In this case, the Gini
appears as the following ane function of a weighted sum of relative incomes
G=1+
1
2
2 M H + 2M H1 + ... + [H 1] M 2 + HM 1 .
H
H
(3.21)
The second formulation expresses the means by which the first was derived.
The Gini index considers all possible pairs of incomes and out of each pair
selects the minimum income level. Summing and normalising provides a formula
equivalent to (3.21) but demonstrating more clearly the process involved
H
G=1+
1 XX
min M i , M j .
H 2 i=1 j=1
(3.22)
The final way of expressing the Gini coecient exploit its relationship to the
Lorenz curve. As illustrated in Figure 3.5, the Gini index is equal to the area
between the Lorenz curve and the line of equality as a proportion of the area
of the triangle beneath the line of equality. If the area of the box is normalised
at 1, the Gini coecient is then twice the area between the Lorenz curve and
the equality line. Expressed in this way, it can be seen that the Gini coecient
can take values lying between 0 and 1. This definition of the Gini index makes
it clear that the Gini can be used to rank distributions when the Lorenz curves
cross since the relevant areas are always well defined. Since the measures are
all stronger than the Lorenz dominance criteria, they must each contain some
additional assumptions over and above those which are present in the Lorenz
curve comparisons. What this extra structure involves will be investigated below
when statistical measures are related to social welfare.
p yp{}p p
ppp p {}{ggg yp
pp j pp g
77
gg }{
p p
Figure 3.5: Relating Gini to Lorenz

The Gini index is a relative index of inequality so that it is independent of
scale and it also satisfies the Pigou-Dalton This can be seen by considering a
transfer from household i to household j of M with the households chosen so
that M i > M j . From the ranking of incomes this implies j > i. Then
G =
2
[j i] < 0,
H 2
(3.23)
as required. It is important to note that the eect of the transfer on the index
depends only on the relative rankings of i and j in the income distribution. For
example, a transfer from i = 1 to j = 11 counts as much as one from i = 151
to j = 161. It might be expected that an index should be more sensitive to
transfers between households low in the income distribution but it is actually
most sensitive to transfers around the modal income.
A final relevant property of the Gini index is that the Gini can incorporate
the negative income observations which often arise in income data. Extensions
of the Gini index are analysed in Donaldson and Weymark (1980) and Weymark
(1979a).
Theils entropy measure
The entropy measure of Theil (1967) is drawn from information theory and,
in particular, the measurement of the average information content of a system
of information. Although there is little apparent economic motivation in the
construction of the entropy measure, it will appear below when decomposable
inequality measures are discussed.
The definition of the (normalised) Theil entropy index, T , is given by
H
H
X
1 X Mh
Mh
Mh
1
1
Mh
T =
log
log
=
log
. (3.24)
log H
H
H
H
H log H
h=1
h=1
78
In respect of the Pigou-Dalton criteria, the eect of an income transfer, d,

between households i and j upon the entropy index is given by
dT
1
Mj
=
log i < 0,
d
H log H
M
(3.25)
so that the entropy measure does satisfy the criteria. However, the change in
the index now depends on the relative incomes of the two households involved
in the transfer. Whether this form of sensitivity to transfers is the correct one
is again open to question.
This concludes the review of the statistical measures of inequality. At a
number of points in the discussion reference has been made to acceptable criteria for an inequality index to possess and to whether the properties of the
indices, such as the manner in which they were aected by transfers, were satisfactory. Obviously, to judge whether criteria are acceptable or not requires
the existence of some underlying notion of distributive justice or social welfare.
Two questions then arise. Firstly, to what extent can income distributions be
compared without the need for an explicit welfare function and, secondly, do
the statistical measures of inequality have implied social welfare functions? The
answers to these questions constitute the subject of the next section.
3.4.3
Statistical measures and welfare
One of the significant developments in the theory of inequality measurement

was the clarification of the link between statistical measures and social welfare.
There are two aspects involved in this link. The first concerns the extent to
which income distributions can be ranked in terms of inequality without specifying a precise social welfare function and the link between such a ranking and
the statistical measures. The essential work in this respect was undertaken by
Atkinson (1970), Dasgupta et al. (1973) and Rothschild and Stiglitz (1973).
The second aspect is rather more subtle and involves the construction of the
social welfare function implied by an inequality measure. This line of enquiry is
exemplified by the work of Blackorby and Donaldson (1978). Throughout this
section it is assumed that the income distributions to be ranked have the same
mean level of income.
To discuss the extent to which distributions can be ranked without specifying
a precise welfare function it is necessary to introduce a number of definitions.
A square matrix is said to be bistochastic if its entries are nonnegative and each
of its rows and column sum to 1. A permutation matrix is a bistochastic matrix
that has a single positive element in each row and column. For example the
matrix
0.5 0.2 0.3

0.1 0.6 0.3 ,
(3.26)
0.4 0.2 0.4

is bistochastic and
0 1 0
1 0 0 ,
0 0 1
79
(3.27)
is a permutation matrix. The permutation matrices form the extreme points of

the set of bistochastic matrices.
The relevance of these concepts are that given a bistochastic matrix Q and
f = QM represents an averaging of M , or
an income vector M , the vector M
equivalently, could be obtained from M by a series of transfers from rich to
c = P M is obtained from M by
poor. If P is a permutation matrix, then M
re-ordering the components of M .
Now consider a social welfare function

f U 1 M 1 , ..., U H M H = W M 1 , ..., M H ,
W =W
(3.28)
where W () is increasing in its arguments. The welfare function is symmetric if
W (M ) = W (P M ) ,
(3.29)
for all permutation matrices P and Schur-concave (or S -concave) if

W (QM ) W (M ) .
(3.30)
The social welfare function is strictly S -concave if the inequality is strict for
all bistochastic matrices which are not permutation matrices. The assumption of
S -concavity is a natural one in the context of inequality since QM can obtained
from M by transfers from rich to poor and a welfare function that is sensitive
to distribution should increase with such transfers. Conversely, an inequality
index is S -convex if I (QM ) I (M ).
The interconnections between these concepts and the inequality indices already discussed are summarised in the following lemma due to Kolm (1969) and
Dasgupta et al. (1973).
= M
1 , ..., M
H and M
H ,
= M
1 , ..., M
Lemma 12 Given two income vectors M
P
1 M
2 ... M
H
h PH M
h , which are ordered so that M
with H
h=1 M =
h=1
1 M
2 ... M
H , then the following four conditions are equivalent:
and M
(i) there exists a bistochastic matrix Q (which is not a permutation matrix)
such that
PM =QhM ; Pk
h for all k H and PH M
h = PH M
h;
(ii) kh=1 M
h=1 M
h=1
h=1
(iii) M can be obtained from M by a sequence of transfers, with all transfers

from richerto poorer households;

>W M
for all strictly S-concave functions W ().
(iv) W M
Proof. The proof that (i) to (iii) are equivalent can be found in Berge (1963).
That (i) implies (iv) follows directly from the assumption of strict S -concavity.
This shows the equivalence of (i) to (iv). ||
80
The interpretation of this lemma rests on the observation that (ii) is simply
lies everywhere inside
the statement that the Lorenz curve for distribution M
that for M . The equivalence of this to (iv) then implies that all S -concave social
welfare functions will assign greater welfare to the distribution that is ranked
higher by the Lorenz criterion. Hence when a ranking can be derived by the
Lorenz criterion, this ranking will be agreed by all S -concave welfare functions.
To complete this discussion, it is noted that S -concavity is a weaker concept
than quasiconcavity; hence any symmetric quasiconcave function is S -concave.
Slightly dierent versions of this result can be found in Rothschild and Stiglitz
(1973) and treatment of the continuous case in Atkinson (1970).
The converse of Lemma ?? is that if the Lorenz curves for two distributions
cross, then quasiconcave social welfare functions can be found that will rank the
two distributions dierently.
Put together, the lemma and its converse show that the Lorenz dominance
criteria provides the most complete ranking of income distributions that is possible given only that social welfare is an S -concave function of incomes. To
provide a complete ranking when Lorenz curves cross requires more restrictions
to be placed upon the structure of the social welfare function. In addition, any
index of inequality is necessarily stronger than Lorenz dominance and, when it
can be derived from a social welfare function, is derived from a social welfare
function that has more structure than simply satisfying S -concavity (and may
not even satisfy that condition).
The preceding discussion has demonstrated that the statistical inequality
indices may represent social welfare functions with restrictive properties. To
formalise this link consider a social welfare function
w = W M 1 , ..., M H = W (M ) ,
(3.31)
that is S -concave and increasing along rays, so that W (M ) > W (M ) , > 1.

Corresponding to this social welfare function is its transformation, or distance,
function D (w, M ) defined by

M
D (w, M ) = max W
w .
(3.32)
The transformation function determines the maximum extent to which the income vector in a given situation can be reduced while social welfare remains on,
or above, a target level. From its construction, it can be seen that D (w, M )
is homogeneous of degree 1 in M and strictly decreasing in w. The distance
function is also an implicit representation of the social welfare function since
solving
D (w, M ) = 1,
(3.33)
generates the initial social welfare function.

Employing the transformation function, Blackorby and Donaldson (1978)
suggest as an index of equality (the converse of an inequality index) the function
E (w, M ) defined by
E (w, M ) =
D (w, M )
,
D (w, e)
81
(3.34)
where e is the vector (1,...,1). The interpretation of (3.34) is that it measures

the proportion total income can be reduced if it is distributed equally while
remaining on the same level surface of the transformation function. If the social welfare function is homothetic then D (w, M ) can be written in the form
(M ) and the index is independent of the base level of welD (w, M ) = (w) D
fare or reference-level-free; in all other cases it will be dependent on its base. In
addition, if social welfare is homothetic then

E w, M
W M
W M
,
E w, M
(3.35)
so that the index has normative significance. It can also be shown that E (w, M )
is homogeneous of degree zero in M and S -concave in M .
To construct a social welfare function from an equality measure, note that
since D (w, M ) is homogeneous of degree 1 in M it follows that
D (w, e) = D (w, e) = (w) .
(3.36)
Equation (3.34) can therefore be rearranged to give

D (w, M ) = (w) E (w, M ) .
(3.37)
In (3.37), (w) implies the form of the transformation function and is essentially
arbitrary except for the requirement that must be decreasing in w. If E (w, M )
is S -concave in M , so is D (w, M ). A social welfare function can then be
generated by solving D (w, M ) = (w) E (w, M ) = 1, with a social welfare
function for each (W ). For this constructed welfare function, the implied
measure of equality is, of course, E (w, M ).
The connection of this result with statistical measures of inequality is given
in the Theorem 3.1.
Theorem 13 (Blackorby and Donaldson) If the index of equality is referencelevel-free then the implied social welfare functions are homothetic. Conversely, if
the social welfare function is homothetic then its measure of equality is referencelevel-free.
Proof. A reference-level-free equality index can be written E (W, M )
E (M ) so that D (w, M ) = (w) D (M ) = (w) E (M ). Solving around 1
h
i1
1
gives w = W (M ) = f D (M ) with f = (w)
. Hence f is increasing in w
and D (M ) is homogeneous of degree 1. This implies W (M ) is homothetic. If
W (M ) is homothetic then W (M ) = W (M ) with W (M ) homogeneous of

degree1. The transformation
function
for W (M ) is then defined by D (w, M ) =

1
max > 0 : W M
W (M ) = (w) W (M ) = (w) D (M )
w
=
82
where D (M ) = max > 0 : W M

w . D (M ) is homogeneous of degree
1 so that the index of equality can be written E (w, M ) =
(w)D(M)
(w)D(e)
D(M)
E (M ) where the normalisation D (e) = 1 has been employed. This proves the
equality measure is reference-level-free. ||
The proof of Theorem 3.1 provides the statement required for the construction of the social welfare functions underlying the statistical measures. It has
been shown that for reference-level-free indices, which all the statistical indices
given above are, the identity D (M ) = E (M ) holds. Given E (M ) the transformation function can be calculated from this identity and then solved to give
the social welfare function.
Blackorby and Donaldson show that the welfare functions corresponding to
the Gini and Theil entropy inequality indices are given respectively by
WG (M ) =
and
1 H
M + 3M H1 + ... + [2H 1] M 1 ,
2
H
#
"
H
X
1
h
h
WT (M ) =
M log M .
H log H
H log H
(3.38)
(3.39)
h=1
The form of WG (M ) is particularly interesting since it shows that the Gini

coecient is implied by a quasi-concave, but not strictly quasi-concave, and
homothetic social welfare function. This observation clarifies a number of the
issues raised in Newbery (1970) and Sheshinski (1972a) concerning the form
that the social welfare function consistent with the Gini must take. In addition,
it can be seen that none of these functions is obviously superior to the others
as a measure of social welfare.
This section has investigated the link between statistical measures of inequality and social welfare. The first lemma demonstrated that the Lorenz ranking
is consistent with that produced by any S -concave social welfare function so
that to proceed beyond the Lorenz criterion it is necessary to provide further
restrictions upon the social welfare function. The statistical measures can be related to social welfare functions that have properties stronger than S-concavity.
It was shown how equality measures could be constructed from a social welfare function and how the converse construction could also be achieved. The
resulting forms of social welfare function for three of the statistical measures
could then be seen to embody very particular, and not necessarily acceptable,
assumptions.
3.4.4
Generalised Lorenz Curves
The relation between the Lorenz curve and the ranking introduced by a social
welfare function summarised by Lemma 3.1 concerned only the ranking of income distributions with the same mean level of income. Analogous results for
the ranking of income distributions with dierent means have been derived by
Shorrocks (1983) using the Generalised Lorenz curve.
83
Assume that the social welfare function is of the form (3.31), is S -concave
and is non-decreasing in each income level so that all h. Given two income
distributions and , then it is clear that for any S -concave and non-decreasing
social welfare function if the Lorenz curve of lies inside that of and the mean
of is greater than that . To proceed beyond these sucient conditions, it is
necessary to introduce the Generalised Lorenz curve.
Generalised Lorenz curve
In the continuous case the coordinates of the Generalised Lorenz curve are
given by
Z m
Z
1 m
(M ) dM,
M (M ) dM ,
(3.40)
0
0
Ph
Mi
h
.)
, i=1
(and in the discrete case by H
H
In the continuous case the Generalised Lorenz curve is a continuous curve
connecting (0, 0) to (1, ). The same is true in the discrete case if the discrete
points are connected by straight lines and the point for h = 1 is connected to
the origin.
Theorem 3.2 determines the relation between the ranking of income distributions with dierent means by the Generalised Lorenz curve and by the social
welfare function.

c W M
f for all non-decreasing S-concave
Theorem 14 (Shorrocks) W M
Ph
ci
M
social welfare functions if and only if i=1

H
c lies inside that for M
f.
Lorenz curve for M
Ph
i=1
fi
M
so that the Generalised
h
f
f
f by M
f =M
fh for h = 1, ..., H 1
Proof. To prove the if part, define M
H
f
f =M
fH + H [b
andM
e]. As the social welfare function is non-decreasing,
i

P h ci
Ph f
f
f
M
i=1 M
e
f W M
f . In addition, since
W M
b=
e and i=1
for all h,
H
H

f
f
c lies inside that for M
f so W M
c W M
f W M
f .
the Lorenz curve for M
To prove the only if define Wh (M ) =
Ph
Mi
, h = 1,
..., H, and
note
c
f for
that Wh (M ) is S -concave and non-decreasing. Then Wh M Wh M
Ph
ci
M
Ph
i=1
fi
M
all h only if i=1

i=1
.
H
H
This theorem demonstrates that a ranking of two income distributions with
dierent means can only be unambiguous if the Generalised Lorenz curves do
not intersect. It also shows that even if the Lorenz curves for two distributions
intersect, the condition on the Generalised Lorenz curve may still be satisfied.
3.4.5
Welfare-theoretic indices
The discussion above has emphasised the unsatisfactory nature of the implicit
welfare assumptions that are embodied in statistical measures of inequality. A
84
response to this is to make the welfare judgements explicit by deriving the inequality measure from a specific social welfare function. One method of achieving this was discussed in the previous section; the present section considers the
method of equally distributed equivalent incomes suggested by Dalton (1920),
Kolm (1969) and Atkinson (1970). With this procedure, social concern for equity can be incorporated in the index by the choice of the utility of income function assigned to households and the function employed for aggregating household
utilities into social welfare.
To simplify, assume social welfare can be represented by a utilitarian social
welfare function
W =
H
X
h=1
U Mh ,
(3.41)
where the household utility of income function, U (M ), is increasing and strictly

concave so that U 0 (M ) > 0 and U 00 (M ) < 0. Utility U (M ) can either be the
true cardinal utility function representing households preferences or it can be
chosen by the state as its evaluation of the utility of income to each household.
In this second interpretation, since social welfare is then obtained by adding the
individual utilities, welfare judgements can be captured in the choice of U (M ).
Increasing the concavity of the utility function eectively places a higher weight
on low incomes.
Before specifying a useful form of U M h , it is possible to derive the measure

of inequality proposed by Kolm (1969) and Atkinson (1970). Define MEDE as
the solution to
H
X
U M h = HU (MEDE ) .
(3.42)
h=1
MEDE is called the equally distributed equivalent income and is that level of
income that if given to all households would generate the same level of social
welfare as the initial income distribution.
Using the definition of MEDE , the Atkinson-Kolm index is given by
A=1
MEDE
MEDE
=
.
(3.43)
The concavity of the household utility function guarantees that index A is nonnegative. In addition, it is independent of the mean level of income and is
therefore a relative index.
For a two-household economy the construction of MEDE is illustrated in
Figure 3.6. The initial income distribution is given by M 1 , M 2 and this determines the relevant indierence curve of the social welfare function. MEDE is
found by moving round this indierence curve to the line of equal incomes. It is
clear the because of the concavity of the social indierence curve MEDE is less
than the mean income, .
The flexibility of this index lies in the freedom of choice of the utility of
income function and it is this function that determines the importance attached
85
Figure 3.6: The Equally Distributed Equivalent Income

to inequality by the index. One form of utility function that was proposed by
Atkinson (1970) is
M 1 0
(3.44)
, U (M ) = M , 6= 1.
1
This function is isoelastic and concave if 0. At = 1 and = 0. the function
takes the following form
U (M ) =
= 1 U (M ) = log M, = 0 U (M ) = M.
(3.45)
The welfare judgements of the policy-maker are contained in the chosen value
of since determines the degree of concavity of the utility function. Increasing
makes the utility function more concave and reduces the importance given to
high income in the determination of social welfare. Utility function (3.44) has
found numerous uses in applications of public economics.
The Atkinson-Kolm index is an attempt to make explicit the welfare judgements that are contained within statistical measures of inequality. Although the
derivation above worked with a utilitarian social welfare function, this is not essential and it can be seen that the definition of equally distributed equivalent
income applies to any form of welfare function. Within this framework, however, there is considerable ambiguity as to the distinction between the form of
social welfare and the form of individual utility. As noted, the utility of income
function can either be the true function capturing households preferences or it
can be the states evaluation of income to households. In the latter case, the
distinction between individual utility and social welfare becomes arbitrary.
86
3.4.6
Axiomatic inequality measures
The aim of the axiomatic approach to inequality measurement is to derive the set
of inequality indices that satisfy a chosen set of axioms. The axioms are chosen
to capture what are viewed as the desirable properties for an index to possess
and the set of derived indices will constitute all indices with those properties.
As the number of independent axioms is increased, the set of indices that satisfy
them will generally decrease. This is advantageous in focusing attention on a
reduced set of alternative indices and in clarifying which properties indices that
do not satisfy the axioms fail to possess.
A number of potential axioms have already been noted informally in the
above discussion. The first of these, and the least questionable, is the axiom
of symmetry. Letting the index of inequality, I (M, H), depend on the income
distribution, M , and population size, H, this axiom can be stated as
Axiom I1 (Symmetry)
I (M, H) = I (P M, H) for all permutation matrices P .
This axiom has also been termed anonymity and requires that the inequality
measure is not aected by the order in which households are labelled. The
second axiom usually invoked relates to S -concavity of social welfare and the
principle of transfers.
Axiom I2 (Transfer)
I (QM, H) < I (M, H) for all bistochastic matrices which are not permutation matrices.
The direction of this inequality implies that the inequality measure is an
S -convex function of incomes.
Axioms other than I1 and I2 do not meet with universal approval. This
can either be on the grounds that they are not strictly necessary, which is the
case for the decomposability axioms introduced below, or because there exists
an alternative property to that expressed by the axiom which could be equally
justified. Such is the case for the next axiom.
Axiom I3 (Relative)
I (M, H) = I (M, H) for all > 0.
Axiom I3 captures the notion of inequality as a relative concept and states
that a proportional increase in all incomes should leave inequality unchanged.
This has been termed rightist by Kolm (1976). An alternative to Axiom I3 is
Axiom I3 (Absolute)
I (M + e, H) = I (M, H) for all such that M + e 0,
which has been termed leftist. An intermediate axiom between I3 and I3,
termed -invariance by Eichhorn (1988), is the following
Axiom I3 (Intermediate)
I (M + [M + [1 ] e] , H) = I (M, H) for all 0 < < 1 and M +
[M + [1 ] e] 0.
Clearly, I3 would reduce to I3 if = 1 and to I3 if = 0.
Fields and Fei (1978) and Foster (1983) note that many of the standard
statistical measures, including C, G, A, and T, satisfy axioms I1 - I3. This
point was implicit in the previous discussion. For the alternative axioms I3 and
87
I3, Eichhorn (1988) classifies the set of functions that satisfy axioms I1, I2 and
one of I3 - I3. This class of functions has the general form
M + [1 ] e
M 1 + ... + M H
I (M, H) =
, (M ) =
,
(3.46)
(M ) + [1 ]
H
where () is Schur-convex and satisfies (e) = 0. An application of these
alternative axioms can be found in Pfingsten (1986) where tax changes that are
inequality-neutral under the alternative criteria are identified.
A second aspect of inequality measurement that has been pursued via the
axiomatic approach is that of decomposability. Let the population be divided
into subgroups 1, ..., g, ..., G with the population in each being H g . An inequality
measure is decomposable if total inequality can be found by a weighted sum of
the inequality within the subgroups and between the subgroups. Defining Mg
as the income vector within group g, g as the mean income of group g and eg
as the vector of 1s with H g entries, decomposability is captured in the following
axiom
Axiom I4 (Theil Decomposability)
P Hg
P
Mh
G
G
P h=1
I (M1 , ..., MG , H) = G
H
h.
g=1 wg I (Mg , H)+I (1 e1 , ..., G eG , H) , wg =
h=1 M
With this definition of decomposability, the weights are the ratio of the total
income in subgroup g to total income in the population. The following theorem
is proved in Foster (1983).
Theorem 15 (Foster) An inequality index satisfies axioms I1, I2, I3 and I4 if
and only if it is a positive multiple of T .
Proof. See Foster (1983).
The surprising conclusion of Theorem 3.3 concerns the emergence of Theils
entropy index, T , as the unique index that satisfies decomposability as defined
in I4. Although this index appears to have little economic motivation, it does
have receive convincing support from an axiomatic derivation. In contrast, other
indices which appear more appealing on the surface fail to satisfy the required
criteria. A more general result on decomposable indices is proved in Shorrocks
(1984). Replacing I4 by
Axiom I4 (Decomposability)
A decomposable index satisfies I (M1 , M2 ) = A I (M1 ) , 1 , H 1 , I (M2 ) , 2 , H 2

where A is continuous and strictly increasing in I (M1 ) and I (M2 ).
Shorrocks proves the following theorem.
Theorem 16 (Shorrocks) An inequality index satisfies I1, I2, I3, I4 and if
and only if there exists a parameter c and function , continuous and strictly
increasing in I with such that
i
PH hh M h ic
1
1
1 if c 6= 0, 1,
h=1
H c[c1]
i
PH h M h
1
Mh
F (I (M ) , H) =
log
if c = 1,
h=1
H
PH
1
Mh
h=1 log if c = 0.
H
88
Proof. See Shorrocks (1984).

The class of inequality measures identified in Theorem 3.4 are termed Generalised Entropy indices and include the Theil index (3.28) as a special case. In
essence, the theorem shows that once decomposability is imposed, the class of
inequality measures reduces to the Generalised Entropy indices.
Further results on decomposable indices can be found in Bourguignon (1979)
and Shorrocks (1980) who consider alternative definitions of decomposability,
and Cowell and Kuga (1981). Pyatt (1985) provides a rather dierent set of
axioms which characterise the Gini index.
3.4.7
Summary
The section began with an introduction to statistical measures of inequality

which, at first sight, appeared to avoid the requirement for welfare analysis.
Closer inspection of these indices indicated that each embodied implicit welfare assumptions and that their failings emphasised basic criteria that an index
should satisfy. The first point was developed by considering the methodology
of Blackorby and Donaldson (1978) for generating inequality measures from social welfare functions and constructing underlying social welfare functions from
inequality measures. The examples given underline the belief that statistical
measures can have unsatisfactory welfare implications. One response to this
has been the suggestion that an explicit social welfare function should form the
basis of the inequality measure and an example of this approach, the KolmAtkinson index, was described. The existence of basic criteria for an index to
satisfy has been developed into the axiomatic approach to inequality. Some of
the directions in which this can be taken were illustrated by the study of relative and absolute measures and decomposability. It was surprising to find that
decomposability led back to one of the statistical measures.
3.5
The measurement of poverty
The measurement of poverty involves may of the same issues as the measurement
of inequality and it can be argued, as most notably in Lewis and Ulph (1988),
that the two are simply dierent aspects of the same phenomenon. As with
inequality, it is easy to appreciate at an abstract level what constitutes poverty
and an increase in poverty but diculties arise once an attempt is made to
provide a quantitative representation.
The presentation of this section is somewhat similar to that on inequality.
After discussing competing notions of poverty and introducing the necessary
definitions, a number of standard poverty measures are discussed and the critical
discussion of these leads naturally into an axiomatic approach. The major
distinction between the measurement of inequality and poverty is that the latter
necessitates the identification of those households in poverty as distinct form
those which are not. Identification, which is discussed in detail in Sen (1979),
is typically based on the idea of a level of income, termed the poverty line,
3.5. THE MEASUREMENT OF POVERTY
89
above which a household is no longer in poverty. The poverty line is therefore

of central importance in the measurement of poverty so that methods to define
the poverty line and attempts to avoid some of its implications will both be
discussed.
3.5.1
Relative or absolute?
Before measuring poverty, it is first necessary to define it. It is obvious that

poverty refers to a situation involving a lack of income and consequent low level
of consumption and welfare. What is not so clear is the standard against which
the level of income should be judged. Two possibilities arise in this context: an
absolute conception of poverty and a relative one. The distinction between these
has implications for changes in the level of poverty over time and the success of
policy in alleviating poverty.
Absolute poverty
The concept of absolute poverty assumes that there is some fixed minimum level
of consumption (or similarly of income) that constitutes poverty and that is
independent of time or place. Such a minimum level of consumption is often
taken to be a diet that is sucient to maintain health and provision of housing
and clothing. From this view, if the incomes of all households rise, there will
eventually be no poverty. Viewed as an absolute concept, it is possible for
poverty to be eliminated.
Although a concept of absolute poverty was probably implicit in early studies
of poverty, such as Rowntree (1901, 1941) and in the claims following Rowntree
and Lavers (1951) that poverty was no longer a problem, the appropriateness of
absolute poverty has since generally been rejected. In its place has been adopted
the notion of relative poverty.
Relative poverty
The concept of relative poverty is not a recent one; it is only the usage of the
concept that has been recently adopted. Its history can be traced back to at
least Adam Smith (1776) who frames the definition of relative poverty as the
lack of necessities, where necessities are defined as what ever the custom of the
country renders it indecent for creditable people, even of the lowest order, to
be without. It is clear from this definition that relative poverty is defined in
terms of the standards and norms of a given society at a given time. As the
standard of living of the society rises and more goods are required to be decent,
the income level required to be out of poverty must increase.
The notion of relative poverty has also been advocated more recently in
Townsends (1979) discussion of participation in society. This approach to
poverty considers whether the household possesses sucient resources to allow it to participate in the activities which are customary for the economy to
which it belongs. The ability to participate is clearly a relative concept and this
90
view captures much the same features as that of Smith. Relative poverty is also
embodied in the measurement of poverty in the U.K. and E.U.
The adoption of either an absolute or relative view of poverty does have
implications for how the poverty line is determined for a particular economy at
a particular time but is of most consequence for comparisons of poverty across
time or across economies and in the properties that it is felt a satisfactory
poverty measure should have. The latter aspect will become evident in the
specification of axioms for poverty measures.
3.5.2
The poverty line
As already noted, the starting point for the measurement of poverty is to determine a poverty line. The poverty line is defined here as that level of income
on or below which a household is defined as being in poverty. It should be
noted that this definition is not universally agreed and the poverty line is often
interpreted as the level of income just sucient to move the household out of
poverty. This procedure is somewhat arbitrary since it is dicult to accept such
a precise cut-o between poverty and non-poverty. It is now standard practice
and is followed throughout almost all the literature.
In practice, poverty lines have often been determined by following the minimum needs approach that was discussed in connection with equivalence scales.
As noted in Sawhill (1988), this is the case with the U.S. poverty line that was
set in 1965 and has been updated for inflation since. The practice in the U.K.
has been to set the poverty line as the level of income which is 120% or 140%
of the minimum supplementary benefit level (see Callan and Nolan (1991)) but,
since this level of benefit is itself determined by minimum needs, this amounts
to a minimum needs poverty line. In addition, since the level of benefits have
tended to rise with increases in average income, this embodies some aspects of
relative poverty.
An alternative approach that has also been used is the Engel method of
employing the proportion of income spent on food as an indicator of welfare
with those in excess of a critical proportion being deemed as living in poverty.
For examples of this approach see Rao (1981). Desai and Shah (1988) argue
that poverty should be measured by the distance of a households consumption
experience from the norm.
Obviously, these do not exhaust the possible methods of defining the poverty
line and debate about what constitutes the poverty line and how it changes over
time cannot be avoided.
3.5.3
Standard measures
Prior to the recent interest in poverty measurement following the work of Sen
(1976), the number of available poverty measures was rather limited. Eectively,
the measurement of poverty would be based upon either the headcount ratio or
some variant of the income gap ratio. These measures are now discussed both for
91
their historical importance and because of the role they play in the development
of later measures.
It is first necessary to introduce the following notation that is used extensively below. The poverty line is denoted by z, so that any income level below or
equal to z represents poverty. For a typical household, h, gh (M ; z) z M h ,
is the income shortfall of household h and measures the extent to which the
household is below the poverty line. Given the poverty line z and an income
distribution M , the number of households in poverty is given by q = q (M, z).
The dimensionality of the income distribution vector determines the size of the
population, H, via the relation H = H (M ).
The headcount ratio measures the extent of poverty by counting the number
of households whose incomes are not above the poverty line. Expressing this as
a proportion of the population, the headcount ratio is defined by
q
E= .
(3.47)
H
This measure of poverty was used by Rowntree (1901) and has been used in
many subsequent studies.
The major advantage of the headcount ratio is its simplicity of calculation.
Its major disadvantage is that it pays no attention to how far the households
fall below the poverty line and therefore gives no indication of how costly it
would be to alleviate the observed poverty. In addition, a transfer from a poor
household to one that is slightly richer does not change E if both households
remain on the same side of the poverty line and will actually reduce it if the
transfer takes the recipients income above z.
The aggregate poverty gap and the income gap ratio both take account of
how far below the poverty line are the incomes of the poor households. They
are defined respectively by
q
X
V =
gh ,
(3.48)
h=1
and
X gh
1
I=
.
q (M, z)
z
(3.49)
h=1
Since both measures take account of income shortfalls, they do provide information on the expenditure needed to eliminate poverty but, because they give
equal weight to all income shortfalls, they are not sensitive to transfers unless
the transfer takes one of the households out of poverty.
3.5.4
Axiomatic approach
As with inequality measurement, these criticisms of the basic measures indicates

the existence of beliefs about the properties that a poverty measure must have.
Having noted this, an axiomatic approach follows naturally. Two alternative
axiomatic developments will now be considered: the original derivation of Sen
(1976) and the axiom system of Foster, Greer and Thorbecke (1984).
92
Denoting a poverty measure as a function P (M ; z), a basic set of axioms

for poverty measurement begin with the focus axiom which restricts attention
to the incomes of those in poverty.
Axiom P1 (Focus)
c is obtained from M
f
If income distribution M
by a change
in incomes of
f; z = P M
c; z .
households above the poverty line then P M
The second axiom expresses the requirement that measured poverty must
increase when the income level of any of the poor households falls. This axiom
is termed monotonicity.
Axiom P2 (Monotonicity)
f by reducing the income of
h
f; z <
household h, M < z, and all other incomes remain unchanged then P M
c; z .
P M
As discussed in connection with inequality, the level of poverty should not

be aected by a re-labelling of the households. This property is captured by the
symmetry axiom.
Axiom P3 (Symmetry)
c = PM
f for some permutation matrix then P M
f; z =
c; z .
P M
These three axioms are perhaps the least contentious, although poverty measures that do not satisfy P1 have been considered, for example Beckerman (1979)
employs the aggregate poverty gap as a proportion of aggregate income. They
are, however, usually adopted as the basis of an axiomatic system. Defining a
regressive transfer to be a transfer from a household to one with a higher income
level, the fourth axiom relates to the eect of transfers upon measured poverty.
Axiom P4 (Transfer)
f by a regressive transfer among
f; z < P M
c; z .
the poor households then P M
This is the first form of the transfer axiom; an alternative will be introduced
below.
Taken together, any poverty measure that satisfies these axioms must be a
strictly decreasing and strictly S -convex function of the incomes of the poor
households. Since the poverty measure is S -convex, it must also agree with
the Lorenz criteria applied to incomes below the poverty line. It can be seen
that neither the headcount ratio nor the income gap ratio satisfies the Lorenz
dominance criterion of 4.2.1 and therefore do not satisfy axioms P1 - P4. To
proceed beyond the Lorenz criteria requires the addition of further axioms.
Sen measure
Let denote the set of households with incomes on, or below, the poverty line;
there are q of these. Sen (1976) proposes that the general form of a poverty
93
measure should be given by a multiple of a weighted sum of income gaps

X
P (M ; z) = A (M ; z)
gh (M ; z) vh (M ; z) ,
(3.50)
h
where the form of A (M ; z) and the weights vh (M ; z) are determined by additional axioms.
To derive the form of Sens measure, two additional axioms are imposed.
The symmetry axiom permits the q households to be ranked with the household
closest to the poverty line given rank of 1 and the poorest household, that
furthest from the poverty line, given rank q. Let the rank of poor household h
be given by rh (M ; z). The next axiom is
Axiom P5 (Ranked deprivation)
The weight assigned to household h is given by their ranking amongst the
poor: vh (M ; z) = rh (M ; z).
The fifth of Sens axioms is derived by arguing that when all the households
in poverty have the same income, the level of poverty is measured by the product
of the headcount ratio and the income gap ratio. This axiom acts much in the
way of a normalisation to determine A (M ; z). Expressed as an axiom
Axiom P6 (Normalisation)
If M is such that M h = M , h , then P (M ; z) = EI.
From these axioms follows the Sen measure
X
q
2
gh rh = E I + [1 I] Gp
S=
,
(3.51)
[q + 1] Hz
q+1
h
where Gp is the Gini index of income inequality amongst the households below
the poverty line. Expressed in the latter form, this poverty measure can be
seen as combining a measure of the shortfall of income of the poor with one of
distribution of income between the poor.
The method of constructing this measure illustrates the alternative directions
that could be taken. The general form given is not the only possibility and
Takayama (1979) presents an example of an alternative form. Using the ranking
of the poor as the weights is also restrictive; an alternative to this will be
considered below. Finally, the normalisation axiom is also arbitrary and any
number of other possibilities could be chosen. Although having many features
in its favour, the poverty measure (3.51) cannot be assigned any particularly
special value since any variation in the particular axioms chosen would lead to
a dierent index and some of the axioms are not entirely compelling.
Decomposability
An important class of poverty measures that constitute an alternative to (3.51)
have been derived on the basis of decomposability amongst subgroups. To
motivate this discussion, consider the poverty measure given by
1 X 2
P (M ; z) =
gh .
(3.52)
Hz 2
h
94
The measure given in (3.52) is an example of the general form of measure given
by (3.50) but with the weights given by the income gap rather than the ranking.
Given the composition of (3.51) in terms of the Gini index, it is interesting to
note that (3.52) can also be written
h
i
P (M ; z) = E I 2 + [1 I]2 Cp2 ,
(3.53)
where Cp is the coecient of variation of income amongst the poor. Hence

(3.52) also captures poverty by combining total income shortfall and an index
of distribution.
To allow the index to be decomposable amongst subgroups, aggregate measured poverty must rise if it increases for any subgroup. This requirement is
captured by the following axiom.
Axiom P5 (Subgroup monotonicity)
f by increasing poverty in subIf income distribution M
group g without aecting
either
the number in poverty or incomes in other
f; z < P M
c; z .
subgroups then P M
An argument that can be levied against the transfer axiom, P4, is that the
eect of the transfer should be dependent upon the incomes of those involved
in the transfer. For instance, a transfer away from the lowest income household should have more eect on measured poverty than a transfer away from a
household closer to the poverty line. This interpretation of the transfer axiom
is captured in axiom P4.
Axiom P4 (Transfer sensitivity)
It a transfer t > 0 of income takes place from a poor household with income
M h to a poor household with income M h + d, d > 0, then the magnitude of the
increase in poverty must be smaller for larger M h .
Although the poverty measure defined by (3.52) does not satisfy P4, its
form suggests a class of measures that do. Consider the following extension of
(3.52)
1 X h gh i
P (M ; z) =
,
(3.54)
H
z
h
and note that P0 = H, P1 = H.I. As a increases, more concern is placed on

the lower income level households which leads to the following proposition.
Theorem 17 (Foster, Greer, Thorbecke) P satisfies monotonicity, P3, for
> 0, transfer, P4, for > 1 and transfer sensitivity, P4, for > 2. In
addition, P satisfies subgroup monotonicity, P7.
Proof. Obvious by inspection.
In applications to data, for instance to Kenya in Foster, Greer and Thorbecke
(1984), P measures have proved popular. However, the U.K. practice has
been to remain with the headcount ratio, as has the U.S., despite its faults
and the limited information it conveys. It is, however, statistically robust and
corresponds to a concept of poverty based on minimum rights, on the latter see
3.6. CONCLUSIONS
95
Atkinson (1985). There is therefore a growing dierence between the first-best

practice of economists and that of government.
3.5.5
Variable poverty line
The notion of the poverty line conveys a switch out of poverty as a households
income crosses the line. Two obvious diculties exist with this notion. First,
it is dicult to accept such a precise line between poverty and non-poverty.
It would seem more natural for there to be a gradual move out of poverty
as income increased. Secondly, the precision of the poverty line also leads to
diculty in determining where it should lie since the level of poverty may be
critically dependent on the precise level chosen.
These diculties can be overcome by observing that often it is not the precise
level of poverty that matters but changes in the level of poverty over time and
across countries. In these instances the poverty value is not too important but
only the rankings. This suggest the procedure of calculating poverty for a range
of poverty lines (Atkinson (1987)). If poverty is higher today for all poverty
lines than it was yesterday, then it seems unambiguous that poverty has risen.
In this sense, the poverty line may not actually be of critical importance for the
uses to which poverty measurement is often put.
3.6
Conclusions
The chapter began with the aim of studying measures of the state of an economy
which were intended to be independent of welfare criteria. To allow income levels to be comparable between households of dierent demographic composition
equivalence scales need to be constructed. Although these have been computed
upon the basis of minimum needs and observed expenditure on food, careful
examination revealed deficiencies in such approaches. It was then shown that,
if the household acts in accord the theory of utility maximisation, exact comparisons could only be made by basing the scales upon a comparison of household expenditure functions and hence upon the underlying utility functions.
A broadly similar conclusion was reached with measures of both inequality and
poverty: statistical measures that appear welfare-free are actually founded upon
implicit assumptions concerning the form of the social welfare function. In any
case, a close examination of the methodology lead back to the need for welfaretheoretic constructions.
The response to this finding can either be to construct the required index
directly form a specified utility or social welfare function, as in the AtkinsonKolm inequality index for example, or to specify the properties that the measure
must possess and derive the measure from these. This latter approach has gained
popularity in the literature on inequality and poverty and has also lead to some
surprising conclusions. There remains further research to be undertaken, not
least in refining the link between inequality and poverty.
Part II
Analysis in the competitive

model
97
Chapter 4
Commodity Taxation
4.1
Introduction
This chapter is the first to consider policy analysis and to arrive at characterisations of optimal policies. The ideas that it surveys have developed over a
considerable period, beginning with the seminal contribution of Ramsey (1927).
One important feature of this development is the gradual increase in generality
and the recent move towards applying the theoretical analysis to data. This has
moved the theory closer to practical application.
The initial literature on commodity taxation focused upon the following
simple problem. There is a given level of government revenue to be raised
which must be financed solely by taxes upon commodities: how should these
taxes be set so as to minimise the cost to society of raising the required revenue?
If a social welfare function is adopted to represent the states preferences, the
problem can be conveniently rephrased as that of choosing the commodity tax
rates to maximise social welfare subject to the revenue constraint.
The first solution to this problem was given by Ramsey (1927) following its
proposal to him by Pigou. This contribution appears to have been overlooked
for the following forty years during which time the less general inverse elasticities rule became a standard feature of textbooks. The results of Ramsey were
rediscovered by Samuelson (1986) in a 1951 memo to the U.S. Treasury. The
theory of commodity taxation was given its modern form by Diamond and Mirrlees (1971) in an analysis that made much use of the emerging duality methods
and results in general equilibrium theory. Diamond and Mirrlees (1971) derived
both single-household and many-household tax rules and proved the Production Eciency Lemma. Developments since the publication of Diamond and
Mirrlees have been concerned with the practical implementation of the methods
of that paper and in extensions of the basic economy away from the standard
competitive framework with constant returns to scale.
It should be noted that there are close connections between the theory of
commodity taxation and that of public sector pricing. In both cases the gov99
100
CHAPTER 4. COMMODITY TAXATION
ernment is choosing the set of consumer prices that maximise welfare subject
to a constraint. Under the commodity taxation interpretation these prices are
achieved by setting the level of tax to be included in each consumer price whereas
with public sector pricing the prices are chosen directly. However the choice of
tax rate is equivalent to the choice of consumer price. In the context of public
sector pricing, the optimal prices are generally known as Ramsey prices. The
constraint on the optimisation with commodity taxation requires the raising of
a specified level of revenue. With public sector pricing this can be reinterpreted
as the need to raise a given level of revenue in excess of marginal cost. The tax
rates of the commodity taxation problem then translate into the mark-up over
marginal cost in the public sector pricing interpretation.
The chapter begins by deriving the single-household Ramsey rule and providing an interpretation of this. It is then shown how the inverse elasticity rule
follows as a special case. The extension to many consumers is made and the
resolution of the equity/eciency trade-o is emphasised. This is followed by
a review of numerical calculations of optimal taxes based on empirical data.
Three more specialised topics are then considered: generalising the production
technology; the status of untaxed goods; and conditions guaranteeing the uniformity of taxes. A discussion of the Diamond-Mirrlees Production Eciency
lemma concludes the chapter.
4.2
Methodology
The analysis restricts the set of feasible policy instruments available to the government to commodity taxes. The use of optimal lump-sum taxes is assumed
to be prevented by the arguments covered in chapter 2: the relevant characteristics for the determination of taxes are preferences and endowments but these
are private information and will not be truthfully revealed under the optimal
tax system. In contrast, the employment of commodity taxes requires only that
the government is able to observe trades in commodities, which is a far weaker
informational requirement. Although it may be possible for the government to
levy a uniform lump-sum tax, and in a one-household economy such a tax would
also be optimal, it is assumed for simplicity that such taxes cannot be employed.
In an economy where the households are not identical, their introduction does
not significantly modify the conclusions.
The standard methodology in optimal commodity tax theory has been to
consider only linear taxes, either additive so that the post-tax price of good i is
given by pi + ti or multiplicative with post-tax price . In the competitive framework of this chapter, the choice is immaterial. The analysis of income taxation
in Chapter 6 will present an analysis of nonlinear taxation. A social welfare
function is then maximised by choice of the tax rates and the first-order conditions for this maximisation are manipulated to provide a qualitative description
of the optimal tax system. The qualitative description is then interpreted in
terms of eciency criteria and the concern for equity embodied in the social
welfare function. Explicit formulae for taxes are rarely calculated and, indeed,
4.3. THE RAMSEY RULE
101
can only be calculated for a number of uninteresting special cases. In response

to this, numerical studies have been employed to provide concrete results.
There are two important points that need to be made here. Firstly, because
of the normalisation rules employed, the actual values of tax rates can be argued
to have little meaning. Instead, it is the real eect of the tax system upon
the equilibrium quantities of each good that is relevant. This point is argued
forcefully in Mirrlees (1976) where the index of discouragement is introduced to
measure the eect of taxes. The standard procedure of deriving the tax rates will
be followed in this chapter but the Mirrlees interpretation, which is introduced
formally below, should always be borne in mind. Secondly, the discussion of net
trades in 2.13 showed how a households budget constraint could be nonlinear
in the presence of commodity taxation. To avoid this, the economies of this
chapter assume that households trade only with firms.
4.3
The Ramsey rule
The Ramsey rule is one of the oldest results in the theory of optimal taxation
and is probably the oldest formally stated result. It is derived from an analysis of
the simplest form of general equilibrium economy, that with a single household.
The single household basis implies that there can be no equity considerations in
the setting of tax rates so that the resulting tax rule describes an ecient tax
system. As the Ramsey rule forms the basis for later results, its derivation is
described in some detail.
4.3.1
The economy
The Ramsey rule is derived within the context of a competitive economy in

which there are available n consumption goods and a single form of labour.
Labour is the only input into production. In addition, each industry is assumed
to produce a single output using a constant returns to scale technology. There is
a single household or, equivalently, a population of identical households, whose
preferences can be represented by an indirect utility function.
The assumptions on production imply that for each good i there is coecient
ci that describes the labour input necessary to produce one unit of that good.
With a wage rate w, the competitive assumption ensures that the pre-tax price
of good i is determined by
pi = ci w, i = 1, ..., n.
(4.1)
The normalisation rule that is adopted is to choose labour as the numeraire

and to fix the wage rate at the constant value w. In conjunction with the
production assumptions and condition (4.1), this normalisation rule provides a
set of eectively fixed pre-tax, or producer, prices for the consumption goods.
Labour is also untaxed; that this does not involve a further restriction on this
system will be demonstrated in Section 6.
102
Employing the competitive assumption again, post-tax or consumer prices

are equal to the pre-tax prices plus the taxes. For good i the consumer price qi
is
qi = pi + ti , i = 1, ..., n.
(4.2)
Writing xi for the consumption level of good i, the tax rates on the n consumption goods must be chosen to raise the required revenue. Denoting the revenue
requirement by R, the revenue constraint can be written
R=
n
X
ti xi .
(4.3)
i=1
To ensure that there is an economy-wide balance in supplies and demands,

the formal interpretation of this constraint is that the revenue raised by the
government is used to purchase a quantity of labour with value R. This labour
is used by the state for some undefined purpose and does not produce any good
that is traded in the economy. One example that satisfies this assumption would
be the use of labour for defence purposes. The use of a revenue constraint, rather
than a production constraint, has been discussed in 2.7 and will be discussed
further below.
The preferences of the single household are represented by the indirect utility
function
U = V (q1 , ..., qn , w, I) .
(4.4)
The form of (4.4) implies that the household consumes the goods produced and
supplies the labour used in production and by the state. The assumption of
constant returns to scale and competitive behaviour imply that the firms earn
zero profits. The household therefore receives no profit income and lump-sum
income, I, is zero.
4.3.2
Derivation
Employing the economy described above, the optimal tax problem can be summarised by the maximisation
max V (q1 , ..., qn , w, I) subject to R =
{t1 ,...,tn }
n
X
ti xi .
(4.5)
i=1
The Lagrangean corresponding to (4.5) is given by

" n
#
X
ti xi R .
L = V (q1 , ..., qn , w, I) +
(4.6)
i=1
From (4.6), the first-order necessary condition for the choice of tax rate on good
k is
#
"
n
X
V
xi
L
= 0,
(4.7)
+ xk +
ti
tk
qk
qk
i=1
103
where the identities

V xi
xi
V
,
qk
tk qk
tk
have been used. Equation (4.7) can be rearranged to give
"
#
n
X
xi
V
= xk +
ti
,
qk
qk
i=1
(4.8)
(4.9)
and a similar condition must hold for all n of the goods. The interpretation
of (4.9) is that for all goods the utility cost of raising the tax rate on good k
should stand in the same proportion to the marginal revenue raised by the tax
rise. Expressed alternatively, additional tax revenue per unit of utility foregone
should be the same regardless of which tax rate is changed to generate that
extra revenue.
From Roys identity it follows that
V
V
=
xk = xk ,
qk
I
(4.10)
where I is the households lump-sum income and is their marginal utility

of income. Lump-sum income, I, should be clearly distinguished from total
income, M , used previously since total income includes both lump-sum income,
income from the sale of endowment and labour income. Substituting (4.10) into
(4.9)
"
#
n
X
xi
ti
.
(4.11)
xk = xk +
qk
i=1
After rearrangement (4.11) becomes
n
X
xi
ti
=
xk .
qk
i=1
(4.12)
The next step in the derivation is to employ the Slutsky equation to note that
xi
xi
= Sik xk
.
qk
I
(4.13)
Substituting from (4.13) into (4.12) gives
xi
ti Sik xk
=
xk .
I
i=1
n
X
or
n
X
i=1
ti Sik
n
X
xi
=
ti xk
xk +
.
I
i=1
(4.14)
(4.15)
104
The right-hand side of (4.15) is now simplified by extracting the common factor
xk which yields
"
#
n
n
X
X xi
ti Sik = 1
ti
(4.16)
xk .
i=1 I
i=1
The symmetry of the Slutsky substitution matrix implies that Ski = Sik . This
symmetry can be used to rearrange (4.16) to give the expression
"
#
n
n
X
X xi
ti Ski = xk , = 1
ti
.
(4.17)
i=1 I
i=1
Equation (4.17) is the Ramsey rule describing a system of optimal commodity
taxes and an equation of this form must hold for all goods, k = 1, ..., n. It
is important to note that the value of q is independent of the particular good
chosen.
Finally, multiplying both sides of the Ramsey rule by tk and summing over
k gives
n
n X
X
ti tk Ski = R.
(4.18)
k=1 i=1
As the Slutsky matrix is negative semi-definite, the left-hand side of (4.18) is

negative so that has the same sign as government revenue. Given the sign of
, it is now possible to provide a descriptive interpretation of the Ramsey rule.
4.3.3
Interpretation
To provide an interpretation of the Ramsey rule the focus upon the typical good
k is maintained. First note that, by definition of the substitution terms,
Ski =
k
,
qi
(4.19)
where k is the Hicksian or compensated demand for good k. Consequently,

starting from a position with no taxes, and noting that is then the change in
the tax rate on good i,
ti Ski = ti k ,
(4.20)
qi
is a first-order approximation of the change in compensated demand for good k
due to the introduction of the tax ti , but with the property that the derivative
is evaluated at the final set of prices and at post-tax utility level. If the taxes
are small, this should be a good approximation. Extending this argument to
the entire set of taxes, it follows that
n
X
i=1
ti Ski ,
(4.21)
105
Figure 4.1: Interpretation of the Ramsey Rule

is an approximation to the total change in compensated demand for good k due
to the introduction of the tax system from an initial no-tax position.
In considering the value of this interpretation, note that because both utility
and prices change when the tax system is introduced, the actual change in
demand is given by
k p, U 0 k p, U 1 ,
(4.22)
where U 0 is the initial utility level prior to the introduction of commodity taxes
and U 1 the final level after taxation. The structure of the approximation is
illustrated in Figure 4.1.
Putting these points together and recalling that if R is positive then so is ,

the Ramsey rule when written in the form
Pn
i=1 ti Ski
= , k = 1, ..., n,
(4.23)
xk
can be interpreted as saying that the optimal tax system should be such that the
compensated demand for each good is reduced in the same proportion relative
to the pre-tax position. This is the standard interpretation of the Ramsey rule.
The importance of this observation is reinforced when it is set against the
alternative, but completely unfounded, view that the optimal tax system should
raise the prices of all goods by the same proportion in order to minimise the
distortion caused by the tax system. What the Ramsey rule is approximately
saying is that it is the distortion in terms of quantities that should be minimised.
106
Since it is the level of consumption that actually determines welfare, it is not

surprising that what happens to prices is unimportant; they only matter so far
as they determine demands. In addition, the importance of working in terms of
quantities rather than prices is highlighted when it is recalled that prices can
only be determined up to a normalising factor. This point is discussed further
in Section 6.
This emphasis upon quantities suggests defining
Pn
ti Ski
dk = i=1
,
(4.24)
xk
where dk , the proportional reduction in demand, is Mirrlees (1976) index of
discouragement. The Ramsey rule states that the tax system is optimal when
the index of discouragement is equal for all goods.
4.3.4
Implications
The Ramsey rule only provides an implicit expression for the optimal tax rates
and precise statements cannot be made without further restrictions. However,
some general comments can be made. Accepting the approximation interpretation, this suggests that since the proportional reduction in compensated demand
must be the same for all goods it can be expected that goods whose demand
is unresponsive to price changes will bear higher taxes. Although broadly correct, this statement can only be truly justified when all cross-price eects are
accounted for. One simple case that overcomes this diculty is that in which
there are no cross-price eects between the taxed goods; this limiting case will
be considered in the next section.
Returning to the general case, goods that are unresponsive to price changes
are typically necessities such as food and housing. Consequently, the implementation of a tax system based on the Ramsey rule would lead to taxes that would
bear most heavily on necessities, with the lowest tax rates on luxuries. This interpretation has been demonstrated more formally by Deaton (1981) under the
assumption of weak separability of preferences. Put into practice, this structure of taxation would involve low income households paying disproportionately
larger fractions of their incomes in taxes. The inequitable nature of this outcome
is simply a reflection of the single household assumption: the objective function
of the maximisation does not care about equity and the solution reflects only
eciency criteria.
The equilibrium determined by the set of optimal taxes is second-best compared to the outcome that would arise if the tax revenue had been collected
via a lump-sum tax. This is because the commodity taxes lead to substitution eects which distort the households optimal choices and lead to eciency
losses. Although unavoidable when commodity taxes are employed, these losses
are minimised by the optimal set of taxes that satisfy the Ramsey rule.
Since the single-household framework is untenable as a description of reality
and leads to an outcome that would be unacceptable on the most minimal of
107
equity criteria, the value of the Ramsey rule is therefore primarily in providing
a framework and a method of analysis that can easily be generalised to more
relevant settings. Contrasting the Ramsey rule tax system with later results will
also highlight the consequences of the introduction of equity considerations.
4.3.5
Inverse elasticities rule
The inverse elasticities rule, discussed in detail in Baumol and Bradford (1970),
is derived by placing further restrictions on the economy used to derive the
Ramsey rule. To be precise, it is assumed that there are no cross-price eects
between the taxed goods so that the demand for each good is dependent only
upon its own price and the wage rate. Invoking this assumption essentially turns
the general equilibrium model into one of partial equilibrium as it removes all
the interactions in demand and, as shown by Atkinson and Stiglitz (1980), the
inverse elasticities rule can be derived from minimising the excess burden of
taxation in a partial equilibrium framework. The independence of demands is
clearly a strong assumption and it is therefore not surprising that a clear result
can be derived.
To derive the inverse elasticity rule, equation (4.11) is taken as the starting
point. Hence
"
#
n
X
xi
xk = xk +
ti
.
(4.25)
qk
i=1
The assumption of independent demands implies that

xi
= 0 for i 6= k.
qk
Employing (4.26), equation (4.25) reduces to
xk
.
xk = xk + tk
qk
(4.26)
(4.27)
Rearranging (4.27) and dividing by qk , where by assumption qk = pk + tk , gives
tk
xk qk
.
(4.28)
=
pk + tk
qk xk
As
xk qk
1
= d,
qk xk
k
(4.29)
where dk is the price elasticity of demand for good k, (4.28) can be written
1
tk
=
.
(4.30)
pk + tk
dk
Equation (4.30) is the inverse elasticities rule. From inspection, it can be seen
that this states that the proportional rates of tax should be inversely related to
108
the price elasticity of demand of the good on which they are levied. Recalling
the discussion in 3.4, this statement can be viewed as an extreme version of
the general interpretation of the Ramsey rule. Its implication is clearly that
necessities, which by definition have low elasticities of demand, should be highly
taxed. In this case it is a clearly defined result and it is not necessary to be
concerned with approximations. However the restrictiveness of the assumptions
invoked to arrive at (4.30) should not be understated.
4.4
Extension to many-households
The objective of this section is to extend the single-household economy of the

Ramsey rule to incorporate further, non-identical, households. This extension
naturally introduces equity considerations into the determination of the optimal
tax rates. The principal paper in this area is Diamond and Mirrlees (1971) in
which was presented the first integrated analysis of this issue. Other important
references are Diamond (1975) and Mirrlees (1975).
The variant of the Diamond-Mirrlees economy studied in this section is a
restriction of the general case and simply involves extending that used to derive the Ramsey rule by adding further households. The restrictions on the
production technology are retained, so that labour remains the only input into
production and technology is constant returns to scale. The original DiamondMirrlees analysis, which permits greater generality of the production technology,
will be described in Section 6. It is worth noting that the restrictions do not
significantly aect the form of the optimal tax structure.
4.4.1
The optimal tax rule
The economy is assumed to consist of H households. Each household h is

described by an indirect utility function
U h = V h q1 , ..., qn , w, I h .
(4.31)
These functions vary amongst the households. If they did not, the economy
would reduce to that of a single household since lump-sum incomes are all zero
under the assumption of constant returns to scale. Labour remains the untaxed
numeraire with wage rate w and all households supply only the single form of
labour service.
Writing xh1 , ..., xhn for the consumption demands from h, the revenue constraint is given by
n X
H
X
ti xhi .
(4.32)
R=
i=1 h=1
Social welfare is determined by a Bergson-Samuelson social welfare function

which is defined on the vector of indirect utilities
W = W V 1 () , ..., V H () .
(4.33)
4.4. EXTENSION TO MANY-HOUSEHOLDS
109
Combining (4.32) and (4.33), the optimal set of commodity taxes solve the
maximisation problem
H
n X
X
max W V 1 () , ..., V H () subject to R =

ti xhi .
{t1 ,...,tn }
(4.34)
i=1 h=1
Two alternative presentations of the solution to (4.34) are now given. The first
solution follows closely that of Diamond and Mirrlees (1971) and emphasises
eciency and equity aspects. The second solution parallels the Ramsey rule
and has formed the basis for numerical implementation.
From the Lagrangean for the maximisation, the first-order condition for the
choice of the tax rate on good k, is
"H
#
H
n X
H
h
X
X
X
W V h
x
+
xhk +
ti i = 0.
(4.35)
V h qk
qk
i=1
h=1
h=1
h=1
Using Roys identity the first term of (4.35) can be written

H
H
X
X
W V h
W h h
=
xk .
V h qk
V h
h=1
(4.36)
h=1
Now define
W h
.
(4.37)
V h
h is formed as the composition of the eect of an increase in household hs
utility on social welfare and the marginal utility of income for h. It can be
interpreted as the increase in social welfare resulting from a marginal increase
in the income of household h. Following Diamond and Mirrlees (1971), h is
termed the social marginal utility of income for household h.
Employing the definition of h , (4.35) becomes
"H
#
H
H
n X
X
X
X
xhi
h h
h
xk =
xk +
ti
.
(4.38)
qk
i=1
h =
h=1
h=1
h=1
Substituting from the Slutsky equation

xh
xhi
h
= Sik
xhk hi ,
qk
I
(4.39)
into (4.38) and rearranging gives the tax rule

Pn
PH
h
i=1
h=1 ti Ski
PH
h
h=1 xk
PH
h h
h=1 xk
PH
h
h=1 xk
1+
PH
hP
xh
n
i
h=1
i=1 ti I h
PH
h
h=1 xk
xhk
(4.40)
In (4.40), the left-hand side has an equivalent interpretation to that of the

Ramsey rule: it is approximately the proportional change in aggregate compensated demand for good k and is the generalisation of the discouragement index
110
of (4.24). When a positive amount of revenue is to be raised, it is natural to

assume that the level of demand will be reduced by the tax system, so that the
left-hand side of (4.40) will be negative. As with the approximation involved
with the interpretation of the Ramsey rule, this has two sources of error. Firstly,
real incomes change in moving to the system of optimal taxes and, secondly, the
gradient of the demand functions are unlikely to be constant.
The right-hand terms indicate that the reduction in demand should be
smaller when: (i) demand for good k is concentrated amongst individuals with
high values of h , these are the consumers regarded as socially important and
(ii) demand is concentrated amongst those whose tax payments change considerably as income changes. The first of these points can be seen as reflecting
equity criteria. An individual will have a high value of h when their personal
marginal utility of income, h , is large and when their social welfare weight
W
is also large. If the social welfare function is concave, both of these will
V h
be satisfied by low utility households with low incomes. Hence the implication
is that concern for such households will reduce the rates of tax levied on the
goods they consume. The remaining term is related to the eciency aspects
of the tax system. If taxation were to be concentrated on goods consumed by
those whose tax payments fell rapidly with reductions in income, then increased
taxation, and consequently greater distortion, would be required to meet the
revenue target.
The optimal tax rule (4.40) can be expressed in an alternative form that is
closer to that of the standard Ramsey rule. To do this, (4.40) is rearranged to
give
"
"H
##
PH
H
n X
n
h h
X
X
X xh
x
h
i h
k
ti Sik
= Hxk h=1
ti
x
,
(4.41)
h k
I
i=1
i=1
h=1
h=1
where
xk =
PH
h=1
xhk
,
H
is the mean level of consumption of good k across the households.
Now define
n
h X xhi
ti h ,
bh =
+
I
i=1
(4.42)
(4.43)
where bh is Diamonds (1975) net social marginal utility of income measured in

terms of government revenue. It is net in the sense that it measures both the
gain in social welfare h due to an increase in income to h and the increase in
tax payments of h due to this increase in income. From reviewing its definition,
it can be seen that bh involves both equity and eciency eects. Using definition
(4.43), equation (4.41) can be rearranged to give
Pn
PH
h
h=1 ti Sik
PH
h
h=1 xk
i=1
"
= 1
H
X
bh xh
k
h=1
H xk
(4.44)
4.4. EXTENSION TO MANY-HOUSEHOLDS
111
Tax rule (4.44) shows that the reduction in aggregate compensated demand
for the k th commodity due to the introduction of the tax system should be
inversely related to the correlation between bh and xhk . In other words, to the
extent that the values of bh reflect equity concerns, equity implies that goods
consumed by those with high bh s should be less discouraged or, eectively, they
should have lower taxes. In general, the reduction in demand is smaller: (i) the
more the good is consumed by individuals with a high bh ; (ii) the more the good
is consumed by individuals with a high marginal propensity to consume taxed
goods.
The two forms of the optimal commodity tax rule for a many-household
economy in (4.40) and (4.44) illustrate aspects of the eciency/equity tradeo by the manner in which the reduction in demand for a good is related to
the social importance of the major consumers of that good and their general
contribution to the tax revenue. The rules also demonstrate that the major
feature of the Ramsey rule, the focus upon changes in quantities, is maintained
in the general setting.
As with the Ramsey rule, the results do not give an explicit statement of
the structure of taxes but only provide a characterisation of the consequences of
the optimal system. This lack of clear results provides some of the motivation
for the numerical studies reviewed in Section 5. That section will also consider
the information required to implement the optimal tax rule.
4.4.2
Reduction to the Ramsey rule
The tax rule given in the form of (4.44) is useful for discovering when the
many-household rule actually collapses to the standard Ramsey rule. In such
circumstances the equity criteria are eliminated. The two examples below illustrate that this can occur either because all households are given the same
social valuation or because the tax system is unable to discriminate between
households. The latter reason motivates an alternative interpretation of the tax
problem as one of inferring the status of the households from the signals they
provide in their choice of demands.
Example 1 : bh = b all h = 1, ..., H.
In this case the bracketed term in (4.44) becomes
# "
"
#
H
H
X
b X xhk
bh xhk
= 1
1
= [1 b] .
(4.45)
H xk
xk
H
h=1
h=1
As (4.45) is independent of k, it implies that the proportional reduction in

demand will be the same for all goods. Recalling the interpretation of bh , this
will occur if all households are valued equally and have the same propensity
to pay tax or, most unlikely, the valuation and propensity sum to the same
for all households. In either case, the equity considerations are eectively lost
and the solution returns to that describing an ecient tax system. It must be
emphasised that the value of bh is endogenous to the system.
xh
Example 2: xkk is the same for all k.
112
This condition implies that no good is consumed disproportionately by rich

and poor, a situation that will arise whenever households have identical Engel
curves and these are lines through the origin. When this occurs there is no
method of subsidising high bh people that doesnt also subsidise low bh since all
households have the same expenditure patterns and pay the same proportion of
income in tax. It should be noted that identical linear Engel curves aggregatperfectly hence the economy acts as if it were a single household. As above, it
xh
should be stressed that the value of xkk is determined endogenously.
This result also highlights that the commodity tax problem can be viewed
as one of signal extraction. From the purchases of each household the state is
attempting to infer the households preferences and endowment in order that it
be taxed according to its circumstances. However, when the purchases carry no
information, as in the case above, no signal can be extracted and redistribution
fails. Without any redistributive aspect, the taxes must be chosen to meet
eciency criteria only.
An example of an economy that has these properties is one in which all
households have identical Cobb-Douglas preferences and dier only in lumpsum incomes. It is not specified how these lump-sum incomes arise but they
could be due to the operation of a non-uniform (but non-optimal) lump-sum
subsidy system by the government. Assuming each household has the utility

P
function U h = ni=1 xhi i `, where ` is labour supply and ni=1 i = 1, the
demand for good k from household h is
k I h + w
h
xk =
,
(4.46)
qk
and
H
H
1 X k I h + w
k X I h + w

xk =
=
= k I +w .
H
qk
qk
H
qk
h=1
(4.47)
h=1
Thus calculating the ratio of the demand of household h to the mean level of
demand provides the expression
h
I +w
xhk
, all k = 1, ..., n,
(4.48)
=
xk
I +w
which clearly satisfies (ii).
4.4.3
A cautionary note
To this point the analysis has proceeded on the implicit assumption that the
first-order conditions for the maximisations in (4.5) and (4.34) accurately characterise the solution. However this need not always be the case. It is a standard
result that an indirect utility function representing convex preferences will be
quasi-convex in prices. That is, the set of prices that lead to less than a specified
4.5. NUMERICAL RESULTS
113
level of utility is a convex set. In addition, with linearity in labour supply the
indirect utility function may even be strictly convex, see Varian (1984). This
poses diculties for many maximisations in public economics.
For the Ramsey rule, the objective function was the households indirect
utility function and hence was quasi-convex. In the many-household economy,
the objective was some concave function of the vector of indirect utility functions. Despite the concavity of social welfare in utility it need not be concave in
the choice variables, the tax rates, due again to the quasi-convexity of indirect
utilities. In addition, the set of tax rates that generate at least the required
revenue may not be a convex set.
For these reasons the standard suciency conditions of quasi-concave programming cannot be appealed too so there is no guarantee that the first-order
conditions actually describe a maximum. This problem occurs throughout public economics where many maximisations are ill-conditioned and has been explored extensively by Mirrlees (1986). This problem is often put to one side
and it is simply assumed that the first-order conditions will correctly describe
the optima. Although unsatisfactory, there is typically little alternative to this.
Some comfort can be taken in the present circumstances by appealing to the
work of Diamond and Mirrlees (1971) who prove that their first-order conditions
do represent the solution to the optimal commodity tax problem.
4.5
Numerical results
Numerical analysis of optimal tax rates has become popular for two reasons.
Firstly, the tax rules derived above suggest general observations about the structure of optimal taxes but they do not have precise implications. Numerical
analysis can be seen as providing a check on the interpretations and a means of
investigating them further. Secondly, the motive for the analysis is to provide
practical policy recommendations. To do this, the tax rules must be capable of
being applied to data and the values of the resulting optimal taxes calculated.
Numerical studies in this context represent the development of a technology for
carrying out this program of work.
4.5.1
Applications
A series of papers by Ray (1986a), Murty and Ray (1987) and Srinivasan (1989)
have presented progressively more refined estimates of optimal commodity taxes
for the many-household economy based on data from the Indian National Sample
Survey. To calculate the optimal tax rates, the first step is to specify the social
welfare function. The procedure used for this is based on the work of Atkinson
(1970), which was discussed in Chapter 3, and involves adopting an additive
social welfare function and defining a social utility of income function for each
household. Together these give the social evaluation of income to the household.
Denote the aggregate expenditure of household h, which is assumed to be
114
equal to income, by h and let the social utility of income to h be given by

)
(
1
K [h ]
, 6= 1,
1
(4.49)
Uh =
K log h , = 1,
for some positive K. With the additive social welfare function
marginal social utility of income to h, h , is given by
h =

U h
= K h
.
h
W
V h
= 1 so the
(4.50)
The households are then ranked according to their expenditures, with the lowest expenditure household first in the ranking. Setting 1 = 1 for the lowest
expenditure household in the data set, it follows that
1
h
=
.
(4.51)
h
Equation (4.51) implies that as h increases relative to 1 , h declines monotonically at rate . Since a higher value of reduces h relative to 1 for all h > 1,
can be seen as the concern for equity as expressed in the social welfare function,
with high values representing the greatest concern for equity.
The advantage of this method of defining the marginal social utilities is that
the h s are fixed exogenously and can be determined by the observed expenditure levels in the data set. In addition, the concern for equity is clearly expressed
and can be varied parametrically. However, it is important to note that (4.50)
is imposing a very particular structure on the product of social preferences and
individuals utilities of income, a point that can be appreciated by reconsidering
the definition in (4.37). Furthermore, it is denying the endogeneity of income
and hence expenditure.
Given the specified structure of the h s, the optimal tax rates are determined
from the equations
#
"H
H
n X
H
X
X
X
xhi
h h
h
, k = 1, ..., n,
(4.52)
xk =
xk +
ti
qk
i=1
h=1
h=1
and
R=
h=1
H
n X
X
ti xhi ,
(4.53)
i=1 h=1
which were derived above as (4.38) and (4.32) respectively. To apply these
equations to data, it is necessary to have knowledge of individual demands and
the demand derivatives. To obtain these, a demand system is specified and then
estimated for the data set. The demand functions estimated are then substituted
together with the definition of h into (4.52) and (4.53). The resulting n + 1
equations are then solved for the n tax rates and the Lagrange multiplier.
115
The first paper to adopt this approach was Ray (1986a). In that paper the
n + 1 equations in (4.52) and (4.53) were taken as simultaneous equations with
the demand levels and demand derivatives held constant at their initial levels.
This constancy implies that the tax rates given by solving the equations are not
fully optimal but only optimal conditional upon the constant demands. A more
advanced approach is given in Murty and Ray (1987) which incorporates the
eects of the taxes on demand and uses an iterative procedure to solve for the
optimum. The demand system used in Murty and Ray (1987) is based on the
indirect utility function
V h () =
Pn
1 2
i=1 i pi wh
nk=1 pk k wh 0
(4.54)
The form of this function permits the evaluation of various separability assumptions via the values of the estimated parameters. There is separability between
goods and leisure in utility if 2 and are both zero and non-separability between dierent goods when 1 diers from unity. The tax rates were calculated
on the basis of the estimated values of the parameters i , i , 1 given in Ray
(1986a), 0 was set to zero and the value of 2 was assumed.
Defining to be the wage as a proportion of expenditure, the value of which
is imposed upon the analysis, a sample of the results of Murty and Ray (1987)
are given in Table 4.1 for = 2 and 2 = 0.025.
Item
= 0.05
Cereals
-0.015
Milk and milk products -0.042
Edible oils
0.359
Meat, fish and eggs
0.071
Sugar and tea
0.013
Other food
0.226
Clothing
0.038
Fuel and light
0.038
Other non-food
0.083
Table 4.1: Optimal tax rates
= 0.1
-0.089
-0.011
0.342
0.083
0.003
0.231
0.014
0.014
0.126
Table 4.1 illustrates that redistribution takes place via these optimal taxes
since cereals and milk products, both basic foodstus, are subsidised. Such
redistribution is a reflection of the concern for equity embodied in a value of
of 2. In addition, the results are not too sensitive to the choice of which
could not be estimated from the data set. However, the analysis is limited by
the degree of aggregation that leads to the excessively general other non-food
category.
Ray (1986b) also used this framework to analyse the redistributive impact
of Indian commodity taxes. The method employed was to calculate the total
payment of commodity tax, T h , by household h relative to the expenditure of
116
that household. The net gain from the tax system for h can then be defined by
Th
.
h
(4.55)
h
The household is gaining from the tax system if Th is positive since this
implies a net subsidy is being received. Contrasting the gains of household h
from the existing tax system with those of the optimal system then provides
an indication of both the success of the existing system and the potential gains
from the optimal system. For demand estimates based on a linear expenditure
system, the calculations for the existing Indian tax system give the gains noted
in Table 4.2.
Rrual Urban
h
h
Expenditure level Th
Th
Rs. 20
0.105 0.220
Rs. 50
0.004 0.037
Table 4.2: Redistribution of Indian commodity taxes
The expenditure levels of Rs. 20 and Rs. 50 place households with these
incomes in the lower 30% of the distribution. In this case there is a net gain to
households at both income levels from the tax system with the lower expenditure
householder making a proportionately greater gain.
From Ray (1986a) the redistributive impact of the optimal tax system for a
household with expenditure level , where is mean expenditure, is given in Table
4.3.
= 0.1 = 1.5 = 5
T 0.07
0.343
0.447
Table 4.3: Optimal redistribution
For = 1.5 or more, it can be seen that the potential gains from the tax
system, relative to the outcome that would occur in the absence of taxation, are
quite substantial. Therefore it can be seen that with sucient weight given to
equity considerations the optimal set of commodity taxes can eect significant
redistribution and that the existing Indian tax system does not attain these
gains.
4.5.2
Further results
Atkinson and Stiglitz (1972) calculated the optimal tax rates satisfying the
Ramsey rule for a single household economy. Two demand systems were considered: the direct addilog with the parameters and data taken from Houthakker
(1960); and the linear expenditure system calculated by Stone (1954). In both
cases, additive separability in labour supply was assumed. The results support the interpretation of the Ramsey rule in 3.3: food and rent (necessities)
bear the highest tax rates and durable goods (luxuries) the lowest and the
4.6. GENERALISING THE PRODUCTION TECHNOLOGY
117
optimal tax system is non-uniform. Deaton (1977) presents numerical results

that rely on an aggregation argument and therefore fall somewhere between the
single-household and a true many-household economy. The major finding of the
analysis is that the tax rates move further from uniformity as the concern for
equity increases.
A further calculation of optimal commodity taxes can be found in Ebrahimi
and Heady (1988) which, following Deaton and Stern (1986), augments the
standard analysis by including publicly observable demographic characteristics.
The observability of the demographic variable, which is the number of children
in the household, allows lump-sum tax payments to be based upon them and
the tax system consists of commodity taxes, a lump-sum payment independent
of demograhics and a child benefit, with the child benefit potentially variable
according to childrens age. The data for the calculation are taken from the
1981 U.K. Family Expenditure Survey and relate to 2126 households with twoadults but various numbers of children. The computational procedure is based
on an algorithm for finding the fixed point of a mapping that was described in
Heady and Mitra (1980). The results illustrate again that optimal taxes will
be non-uniform. This remains true even with the considerable flexibility in the
setting of lump-sum transfers permitted by the use of the lump-sum transfer
and two distinct child benefits. The tax rates, the lump-sum transfer and the
benefits are all large in value relative to the taxes and transfers seen in practice.
An alternative approach to computation is given in Harris and McKinnon
(1979) who develop an algorithm for solving the necessary conditions for optimal taxes. The algorithm finds the fixed point of a mapping and is applicable
even when supply is a set-valued function of prices rather than a point-valued
function. Although Harris and McKinnon present the results of some simulation experiments, their algorithm does not yet appear to have been applied to
empirical data.
The results above indicate conclusively that the optimal tax rules can be
operationalised, albeit in somewhat restricted settings. The numerical analyses
all have small commodity groupings and none truly solves the problem in its
full generality since producer prices are generally taken as fixed. However the
results do show that progress is being made and illustrate a fruitful direction
for future research.
4.6
Generalising the production technology
The economies analysed in Sections 3 and 4 assumed that labour was the only
input into production and that there were constant returns to scale. These
assumptions were adopted to provide the simplest basis on which to develop the
results. Both assumptions can be relaxed and this section briefly considers the
consequences of so doing.
Retaining the assumption of constant returns but allowing all goods to be
potential inputs into production alters the method of analysis but does not aect
the structure of the results derived. The argument in this case, developed in
118
Diamond and Mirrlees (1971), relies on the disconnection between households

and firms. As there are constant returns to scale, all firms earn zero profits and
households incomes are derived solely from the sale of goods. All these sales
take place at consumer prices, q. Hence households actions are determined only
by consumer prices. In contrast, firms transact at producer prices, p, and are
not influenced in their decisions by the consumer price level. Eectively, the
two sets of prices can be chosen separately and the dierence between the two
determines the tax rates.
Expressed formally, the government can face households with a price vector
(q1 , ..., qn ) which then entirely determines their actions. Aggregating the individual firms production sets determines the economys production possibility
set. The choice of a producer price vector (p1 , ..., pn ) then determines supply
as the profit maximising choice at these prices over all points in the aggregate
production possibility set. Since constant returns to scale are assumed, profit
maximisation determines a supply correspondence rather than a supply function. For the reasons noted in the discussion of the Second Theorem of Welfare
Economics, it is not necessary to worry about the maximisation of individual
firms. Summarising this discussion, the behaviour of the agents in the economy
is described by
n
n
X
X

Households: max U h xh subject to
qi xhi =
qi hi ,
{xh }
i=1
(4.56)
i=1
with optimal choices

xhi = xhi (qi ) , i = 1, ..., n, h = 1, ..., H,
(4.57)
Production: max py subject to y Y,
(4.58)
and
{y}
where Y is the aggregate production possibility set, with supply correspondence

y (p) .
(4.59)
Equations (4.63) and (4.65) capture the independence of household choices from
the producer price vector and of the production choice from consumer prices.
From the maximisations (4.62) and (4.64) it should be noted that both
demands and supplies are homogeneous of degree zero in the relevant price
vectors. This implies
Xi (q) = Xi (q) , > 0, i = 1, ..., n,
(4.60)
y (p) = y (p) , > 0.
(4.61)
and
.Given a set of producer and consumer prices, the tax rates are determined
implicitly by the identity
t = q p.
(4.62)
4.6. GENERALISING THE PRODUCTION TECHNOLOGY
119
However, the homogeneity in (4.66) and (4.67) implies that an equivalent tax
system to that in (4.68) is given by
t0 = q p,
(4.63)
for any , > 0. The equivalence of t and t0 follows from noting that the prices
supporting both tax systems elicit the same production choice and the same level
of demand from consumers. In this sense, the values of the tax rates themselves
are of limited interest since these are only determined up to the choice of the
arbitrary constants , .
This degree of freedom in the choice of tax rates is the basis upon which
Mirrlees (1976) argues for considering the index of discouragement and its generalisations, which are defined in terms of quantities rather than prices, since
it is the eect upon equilibrium quantities that is the real property of the tax
system. It should be recalled that it was reasoning in terms of quantities that
provided the basis for the interpretations of the tax rules given above. The
indeterminacy in the values of the tax rates is usually resolved by adopting the
standard normalisation rule of setting the tax on one of the goods to zero. This
is eectively what was done with the wage rate above, which is why there was
no loss of generality in using that assumption. There has been much confusion
on this point, see particularly the remarks of Mirrlees (1976) which attempt to
provide clarification and the discussion in Section 7.
Returning now to the analysis of the more general economy, the optimal
tax system is found by choosing the consumer price vector q to maximise welfare subject to there being some producer price vector p that will generate the
appropriate level of supply from firms in order to match demand with supply,
with demand being the sum of household and government demand. As this procedure guarantees that all markets are in equilibrium, the government budget
constraint does not need explicit recognition; see 2.6. Writing the production
possibility set in implicit form as
F (X(q) + xG ) 0,
(4.64)
where xG is the fixed vector of government demands, the Diamond and Mirrlees
formulation is summarised by the maximisation
max W V 1 () , ..., V H () subject to F (X(q) + xG ) 0.

(4.65)
{q}
With profit maximising behaviour, producer prices can be determined by noting

that they will be proportional to the normal to the production possibility set
at the equilibrium point. Such a price vector will guarantee zero profit for the
firms. Hence p is proportional to F evaluated at X(q) + xG . Although this
problem appears distinct from that described in (4.34), it does have the same
solution which is given by (4.40) or (4.44).
If the assumption of constant returns to scale is relaxed and decreasing
returns are permitted, those firms with decreasing returns will earn positive
profits. If profits are assumed to be taxed at a rate of 100% then all firms
120
profits are returned to the government and the separation between households
and firms remains. The optimal tax problem is then still described by (4.71),
with suitable assumptions on the production possibility set, and the DiamondMirrlees results still apply. When profits are not taxed at 100%, some profit
income will accrue to households through their shareholdings in firms. Once
this occurs, the separation between households and firms no longer applies.
To show the eect of profits, the simplest case is chosen and it is assumed
that profits are entirely untaxed. Profits will then be distributed to households
in accordance with the distribution of shareholdings and will modify the typical
households decision problem to
n
n
m
X
X
X

max U h xh subject to
qi xhi =
qi hi +
hj j .
{xh }
i=1
i=1
(4.66)
j=1
Since the profit levels of the firms are determined by producer prices, as shown
in (2.15), the solution of (4.72) is now dependent upon p so the household is
aected by the level of producer prices. Household demands are therefore given
by
xhi = xhi (q, ) = xhi (q, p) , i = 1, ..., n, h = 1, ..., H,
(4.67)
since must depend on p and q. As shown by Munk (1978), this dependence

necessitates that a dierent solution procedure is employed.
To derive the single-household tax rule in this case, first let the aggregate
production constraint be written in the form
F (y) = y1 f (y2 , ..., yn ) = 0.
(4.68)
Profit maximisation subject to this constraint implies that

pi = p1
f
.
yi
(4.69)
Using (4.73), the demands from the single household can be written
xi = xi (q, ) ,
(4.70)
where is the aggregate level of profit given by = py.

Now assume that the government demands a fixed vector xG of commodities.
Equilibrium occurs when demands equal supplies or
x + xG = y + .
(4.71)
y = x + xG .
(4.72)
From (4.77)
Now using (4.75), (4.76), (4.78) and the definition of aggregate profit gives the
equations
n
X
=
pi x (q1 , ..., qn , ) xG ,
(4.73)
i=1
4.7. UNTAXED GOODS

pi = p1
121
G
f x2 (q, ) + xG
2 2 , ..., xn (q, ) + xn n
, i = 2, ..., n. (4.74)
yi
Normalising p1 at unity, the n equations in (4.79) and (4.80) will be sucient,

when the conditions of the implicit function theorem are satisfied, to provides
solutions
pi = pi (q) , i = 2, ..., n, = (q) ,
(4.75)
for the producer prices and the profit level conditional upon the vector of consumer prices. Any consumer price vector, q, that satisfies the market equilibrium
condition
x (q, (q)) + xG = y (p (q)) + ,
(4.76)
where p (q) is the vector (p1 , p2 (q), ..., pn (q)), is an admissible price vector. The
set of admissible equilibrium consumer price vectors is denoted Q.
With this formulation, the optimal tax problem involves choosing the vector
of consumer prices to solve the maximisation
max V (q, (q)) subject to [q p (q)] x (q, (q)) = p (q) xG and q Q,
{q}
(4.77)
where the first constraint in (4.83) is the government budget constraint. Dierentiating and using Roys identity, the optimal tax rule is
n
X
xi
xi
ti
+
=
, k = 1, ..., n.
(4.78)
xk +
qk
qk
qk
i=1
Hence the reduction in demand is now related to the eect of consumer prices
upon profits. Further analysis of this result can be found in Munk (1978).
4.7
Untaxed goods
The role of normalisation procedures and of the untaxed good was discussed at
some length in the previous section. The importance of applying normalisations
correctly has been emphasised in the literature on optimal commodity taxation
by the number of cases in which they have been misunderstood. This section
notes the misunderstandings that have arisen and illustrate their origins.
It has been shown that in an economy with constant returns to scale, consumer and producer prices can be normalised separately and that the standard
procedure is to make one good the numeraire and set its consumer and producer
prices equal. This normalisation also has the eect of setting the tax on that
good to zero. The latter fact is clearly seen to be of no consequence whatsoever
since the zero tax is just a result of the normalisation rule. In particular, the
zero tax carries no implications about the nature of the good nor about the ability to tax that good. This follows since the good with zero tax can be chosen
arbitrarily from the set of available goods.
Unfortunately, this reasoning has not been as clearly appreciated in some
of the literature as it should have been. The reason for this has been the
122
convention, as adopted in sections 3 and 4 of this chapter, of taking labour, as the

untaxed commodity. Since labour is often viewed as the negative of leisure, it has
been inferred from this that, since leisure cannot be measured in the same way
that purchases of other commodities can, the zero tax on leisure is a restriction
on the permissible tax system brought about by an inability to tax leisure. In
addition, the further inference is usually made that the optimal tax system
aims to overcome the missing tax on leisure by taxing goods complementary to
leisure. Particular examples of this are found in Corlett and Hague (1953) By
taxing those goods complementary with leisure, one is to some extent taxing
leisure itself (p.26) and Layard and Walters (1978) The theory of second
best tells us that it we cannot tax leisure, we can do better than by taxing
all other goods equiproportionately. (p.184). Many other instances of similar
statements could easily be given. This, of course, is a false interpretation. When
real restrictions upon the permissible range of tax instruments are introduced
the results obtained are aected. A number of such restrictions are considered
in Munk (1980) where it is shown that the resulting optimal tax structure is
sensitive to the precise restrictions imposed.
A further mistake that has arisen in this context can be found in Dixit (1970)
and Lerner (1970). In a single-household economy, any required revenue can be
raised most eciently by a lump-sum tax on the household equal to the value of
the revenue. Noticing this, it has been suggested that a set of commodity taxes
which raise the price of all goods by the same proportion will have the same
eect as the lump-sum tax and therefore that when all goods can be taxed, the
optimal system has the same proportional tax on all goods. This conclusion is
clearly in contrast to that of the Ramsey rule. The mistake in the reasoning
was pointed out by Sandmo (1974) who demonstrated that such a proportional
tax system would raise no revenue. This follows since households both demand
goods and supply labour. A proportional tax then taxes demands but subsidises
supplies and, since the value of household demand equals the value of supply,
the proportional tax is just oset by the proportional subsidy. Eectively, the
proportional tax on all commodities is just a rescaling of the consumer price
vector which does not aect household choices.
Returning to untaxed goods, as the analysis that will now be given shows,
it is correct to state that if leisure is untaxed then goods complementary with
leisure should be taxed more highly but it is not correct to infer from this either
that not taxing leisure is a restriction on the system or that the taxes aim to
compensate for the missing tax. Taking the Ramsey rule as defined in (4.17) for
a two-good and labour economy gives the necessary conditions for an optimal
tax system of
t1 S11 + t2 S12 = x1 ,
(4.79)
t1 S21 + t2 S22 = x2 .
Solving these equations, the tax rates are implicitly characterised by

t1 =
[S12 x2 S22 x1 ] ,
S
(4.80)
(4.81)
4.7. UNTAXED GOODS
123
t2 =
[S21 x1 S11 x2 ] ,
S
(4.82)
where S = S11 S22 S12 S21 can be assumed to be negative by the negative semidefiniteness of the Slutsky matrix. Now define the elasticity of compensated
demand by
qj Sij
cij =
,
(4.83)
xi
and note that by dividing (4.87) by (4.88), multiplying by qq21 and normalising
both producer prices at unity (which can be done by selecting the units in which
the commodities are measured), provides the relation
t1
1+t1
t2
1+t2
c12 c22
.
c21 c11
(4.84)
Representing leisure by good 0, it follows from Eulers theorem and the

homogeneity of degree 1 of the expenditure function in prices that
2
X
qj Sij = 0,
(4.85)
j=0
and hence that after dividing by xi

ci0 + ci1 + ci2 = 0.
Therefore (4.90) can be written in the form
t2
[c11 + c22 ] c10
t1
=
.
1 + t1
1 + t2 [c11 + c22 ] c20
(4.86)
(4.87)
Equation (4.93) states that if the two goods stand in the same relation
to labour, so that c10 = c20 , then they should be taxed at the same rate.
Otherwise the good with the lower value of ci0 , and hence more complementary
with leisure, should be taxed at the higher rate. This analysis is the basis for
claims that the good complementary to the untaxed good should be taxed more.
However, it should be noted that this conclusion is simply an artefact of the
homogeneity of the expenditure function rather than any fundamental result
about the optimal taxes trying to compensate for the missing tax on leisure.
The discussion of Section 6 has already shown that the non-taxation of labour
is simply a harmless normalisation and that any good could have been chosen as
the untaxed commodity. In other words, good 0 could be chosen to be one of the
consumption goods and good 1 labour. In that case, the direct link with leisure
is then lost from (4.93). Additionally, equation (4.90) presents the tax rule in
terms of own-price and cross-price elasticities of the two taxed goods without
any explicit mention of the untaxed good and provides an equally acceptable
point from which to seek interpretations.
124
This section has noted misunderstandings that have arisen in connection

with the interpretation of the position of the untaxed good. Although the tax
rule can be written so as to depend on the complementary with the untaxed
good, there is nothing fundamental in this observation. In addition, arguments
that an optimal system would tax all goods at the same rate are unfounded
since such a system would raise no revenue.
4.8
Uniform taxes
The numerical results reported in Section 5 have demonstrated that in general

the structure of optimal commodity taxes will be far from uniform. However,
uniform taxes are not without their supporters, see for example Hatta (1986),
and it is natural to consider whether there are any circumstances in which the
optimal structure should be uniform.
Conditions guaranteeing uniformity have been derived in papers by Deaton
(1979, 1981) and Besley and Jewitt (1990). These papers have used a variety
of representations of preferences in alternative formulations of the optimal tax
problem. To present the central result in the manner closest to the analysis
above, this section will present the problem in terms of ad valorem taxation
with an indirect utility function capturing preferences.
With ad valorem taxation the consumer price of good i is given by and the
optimal commodity tax problem becomes
max V (q1 , ..., qn , w, I) subject to R =
{t1 ,...,tn }
n
X
ti pi xi .
(4.88)
i=1
For the choice of tax rate on good k the maximisation in (4.94) has the first-order
condition
"
#
n
X
V
xi
pk + pk xk +
ti pi
pk = 0, k = 1, ..., n.
(4.89)
pk
qk
i=1
Eliminating the common factor pk from (4.95) and using the Slutsky equation,
the Ramsey rule in this case takes the form
n
X
i=1
ti pi Sik = xk , k = 1, ..., n,
(4.90)
P
i
with = 1 ni=1 ti pi x
I .
Treating labour as good 0, the homogeneity of the expenditure function can
be used to write
n
X
qi Sik + wS0k = 0,
(4.91)
i=1
where (4.97) is the generalisation of (4.91). Normalising w at 1 and using (4.97)

the optimal tax rule becomes
S0k = xk , k = 1, ..., n.
(4.92)
4.8. UNIFORM TAXES
125
In this form of the Ramsey rule the tax rates are set so that the demand for
each good is proportional to the rate of substitution between that good and
labour. Using Shephards lemma and the relation of the substitution terms to
the expenditure function, (4.98) is equivalent to
C0k (q1 , ..., qn , w, U ) = Ck (q1 , ..., qn , w, U ) ,
(4.93)
with w = 1.
When a uniform system of taxation is optimal, so that ti = t all i, there
must be some t that solves
C0k ([1 + t] p1 , ..., [1 + t] pn , w, U ) = Ck ([1 + t] p1 , ..., [1 + t] pn , w, U ) . (4.94)
Since the expenditure function is homogeneous of degree 1, its first derivative
is homogeneous of degree 0 and its second of degree -1. Therefore (4.100) is
equivalent to
w
w
1
C0k p1 , ..., pn ,
, U = Ck p1 , ..., pn ,
,U ,
1+t
1+t
1+t
(4.95)
or
C0k (p1 , ..., pn , w0 , U ) = 0 Ck (p1 , ..., pn , w0 , U ) ,
(4.96)
w
. For uniform taxation to be always optimal,
where 0 = [1 + t] and w0 = 1+t
(4.102) must hold for all producer price vectors p1 , ..., pn and wage rates w0 .
Condition (4.102) places restrictions upon the forms of cost function, and hence
preferences that they represent, that will always lead to uniform taxation.
From Deaton (1979, 1981), one form of cost function that satisfies (4.102) is
C (p1 , ..., pn , w, U ) = C (c (p1 , ..., pn ) , w, U ) ,
(4.97)
which is called the implicitly separable cost function. More generally, Besley
and Jewitt (1990) have shown that any preferences that can be defined in the
form
f ( (x, x0 , U ) , x0 , U ) = 1,
(4.98)
where x0 is labour supply and (x, x0 , U ) is homogeneous of degree zero in

(x, x0 ) will result in uniform commodity taxes. A simple example of (4.104) is
the utility function

x
U =U
, x0 .
(4.99)
x0
The important conclusion to draw from this analysis is that the conditions
implying uniform taxation are restrictive and there is no reason why they should
be satisfied in practice. Therefore there are no grounds for believing that the
optimal tax system should be uniform.
126
4.9
Production eciency
Production eciency occurs when an economy is maximising the output attainable from its given set of resources. This requires the economy to be on the
boundary of its production possibility set. When such a point is attained, reallocation of inputs amongst firms cannot increase the output of one good without
reducing that of another. In the special case in which each firm employs some
of all of the available inputs, a necessary condition for production eciency is
that the marginal rate of substitution (MRS ) between any two inputs is the
same for all firms. Such a position of equality is attained, in the absence of
taxation, by the profit maximisation of firms in competitive markets. Each firm
sets the marginal rate of substitution equal to the ratio of factor prices and,
since factor prices are the same for all firms, this induces the necessary equality
in the MRS s. The same is true when there is taxation provided all firms face
the same post-tax prices for inputs, that is, inputs taxes are not dierentiated
between firms.
In the context of commodity taxation, Diamond and Mirrlees (1971) proved
the Production Eciency Lemma. Assuming the economy is competitive, the
lemma states that the equilibrium with optimal commodity taxation should be
on the frontier of the aggregate production set. This can only be achieved if
private and public producer face the same shadow prices and if input taxes are
not dierentiated between firms. In addition, since the competitive assumption
implies that any set of chosen post-tax prices can be sustained by the use of
taxes on final goods alone, the latter statement also carries the implication that
intermediate goods should not be taxed.
This result was seen as surprising at its time of publication because it
was clearly in sharp contrast to the predictions of the Lipsey-Lancaster (1956)
Second-Best theory that was being widely applied. Application of Second-Best
theory, which typically suggests that one distortion should be oset by others, would imply that the distortion induced by the commodity taxes should be
matched by a similar distortion in input prices. Commodity taxation is therefore
a special case for which the general reasoning requires careful application.
The eciency lemma, and the structure of the optimal commodity tax problem, can easily be explained diagrammatically for a single household two-good
economy. In Figure 4.2 the horizontal axis measures input use and the vertical axis output. The shaded area is the production set for the economy and
the horizontal distance of the production set from the origin represents the tax
revenue required in units of the input good. It is assumed that the household
supplies the input and consumes the output, so that the supply of more input
from the household permits the purchase of more output. The households budget constraint is therefore upward sloping and, in the absence of lump-sum taxes
or income, must pass through the origin. Denoting the optimal set of post-tax
prices by q, the budget constraint corresponding to this price vector is illustrated. Since supplying the input causes the household disutility, an increase
in input supply must be compensated for by further consumption of output in
order to keep utility constant. The households indierence curves are therefore
4.9. PRODUCTION EFFICIENCY
127
Figure 4.2: Production Eciency
downward sloping.
The optimal tax equilibrium is given by the highest point on the households
oer curve that is in the production set; in Figure 4.2 the optimum is indicated
by point e. At this point the household is on indierence curve I0 I0 and
production is ecient. To ensure that firms earn zero profits when producing at
the equilibrium point, producer prices, p, are determined by the normal to the
production set. For both producer and consumer prices, the price of the input
is normalised at 1. The dierence between the two sets of prices, t = q p,
determines the optimal set of tax rates. It should also be noted that there are
points above the indierence curve I0 I0 which are preferred to e and which
are productively feasible. These can only be reached by the use of lump-sum
taxes and their existence illustrates the second-best nature of the commodity
tax optimum. With the use of an optimal lump-sum tax, the household would
reach indierence curve I1 I1 .
To see that the optimum must be on the frontier of the production set, consider points f and g. If the equilibrium were at f , the households welfare could
be raised by reducing the use of the input whilst keeping output constant. Since
this is feasible, f cannot be an optimum. From g, output could be increased
without employing more input so that g cannot be an optimum. Since this
reasoning can be applied to any point that is interior to the production set, the
optimum must be on the boundary.
Although Figure 4.2 was motivated by considering the input to be labour, a
slight re-interpretation can introduce intermediate goods. Assume that there is
an industry that uses one unit of labour to produce one unit of an intermediate
good and that the intermediate good is then used to produce final output. Figure
128
4.2 then depicts the intermediate good (the input) being used to produce the
output. Although the household actually has preferences over labour and final
output and acts only on the markets for these goods, the direct link between
units of labour and of intermediate good allows preferences and the budget
constraint to be depicted as if they were defined directly on those variables.
The production eciency argument then follows directly as before and now
implies that intermediate goods should not be taxed since this would violate
the equalisation of MRS s between firms.
Moving now to a many-household competitive economy with constant returns to scale, the formal statement of the Production Eciency Lemma can be
given.
Lemma 18 (Diamond and Mirrlees) Assume that social welfare is strictly increasing in the utility level of all households. If either
b
(i) for some i, xhi 0 for all h and xhi < 0 for some b
h;
or
b
h;
(ii) for some i, with qi > 0, xhi 0 for all h and xhi > 0 for some b
then if an optimum exists, the optimum has production on the frontier of the
production possibility set.
Proof. Assume the optimum is interior to the production set. In case (i),
increasing qi would not reduce the welfare of any household and would strictly
raise that of b
h. Such a change is feasible since the optimum is assumed interior
and the aggregate demand function is continuous. The change would raise social
welfare, thus contradicting the assertion that the initial point was optimal. The
same argument can be applied in case (ii) for a reduction in qi .
When decreasing returns are permitted, Dasgupta and Stiglitz (1972) conclude that production eciency is only desirable if the range of government
instruments is suciently great, in eect, only if profits can be taxed at appropriate rates. Mirrlees (1972) provides further clarification of the relation of
profits and production eciency. These findings show that the constant returns
to scale assumption can be relaxed. Whilst retaining the competitive assumption, one partial exemption to the Diamond-Mirrlees rule has been identified by
Munk (1980) and Newbery (1986). If there are restrictions on taxes on final
goods, then production eciency is no longer necessarily desirable. In detail,
Newbery demonstrates that if there are some goods whose optimal tax would
be positive but the goods cannot be taxed, then input taxes should be used
as partial substitutes for the missing final taxes. Similar results to those of
Newbery are also given by Ebrill and Slutsky (1990), although their analysis is
phrased in terms of regulated industries.
The Diamond-Mirrlees lemma therefore provides an argument for the nontaxation of intermediate goods and the non-dierentiation of input taxes between firms. As noted, it has been extended from its original constant returns
to scale setting. However, except for some special cases, imperfect competition
invalidates the lemma and taxes on intermediate goods will raise welfare. This
result will be considered in Chapter 11.
4.10. SUMMARY
4.10
129
Summary
This chapter has reviewed the major contributions to the large literature on optimal commodity taxation in a competitive economy. The Ramsey rule, which
represents the starting point for the modern analysis of commodity taxation,
has been introduced and its standard interpretation has been given. Although
ecient, the tax system the tax system this describes would be inherently inequitable. To introduce equity considerations, the economy was then extended
to incorporate many households following the work of Diamond and Mirrlees.
This extension clarified the eects of equity upon the optimal rates of tax and
demonstrated how the equity/eciency trade-o was resolved. The economy
was then generalised further and the Diamond-Mirrlees Production Eciency
Lemma was proved. Contrary to the expectations of Second-Best theory, this
lemma showed that production eciency is desirable in conjunction with the
optimal set of commodity taxes.
The motivation behind the theory of commodity taxation is to provide practical policy recommendations. Although the analysis remains some way o fully
achieving this aim, a number of interesting studies have applied these optimal
tax rules to data. This line of work is still in its infancy but gradual improvements in generality have been achieved. The numerical results indicate that the
optimal set of taxes may be able to achieve significant redistribution via the
subsidisation of necessities.
Chapter 5
Income Taxation
5.1
Introduction
The taxation of income is a major source of revenue in most developed countries. It is also one of the most contentious. From one point of view, an income
tax is seen as a direct means of eecting redistribution in order to meet objectives of equity. From another, the imposition of an income tax is viewed as a
major disincentive to eort and enterprise particularly when the marginal rate
of tax increases with income. The theory of income taxation shows how these
competing views influence the design of the optimal tax and how the competing
trade-os are resolved.
The analysis of income taxation that is undertaken below follows from the
initial contribution of Mirrlees (1971). Prior to that, there had been no formal
analysis of the structure or determinants of an income tax schedule that fully
captured the eciency/equity trade-o involved in income taxation. In addition,
the Mirrlees analysis also embodied the fact that the truly relevant characteristics for taxation, the unobservable ability levels of the households, can only be
inferred indirectly from observed behaviour. This implies that the structure of
the income tax must be compatible with the revelation of this information by
households.
The chapter begins by providing a general description of the Mirrlees economy which is the basis for the analysis. The major theoretical results that have
been derived both for the general case and for the restricted constant marginal
tax rate case are described. This is followed by a review of the results of a
number of numerical studies. A critical analysis of the assumptions on which
the simulations are based is then given and the implications of modifying these
is illustrated by a brief discussion of labour supply decisions and some attempts
at the direct assessment of labour supply eects. The chapter is concluded by
reviewing extensions of the basic analysis, including the design of tax systems
with both linear and nonlinear taxes, and issues that it omits.
131
132
5.2
CHAPTER 5. INCOME TAXATION
The Mirrlees economy
The value of the Mirrlees economy in the analysis of income taxation is due to
the manner in which it captures the most important features of the tax design
problem. These features are that the no-tax equilibrium of the economy must
have an unequal distribution of income in order to introduce equity motivations
for taxation. The income distribution must also be generated endogenously by
the model, with households diering in the income they earn, and the income tax
must aect the labour supply decisions of the households in order to introduce
eciency considerations. The economy must also be suciently flexible that no
prior restrictions are placed on the tax functions that may be solutions. The
Mirrlees specification is the simplest that satisfies all these requirements.
To simplify and focus the analysis, it is assumed that the economy is competitive and that households in the economy dier only in their levels of skill
in employment. A households level of skill determines their hourly wage and
hence their income. The skill level is private information and is not known to
the government. The only tax instrument of the state is an income tax. An
income tax is employed both because lump-sum taxes are infeasible and because
it is assumed that it is not possible for the state to observe separately hours
worked and income per hour. Therefore, since only total income is observed, it
has to be the basis for the tax system. The content of this restriction is best
understood by considering the consequences of its relaxation. If it were relaxed,
a tax could be levied that was based on income per hour; in many cases this
would be a better guide to a households potential earning power than actual
income. Indeed, in the economy employed below, income per hour is precisely
the ultimate target of taxation. Despite this, it does not seem unreasonable to
assume that, as in practice, only total income is observed.
The income tax function is chosen to maximise social welfare subject to
achieving the required level of revenue. The generality of the analysis, and the
source of many of the diculties involved in carrying it out, derives from the
fact that no restrictions are placed at the outset on permissible candidates for
the optimal tax function. It is intended by this that the economy determines
the structure of the tax function rather than important aspects of the function
being determined by a priori assumptions.
5.2.1
The optimisation
The optimal income tax function is chosen to maximise social welfare. This
maximisation is subject to two constraints. The first constraint is that the income tax function must lead to an outcome that satisfies productive feasibility
or, equivalently, meets the governments revenue requirement. The second constraint that must be satisfied is rather more complex and the way in which it is
handled is of central importance for the analysis. To best understand the nature
of this constraint, an alternative interpretation of the optimisation is helpful.
Rather than viewing the government as choosing an income tax function, it can
be seen as assigning to each household a pre-tax income - consumption pair.
5.2. THE MIRRLEES ECONOMY
133
The additional constraint is then that each household must find it in their own
interest to choose the pre-tax income - consumption pair that the government
intends for them rather than a pair assigned to a dierent household. In other
words, the intended pair be must utility-maximising for the household over the
set of available pairs. Due to its nature, this is termed the self-selection constraint. The nature of the self-selection constraint will be discussed in 2.4.
It is worthwhile noting at this point some of the diculties involved in the
analysis of the general problem. The tax function is not restricted in form so
that for most tax functions the budget constraints of the households will be
nonlinear. In those cases for which the budget set is non-convex, there may
be non-uniqueness in the solution to the individual households maximisations
and the solutions need not form a convex set. These facts prevent the behaviour of the individuals being expressible by demand functions, in contrast to
the commodity tax problem of Chapter 5, and therefore introduce considerable
mathematical complication. It is also the reason why the self-selection constraint must be made explicit in the analysis of income taxation but not in that
of commodity taxation. Mirrlees (1986) provides a thorough discussion of these
issues.
5.2.2
Basic structure
Turning now to further details of the economy, it is assumed that there are two
commodities: a consumption good and a single labour service. A households
supply of the labour service is denoted by ` and consumption of the good by x.
The supply of labour is limited by 0 ` 1 and x 0. When the distinction
becomes necessary, the demands of dierent households will be distinguished by
an additional index. Each household is characterised by their skill level s The
value of s gives the relative eectiveness of the labour supplied per unit of time,
so a high s household is more eective in production.
If a household of ability s supplies ` hours of labour, they provide a quantity
s` of eective labour. The analysis is simplified by assuming that the marginal
product of labour is constant and, for a worker of ability s, is given by s.
The total productivity of a worker during the ` hours at work is then equal
to s`. Denote the supply of eective labour of a household with ability s by
z (s) s` (s). Normalising the price of the consumption good at 1, z (s) is
also the households pre-tax income in units of consumption. Denoting the tax
function by T (z) and the consumption function by c (z), a household that earns
z(s) units of income can consume
x (s) c (z (s)) = z (s) T (z (s)) ,
(5.1)
units of the consumption good.

The ability parameter, s, is continuously distributed throughout the population with support S, which may be finite with S = [S1 , S2 ] or infinite with
S = [0, ]. Typically the finite support will be used, with the infinite support introduced where appropriate. The cumulative distribution of s is given
134
by (s), so there are (s) households with ability s or less. The corresponding
density function is denoted (s).
All households have the same strictly concave utility function, an assumption that permits interpersonal comparability. This common utility function is
denoted
U = U (x, `) .
(5.2)
Each household makes the choice of labour supply and consumption demand to
maximise utility subject to the budget constraint. Hence a household of ability
s chooses x(s), `(s) to
max U = U (x, `) subject to x (s) c (s` (s)) .
(5.3)
Define u(s) = U (x(s), `(s)) as the maximised level of utility at the optimal
choices.
In the absence of income taxation, a household of ability s would face the
budget constraint
x s`.
(5.4)
From (5.4) it can be seen that the budget constraint in (`, x)-space diers with
ability.
For the purposes of tax analysis, it is the budget constraint after tax, or
equivalently the imposed consumption function, that is of interest. The analysis
is simplified if all households face the same budget constraint and this can be
achieved by setting the analysis in (z, x)-space. In this space, the pre-tax budget
constraint is given by the 45o line for households of all abilities. With income
taxation, the gradient of the consumption function (6.1) is equal to 1 minus the
marginal rate of tax, where the marginal rate of tax, T 0 , is defined as Tz(z) ,
and the vertical distance between the consumption function and the 45o line
represents the total tax paid. This is illustrated in Figure 5.1 for the consumer
earning units of income.
5.2.3
The structure of utility
As already noted, the households have identical preferences over consumption

and leisure. The utility function is continuously dierentiable, strictly increasing
in consumption and strictly decreasing in leisure. In addition, it satisfies
Us > 0, U` < 0, Uxx < 0,
(5.5)
U` (x, `) as ` 1.
(5.6)
and
Condition (5.6) implies that each household will endeavour to avoid corner solutions with ` = 1. It should be noted that the quantity of labour supplied
is measured positively in order to conform with the literature. The indierence curves of the utility function are illustrated in Figure 5.2, in which utility
increases to the north west.
Figure 5.1: After-Tax Budget Constraint
Figure 5.2: Preferences
135
136
Figure 5.3: Translation of Indierence Curves

To allow preferences and the budget constraint to be depicted on the same
diagram, the utility function can be written
z
U = U (x, `) = U x,
u (x, z, s) .
(5.7)
s
The indierence curves of u (x, z, s), drawn in (z, x)-space are dependent upon
the ability level of the household since it takes a high-ability household less
labour time to achieve any given level of income. In fact, the indierence curves
are constructed from those in (`, x)-space by multiplying by the relevant value
of s. This construction is shown in Figure 5.3 for the single indierence curve
I0 and households of three dierent ability levels.
A number of the results below will require additional structure to be placed
upon preferences. This involves relating the gradient of the indierence curves
through a given consumption-income point for households of dierent abilities
to the ability levels. The required assumption, termed agent monotonicity, was
introduced by Mirrlees (1976) and named by Seade (1982).
Agent Monotonicity
The utility function (5.7) satisfies agent monotonicity if is uuxz a decreasing
function of s.
It should be noted that uuxz is the marginal rate of substitution between
consumption and pre-tax income and that agent monotonicity requires s
U`
s < 0. An equivalent definition of agent monotonicity is that ` Ux is an

U` (x, z )
increasing function of `. This can be seen by noting that uuxz = sU x,sz .
x(
s)
i
h
U
Calculating
`
` Ux
and s shows
h
i
U`
`
Ux
1
.
s = 2
s
`
(5.8)
Agent monotonicity is equivalent to the condition that, in the absence of taxation, consumption will increase as the wage rate increases. This equivalence is
137
Figure 5.4: Agent Monotonicity

easily demonstrated by calculating the derivative in (5.8) and contrasting this
to the rate of change of consumption with respect to the wage rate derived from
the comparative statics of utility maximisation. A sucient condition for agent
monotonicity is that consumption is not inferior, i.e. it does not decrease as
lump-sum income increases. This result can also be derived from the contrast
between the comparative statics of utility maximisation and the derivative in
(5.8).
Since the marginal rate of substitution is the gradient of the indierence
curve, agent monotonicity implies that at any point in (z, x)-space the indifference curve of a household of ability s1 passing through that point is steeper
than the curve of a household of ability s2 if s2 > s1 . This is illustrated in
Figure 5.4. The name for the condition follows from this monotonicity property
that it imposes upon preferences. Agent monotonicity also implies that any two
indierence curves of households of dierent abilities only cross once.
Theorem 5.1 shows that when the consumption function is a dierentiable
function of labour supply, agent monotonicity implies that gross income is an
increasing function of ability. It is important to stress that no other assumption
than dierentiability is placed on the consumption function to derive this result.
Theorem 19 (Mirrlees ) When ` > 0 and the sucient conditions for utility
maximisation are satisfied, z(s) is an increasing function of s.
Proof. Writing the utility functions as U (c (s` (s)) , ` (s)), the first-order
condition for utility maximisation, U` + Ux sc0 (s`) = 0, is equivalent to
U`
U`
= 0,
s`c0 (s`) + `
s`
Ux
or, since
U`
s`
(5.9)
< 0,
zc0 (z) + `
U`
= 0.
Ux
(5.10)
138
The second-order condition then becomes

h
i
zc0 (z) + ` UUx`
z
0.
(5.11)
The total dierential of the first-order condition (5.10) with respect to s is

given by
i
i
h
h
zc0 (z) + ` UUx`
` UUx` z
= 0.
(5.12)
z 0 (s)
z
` s2
i
h
U`
` Ux
Since agent monotonicity implies ` is negative, it follows from the secondorder condition (5.11) that z 0 (s) > 0.
This theorem can be reinterpreted as saying that if agent monotonicity holds
and the implemented tax function has pre-tax income increasing with ability,
then the second-order condition (5.11) for utility maximisation must hold. In
other words, with agent monotonicity, the second-order condition and the condition that z 0 (s) 0 are equivalent. The importance of the condition z 0 (s) 0
will appear again in the discussion of self-selection.
5.2.4
Self-selection
In choosing the optimal tax function, it must be the case that each household
will find it in their interest to choose the income-consumption pair that the government intends them to select. This self-selection constraint is now introduced
formally and a convenient representation is derived.
Let x (s) and z (s) now represent the consumption and income levels that the
government intends a household of ability s to choose. The household of ability s
will choose (x (s) , z (s)) provided that this pair generate at least as much utility
as any other choice. This condition must apply to all consumption-income pairs
and to all households. Written using (6.7) the self-selection constraint is as
follows.
Self-selection
The self-selection constraint is satisfied if u (x (s) , z (s) , s) u (x (s0 ) , z (s0 ) , s)
for all s, s0 .
It should be noted that the utility arising from the pair (x (s) , z (s)) is compared with that from all other possible combinations. It is this that leads to
some diculties in applying the self-selection constraint. Since the economy has
a continuum of ability levels, there is an uncountable infinity of self-selection
constraints for each household when expressed in the form above. Clearly, such
a set of constraints cannot be easily incorporated as a restriction on a maximisation. An important aspect of the analysis is to construct an alternative
representation of the self-selection constraint that can be more easily accommodated into a maximisation problem.
In the case of linear taxation, each household pays (or receives) a fixed
lump-sum and the marginal rate of tax on income is constant. The imposed
consumption function is linear and each household faces a convex budget set.
139
This structure eliminates the requirement to consider the self-selection constraints since the behaviour of the household can be determined as a function
of the two parameters that describe the tax function: the lump-sum payment
and the marginal rate of tax. The derivation of the optimal linear income tax
is therefore considerably simpler than that of the nonlinear tax and resembles
the analysis of commodity taxation.
When nonlinear taxation is considered, the self-selection constraints must
be included. This is achieved by noting that the satisfaction of the self-selection
constraint is equivalent to achieving the minimum of a certain minimisation
problem. If the sucient conditions for the minimisation are satisfied by the
allocation resulting from the tax function, then the self-selection constraint is
satisfied.
To derive the required minimisation problem, let u (s) u (x (s) , z (s) , s)
represent the maximised level of utility for a consumer of ability s resulting
from (5.3). Now note that
0 = u (s) u (x (s) , z (s) , s) u (s0 ) u (x (s) , z (s) , s0 ) ,
(5.13)
so that s0 = s minimises u (s0 ) u (x (s) , z (s) , s0 ). Hence

u0 (s) = us (x (s) , z (s) , s) .
(5.14)
From the definition of u (s) it follows that

ux x0 (s) + uz z 0 (s) = 0,
(5.15)
is equivalent to (5.14). Condition (5.14), or equivalently (5.15), is the necessary

condition (or first-order condition) for the self-selection constraint to be satisfied.
Its advantage over the direct formulation of self-selection is that it provides
a single condition for each value of s rather than an uncountable infinity of
conditions.
The second-order condition for the self-selection constraint to be satisfied is
found from the second derivative of u (s0 ) u (x (s) , z (s) , s0 ) with respect to s0
to be
(5.16)
u00 (s) uss (x (s) , z (s) , s) 0.
Again using the definition of u (s)
u00 (s) = usx x0 (s) + usz z 0 (s) + uss ,
(5.17)
which gives, by using (5.16),

usx x0 (s) + usz z 0 (s) 0.
Eliminating x0 (s) using (5.15) provides the final condition
s
uz 0
z (s) = z 0 (s) 0,
usz usx
ux
ux
(5.18)
(5.19)
140
where is the marginal rate of substitution introduced in the discussion of

agent monotonicity. With agent monotonicity s < 0, so that satisfaction of
the second-order condition for self-selection is equivalent to z 0 (s) 0. Any tax
function that leads to an outcome satisfying (5.15) and will therefore satisfy the
self-selection constraint.
It is also worth noting a further result that can be derived directly from
the conditions already stated. Since ux > 0 and u` < 0, z 0 (s) 0 implies
from (5.15) that any tax function satisfying the self-selection constraint must
result in x0 (s) 0. The self-selection constraint therefore requires both pre-tax
income and consumption to be non-decreasing with s.
5.3
Characterisation of optimal tax function
The aim of the theoretical analysis is to provide a characterisation of the properties that the optimal tax function will have, given the specification and assumptions adopted. It will clearly not be possible to calculate the function
without precisely stating the functional forms of utility, production and skill
distribution. What will be achieved is the derivation of a set of restrictions that
the optimal function must satisfy. This is undertaken firstly for the case where
the marginal rate of tax is assumed constant. As already noted, this implies
convexity of the individual budget set and allows behaviour to be expressible
via demand functions which considerably simplifies the resulting analysis. The
general problem is then considered. Results are derived using both the necessary conditions for the maximisation of social welfare and directly using a
diagrammatic framework.
5.3.1
The general problem
Using the individual demand and supply functions and integrating over the
population, it is possible to define total eective labour supply, Z, by
Z
Z=
z (s) (s) ds,
(5.20)
0
and aggregate demand, X, where

Z
X=
x (s) (s) ds.
(5.21)
The optimal tax function is then chosen to maximise social welfare, where welfare is given by the Bergson-Samuelson function
Z
W =
W (u (s)) (s) ds,
(5.22)
0
with W 00 0.
5.3. CHARACTERISATION OF OPTIMAL TAX FUNCTION
141
There are two constraints placed upon the maximisation of (5.22). The first
is that the chosen allocation must be productively feasible. This requirement
can be denoted in terms of quantities by
X F (Z),
(5.23)
where F () is the production function for the economy. This definition of

productive feasibility can incorporate the government revenue requirement, expressed as a quantity of labour consumed by the government z G , by noting that
(5.23) can be written X Fb(Z z G ) = F (Z). Alternatively, the government
revenue constraint can be used in place of the production constraint to express
the restriction in value terms. Denoting the level of revenue required by R,
R z G , the revenue constraint can be written
Z
R
[z (s) x (s)] (s) ds.
(5.24)
0
The second constraint upon the optimisation is that it must satisfy the selfselection constraint. This has already been described in 2.4.
5.3.2
Linear taxation
The complexity of the general model of income taxation has lead to considerable
interest in the restricted case of linear taxation. With linear taxation the marginal rate of tax is constant and there is an identical lump-sum tax or subsidy
for all households. The advantages of this restriction is that it ensures that the
budget sets of all households are convex so that optimal choices will be unique
when preferences are strictly convex. In addition, the tax system is described
by just two parameters: the marginal tax rate and the lump-sum subsidy. The
choice of optimal policy therefore corresponds to a standard maximisation problem. In addition, the linear tax structure corresponds to proposals for negative
income tax schemes, in which all households below a given income level receive
a subsidy from the tax system, and to the tax reform proposals of a number of
countries that have reduced the number of tax rate bands.
Using the notation set out above, under a linear tax system a household
with ability s supplying ` units of labour will pay tax of amount
T (s`) = + ts`,
(5.25)
where t is the marginal rate of tax and is a lump-sum subsidy if positive and a
tax if negative. Denoting [1t] by , the consumption function of the household
is
x = + s`.
(5.26)
Each household chooses consumption and labour supply to maximise utility
subject to (5.26), with utility represented by (5.2). Assuming the solution to
the individual maximisation is interior, the first-order conditions can be reduced
to
U`
= s.
(5.27)
Ux
142
Combined with the consumption function (5.26), (5.27) implies labour supply
and consumption demand functions of the form
` = ` (, , s) , x = + s` (, , s) .
(5.28)
Substituted into the utility function, these determine the indirect utility function
U = U ( + s` (, , s) , ` (, , s)) = V (, , s) ,
(5.29)
V
V
= Ux ,
= Ux s`,
(5.30)
with
where V
is equal to the marginal utility of income.
The governments optimisation problem is to choose the parameters of the
tax system to maximise social welfare subject to raising the required revenue,
R. Using (5.26) and (5.29), the optimisation can be expressed as
Z
max
W (V (, , s)) (s) ds,
(5.31)
{ ,}
subject to
[ + [1 ] s` (, , s)] (s) ds R.
(5.32)
Using (5.30) and defining the social marginal utility of income for a household
of ability s by
V (, , s)
(s) = W 0 (V (, , s))
,
(5.33)
the necessary conditions for the choice of and respectively are
Z
Z
z
(s) ds = H
[1 ]
(s) ds ,
(5.34)
0
0
and
z (s) ds =
z [1 ]
z
(s) ds,
(5.35)
where H is the population size, .

The first use of these necessary conditions is to derive a simple expression
for the optimal marginal tax rate, t, that is due to Tuomala (1985). To do this,
x
divide (5.35) by (5.34) and denote by a bar terms of the form H
. This gives
R
R
z 0 [1 ] z
z (s) ds
(s) ds
R0
.
(5.36)
=
R
z
(s) ds
1 0 [1 ]
(s) ds
0
The term on the left-hand side of (5.36) is now denoted by z() and can be
interpreted as the welfare-weighted average labour supply. From totally dierentiating the government revenue constraint whilst holding R constant, it can
be found that
R
z + 0 [1 ] z
d
(s) ds
.
(5.37)
|R const =
R
d
1
[1 ] z (s) ds
0
143
Hence from (5.36) and (5.37)

z() =
d
|R const .
d
(5.38)
Since averaging over the population must give z = z ( , ), it follows from (5.38)
that, holding revenue constant,
dz
dz
dz d
dz
dz
|R const =
+
|R const =
z().
d
d
d d
d
d
Therefore, recalling that t = 1 , (5.36) can be written in the form
dz
dz
dz
z()
= t |R const ,
z() z = [1 ]
d
d
d
or
t=
z() z
,
dz
dt |R const
(5.39)
(5.40)
(5.41)
where the derivative is taken with revenue constant.

Although the tax rule in (5.41) only provides an implicit expression for t, it
can be used to assess the eects of various parametric changes. If it is assumed
that z() z is positive and dz
dt |R const negative, then an increase in the disincentive eect of taxation (a rise in dz
dt |R const ) would reduce the optimal tax rate.
A reduction in the optimal tax would also occur, with a decreasing function
of s and z an increasing function of s, if the welfare weights were increased on
the high-s households so that equity was given less weight. The eect of other
variations can be addressed by similar reasoning.
From the first-order conditions it is also possible to provide general results
on the sign of the tax rates. The simplest result of this form is due to Romer
(1976) and Sheshinski (1972b) and is stated as Theorem 5.2.
Theorem 20 (Sheshinski/Romer) If (i)
t > 0 and > 0.
< 0, (ii)
0 and R = 0, then
Proof. It is clear form (5.34) and (5.33) that > 0 and that > 0 (or else
x, ` = 0 for all s). Now assume that > 1. Then (5.34) and (i) imply that
Z
[ ] (s) ds 0.
(5.42)
0
Now since z is an increasing function of s in order that the second-order condition for utility maximisation is satisfied and is a decreasing function
Z
[ ] z (s) (s) ds 0.
(5.43)
0
From (5.35) and (ii) however, inequality (5.43) cannot be satisfied if 1.

Hence < 1 which implies that the marginal tax rate t > 0. The positivity of t
then follows from the zero revenue condition.
144
The assumptions of this theorem have been weakened by Hellwig (1986) to

those of agent monotonicity and non-inferiority of leisure.
The linear model has also been used to find upper bounds upon the optimal
tax rate. Hellwig (1986) and Svensson and Weibull (1986) show that the optimal
rate is bounded above by the optimal tax under a maxi-min social welfare function and this rate is bounded by the revenue-maximising tax rate. Svensson and
Weibull (1986) also show that the upper bound given by the revenue-maximising
tax rate holds under a Pareto-ordering of states alone. Finally, Helpman and
Sadka (1978) consider some comparative statics eects of changes in parameters. An increase in the concavity of W () is shown to raise the marginal rate
of tax (as suggested after (5.41)). In contrast, the eect of a mean-preserving
increase in the spread of the skill distribution cannot be unambiguously signed.
5.3.3
Nonlinear taxation
With nonlinear taxation the self-selection constraint must be taken fully into
account. As shown in 2.4 this consists of (5.14) (the first-order condition) and
z 0 (s) 0 (the second-order condition). Most analyses of optimal income taxation, such as Mirrlees (1971) and Seade (1977), have incorporated only the
first-order condition and adopt what has become known as the first-order approach. As shown in Mirrlees (1986) this may lead to incorrect results. In fact,
Ebert (1992) presents an example in which the solution with only the first-order
condition included leads to a consumption function with an interval over which
it is decreasing. In contrast Brito and Oakland (1977) and Ebert (1992) impose
both the first- and second-order conditions as constraints upon the maximisation and so avoid these diculties. Given these observations, the analysis below
will adopt the second-order approach and impose both the first-order condition
and the constraint z 0 (s) 0 upon the optimisation.
Formal optimisation
The maximisation problem has been analysed in two distinct ways. The analysis
of Seade (1977) formulated the maximisation in terms of the calculus of variations and employed as choice variables the functions x(s) and `(s). In contrast,
Mirrlees (1971) and Ebert (1992) employ a formulation based on Pontryagins
maximum principle. Naturally, there are slight dierences between the two, but
the final characterisation is the same. The optimal structure of income taxation is characterised here by applying Pontryagins maximum principle, a good
introduction to which is given in Intriligator (1971). In this framework, the
level of utility, u(s), pre-tax income z (s), and the tax payments of households
of ability s or less, given by
Z s
R (s) =
[z (s0 ) x (s0 )] (s0 ) ds0 ,
(5.44)
s1
are taken as state variables and the derivative of gross income, (s) dz
ds , is
taken as the control variable. The level of consumption can then be found by
145
solving u (s) = u (x (s) , z (s) , s).

Adopting a utilitarian objective, the control variable is chosen to maximise
Z s2
u (s) (s) ds,
(5.45)
s1
subject to
dR
= [z (s) x (s)] (s) ,
ds
R (s1 ) = R (s2 ) = 0,
du
= us (x (s) , z (s) , s) ,
ds
dz
= (s) ,
ds

dz
= ( (s)) 0.
ds
(5.46)
(5.47)
(5.48)
(5.49)
(5.50)
The revenue constraint is captured by (5.46) and (5.47). To simplify, it is

assumed that zero revenue is to be collected and, from (5.44), this is reflected in
the upper end-point condition in (5.47). The rate of change in revenue, (5.46),
is derived directly from (5.44). The self-selection constraint is represented by
(5.48) - (5.50). The first-order condition (5.14) is written as (5.48) and the
second-order condition is included as (5.49) and (5.50). Condition (5.49) defines
the rate of change of the state variable . The interpretation of (5.50) is that the
straightforward inclusion of the second-order condition (s) 0 would lead to
singularity when combined with (5.49). The second-order condition is therefore
transformed to (5.50) where the dierentiable function ( (s)) satisfies (0) = 0
and 0 () > 0.
Introducing the adjoint variables (s), (s), (s) and (s), the Hamiltonian
for the optimisation is
H = u (s) (s)+ (s) [z (s) x (s)] (s)+ (s) us (z (s) , x (s) , s)+ (s) (s)+ (s) (s) ,
(5.51)
and the necessary conditions are
dH
= + 0 () = 0,
d
[[z (s) x (s)] (s)]
us (z (s) , x (s) , s)
dH
=
+
= 0 ,
dz
z
z
[[z (s) x (s)] (s)]
us (z (s) , x (s) , s)
dH
= (s) +
+
= 0 ,
du
u
zu
dH
= 0 = 0,
dR
dz
= 0, 0,
ds
(5.52)
(5.53)
(5.54)
(5.55)
(5.56)
146
with transversality conditions

(S1 ) = (S2 ) = 0, (S1 ) = (S2 ) = 0
(5.57)
To derive the form of these conditions that will be used below, note that
from the identity u (s) = u (z (s) , x (s) , s) it follows that
and
x
ux
= ,
=
z
uz
(5.58)
x
1
.
=
u
ux
(5.59)
us (x (s) , z (s) , s)
x
= usx
+ uz = ux ,
z
z
(5.60)
In addition,
and
usx
x
us (x (s) , z (s) , s)
.
(5.61)
= usx
=
u
u
ux
Now denoting (s) = (s) 0 ( (s)), (5.52) - (5.57) can be given the alternative
formulation
ux s + [1 ] 0 = 0,
(5.62)
usx
+ 1
= 0,
(5.63)
0 +
ux
ux
dz
= 0, 0,
(5.64)
ds
(5.65)
(S1 ) = (S2 ) = 0, (S1 ) = (S2 ) = 0,
where the second part of (5.65) follows from (5.52).
The interpretation of these necessary conditions is best undertaken by considering separately the case in which is zero for all s and that in which it is
positive for some s. In the first case, since is zero the second-order condition
for the satisfaction of the self-selection constraint is not binding and pre-tax
income is a strictly increasing function of ability. When this is true, the results derived from the first-order approach of Mirrlees (1971) and Seade (1977)
are identical to those of the second-order approach. In particular, (5.64) and
the second part of (5.65) become irrelevant, and 0 is eliminated from (5.62).
The two remaining equations are then the standard conditions of the first-order
approach describing the optimal tax function.
Now consider the case where there are some values of s for which is not
zero. For such s, z 0 = 0. Hence if is positive over the interval [s0 , s1 ], all
households with abilities falling in that interval earn the same pre-tax income.
These households are therefore bunched at a single income level. Furthermore,
because of (5.15), they must also have the same level of consumption. A point
where bunching occurs can be identified by a kink in the consumption function.
It is also interesting to note that although pre-tax income and consumption are
identical, utility is increasing with s over the bunched households since those
with higher s have to undertake less work to obtain the common income level.
147
Characterisation of optimal tax function

The first theorem on the structure of the optimal tax function is concerned with
demonstrating that if any households are unemployed at the optimum, then it
will be the lowest ability households.
Theorem 21 (Mirrlees) If there exists an ability level s0 S such that ` (s0 ) =
0, then ` (s) = 0 for any s < s0 .
Proof. Assume ` (s0 ) = 0 and that there is some s, with s < s0 , such that
` (s) > 0. It follows that

s
s
U (c (s` (s)) , ` (s)) < U c s0 ` (s) , ` (s) ,
(5.66)
s0
s0
since ss0 < 1 and therefore the right-hand side of (5.66) represents the utility
derived with less labour supply but retaining the same level of consumption. In
addition,

s
s
U c s0 ` (s) , ` (s) < U (c (s0 ` (s0 )) , ` (s0 )) ,
(5.67)
s0
s0
since ` (s0 ) is utility maximising for the household of ability s0 . Hence if ` (s) >
0, combining these inequalities gives
u (s) < u (s0 ) .
(5.68)
However, if ` (s) = 0, it is clear that

u (s) = u (s0 ) .
(5.69)
Therefore ` (s) > 0 is not utility maximising and ` (s) = 0 for s s0 .

Theorem 5.3 is an interesting result since it shows, even without specifying
the tax function, that the optimal tax system may generate unemployment in
the sense that it results in low ability households choosing to do no work. As
these households are productive whenever their ability level is non-zero, and
output would increase if they did work, this carries important implications for
the relation between optimal taxes and the achievement of maximum potential
output.
The second result demonstrates that a marginal tax rate in excess of 100%
will never be optimal.
Theorem 22 (Mirrlees ) The consumption function c(z) will be increasing in
z.
Proof. Self-selection requires that ux x0 + uz z 0 = 0 and z 0 0, therefore
x0 0. By definition, x (s) = c (z (s)) so that x0 = c0 z 0 . Hence c0 0. Since
c(z) = z T (z), the fact that c0 0 implies that c0 = 1 T 0 0 T 0 1, so
the marginal tax rate is always less than or equal to 1.
148
Figure 5.5: A Tax Rate Greater than 100%

To illustrate the reasoning behind Theorem 5.4, consider a consumption
function that has a section with gradient less than zero, that is a marginal tax
rate greater than 100%. Such a consumption function is shown in Figure 5.5
with the decreasing section A B. Because the indierence curves represent
convex preferences, no household will locate between A and C, although with
Leontief indierence curves there may be some households indierent between
A and C, and the consumption function with the flat section joining A and
C produces an identical final allocation. Therefore downward sloping sections
are redundant and there is no gain from a consumption function that has a
decreasing section.
The next theorem proves that the marginal rate of tax must always be
non-negative. The theorem was first proved for an additively separable utility function by Mirrlees (1971) and for a linear tax system by Sadka (1976).
The generalisation of the theorem to nonlinear taxes and non-separable utility
was given in Seade (1982). The version of the theorem proved here relies upon
strong assumptions; in particular it is assumed that there is no bunching. An
extension of the theorem to take account of the possibility of bunching is contained in Ebert (1992). The theorem can also be extended to prove that the
marginal tax rate must be strictly positive except for the households with the
highest and lowest abilities, again see Ebert (1992).
To develop the theorem it is first necessary to note that when there is no
bunching 0 = 0 so that (5.62) can be written as
1=
ux s
.
(5.70)
Under the adopted normalisation, 1 (s) is the marginal tax rate facing the
household of ability s: the pre-tax price of consumption relative to income is 1
and (s), being the post-tax marginal rate of substitution, is equal to post-tax
relative prices. Given that will be proved to be positive below, the sign of
the marginal rate of tax is then the opposite of that of . To establish that the
149
tax is non-negative, it is therefore necessary to establish that is non-positive.

This is the basis of the proof of the following theorem.
Theorem 23 (Seade) Assuming agent monotonicity, if leisure is not an inferior good and uzx 0 then the marginal tax rate is positive.
Proof. It is first shown that non-inferiority of leisure and uzx 0 imply
that ux is a decreasing function of s. This follows since
ux uxz
dux
(5.71)
+ uxs ,
= uxx x0 + uxz z 0 + uxs = x0 uxx
ds
uz
where the second equality follows from the first-order condition (5.15). Noninferiority of leisure implies uxx uxuuzxz 0 and, since uxs = zusxz , the
x
assumption that uzx 0 (and uxx < 0) result in du
ds 0.
Solving the dierential equation (5.63) gives
"Z 0
#
Z S2
s
usx 00
0
(s) =
ds ds0 .
(5.72)
1
(s ) exp
u
u
x
x
s
s
Since the transversality conditions require (S1 ) = (S2 ) = 0 and ux > 0, it
follows that > 0. In
(5.55) implies that is a constant. Combining
h addition,
i
these observations, 1 ux cannot be always positive or always negative. In

fact, it must be negative for low values of s and positive for high values. This
implies that (s) is decreasing for s less than some s and increasing for s greater
than s. Since (0) = 0, (s) is then seen to be non-positive for all s, which
completes the proof when combined with (5.70).
The theorems stated to this point have been concerned with the general
structure of the tax function, rather than with particular properties. One particularly relevant property of an income tax function is progressivity. Although
there are several possible definitions of progressivity, see the discussion in Lambert (1989), the one that is adopted here is to say the tax system is progressive if
it has a marginal rate that increases with income. This requirement is equivalent
to the condition
T 00 0.
(5.73)
Progressivity is a feature that almost all actual income tax systems possess, an observation that illustrates the importance of investigating whether
this property is always optimal.
The following theorem has obvious implications for the issue of progressivity
of the tax function. It is proved both by a direct argument and by appeal to
the necessary condition in (5.70).
Theorem 24 (Sadka/Seade) Let the upper bound on ability, S2 , be finite. Then
the marginal rate of tax must be 0 for a household of ability S2 .
150
Figure 5.6: Zero Rate at the Top

Proof.
Let ABC in Figure 5.6 be the initial graph of c(z) = z T (z) and HBE be
an indierence curve of a household with ability S2 . Since it is assumed that
T 0 > 0 for ABC, it follows c0 < 1, hence the gradient of HBE is < 1 at B. Now
define a new tax T1 as follows
T1 (z) = T (z) for z z(S2 , T ),
(5.74)
T1 (z) = T (z(S2 , T )) all z z(S2 , T ).
(5.75)
and
Under the new tax T1 , the tax payment is held constant above income level
z(S2 , T ).
The graph of c1 (z) = z T1 (z) is then ABD where the section BD has a
zero marginal rate. A household of ability S2 will move to K and is evidently
better o. Revenue has also not changed. Therefore this must represent an
improvement in welfare and T 0 (S2 ) > 0 cannot be optimal.
The alternative proof is to note that the upper bound on ability implies that
(S2 ) > 0 and the transversality condition implies (S2 ) = 0. Therefore, from
(5.62),
0
1=
.
(5.76)
However, (5.64) and (5.65) imply 0 is non-positive at S2 , giving 1 0 at

S2 . Combining this inequality with Theorem 5.5, it follows that 1 (S2 ) = 0
and the theorem is proved.
For this argument to work in the case of the infinite support, where it is
natural to assume lims (s) = 0, further restrictions are necessary in order
to evaluate the limiting tax rate. Details of these can be found in Seade (1977).
Theorem 5.6 is purely local in the sense that it relates only to the household
of highest ability and, by continuity, to the households of similar ability. The
151
Figure 5.7: Initial Schedule
Figure 5.8: Pareto-Improving Schedule

next result of Seades is concerned with extending the improvement further down
the ability scale.
Theorem 25 (Seade) For a population with bounded ability, any income tax
schedule with a positive marginal rate at the top of the scale can be replaced by
one that leaves all households better o, inducing them to earn more income but
paying the same tax.
Proof. Start with the initial schedule in Figure 5.7 and the optimal
choices, illustrated for households of three dierent abilities, given by the tangency points a, b and c.
It is first noted, as before, that any movement along a 45o line does not aect
tax revenue since T 0 = 0 and no net payments are made out of the extra income.
Further, any relocation locally up the relevant line 45o for each household will
improve welfare. Hence the tax system b
drawn in Figure 5.8 improves welfare.
The tax function b
is constructed by first selecting b
a for the household of
highest ability and then moving the tax function around their indierence curve
and attempting to cross the 45o line through b (the location of the household
152
Figure 5.9: Final Schedule

with second highest ability) at the highest point, then continuing downwards.
This will obviously lead to a general improvement in welfare.
Such improvements can be continued until the schedule in the Figure 5.9
is reached.
Point a obviously represents the maximum increase in utility possible without altering the level of the tax payment. The resulting schedule is then constructed by working left from a .
The previous two results have been concerned with the shape of the tax
schedule primarily at the top end of the ability scale. The next result of this
section is due to Seade (1977) and determines the tax rate facing the lowest ability household under the assumption that there is no bunching. This assumption
guarantees that the bottom of the tax function will only apply to the lowest
ability household and that they will be at an interior solution rather than at a
corner
Theorem 26 (Seade) If there is no bunching at the lowest income, the optimal
marginal rate for the household of lowest ability is zero.
Proof. From the no-bunching condition the lowest tax rate only applies to
the person of lowest ability. Applying the transversality condition (5.65) and
(5.70) then gives the result.
Combining Theorems 5.6 and 5.8 it is clear that, when there is no bunching
at the lowest income, the tax function cannot be progressive throughout since
it must begin with a zero rate and end with a zero rate. When bunching does
occur at the lowest income, the outcome is dierent. This is summarised in
Theorem 5.9.
Theorem 27 (Ebert) If there is bunching at the lowest income, the marginal
rate of tax is strictly positive at the end of the bunching interval.
Proof. From (5.62) and (5.65) at s = S1 ,
[1 (S1 )] (S1 ) = (S1 ) ux s + 0 (S1 ) = 0.
(5.77)
153
With bunching at the lowest income, x (s) and z (s) are constant over some
interval. Hence over this interval
d [1 (s)]
= x x0 z z 0 s > 0,
ds
(5.78)
since x0 = z 0 = 0. Therefore 1 (s) must be positive at the end of this

interval.
The same argument can be used to prove that there is no bunching at the
highest income level.
Theorem 28 (Ebert) There is no bunching at the highest income level.
Proof. Theorem 5.8 has already shown that the marginal tax rate on the
highest income is zero. Over the interval of bunching, 1 (s) must increase as
s decreases and therefore becomes negative for some s < S2 . This contradicts
Theorem 5.6.
5.3.4
Summary
The major theoretical results for the analysis of the optimal nonlinear tax have
now been derived. These show that the optimal marginal rate of taxation must
lie between zero and one. At the highest and lowest abilities, the tax rate
must be zero. The latter finding shows that the optimal tax function cannot be
progressive; a result that runs counter to observed income tax functions. It may
also be optimal to force some households to choose to undertake no labour. If
this is the case, it is the lowest ability households that will not work. Finally,
pre-tax income and consumption must both be increasing functions of ability.
To obtain further details of the structure of optimal taxes, it is necessary to
consider numerical analyses.
5.4
Numerical results
The standard analysis of optimal income taxation has been introduced above
and a number of results have been derived that provide some characterisation of
the shape of the tax schedule. It has been seen that the marginal rate is between
0 and 1 but as yet no idea has been developed, except for the endpoints, of
how close it should be to either. Similarly, although equity considerations are
expected to raise the marginal rate, this has not been demonstrated formally nor
has consideration been given to how eciency criteria, particularly the eect
of taxation upon labour supply, aects the choice of tax schedule. Due to the
analytical complexity of the nonlinear model, these questions are best addressed
via numerical analysis.
154
5.4.1
Simulations
To generate numerical results, Mirrlees (1971) assumed that the social welfare
function took the form
R 1 U
(s) ds, > 0,
e
W = 0R
(5.79)
U
(s)
ds, = 0.
0
The form of (5.79) permits parametric variations of the form of the social welfare
function by changes in . Higher values of represent greater concern for equity,
with = 0 representing the utilitarian case. The individual utility function was
assumed to be Cobb-Douglas
U = log x + log [1 `] ,
(5.80)
and the skill distribution log-normal

"
#
1
[log (s + 1)]2
(s) = exp
,
s
2
(5.81)
with a standard deviation, , of 0.39. This value of the standard deviation was
derived from data on the distribution of income given in Lydall (1968). There is
thus an implicit assumption that the skill distribution can be inferred directly
form an observed income distribution. Furthermore, this assumption implies
that the skill distribution is unbounded. A selection of the numerical results of
Mirrlees (1971) are given in Table 5.1.
Income
(a) z G = 0.013
0
0.055
0.10
0.20
0.30
0.40
0.50
(b) z G = 0.003
0
0.05
0.10
0.20
0.30
0.40
0.50
Consumption
=0
0.03
0.07
0.10
0.18
0.26
0.34
0.43
Average Tax (%)
Marginal Tax (%)
-34
-5
9
13
14
15
23
26
24
21
19
18
16
=1
0.05
0.08
0.12
0.19
0.26
0.34
0.41
Table 5.1: Optimal
-66
-34
7
13
16
17
tax schedule
30
34
32
28
25
22
20
155
The most important feature of the results in Tables 5.1a and b are the
generally low marginal rates of tax, with the maximal rate being only 34%.
There is also limited deviation in these rates. Relating the results to Theorem
5.6, the marginal rates do become lower at high incomes but do not reach zero
because the skill distribution is unbounded. The average rate of tax is negative
for low incomes in both Tables 5.1a and b so that low income consumers are
receiving an income supplement from the government.
The set of results reported in Table 5.2, also from Mirrlees (1971), show the
eecting of increasing the dispersion of skills. This raises the marginal tax rates
but these remain fairly constant across the income range. This occurs despite
the greater inequality of skills leads to a greater possible role for redistribution
via the income tax. Further results on the eect of varying the dispersion of
skills are given in Kanbur and Tuomala (1994). These support the finding that
an increased dispersion of skills raises the marginal tax rate at each income
level but show that if also has the eect of moving the maximum tax rate up
the income range, so that the marginal tax rate is increasing over the majority
of households.
Income Consumption Average Tax (%) Marginal Tax (%)
0
0.10
50
0.10
0.15
-50
58
0.25
0.20
20
60
0.50
0.30
40
59
1.00
0.52
48
57
1.50
0.73
51
54
2.00
0.97
51
52
3.00
1.47
51
49
Table 5.2: Optimal tax schedule: increased dispersion of skills (z G = 0.013,
n = 1, = 1)
The nature of the tax rate at the upper end of the income scale and the
implications of the zero-endpoint result for nearby incomes has been investigated
by Tuomala (1990). Tuomalas numerical results show that the marginal tax
rate may be far from zero on incomes close to the maximum. The zero-endpoint
result is therefore only a local conclusion and does not necessarily imply the
incomes near the maximum must also be subject to low marginal tax rates.
Although the numerical simulations of Mirrlees (1971) made a number of
variations in specification, these were very restricted in comparison to the potential range of formulations. In attempting to progress further, Atkinson (1972)
considered the eect of changing the social welfare function to the extreme maximin form
W = min{U },
(5.82)
which places the greatest possible emphasis on equity considerations. The reasoning for doing this follows from contrasting Tables 1a and b. From these it
can be seen that increased concern for equity, going from 0 to 1, increased
the optimal marginal tax rates. The natural question would be: could strong
156
equity considerations lead to high marginal rates? The interest in this issue
can be appreciated by recalling that the top British tax rate in the 1970s was
approximately 70% - 98%, dependent on the precise source of income. From
this perspective the rates derived by Mirrlees were relatively low.
The eect of changing to a maxi-min social welfare function can be seen from
considering Table 5.3 which is based on Atkinson and Stiglitz (1980). From the
table it is possible to conclude that the maxi-min criterion leads to generally
higher rates. However they are again highest at low incomes and then decline.
In addition, they remain relatively low.
Utilitarian
Maxi-min
Level of s
Average rate Marginal rate Average Rate
Median
6
21
10
Top decile
14
20
28
Top percentile
16
17
28
Table 5.3: Contrast of utilitarian and maxi-min
5.4.2
Marginal Rate
53
34
26
Choice of specification
The numerical results discussed above have concentrated upon examples with
the same log-normal distribution of skill and Cobb-Douglas individual utility
function. Following the analysis of Stern (1976), which is devoted to considering
whether these assumptions are appropriate and the consequences of modifying
them, each of these assumptions is now discussed in turn.
Estimation of skill distribution
The first point to consider is the determination of the skill distribution. It is
natural to assume that this cannot be directly observed and must be inferred in
some way from observable data. The distribution employed by Mirrlees (1971)
was taken directly from an analysis of income data and therefore embodies the
assumption that skill and income have the same distribution. The question that
must be asked, considering the importance of the distribution demonstrated in
Table 5.3, is whether this assumption is justified.
Stern (1976) makes the point that it is generally not possible to pass from
knowledge of the observed income distribution to knowledge of the underlying
skill distribution unless there is full knowledge of the utility function and the tax
function. This point is illustrated by the utility function U (x, `) = 1` if x x,
= otherwise. If is unobserved, all that will be observed is that all consumers
achieve equal incomes and have consumption level x. However, consumers of
high ability use little labour time to obtain this consumption level but low
ability consumers require a considerable amount. From this example, it can be
seen that the income distribution need not directly reflect the skill distribution.
Whether the income distribution accurately reflects skills is dependent upon the
other components of the economy and is not an exogenous property.
157
Labour supply
The choice of optimal income tax is concerned with maximising social welfare
by reaching an ecient trade-o of equity against eciency. The factor that is
most intimately linked with the eciency cost of taxation is the wage elasticity of
labour supply or, alternatively, whether labour supply will be greatly aected by
high marginal rates of taxation. This subsection briefly reviews some evidence
on the eect of taxation on labour supply and then considers the implications
of alternative formulations of the labour supply function.
Empirical evidence on the eect of income taxes can be found in both the
results of surveys and of econometric estimates of labour supply functions. Break
(1957) conducted a survey of the disincentive eect of high tax rates upon
solicitors and accountants in the UK, 63% of whom where subject to marginal
tax rates above 50%. The survey concluded that as many respondents were
working harder because of the tax rates as were working less hard. A similar
conclusion was obtained by Brown and Levin (1974) in a survey of the eect
of income taxation on the level of overtime worked by a sample of weekly paid
workers; little net eect of taxation on working hours was found.
Econometric evidence has also produced comparable results. Burtless and
Hausman (1978) employ data from the Gary Negative Income Tax experiment
to estimate a labour supply function and find a wage elasticity of labour supply
of 0. These results, which relate primarily to the labour supply of males, suggest
that there are grounds for believing the disincentive eect of income taxes for
males to not be great. In contrast, the elasticity of labour supply from married
females may be much higher and for this group the participation eect is also
relevant. This suggests that the analysis should really treat the two groups
separately. Further discussion of the relation of labour supply to taxation is
given in Hausman (1985) and the evidence for the U.K. on female supply is
surveyed in Blundell (1992).
With respect to the response of labour supply to taxation, the specification
adopted for the simulations generating Tables 5.1 - 5.3 was very restrictive due
to its imposed unit elasticity of substitution between leisure and consumption.
As will now be shown, alternative formulations can produce somewhat dierent
results.
First, consider the Cobb-Douglas utility as used by Mirrlees (1971) and
Atkinson (1972). The elasticity of substitution, , between leisure, `, and consumption, x, is defined as
U
x
U
[1`]
1`
x
1`
x
U
x
U
[1`]
(5.83)
This elasticity is a unit-free measure of the rate at which consumption can

be exchanged for leisure while keeping utility constant. For the Cobb-Douglas
function (5.80), = 1. Hence the simulations of Atkinson (1972) and Mirrlees
(1971) were based on a constant elasticity of substitution equal to 1. Given the
158
restrictiveness of this formulation, it is natural to question its consequences and

whether it is justified by data on labour supply.
Stern (1976) investigated the more general form of preferences described by
the constant elasticity of substitution (C.E.S.) utility function
h
i 1
U = [L `] + [1 ] x
,
(5.84)
where L is full leisure time available. L was fixed at unity in (5.80). For (5.84)
the elasticity of substitution is given by
=
1
.
+1
(5.85)
To understand the consequences that dierent values of may have, consider a consumer maximising the utility function (5.84) subject to the budget
constraint
x = A + s`,
(5.86)
where s is the net wage and A is the level of lump-sum grant, which may be
negative. The first-order condition for utility maximisation is
+1
L`
=
,
(5.87)
A + s`
[1 ] `
which gives an implicit expression for `. From dierentiating (5.87) it can be
found that
[A s`] [L `]
d`
=
.
(5.88)
ds
s [ + 1] [A + s`]
Now from (5.85) it follows that = 1 1 so, for positive, it is possible that A
d`
, may become negative and thus the labour supply curve bends
s`, and hence ds
backwards. The importance of a backward bending labour supply function is
that tax increases at high incomes will actually increase labour supply; hence
the equity and eciency factors are not directly competing. For low values of
it is therefore to be expected that the optimal marginal rate of tax will be higher.
This possibility could not arise in the specification based on the Cobb-Douglas
utility function model with = 0.
Applying the model based on (5.84) to the data of Ashenfelter and Heckman
(1973) from an analysis of 3,203 American males, Stern (1976) calculated the
elasticity of substitution and found that = 0.408. This value of is substantially less than that used by Mirrlees in the numerical simulations.
Adopting (5.84) as the individual utility function and a social welfare function of the form
Z
1
W =
[U (x, `)] (s) ds,
(5.89)
0
Stern (1976) presented estimates of the optimal linear income tax. Retaining
the log-normal distribution of ability with the standard deviation, the general
pattern of results is summarised by the Figure 5.10.
5.5. EXTENSIONS AND OMISSIONS
159
Figure 5.10: Optimal Tax Rate and Equity

From Figure 5.10 it can be seen that the Mirrlees specification = 1, = 1
gives the lowest possible rates of taxation and that tax rates increase to 100% as
tends to zero; a proof that this always occurs is given by Stern (1976). What
is important to note is that the tax rates are high for = 0.0408 and = 1.
Therefore high tax rates can be justified without necessarily appealing to the
maxi-min criterion.
5.5
Extensions and omissions
The basic Mirrlees economy described above has been extended in a number of
directions and two of these are now considered. The first extension introduces a
second form of labour service which allows the income tax to have indirect distributional eects via the changes in relative wages. Following this, the nonlinear
income tax is combined with linear taxes upon commodities and conditions are
found for which commodity taxes are unnecessary. The section is completed by
noting some relevant issues that are not addressed in the formal analysis.
Two forms of labour
The relevance of introducing a second form of labour service is that the economy
can be designed so that the population is partitioned with those of low skill
supplying a form of labour with a low wage and those of high skill supplying
labour with a higher return. To obtain such a partition it is only necessary to
assume that there are two distinct levels of skill: high and low. The new factor
introduced by the existence of two wage levels is that the income tax can alter
the relative values of these wages and, in doing so, alters the distribution of
income between the two skill groups. This second route for redistribution will
clearly be important in the determination of the optimal income tax.
In the first analysis of a model of this form Feldstein (1973) employed numerical techniques to investigate the eect of the relative wage variation upon the
value of the optimal linear income tax. With the two forms of labour entering
160
into a Cobb-Douglas production function, Feldstein concluded that there was

little dierence between the optimal tax with fixed wages and that with variable
wages.
The conclusion of Feldstein was reconsidered in Allen (1982) employing an
analytical, rather than a numerical, approach. The significant extension made
by Allen was to use a more general form of production technology. This demonstrated that in the Cobb-Douglas case the redistribution via the income tax
was reinforced by the adjustment of relative wages thus explaining the Feldstein
(1973) results. However, if the elasticity of substitution between the two forms
of labour is low and the labour service earning the higher wage has a negative
supply elasticity, then the relative wage eect operates in the opposite direction
of the income tax eect and may outweigh it. When the indirect eect does outweigh the direct eect, the optimal policy becomes a combination of lump-sum
tax and a negative marginal rate of income tax.
These results indicate that the analysis of income taxation becomes rather
more dicult and can generate surprising conclusions when more than one form
of labour service is introduced into the model. This assumption of the standard
economy may therefore be more restrictive than it at first appears.
5.5.1
Income and commodity taxes
The analysis of a combined tax system of linear commodity taxes and a nonlinear
income tax has been studied in a number of papers, most notably Atkinson and
Stiglitz (1976), Mirrlees (1976) and Revesz (1986). The major results of this
line of study are now briefly described.
It is assumed that there are n goods available with labour denoted as good
1 and the wage rate is w. As a normalisation rule the linear tax on good n is
set at zero. With these conventions, the budget constraint facing a consumer of
ability s takes the form
n
X
i=2
qi xi = swx1 T (swx1 ) .
(5.90)
To simplify the derivation, the production technology is taken to be linear so

that production possibilities are constrained by the relation
n Z
X
i=2
xi (s) (s) ds
swx1 (s) (s) ds z G .
(5.91)
With the linear technology, it is possible to take the producer price of each good
2, ..., n to be 1.
The optimal taxes can be found by treating U (s) as the state variable and
xi (s), i = 1, .., n 1 as the control variables, with xn (s) determined from the
identity U (s) = U (x1 (s), ..., xn (s)). The first-order condition for self-selection is
derived from (5.14) by using the fact that us = Us`2z = Us` ` or, in the present
5.5. EXTENSIONS AND OMISSIONS

U
161
notation, us = xs1 1 . Employing the first-order approach, the Hamiltonian

for the maximisation can be written using (5.91) as
"
"
##
n
X
Ux x1
H = U + swx1
xi (s) 1 .
(5.92)
s
i=2
For the choice of xk (s), k = 2, ..., n 1, using the fact that
Ux
xn
= k,
xk
Uxn
(5.93)
the necessary condition for optimality is
Ux
x1
Ux
1 k
Ux1 xk Ux1 xn k = 0, k = 2, ..., n.
Uxn
s
Uxn
(5.94)
From the necessary conditions for household utility maximisation

Uxk
1 + tk
=
.
Uxn
1
(5.95)
Substituting (5.95) into (5.94), the optimality condition (5.94) can be written
after some rearrangement as
i
h
Uxk
x1 Uxk d log Uxn
, k = 2, ..., n 1.
(5.96)
tk =
s
dx1
d log
hU
xk
Uxn
The result in (5.96) reveals two facts. Firstly, if

= 0, for all k =
dx1
2, ..., n1, which holds if the utility function is weakly separable between labour
and all other commodities, then tk = 0 for all k = 2, .., n 1. This is the major
result of Atkinson and Stiglitz (1976). In these circumstance the commodity
taxes are unnecessary and the income tax is sucient for achieving welfare aims.
This result derives from the tax system attempting to tax the innate ability of
the households but, when the separability holds, there is insucient correlation
between consumption choice and ability for commodity taxation to have any
eect. The second consequence of (5.96) is, holding all other variables constant,
that the tax rate on a good should be positively related to the rate of change
of the marginal rate of substitution between that good and labour as labour
supply increases. Therefore, those goods relatively preferred by the consumers
supplying most labour should be taxed more. Using a more general framework,
Mirrlees (1976) strengthens this conclusion to show that the commodity tax
rates should be highest on those goods for which the high ability households
have the relatively strongest preference.
An alternative perspective upon the combination of income and commodity
taxes has been provided by Christiansen (1984). Christiansen takes as the starting point a situation in which the income tax has been optimised but with no
162
commodity taxes and then determines the welfare eect of introducing commodity taxes whilst holding revenue constant. The analysis provides the conclusion
that goods for which demand increases if more leisure is obtained, but with
no change in income, should have positive commodity taxes introduced. If no
change in demand follows from the change in leisure, the tax should be zero and
it should be negative if demand falls.
The results noted have determined some relations between income and commodity taxes. However, the number of alternative perspectives from which the
model can be approached and the richness of the model prevent any simple and
summary statement being given.
5.5.2
Omissions
The economy that has been studied was, by necessity, highly stylised. Although
this brings undoubted analytical benefits, it does eliminate from consideration
many issues that are of practical interest. Some of these are now briefly discussed.
The economy involved only a single form of labour service but with dierences in the ability of households to perform this service. In reality, there are
many dierent forms of labour in an economy which dier in the skills they require and in the working conditions they impose. The actual monetary payment
for the supply of labour may only be part of the package of remuneration, and
some of the return (or cost) may be entirely psychic in nature. A income tax
policy designed to maximise welfare would need to take account of the entire
package of characteristics that constitutes labour supply. The labour supply
decision also involves more than simply the determination of the number of
hours to work. As occupations dier in their characteristics, the choice between
occupations is important and this choice will be aected by income taxation.
For instance, an increase in taxation will be detrimental to occupations where
the return is predominantly monetary. There are also intertemporal aspects to
the labour supply decision such as the timing of entry to the labour force and
the timing of retirement. As an income tax will introduce a distortion into such
decisions, this increases the potential eciency loss. Some of these issues are
addressed in Christiansen (1988).
The preferences of the households have been taken as identical. This need
not be the case and dierences in preferences may arise as in Chapter 4. Furthermore, the household has been viewed as supplying an homogeneous form
of labour but in practice the total labour supply is often the sum of male and
female components. The nature of these is often very dierent and empirical
evidence suggests that they have markedly dierent responses to taxation. This
observation implies that the analysis should be based upon greater detail of the
structure of households. Furthermore, it also raises issues concerning the tax
treatment of the individuals that constitute the household such as whether they
have the option or not of being tax as separate individuals or whether a joint
household return is compulsory.
5.6. SUMMARY
5.6
163
Summary
This chapter has reviewed the formal analysis of the optimal income tax problem. The problem has been treated as one of choosing the income tax function to
balance equity and eciency considerations in an economy characterised by inequality in the distribution of income. The standard economy of Mirrlees (1971)
has been described and it has been noted that this is the simplest for which the
income tax problem has any real content. The major theoretical results have
been derived and, although they do not provide a precise characterisation of the
optimal tax schedule, they do suggest its most important properties, some of
which are in conflict with the structure of observed tax schedules.
Numerical analyses have been used to further investigate the nature of the
tax schedule and its dependence upon the assumed structure of the problem.
Results from some of the most noteworthy of these analyses have been given
above. In particular they demonstrate that the marginal rates of tax may be
rather low but do increase with concern for equity. However, the conclusions
are sensitive to the assumptions invoked by the model. The implications of the
modification of some of these assumptions was considered.
The chapter was concluded by a consideration of two extensions. The introduction of a second form of labour service could lead to a marked change in the
form of the optimal tax function if the indirect eect of taxation upon relative
wages outweighed the direct eect. The optimal combination of income and
commodity taxes was then considered and it was shown that weak separability
of leisure would make commodity taxes redundant.
The methods of analysis can be adapted to treat other forms of nonlinear
taxation. It needs only a minor revision to turn the income tax into an expenditure tax with a suitable re-interpretation of the skill variable as a preference
parameter. Many of the theorems derived then apply directly to this new setting. In fact, several of the studies cited, such as Mirrlees (1976) and Seade
(1977) are concerned with general nonlinear taxes rather than income tax per
se.
Chapter 6
Policy Reform
6.1
Introduction
The previous two chapters have considered the determination of optimal commodity and income taxes. In practice, if the derived tax rules were to be implemented it would be likely that a major upheaval of the fiscal structure would be
required. To be willing to enact such a major change would require the policy
maker to have considerable faith in the accuracy of the policy advice. Taking
this into account, many countries have opted in favour of gradual policy reforms
which involve slowly phasing in some taxes and removing others. The design of
such reforms will be the subject of this chapter.
The theoretical literature on policy reform has been concerned with characterising when there exist feasible reforms that satisfy the policy makers objectives and with determining the optimal direction of reform. For the purpose
of formal analysis, reforms are always interpreted as dierential changes in the
vector of policy instruments. This is the limiting interpretation of the reforms
being small.
This chapter will review the standard analysis that has been developed for
determining the existence of worthwhile reforms in the vector of consumer prices.
The inverse optimum problem, which calculates the welfare weights of households implied by a given set of policy parameters, will be related to this analysis
and the concept of marginal social cost will be discussed. Several applications of
these methods, including empirical investigations, will be described. In considering the practical implications of these results, it should be noted that administration costs are not considered. If such costs are significant they will reduce
the potential attractiveness of a series of small reforms.
6.2
The reform problem
The standard procedure is to take as given some initial vector of consumer

prices, supporting producer prices and implied tax rates. Given these values,
165
166
CHAPTER 6. POLICY REFORM
the equilibrium of the economy is calculated. It is clear that the values of the
policy variables will determine the welfare properties of the equilibrium. Small
changes in consumer prices are then considered and the question is asked: is
there a feasible change that raises welfare? This is the policy reform problem.
It is implicit in this description that as consumer prices change, producer prices
are adjusted to maintain equilibrium.
The welfare objective has typically been either the strict Pareto principle
under which an acceptable reform must raise all households welfare or a social welfare function in which case the policy reform must raise social welfare.
Both of these welfare criteria are considered below. An alternative objective,
a unanimous Wicksellian criteria, is studied in Weymark (1981). Feasibility is
defined in terms of the satisfaction of an aggregate production constraint or
alternatively, where permissible, a government revenue constraint. When the
production constraint is employed, a feasible change must result in a demand
vector that is in the production set. With the revenue constraint, a given level
of revenue must be collected before and after the change. Further discussion of
the structure of the reform problem can be found in Feldstein (1976a).
6.2.1
Productive feasibility
The analysis is set within a competitive economy that has H households and
n goods. The government policy variables are the n consumer prices denoted
by the vector q T = (q1 , ..., qn ), with the superscript T denoting the transpose.
All vectors are written as columns and, for vectors, the notation x > y implies
xi > yi all i, x y implies xi = yi all i and xi > yi some i, and x = y implies
xi yi all i. Producer prices of the n goods are written pT = (p1 , ..., pn ).
To guarantee a separation between the production and consumption sectors, a
100% profits tax is assumed as discussed in 4.6.
Formally, the reform problem is concerned with dierential changes in the
policy parameters from an initial position with policy vector q T = (q1 , ..., qn ) to
a new position with policy q T + dq T = (q1 + dq1 , ..., qn + dqn ). The direction of
change dq is restricted to be normal to q; hence q T dq = 0. The set of normals
to q is denoted T (q) and, until indicated otherwise, all consumer price changes
dq are assumed to belong to T (q). Since aggregate demand is homogeneous
of degree zero in consumer prices, this is simply a normalisation and does not
impose any real restriction on the set of reforms. Figure 6.1 illustrates this
procedure for a two-good economy. Given initial price vector q, T (q) is a line
through the origin orthogonal to q. The change in policy dq T (q) is shown to
move the policy vector to q + dq. The homogeneity of aggregate demand implies
that it is only the direction of the price vector that is relevant, not its length.
It can be seen in Figure 6.1 that by moving along T (q), can be made to point
in any direction in the positive orthant.
Figure 6.1: The set of policy changes
The first step is to identify when a potential reform is productively feasible.
For the present, feasibility will be defined by satisfaction of an aggregate pro-
6.2. THE REFORM PROBLEM
167
Figure 6.1:
duction constraint. The production constraint is given in implicit form by the
strictly quasi-convex and dierentiable function
F (X (q)) 0,
(6.1)
where
X (q) =
H
X
xh (q) ,
(6.2)
h=1
is the aggregate demand function. The assumption of strict quasi-convexity

ensures that at each point on the boundary of the production set there is a
unique (up to a scalar multiple) price vector that will make that point the profitmaximising choice. Dierentiability then implies, as discussed in connection
with production eciency in chapter 4 section 8, that this price vector will be
proportional to the gradient vector, F , of F (). This assumption is relaxed
in Weymark (1979b) who considers non-dierentiable production constraints.
Given producer prices p, the supply vector is denoted by Y (p). An equilibrium
is termed tight if X (q) = Y (p) and non-tight if X (q) Y (p).
Feasibility of the policy reform requires that after the change in policy the
resulting demand vector remains within the aggregate production set. It is
assumed that the economy is initially on the boundary of the production set so
the policy change must satisfy
F T Xq dq 0,
where
Xq =
X1
q1
Xn
q1
X1
qn
Xn
qn
(6.3)
(6.4)
168
Figure 6.2:
is the Jacobian of the aggregate demand function. Employing the fact that
producer prices are proportional to F , the feasibility constraint can be written
in the form
(6.5)
pT Xq dq 0.
Hence, as noted by Guesnerie (1977), feasible reforms are those which cause a
change in aggregate demand that has a non-positive value at initial producer
prices. To simplify the expressions below, the expression pT Xq will be denoted
Z T .
After a change in the consumer price vector, producer prices will also adjust
in order to maintain equilibrium and it is assumed that such adjustment is
in a direction that is normal to p. The following result of Guesnerie (1977)
proves that, when the boundary of the production set is suciently smooth,
the change in demand due to a dierential change in the consumer price vector
can be met by a supply change brought about by a dierential modification of
producer prices. The proof is based on the observation that when the boundary
of the production set is smooth, its gradient changes continuously. In turn,
the normal to the production set changes continuously around its boundary.
This implies that the producer price vectors that support two nearby points,
which are proportional to the normals at those points, will be similar. Figure
6.2a shows a smooth production set for which a small change in demand can
be met by a dierential change in the producer price vector. In Figure 6.2b
the production set has a kink at y = (y1 , y2 ) and to move production from
ye = (y1 + , y2 ) to yb = (y1 , y2 + ) requires a discrete change in producer
prices.
(a) Continuous change in normal (b) Discontinuity
Figure 6.2: Smooth and non-smooth production sets
As supply is homogeneous of degree 1 in producer prices, the Jacobian of the

supply function cannot have rank n. Smoothness of the production set therefore
6.2. THE REFORM PROBLEM
169
manifests itself in the supply function having rank n 1. Any lower rank indicates that there are flat sections or ridges in the production set, either of which
will remove the property that the normal to the production set changes continuously around the boundary of the production set. Following these preliminaries,
the formal statement of the lemma can now be given.
Lemma 29 (Guesnerie) If the Jacobian of the aggregate supply function has
rank n 1 then for any change in consumer prices such that F T Xq dq 0,
there is at least one direction of change, dp, of producer prices such that (dq, dp)
is equilibrium preserving. If F T Xq dq = 0 the change dp is unique and (dq, dp)
is tight equilibrium preserving.
Proof. Since aggregate supply is homogeneous of degree 0 in p, Yp p = 0 and
it follows from the symmetry of Yp , where Yp is the Jacobian ofaggregate supply,
that Yp p = 0 pT Yp = 0T . Now define the set V by V = u : pT u = 0 . As

pT Yp u = 0, it follows that Yp u V . Hence the linear mapping, Yp , maps from
V to V . To see that the linear mapping is also one-to-one, assume that it is
1
2
1
2
not so that
1 there
exist u V and u V such that Yp u = Yp u = u . From
2
this, Yp u u = 0. As the Jacobian of Yp is of dimension n 1, the set
W = {w : Yp w = 0} is ofdimension 1 and,
sinceYp p = 0, if w W , w = p for
some . Hence u1 u2 = p and pT u1 u2 = pT p 6= 0. However, as u1
and u2 V , pT u1 = pT u2 = 0 pT u1 u2 = 0. This contradiction shows
the mapping must be one-to-one and onto. It then follows that on V , Yp has
an inverse Yp1 . Hence given a price change dp, Yp1 dp, where dY is a vector of
supply changes normal to dp: dpT dY = 0. Thus any vector of supply changes
normal to p can be obtained by a change in producer prices which is also normal
to p.
Now let dq be such that F T Xq dq = 0. It follows that pT Xq dq pT dX = 0.
Hence dx V and there exists p = Yq1 dX. Then dY = Yp dp = Yp Yp1 dX =
dX, p is also unique since dX V . Therefore, if the consumer price change leads
to a demand change, dX, on the boundary of the production set a producer price
change can be found that matches the change in demand with an equal change,
dY , in supply and is tight equilibrium preserving.
If dq satisfies F T Xq dq < 0, then pT Xq dq pT dX < 0 and the argument
can be repeated with U = dX + , > 0 and pT U = 0 to show dY = dU > dX
so the change preserves equilibrium but not tight equilibrium.
The value of Lemma 6.1 is that it allows the focus to be placed upon consumer prices in the knowledge that any change satisfying (6.3) can be accommodated by adjustment of producer prices.
6.2.2
Improving reforms
Let V h (q1 , ..., qn ) denote the indirect utility function of the hth household. The
level of social welfare is assumed to be determined by a Bergson-Samuelson
social welfare function
W = W V 1 (q1 , ..., qn ) , ..., V H (q1 , ..., qn ) .

(6.6)
170
Using (6.6), the eect of the change in q upon welfare is given by

dW = W T dq,
where
W
H
H
X
X
W V h
W V h
,
...,
h
V q1
V h qn
h=1
h=1
(6.7)
!
(6.8)
The reform problem, in terms of finding a reform that raises social welfare,
can now be introduced.
Welfare-Improving Reform (WI)
Does there exist a vector dq of reforms such that
(i) W T dq > 0,
and
(ii) Z T dq 0?
Condition (i) requires the reform to raise welfare and (ii) that it is productively feasible.
To formulate the Pareto-improving reform problem, define the vectors
h
V
V h
hT
V
=
, ...,
, h = 1, ..., H.
(6.9)
q1
qn
It is assumed that V h 6= 0 for all h. From these vectors the matrix P is formed
by using the vector V hT as the hth row of P . This construction leads to the
following matrix
V 1T
.
(6.10)
P =
V HT
Using P , the strict Pareto-improving reform problem can be stated formally.

Strict Pareto-Improving Reform (PI)
(i) P dq > 0,
and
(ii) Z T dq 0?
Given the definition in (6.9), condition (i) is equivalent to V hT dq > 0 for
all h, so that the reform must raise the welfare of all the households. The motive
for employing the strict Pareto reform, as opposed to a weak version where the
reform must benefit at least one household while not harming any other, is that
it leads to a simpler statement of later results.
WI and PI represent the two standard policy reform problems and it should
be noted that both take the form of a system of linear inequalities. This has
important implications for the analysis of reforms. It is now possible to proceed
to a characterisation of when a solution, in terms of a non-zero vector that satisfies the relevant inequalities, exists to the problems and what can be concluded
when such a vector does not.
6.3. CHARACTERISING POSSIBILITIES
6.3
171
Characterising possibilities
The first approach to characterisation is to apply a Theorem of the Alternative.

This general class of theorems states that either a set of linear inequalities has
a solution or that there is a linear relationship between the component parts
of the inequalities. Applied to the reform problems, this implies that either
there is a solution to the reform problem or else there exists a set of weights
such that the initial point is optimal with respect to the objective function
determined by these weights. This section introduces the necessary Theorems
of the Alternative and then discusses the interpretation of these for the two
reform problems.
6.3.1
Solution via Theorems of the Alternative
The solution of the reform problem WI can be characterised by appeal to the

Tucker Theorem of the Alternative. This, and other similar theorems, are described in detail in Mangasarian (1969).
Theorem 30 (Tucker) Given a matrix A and a vector B, with the number of
columns of A equal to the dimension of B, exactly one of the following holds
(a) there exists a vector x such that Ax 0, Bx = 0,
(b) there exist vectors y1 and y2 with AT y1 + B T y2 = 0, y1 > 0, y2 0.
Proof. See Mangasarian (1969).
For WI, the matrices A and B represent W T and Z T respectively, both
are 1 by n. Condition (a) describes the case in which a feasible reform exists.
Since A W T , Ax is a scalar and the vector inequality Ax 0 implies that
this scalar is strictly positive. If no reform exists, so (a) cannot be satisfied,
the theorem asserts the existence of a pair of vectors such that (b) is satisfied.
The vectors y1 and y2 in (b) can be interpreted as the shadow variables in
the maximisation of welfare subject to the production constraint since, if no
reform exists, the initial point must be optimal. The maximisation problem
that generates (b) can be clearly seen in Theorem 1D9 in Takayama (1984) and
the solution leads to the optimal tax rules of Chapter 4. Employing (6.5) and
(6.8), the optimality condition (b) can be written as
y1
H
n
X
X
W V h
Xi
y
pi
= 0, k = 1, ..., n.
2
V h qk
qk
i=1
(6.11)
h=1
This is simply P
an alternative representation of (4.35) since the aggregate
budPn
n
i
get constraint i=1 qi Xi and the relation qi = pi + ti imply i=1 pi X
qk =
Pn
Xi
Xi +. i=1 ti qk
The solution of PI requires the use of Motzkins Theorem of the Alternative.
Theorem 31 (Motzkin) For two matrices A and B with the same number of
columns, exactly one of the following holds
(a) there exists a vector x such that Ax > 0, Bx = 0, with A 6= 0,
(b) there exist vectors y1 and y2 with AT y1 + B T y2 = 0, y1 0, y2 0.
172

In terms of the reform problem PI, A represents the matrix P and is therefore
H by n and B is again Z T which is 1 by n. If there is no solution then (b)
applies and the H-vector y1 /y2 defines the implicit welfare weights attached
to the households if the initial state is optimal. This point is returned to in
the discussion of the inverse optimum problem below. Substituting into (b) of
Theorem 6.2,
H
X
y1h
h=1
or, since
V h
qk
n
X
V h
Xi
y2
pi
= 0, k = 1, ..., n,
qk
qk
i=1
(6.12)
= h xhk and h > 0,

H
X
h xhk + y2
h=1
n
X
i=1
pi
Xi
= 0, k = 1, ..., n,
qk
(6.13)
where h = h y1h 0. This condition provides a characterisation of the Pareto

optimal allocation for the chosen weights upon the households.
In summary, application of a Theorem of the Alternative to the welfareimproving and Pareto-improving reform problems demonstrates that either the
initial position is an optimum or else an improving reform can be found. When
a reform cannot be found, part (b) of the theorem provides a characterisation of
the optimum. For the welfare-improving reform problem, the characterisation
of the optimum has been discussed in Chapter 4.
6.3.2
Geometric analysis
An alternative perspective on the existence of a solution can be provided by geometric considerations. The vector Z T defines a half-space Q (p, q), or simply
Q, with x being in Q implying that
Z T x 0.
(6.14)
In terms
of (6.5), the set Q, its interior Int Q and
frontier FrQ are defined by
Q = x : pT Xq x 0 , Int Q = x : pT Xq x < 0 and Fr Q = x : pT Xq x = 0 .

Therefore any vector in Q leads to a change in consumption which has nonpositive value at initial producer prices and that satisfies the production constraint.
Welfare-improving reform
Taking the problem WI first, the vector W T similarly defines a half-space S
in which all vectors, x, located in S are such that
W T x 0.
(6.15)
173
Figure 6.3: Welfare Improvement and Production Feasibility

That is, all the vectors in the half-space S satisfy the requirement that welfare
is not decreased by a change of prices in that direction. If there is a solution
to the reform problem
it must be a vector that lies in the intersection of Int
S = x : W T x > 0 and Q. A solution will therefore exist if Int S Q is
non-empty. In fact, only if the boundaries of S and Q are coincident can there
be no solution. This discussion is summarised in Theorem 6.3.
Theorem 32 (Guesnerie/Weymark) If Int S Q 6= then the solution set to
reform problem WI is non-empty. If the solution set is non-empty, a member
of the solution set, dq, will lead to a tight equilibrium if dq Fr Q and to a
non-tight equilibrium if dq Int Q.
Proof. Directly from the preceding discussion.
In the case that the boundaries of Q and S are coincident, so that there is
no solution, it follows that
Z T = W T ,
(6.16)
for some scalar . Equation (6.16) again represents the necessary condition for
an optimum and is an alternative way of presenting (b) of the Tucker Theorem.
Figures 6.3 illustrates the vectorsZ T and W T and their implied halfspaces for a three-good economy. With three goods, the set T (q) of price changes
normal to q is a plane and, to draw the diagram, this plane has been aligned
with the page. The vectors and are both shown as lying in T (q). For this follows
T h
T
since
is subject to the budget
hPeach household
i
hP constraint iq x = 0 and W =
H
H
h h
h T h
T
= 0 so W T (q).
h=1 x . From this q W =
h=1 q x
T
T
Recalling (6.5), Z T = pT Xq so that q T Z = q T pT Xq = [Xq q] p.
The aggregate demand function is homogeneous of degree zero in q so Xq q = 0
and hence q T Z = 0. is therefore in T (q). Hence, in a three-good economy,
W and Z are three-dimensional vectors that lie in the plane T (q) which is
coincident with the page.
Figure 6.4a depicts the existence of a welfare-improving reform. Here the
interior of the intersection of S and Q is non-empty. The solution set is shown
as a subset of T (q). In contrast, Figure 6.4b shows the arrangement of vectors
that result in the solution set being empty.
174
Figure 6.4: Welfare-Improving Reform
Figure 6.5: Pareto-Improving Set

Pareto reform
The Pareto reform problem is analysed in two stages. In the first stage it is
determined whether there exist any policy changes that will increase the welfare
of all households. If such changes exist, the second stage is to check whether
any are productively feasible. Only when changes exist that satisfy both stages
does the Pareto reform problem have a solution.
Considering the first stage of the Pareto
reform problem,
each row of P
determines an open half-space, P h (q) = x : V hT x > 0 , of reforms that
would benefit the consumer whose preferences that row represents. As it is
required that the reform must increase the utility of all households, such a
h
reform will only exist if the intersection H
h=1 P (q) of these open half-spaces
h
is non-empty. A set of vectors V , h = 1, 2, 3, that generate a non-empty
Pareto-improving set is shown in Figure 6.5a and the implied set of Paretoimproving reforms in Figure 6.5b. Figure 6.5 is drawn by again aligning T (q)
with the plane of the page and utilising the fact that V h T (q) for all h,
which follows since V h = h xh by Roys identity and q T xh = 0.
The intersection of the half-spaces corresponding to the vectors V h will be
non-empty whenever the cone generated by taking all positive linear combinations of these vectors (the rows of P ) is pointed, where a set K is a cone if k K
implies k K for all 0 and a cone is pointed if K (K) = {0}. That
the vectors generate a pointed cone can be viewed as capturing the fact that
175
Figure 6.6: Cones

the preferences of the consumers are not be too dissimilar. The importance of
pointed cones in this context was first noted by Weymark (1979b). To illustrate
these points, the vectors in Figure 6.6a generate a pointed cone but those in
Figure 6.6b do not.
n
o
P
h
Denote by (q) the cone generated by the rows of P , where (q) = : = H
h=1 h V (q) , h 0
and define the positive polar cone to (q) by (q) = : T 0 for all (q) .
If (q) is pointed, (q) has a non-empty interior and the interior of (q), Int
h
(q) = H
h=1 P (q), is the set of strictly Pareto improving consumer price
changes. These sets are illustrated in Figure 6.5.
An alternative way of expressing this is to employ Roys identity to note that
V hT dq > 0 is equivalent to xhT dq < 0. The reform is then a strict Pareto
improvement if it reduces the cost of the initial consumption choice for all households. Denotenby (q) the cone generated byothe household demand vectors,
P
h
hence (q) = : = H
h=1 h x (q) , h 0 . Let P (q) be the negative po
lar cone of (q), P (q) = : T 0 for all (q) . P (q) is the set of
price changes that reduces the cost of all households demands and the set of
strict Pareto improving reforms will be non-empty if P (q) has a non-empty interior. This will be the case precisely when (q) is pointed. That (q) must be
pointed for the set of reforms to be non-empty again captures the idea that the
preferences, and hence demands, of the households must be similar. It should
h
be noted that (q) = (q) , (q) = P h (q) and Int P h (q) = H
h=1 P (q) .
These constructions are shown in Figure 6.7.
The second stage of the Pareto-improving reform problem is undertaken
by combining the Pareto-improving set with the half-space Q related to the
production constraint. It then follows that a feasible improving reform exists
if the cone generated by the rows of P , (q), is pointed, so that the Paretoimproving set is non-empty, and the intersection of Int (q) with the half-space
Q is non-empty. This reasoning is summarised as Theorem 6.4.
Theorem 33 (Guesnerie/Weymark) If (q) is not pointed then Int (q) =
and the solution set is empty. If (q) is pointed then Int (q) 6= and the
solution set to PI will be non-empty if Int (q) Q 6= . When the solution set
176
Figure 6.7: Representation via Demands

is non-empty, a member of the set, dq, will lead to a tight equilibrium if dq Fr
Q and to a non-tight equilibrium if dq Int Q.
Proof. The theorem follows from the preceding discussion.
To determine whether the equilibrium leads to a tight or non-tight equilibrium it is sucient to consider the location of the vector pT Xq relative to (q),
the cone generated by the demand vectors. This is summarised in the following
theorem.
Theorem 34 (Guesnerie) Assume pT Xq 6= 0 then: (a) if pT Xq (q) there
exist no Pareto improving changes; (b) if pT Xq (q) there exist Pareto imC
proving changes but none is tight equilibrium preserving; (c) if pT Xq ( (q) (q)) ,
where the superscript C denotes the complement, there exist Pareto improving
changes that are tight equilibrium preserving.
Proof. A formal proof can be found in Guesnerie (1977) but the theorem
follows directly from the definitions and is easily demonstrated by drawing the
implied diagrams, see Figure 6.8.
The proof of Theorem 6.5 is illustrated in Figure 6.8. In (a), it can be seen
that there is no improving reform. The improving reform in (b) leads to a nontight equilibrium because any point interior to (q) lies inside Q. Finally, in
C
(c) the shaded area is ( (q) (q)) and there are reforms, such as dq , that
are tight equilibrium preserving.
The importance of Theorem 6.5 is that it demonstrates that a Pareto improving reform may lead from an ecient point to a non-tight equilibrium with
the demand vector interior to the production set. Production ineciency may
therefore be introduced as part of the reform process. This aspect of reform will
be considered further in Section 4.
6.3.3
Informational requirements
It is worth contrasting the information required to implement the Pareto improving reform with that necessary for the welfare-improving reform and for
6.4. PRODUCTIVE EFFICIENCY AND MAXIMISING REFORMS
177
Figure 6.8: Explanation of Theorem

optimal taxation. The Pareto improving reform is based upon the gradient vectors of the households but, to determine the Pareto improving set of reforms, it
is necessary only to possess knowledge of the direction of these vectors not the
length. From Roys identity, the direction can be determined by knowledge of
individual household demands; there is no requirement to observe the marginal
utility of income. Such information is, in principle, directly observable. To determine whether a change satisfies productive feasibility it is sucient to know
initial producer prices and the derivatives, or elasticities, of aggregate demand.
Although this information cannot be directly observed, it can be easily inferred
from market data. The Pareto reform analysis can therefore be applied without the need for specifying the social welfare function. An example of such an
application can be found in Ahmad and Stern (1984).
In contrast, the welfare-improving reform does require specification of the
social welfare function and the individual marginal utilities so that the social
welfare weights can be derived. One methodology for achieving this has been
described in 4.5 but, as noted there, that involved imposing significant restrictions upon the structure of the social marginal utilities of income. However, it is
worth emphasising that it is only the direction of the welfare vector that needs
to be known and not its length. The other informational requirements are as
for the Pareto reform.
6.4
Productive eciency and maximising reforms
The analysis above has identified when improving reforms exist but has left two
important questions unanswered. Firstly, if the solution set is non-empty, which
of the reforms in the solution set should be chosen? Secondly, can any insight
be obtained into when a reform will lead to an interior point of the production
set?
From the definitions of the previous section, it should be noted that a reform
will maintain eciency if leads to a tight equilibrium. Producer prices are always
chosen to decentralise a point on the boundary of the production set so that at
178
a tight equilibrium, with X (q) = Y (p), consumption is also on the boundary

and eciency occurs. When the equilibrium is non-tight, consumption is within
the production set and the equilibrium is inecient. That inecient equilibria
can arise due to the reform cannot therefore be ruled out and, in fact, appear
to be as likely as ecient outcomes.
6.4.1
The possibility of ineciency
To proceed beyond these generalities the approach of Dixit (1979) will be followed. The focus will be placed upon the welfare-improving reform problem
and the optimal reform will be derived as that from the solution set that maximises the increase in welfare. It is first necessary to restrict the set from which
solutions may be drawn since, in general, the solution set is a cone: if reform
x raises welfare by dW , then x, > 0, which is also a solution, raises welfare
by dW , hence no maximising reform exists. A procedure introduced by Dixit
(1979) is to restrict consideration to vectors dq that lie on the unit sphere. This
normalisation rule on the set of feasible reforms is now adopted in place of the
restriction used in the previous sections. The reforms must therefore satisfy
n
X
dqi2 = 1.
(6.17)
i=1
To reflect the fact that the reforms in question are dierential, the interpretation
of this normalisation is that units are such that 1 is small.
The optimal reform is identified as that reform on the surface of the unit
sphere which maximises the increase in welfare. Writing
F (q1 , ..., qn ) 0,
(6.18)
for the production constraint, where the function has been redefined to suppress
the dependence upon aggregate demand, the optimal choice of reform is the
solution to
max dW = W T dq,
(6.19)
{dq}
subject to the pair of constraints

dF = F T dq 0,
n
X
dqi2 = 1.
(6.20)
i=1
The Lagrangean for this maximisation is

T
"
L = W dq F dq + 1
n
X
i=1
dqi2
(6.21)
From (6.21), the first-order condition for the choice of the ith element of the
reform vector is
(6.22)
Wi Fi 2dqi = 0.
6.4. PRODUCTIVE EFFICIENCY AND MAXIMISING REFORMS
179
Substituting (6.22) into the constraint on the size of the reforms provides the
restriction that
2
n
X
Wi Fi
= 1.
(6.23)
2
i=1
The focus of interest here is in the optimal reform, so it is assumed a non-zero

solution exists to (6.19)-(6.20) and hence that > 0. It follows that
dqi = h
Pn
Wi Fi
j=1
[Wj Fj ]
i 12 .
(6.24)
As the denominator of (6.24) is positive, substitution from (6.24) into the production constraint shows that productive feasibility will be satisfied if
F T W F T F 0.
(6.25)
Solving (6.25) for m, the multiplier on the production constraint, gives
F T W
.
F T F
(6.26)
It is from (6.26) that production eciency can be addressed. If F T W >

0 then > 0 and, from the complementary slackness conditions, production
eciency is preserved. Conversely, F T W < 0 implies that = 0 so that the
reform will lead to a new equilibrium with productive ineciency.
The importance of this result is the fact that for many situations, as captured
by the Diamond-Mirrlees Production Eciency lemma (Theorem 4.1) and its
extension by Hahn (1973), production eciency will be desirable at the policy
optimum. It would then be expected that it would also be required along the
reform path. In contrast, these results show that this belief is not correct and
that reform may lead to ineciency until the optimum is reached.
The geometric interpretation of the discussion of (6.26) is that F T W < 0
implies, in two dimensions, that the vectors F and W make an angle of more
than 90o . The vector of reforms is chosen to make the narrowest feasible angle
with W as this will maximise the increase in welfare. If F T W < 0, the
maximising reform lies along W itself and can be defined by dq = W , where
(W )T (W ) = 1. This situation is depicted in Figure 6.9 for a two-good
economy with the change in prices is restricted to be on the unit circle.
Production eciency remains after the reform when F T W > 0. The
maximising reform is then the vector orthogonal to F , thus satisfying the
production constraint, that makes the smallest angle with W . This represents
a reform that moves the economy around the boundary of the production set.
This is illustrated in Figure 6.10.
To summarise the discussion, productive ineciency may occur after the
reform has taken place. Whether it occurs is dependent upon the divergence
between the vector of derivatives of the welfare function and those of the production constraint. When these are similar, production eciency is retained.
180
Figure 6.9: Production Ineciency After Reform
Figure 6.10: Production Eciency

These results have been extended by Guesnerie (1977) and Fogelman, Quinzii
and Guesnerie (1978) who consider policy reform as a dynamic process. At each
moment in time the process is such that the reform must be Pareto-improving so
that individual utilities must be increasing along the reform path. In addition,
the changes in demand must be capable of being matched by supply changes
and consequent modification of producer prices.
The major finding of the analysis is, that to be successful, the reform process
must be allowed to pass through points interior to the production set. If it is not,
it may terminate at a point which is not a local Pareto optimum. Despite any
ineciencies that may occur along the path, the process will always terminate
at an ecient point.
6.4.2
Interpretation
There are two interpretations of this result that are worth noting. The first
interpretation is that the production ineciency arises because the reform is
seeking the maximum increase in welfare and there is no reason why this needs
to be along the frontier. Although the initial point for the reform process and
the final point are both on the frontier, the path of greatest increase in welfare
is more likely to go through the production set than to follow its boundary.
The alternative interpretation rests on the observation that the initial point
is not derived as the outcome of a social maximisation but is, in a sense, arbitrary. As a consequence, there is no reason to expect any relationship to hold
6.5. LUMP-SUM TAXES
181
between the gradient vectors Z and W , or between Z and the gradient vectors of
individual utilities. This should be contrasted to the position at the optimum,
shown in Figure 6.4b, where the gradient vectors Z and W point in opposite
directions. Given this arbitrary relation, it is only natural that production ineciency may arise.
6.5
Lump-sum taxes
The analysis to this point has been concerned only with the eects of changes in
the consumer price vector. It is reasonable to expect that other policy variables
will be available to the government such as income taxes or methods of eecting
transfer payments and the eect of these upon the reform problem needs to be
considered.
Although optimal lump-sum taxes can be ruled out for the reasons discussed
in Chapter 2 Section 4.1, a uniform lump-sum tax or subsidy cannot be dismissed
so easily. As shown by Smith (1983), the incorporation of a uniform lump-sum
subsidy has important implications for the reform problem. To understand the
nature of these implications, it is helpful to refer back to the discussion of 4.3.
Production ineciency may arise because the gradient vectors of the indirect
utility functions of the individual households are essentially arbitrary. When
a lump-sum subsidy is introduced, the gradient vector of indirect utility with
respect to the set of policy instruments will have a positive entry for every consumer for the eect of the lump-sum subsidy. This observation ensures a degree
of similarity between the indirect utility functions of the separate households
and significantly modifies the conclusions.
Consider now the consequence of introducing a lump-sum subsidy, I, to all
households. From an initial point with I = 0, a reform in I and q will lead to a
Pareto improvement if
dI
hT
V
> 0, all h,
(6.27)
dq
or, using Roys identity and eliminating the (positive) marginal utility of income,
if
dI xhT dq > 0.
(6.28)
The change will be productively feasible if the change in demand has negative
value at initial producer prices which can be expressed as
H
X
h=1
pT xhI dI + pT Xq dq 0,
(6.29)
where xhI is the derivative of household hs demand with respect to lump-sum

income. Since I is initially zero, the additional condition dI 0 must also be
satisfied.
182
Referring back to Theorem 6.5, a reform will lead to ineciency if pT Xq

(q). From the definition of (q), and the assumption that pT Xq 6= 0, this is
equivalent to the existence of h > 0, all h, and > 0 such that
H
X
h=1
h xh pT Xq = 0.
(6.30)
An improving reform that requires ineciency will only exist if there exist h >
0, all h, 0 and 0 such that (6.30) is satisfied and
H
X
h=1
h + +
H
X
pT xhI = 0.
(6.31)
h=1
PH
Since (6.30) requires > 0, (6.31) can only be satisfied if h=1 pT xhI < 0. That
is, to obtain production ineciency it is necessary
that the aggregate income
P
T h
eect must be negative. If, as is more likely, H
h=1 p xI > 0 then production
ineciency cannot arise. Hence, if all goods are normal, the use of a lump-sum
subsidy will eliminate production ineciency along the reform path.
The reasoning behind this argument is simple. If the government budget is
balanced at the initial, ecient, point, it will be in surplus after moving to an
inecient point. The lump-sum subsidy can be used to disburse this surplus. If
demand rises in response to the subsidy, which it will when goods are normal,
a tight equilibrium will be achieved and eciency restored. This represents a
Pareto-imporvement
over the inecient position. This process will not work
P
T h
when H
p
x
<
0
since the lump-sum subsidy will then reduce the value of
I
h=1
demand further and ineciency will increase.
A final point worth noting is that the ability to use a lump-sum subsidy
ensures that the Pareto-improving set is non-empty. This follows because each
vector V hT has a positive entry for the marginal eect of an increase in lumpsum subsidy so that the cone generated by the vectors must be pointed. With a
lump-sum subsidy only the second stage of the Pareto-improving reform problem
needs to be addressed.
6.6
The inverse optimum
The motivation for the inverse optimum is derived from noting that when no
feasible improving-reform exists it follows from the Tuckers and Motzkins theorems that the initial position is optimal given the social welfare weights. This
result can have an alternative interpretation as follows. Given any initial position, it is possible to consider which set of welfare weights would make the
initial point an optimum. If these weights are calculated and they do not conform with those of the policy maker, this is equivalent to saying a reform exists.
The major work in this area is Ahmad and Stern (1984) who introduced the
inverse optimum and applied the methodology to Indian data.
6.6. THE INVERSE OPTIMUM
183
Figure 6.11: The Inverse Optimum
6.6.1
Theory
To formalise the inverse optimum, return to the welfare improving reform problem. It is convenient to adopt as a normalisation rule that the tax on good 1
is zero and that q1 = p1 = 1; hence dq1 = 0. The vector W T can then be
written as
H
! H
!
H
H
X W V h
X
X h
X
W V h
h h
T
h
, ...,
x2 , ...,
xn ,
=
W =
V h q2
V h qn
h=1
h=1
h=1
h=1
(6.32)
where the second equality in (6.32) follows from (4.37). For the remainder of
this section, 1 , ..., h are termed the social welfare weights of the households.
They remain, of course, a composition of social and private marginal utilities.
Employing (6.32), the first form of the inverse optimum problem can now
be stated.
Inverse Optimum
What must the welfare weights 1 , ..., h be in order that the equations
W T dq > 0,
(6.33)
Z T dq 0,
(6.34)
and
have only the zero vector as solution?
The interpretation of this formulation is that the s are being treated as
variables and the aim is to discover the set of s that justify the initial policy
as optimal. Given these values of the social weights they can be assessed in
terms of their suitable representation of social preferences. If they are obviously
unacceptable, the initial policy can be seen as non-optimal.
Recalling the discussion following (6.16), the diagrammatic explanation of
the inverse optimum is that is being adjusted until the vector W is such
that the boundary of the halfspace, S, is coincident to the boundary of Q(p, q).
In Figure 6.11, the set of weights 00 would solve the inverse optimum.
184
Figure 6.12: Solution of Inverse Optimum

Applying the same analysis to the Pareto-improving reform, part (b) of the
Motzkin theorem provides a set of non-negative weights under which no reform
can be found. These can also be interpreted as the implicit social welfare weights
of the households constituting the economy. Evidently, this can only be achieved
if the number of goods is equal to the number of households. With too many
households the s will not be uniquely identified and with too few the equations
need not have a solution. When the inverse optimum is solved it is possible that
some of the values may be negative. This indicates that a Pareto improvement
may be made.
6.6.2
Applications
In their application of the inverse optimum analysis, Ahmad and Stern (1984)
employ the same commodity groupings and Indian data set as in the work of
Murty and Ray reported in Chapter 4. Their first step is to take the revenue
requirement as given and to solve the inverse optimum problem. As their data
set has more households than goods, this is achieved by grouping the households
into 9 income bands to give the same number of households as goods. Table 6.1
describes the outcome of this procedure.
As can be seen from Table 6.1, there is considerable variance in the welfare
weights attached to the dierent income groups. There are also a number of
negative weights which indicate that Pareto-improving reforms are possible.
Having identified the possibility of making Pareto-improving reforms, Ah-
6.6. THE INVERSE OPTIMUM
185
Figure 6.13: Tax Changes

mad and Stern proceed to calculate a vector of such reforms. This vector is
restricted by the assumption that the tax changes must not increase or decrease
the revenue raised by more than one rupee. This normalisation rule is an alternative to (6.17). The resulting tax changes are shown in Table 6.2 and the
welfare eects of these in Table 6.3.
The welfare change table shows clearly how welfare has increased for some
groups and decreased for none. Tax changes could obviously be undertaken
until the gains from reform were exhausted.
An alternative approach to the inverse optimum problem is taken by Christiansen and Jansen (1978). They approach the analysis from the perspective
that the welfare weights are determined by political considerations and the form
of the welfare function can be inferred from study of the political process. A
vector of income changes is defined as distibutively neutral if it leaves all relative
welfare weights unchanged and the political process is deemed to view an equal
relative increase of all incomes as neutral. Imposing an additive social welfare
function
H
X
W =
W Mh ,
(6.35)
h=1
and defining the welfare weight of h as a function of income by
W M h
Mh =
,
M h
the political view of distributional neutrality implies that
Mh
, h 6= b
h,
M bh
(6.36)
(6.37)
186
Figure 6.14: Welfare Changes
is homogeneous of degree 1 in M h , M h . From this it follows that the welfare

weight function must take the form
(M ) = KM ,
(6.38)
which is identical to the form in (4.51).

This analysis is then integrated within a study of the Norwegian commodity
tax system. In addition to the isues addressed in Chapter 4, the design of taxes
also takes into account the assumption that four of the fifteen commodities
groups may have significant externality eects. When the externality eects are
ignored, a value of = 1.706 rationalises the choice of observed tax rates. In
contrast, incorporating the externality eects provides a value of = 0.88. At
this value of the social welfare function (6.35) is approximately log-linear. The
dierence between the two values of show that the results of inverse optimum
calculations may be highly sensitive to the correct choice of framework.
6.7
Marginal social cost of taxation
The marginal social cost has been developed as a means of assessing the potential
optimality of a commodity tax system and, if the existing system is not optimal,
of indicating the direction in which it should be reformed. If carefully applied,
the technique is also useful for assessing other policy reforms. As above, the
existence of an improving reform follows from the non-optimality of the initial
position but the direction of reform is determined by assuming certain regularity
conditions for the economy, where the required regularity is clarified below.
6.7. MARGINAL SOCIAL COST OF TAXATION
6.7.1
187
Derivation
To introduce marginal social cost the Diamond-Mirrlees economy of Chapter

4 is employed. Recalling that analysis, the consumers are described by their
indirect utility functions
U h = V h q1 , ..., qn , w, I h ,
(6.39)
and the tax revenue that must be raised is given by

R=
H
n X
X
ti xhi .
(6.40)
i=1 h=1
Social welfare is determined by a Bergson-Samuelson social welfare function

which is defined on the vector of indirect utilities
W = W V 1 () , ..., V H () .
(6.41)
Starting from an initial vector of taxes (t1 , ..., tn ), the eect on social welfare
of a change dtk in the tax tk on good k is given by
#
"H
X W V h
(6.42)
dtk .
V h qk
h=1
The eect of the same change in tk on tax revenue is

#
"H
H
n X
X
X
xhi
h
xk +
ti
dtk .
qk
i=1
h=1
(6.43)
h=1
Using these two terms, the marginal social cost of raising a unit of revenue by
increasing the tax on good k, k , is defined by (minus) their ratio
k =
PH
W V h
h=1 V h qk
P
Pn PH
xh
H
h
i
h=1 xk +
i=1
h=1 ti qk
(6.44)
It is clear from the first-order conditions for the optimal tax problem that at
an optimum all the marginal social costs for goods k = 1, ..., n should be equal.
If they are not equal, the existing tax system is not optimal. In the case that
the s are not equal, those goods with the largest values of should have their
taxes reduced since collecting revenue via these goods leads to a larger loss in
social welfare for a unit of revenue than would taxes on low goods.
This procedure can be interpreted as being the first step in an iterative
algorithm to find the maximum. The specific requirement that the goods with
high values of k should have their taxes reduced is based on the assumption
that the problem has sucient regularity so that the direct eect of tk on k
dominates the cross-eects caused by the adjustment of other tax rates. If this
188
restrictions is satisfied the variation in the s indicates the correct direction of

reform. Typically, the non-concavity of the tax problem prevents the application
of standard theorems on the convergence of algorithms.
An interesting application of the social marginal cost is made in Newbery
(1986) where the issue of taxes on intermediate goods when there is not a complete set of final goods taxes is addressed. The argument proceeds by evaluating
the marginal social costs of the existing set of taxes and contrasts these with
the marginal social costs of taxes on intermediate goods. If the latter costs are
lower, and a condition is given by Newbery for when they will be, intermediate
goods taxes should be introduced. This result also indicates the importance of
a full set of commodity taxes in the proof of the Diamond-Mirrlees eciency
lemma. The marginal social cost also forms the basis of the Murty-Ray algorithm, described in hapter 4, for calculating optimal commodity taxes.
6.7.2
Empirical application
An empirical application of the use of the marginal social cost is given by Decoster and Schokkaert (1989). The analysis is applied to Belgian data with a
twelve commodity breakdown. The welfare weights for the social welfare function are defined in the manner described in Chapter 4, thus
1
.
(6.45)
h =
h
The marginal social costs will evidently depend upon the chosen value of . A
selection of the results is given in Table 6.4.
Table 6.3 illustrates that the Belgian tax system is not optimal for any value
of . However, for all values tobacco has the highest marginal social cost and
services the lowest. The policy recommendation from this would be that the tax
on tobacco should be lowered and that on services raised. Furthermore, heating
also has a generally high marginal social cost. Decoster and Schokkaert also
present the social marginal costs for dierent income groupings from which can
be formed general views on the importance of various commodities to dierent
income groupings. For instance food appears important to the low income
groups with transport of no consequence. The ranking is exactly reversed for
the highest income group.
6.8
Political Constraints
The policy reform problem has been extended by Kanbur and Myles (1992) to
include political constraints upon directions of reform. These constraints are
intended to capture the fact that policy makers are invariably constrained in
their choice of policy by the need, for example, to win elections, avoid riots and
civil disorder and satisfy influential citizens.
The general form of a political constraint consists of a set of S functions
Ts (q) : <m <, s = 1, ..., S such that any policy change dq that satisfies the
6.8. POLITICAL CONSTRAINTS
189
Figure 6.15: Marginal Social Costs

political constraint has the property that
where
T dq 0,
(6.46)
T1T
T = ... ,
TST
(6.47)
W T dq > 0, Z T dq 0, T dq 0?
(6.48)
is the matrix formed from the gradient vectors of the Ts functions. One simple
example arises in the case of a political elite whose welfare must not be reduced
by the reform. The functions Ts (q) are then the indirect utility functions of the
households comprising this elite and T is formed from the gradient vectors of
the indirect utility functions.
Incorporating the political constraints into the WI gives the following reform
problem
Constrained Welfare-Improving Reform (CWI).
The structure of this problem can also be characterised by the use of a

theorem of the alternative. From Mangasarian (1969)
190
Theorem 35 Either there exists a vector x with

(a) Ax 0, Bx = 0, Cx = 0,
or there are three vectors such that
(b) AT y1 + B T y2 + C T y3 = 0, y1 > 0, y2 0, y3 0.
Hence either a solution exists (case a) or there is some set of weights (case
b), with the vector y3 proportional to the Lagrange multiplier on the political
constraints, for which the initial state is optimal.
The size of the solution sets to PI, WI and CWI can be contrasted. Under
the normalisation rule in (6.17) that the reform vector must lie on the surface
of the unit sphere, the size of a solution set can be defined as the area it covers
on the surface of the sphere. Denote S P , S C , S W to be the solution sets to PI,
CWI and WI respectively. The relative sizes of these solution sets are then as
given in Theorem 6.7.
Theorem 36 (Kanbur and Myles)
a) S C S W , S P S W ,
and, when the constraint is the non-reduction of the utilities of a political
elite,
b) S P S C S W .
n
o
P
hT
V
,
0
Proof. To show that S P S W define KH = x : x = H
h
h
h=1
W
which is the cone generated by the vectors V h , h = 1, ..., H. As V
h 0,
all h, W
K
.
Next,
take
H
and
H
to
be
the
halfspaces
defined
by
H
Z
W
HZ = x : V T x 0 and HW = x : W T x 0 . Using these defini +

/0 B1 where a
tions, S W = HZ Int HW B1 and S P = HZ KH
superscript + indicates the
positive
polar
of
a
cone
and
B
1 denotes the unit
+
sphere. But, as W KH , KH
/0 Int HW . Hence S P S W .
from using
The proof that S C S W follows
the rows of the matrix T to

define the halfspaces HTs = x : TsT x 0 , s = 1, ...S. From this it can
be seen that S W = HZ Ss=1 HTs Int HW B1 . It is then clear that
S C S W . This proves (a).
To prove (b), note that if the utility levels of a political elite constraint
policy then T is given by the gradient vectors of the indirect utilities of the
members
elite. Index the members
of the elite by c = 1, ..., C and let
n of the P
o
cT
KC = x : x = C
V
,
0
be
the
cone generated by the political
c
c=1 c
+
constraints. Evidently, KC KH and S W = HZ KC
Int HW B1 . Also,
+
+
+
+
as KC KH it follows that
K
K
and
thus
K
H
C
H /0 KC . From this
+
+
P
C
W
KH /0 KC Int HW Int HW and S S S .
The result in (a) arises because W is a positive linear combination of the
rows of P and (b) is explained by the fact that in the case described the cone
generated by T is a subset of that generated by P . This theorem formalises
some of the comments in Ahmad and Stern (1984).
6.9. CONCLUSIONS
191
In the presence of the political elite, the optimal tax problem is
max W = W V 1 (q1 , ..., qn ) , ..., V H (q1 , ..., qn ) ,

{q1 ,...,qn }
(6.49)
subject to the political constraint
V c (q1 , ..., qn ) V , all c C,
(6.50)
where C is the set of elite households, and the production constraint

F (X (q1 , ..., qn )) 0.
(6.51)
Solving (6.49) - (6.51), the optimal policy is characterised by

H
C
X
W V h X c V c
X i
+
F T
= 0, i = 1, ..., n,
h
V qi
qi
qi
c=1
(6.52)
h=1
and
h
i
c
F = 0, 0, c V c V = 0, c 0, c = 1, ..., C.
(6.53)
From (6.52) and (6.53), it can be seen that the eect of the constraint is
W
to raise the eective welfare weight on the constraining households from V
c
W
c
to V
c + , with the complementary slackness conditions, (6.53), guaranteeing
that c is non-negative. From this it can be seen that the implicit eect of the
constraint is to shift the optimal policy in favour of the constrained household
by eectively giving greater concern in social welfare to them.
This result does have important implications for the analysis of the inverse
optimum problem. Working back from the solution to (6.52) would generate
W
c
the eective weights V
which, whenever a constraint is eective, will
c +
W
dier from the true weights V c . Therefore whenever political constraints are
binding, it is not possible to recover the welfare function from observed policy.
The inverse optimum can only recover the combined values of the welfare weights
and the additional multipliers.
6.9
Conclusions
This chapter has reviewed the economic analysis of policy reform and has introduced some of the important analytical tools and results. The importance
of Theorems of the Alternative have been stressed since these theorems show
that either an improving reform is possible or that the initial point is optimal.
An alternative insight into the existence of reforms was provided by a geometric
approach that relied on the properties of convex cones.
One of the most important conclusions to emerge from the analysis was that
temporary ineciency may arise during a process of reform. This finding should
be contrasted to the eciency lemma of Chapter 4 which showed the desirability
192
of production eciency at the optimum. It should also be stressed that although

the theory was developed in the context of price reform in a Diamond-Mirrlees
economy, the method of considering the eects of dierential policy reform can
be usefully applied beyond this particular framework.
Chapter 7
Risk
7.1
Introduction
Risk is a factor that is evident throughout economic activity. Firms must choose
between investment plans for which both the cost and the return cannot be
known with certainty, households purchase goods whose value in use is determined by the state of nature and the government receives uncertain revenues
and allocate funds to projects with unknown outcomes. Although the ArrowDebreu economy is capable of incorporating risks of these kinds, so that they
can be viewed as having already been covered by previous analysis, the special
features involved with risk justify a separate chapter devoted to the subject.
The interpretation of the Arrow-Debreu economy in the presence of risk is
discussed first and the Pareto optimality of equilibrium is reconsidered with
particular focus placed upon the number of markets necessary to sustain optimality. This analysis is at the level of generality of previous chapters. The
reasons why there may be too few markets to sustain optimality and whether
this may justify government intervention are also considered. Individual attitudes to risk, in terms of measures of risk-aversion are then contrasted to social
attitudes. Alternative perspectives on social attitudes, including the ArrowLind theorem supporting risk-neutrality of government, are contrasted. A more
general framework is then presented which shows how social attitudes to risk
can be derived from the social insurance eects of projects and the weighting of
households in the social welfare function.
A more specific interpretation of risk in terms of assets with random returns
is then adopted and household maximisation is analysed in further detail. Reactions to taxation are determined under various assumptions about the loss-oset
provisions of the tax system, the return on the safe asset and the number of
risky assets available. Two alternative perspectives on the nature of risk and
the interaction with taxation are then described. The first is the standard utility maximisation analysis of labour supply but extended to include uncertainty
about the wage. This is followed by a consideration of choice between occu193
194
CHAPTER 7. RISK
pations where one occupation has a known return and the other an uncertain
pay-o.
Several dierent approaches to the design of the income taxation with risk
are described. The first set of analyses consider risks which are purely individual:
some households may do well and others badly. The unifying feature of these
is shown to be the welfare gains that can be achieved by the use of a distorting
income tax due to the social insurance that the tax can provide. The social
insurance eect is such that it preferable to raise revenue from a set of ex
ante identical households using the optimal income tax rather than a uniform
lump-sum tax. These results are contrasted to those that apply under purely
aggregate risk where the entire population either gains or loses. In the latter
case, taxation cannot provide social insurance.
Before proceeding, two points are worth noting. In common with most
recent literature, no distinction is made here between risk and uncertainty; the
two terms are employed interchangably. Secondly, the analysis of tax evasion is
reserved for later analysis in Chapter 12. Although tax evasion is an example
of choice with risk, it is somewhat special because of its illegality. In addition,
the substantial literature on tax evasion merits separate consideration.
7.2
General equilibrium with risk
Risk can be incorporated into the Arrow-Debreu economy with very few formal
modifications; eectively all that needs to be done is to increase the number of
goods and prices in an appropriate manner. The results of the previous analysis
then follow as in Chapter 2. The classic presentations are given in Arrow (1963)
and Debreu (1959). Radner (1985) provides an extensive survey of the relevant
literature.
The major focus here will be placed on the simplest case in which all contracts are formed in period 0 and the uncertainty is about the state of the world
that will occur in the only other period, period 1, in which contracts are fulfilled
and consumption takes place. In this setting it is not necessary to distinguish
commodities by their time of availability. The extension to many time periods
(though retaining a single period in which contracts are formed) will be briefly
discussed; none of the essential conclusions is modified.
7.2.1
Risk in the Arrow-Debreu economy
To formalise the notion of risk, a set of alternative states of the world is introduced. The set of states is intended to be an exhaustive list of all potential
states that dier in any economically relevant way. The relation of these states
to risk is that any realisation of random events will lead to one of the states, s,
where the states are indexed s = 1, ..., S, and that there are sucient states to
cover all possible dierent realisations.
Each good in the economy is dierentiated by its place and state of availability. Contracts are made prior to the realisation of the state and will cor-
7.2. GENERAL EQUILIBRIUM WITH RISK
195
respondingly specify the place of delivery and the state in which delivery will
be made. Since goods will only be delivered if the relevant state occurs, the
term contingent commodities is often used since delivery is contingent upon the
specified state occurring.
The production set of each firm is extended, in a manner compatible with the
extended list of commodities, to include all feasible input-output combinations
of contingent commodities. A production plan is then a list of inputs and
outputs of all goods in all states. Denoting the price vector of the n goods in
state s by ps = (p1s , ..., pns ), the extended
p = (p1 , ..., pn ).
list of prices becomes
j
j
j
j
Firm j chooses a production plan y = y11 , ..., yn1 , ..., ynS to maximise profits
j = py j . With a known price vector this provides a clearly defined level of

profits. The structure of this profit maximisation problem is formally identical
to that introduced in chapter 2 and, given strict convexity of the production set,
the outcome will be continuous supply and profit functions. It should be noted
that these are supplies of contingent commodities which will only be delivered
if the relevant state occurs.
The preferences of each household h are assumed to be representable by a
von Neumann-Morgernstern utility function
Uh =
S
X
s=1

hs Ush xhs ,
(7.1)
h
h
where
s = x1s , ..., xns is the vector of consumption of the n goods in state s
x
and h1 , ..., hn are the probabilities that household h assigns to the possibility
of states 1 to S occurring. The structure of (7.1) allows each of the within-period
utility functions to be state-dependent.
If the endowment vector of h in state s is hs , their budget constraint is
S
X
s=1
ps xhs
S
X
s=1
ps hs +
m
X
hj j .
(7.2)
j=1
Once the price vector is known, the budget constraint in (7.2) is clearly determined. The household will then choose their demand for each contingent
commodity to maximise utility subject to (7.2) and subject to the demand being in the consumption set of the household; the consumption set being defined
in terms of contingent commodities. As in Chapter 2, if (i) the utility function is
continuous and strictly quasi-concave, (ii) the consumption set is closed, convex
and has a lower bound and (iii) the endowment is interior to the consumption
set, this maximisation will generate continuous demand functions for all goods.
It should be noted that in this context quasi-concavity of the utility function
implies risk-aversion; see Arrow (1963) and Debreu (1959).
An equilibrium for the economy is a price vector such that demand is equal
to supply for all contingent commodities. Whichever state arises, it is then the
case that the demand for commodities in that state is equal to the supply. Since
the economy is only a reinterpretation of that without uncertainty, the method
196
CHAPTER 7. RISK
of establishing the existence of equilibrium remains the same. In particular, the

same set of assumptions applied to the extended definitions of the production
and consumption sets remain sucient for the existence of equilibrium.
7.2.2
Eciency
In considering the eciency of the equilibrium that will result from trade, it is
necessary to distinguish between ex ante and ex post eciency. An equilibrium
is ex ante ecient if no redistribution of contingent consumption levels can be
found that is Pareto-improving in terms of expected utilities. Conversely, ex
post eciency would be achieved if no reallocation could take place of actual
consumption levels in the state that occurred that would be Pareto-improving
in terms of realised utilities.
The formal equivalence between the economy with certainty and that with
uncertainty implies immediately that the equilibrium will be Pareto ecient in
the ex ante sense; the proof of the First Theorem of Welfare economics is directly
fsh ps xhs be the equilibrium value of the contingent
applicable. Now let M
commodities
demanded
by h in state s. Then, relative to the vector of income
1
h
f
f
levels Ms , ..., MS , the equilibrium allocation in any state s is also Pareto
ecient in the ex post sense: once the state has been realised no reallocation
of consumption exists that is Pareto improving. This result
is explained
by
1
h
f
f
viewing households being endowed in state s with incomes Ms , ..., MS and
then equilibrium being reached given these incomes. The outcome must be
Pareto-ecient.
A further observation merits attention. If no trade were to take place until
csh , would be
the state were realised, the income of household h in state s, M
h
cs pbs hs , where pbs is
given by the value of the state-contingent endowment, M
the equilibrium price vector that would be realised once state s had occurred. It
is also correct to observe that the equilibrium then reached in state s would be
Pareto ecient in
an ex post sense.
However, there is noreason to expect
that
1
h
1
h
f
f
c
c
the income vector Ms , ..., MS is a simple rescaling of Ms , ..., MS nor that
the equilibria reached with the two income distributions are identical. Both are
Pareto ecient but they will generally be based upon dierent income distributions. The dierence between the two income distributions arises because the
trade in contingent commodities allows households to reallocate their purchasing power between states on the basis of their attitudes to risk and assessment
of the probabilities of the states occurring. Further discussion of this issue can
be found in Dreze (1970-1).
The equivalence between the economy with uncertainty and that with certainty also justifies the direct application of the Second Theorem of Welfare
Economics. When the required convexity and continuity assumptions are satisfied, any allocation that is Pareto ecient in an ex ante sense can be decentralised as a competitive equilibrium with appropriate lump-sum transfers. If
there are no restrictions upon policy instruments, the government can therefore
197
Figure 7.1: Time and Uncertainty

achieve a socially optimal outcome by lump-sum redistribution and operation
of competitive markets. It should be noted that the lump-sum transfers will be
in terms of contingent commodities and will generally be state-dependent.
7.2.3
Including time
The introduction of time in addition to uncertainty can be achieved following

the procedure due to Debreu (1959). Assume that there are T time periods,
indexed t = 1, ..., T , during which consumption can take place and a further
period, labelled 0, in which all trades are enacted. Period 0 occurs prior to
periods during which consumption can take place. At time T there is a set of
possible states of the world that may occur; these are indexed sT = 1T , ..., S T .
It is assumed that there is a unique series of states from period 0 through
to T by which state sT can be realised. Therefore, given a realisation of sT ,
the previous states of the economy are determined. These assumptions are
illustrated in figure 7.1. The states of the economy at each time period can be
represented by the vertices of a tree. If T = 3, it can be seen that the realisation
of state 13 implies that states 12 , 11 must also have occurred.
A commodity can now be defined by its place of availability and by the vertex
of the tree at which it is available. Specifying the vertex uniquely identifies the
time and state in which the commodity is available. Since no delivery will
take place unless the relevant vertex occurs, these commodities again represent
contingent claims for delivery. Profit and utility maximisation can be defined
in terms of these contingent commodities and continuous supply and demand
functions derived.
It should be appreciated that this formulation of time and uncertainty does
not alter any of the essential structure of the economy. The previous reasoning
therefore continues to apply: under the standard assumptions an equilibrium
will exist, a competitive equilibrium will be Pareto ecient and any Pareto
ecient equilibrium can be decentralised. In an economy with this structure,
both time and uncertainty can be incorporated without the eciency of the
competitive equilibrium being destroyed.
198
7.2.4
CHAPTER 7. RISK
Complete markets
When assessing the relevance of these results there are several issues that have to
be considered. In addition to the points raised in chapter 2 concerning the strict
applicability of the economy and the use of lump-sum taxation, it is now also
important to consider the number and arrangement of markets in the economy.
The economies described above have possessed a complete set of markets, that is,
when markets open at the start of the economy trades can be organised for every
good in any state and, in the extended economy, for any time. In short, the
economy is assumed to possess markets for all state-contingent commodities.
This requirement is naturally far too demanding for any actual economy to
satisfy; it is not even likely that the products that will be available in the future
will be known today. In reality there are well-organised futures markets in a
small number of financial, industrial and agricultural products but certainly not
for the majority of products. Furthermore, even when there are futures markets
there is no necessary implication that state-contingent contracts can be written
and enforced. Hart (1975a) demonstrates that if markets are incomplete the
outcome of competitive trade will not be Pareto ecient and may even fail
to be ecient relative to the set of markets that are open. In eect, when
markets are incomplete the economy can become trapped in an equilibrium
which is Pareto-dominated. It is consequently unlikely that assumptions of this
economy can be satisfied and hence its welfare properties cannot be appealed
to as descriptive of reality.
There is, however, an alternative form of the economy, due to Arrow (1963),
that is rather more tenable. This alternative interpretation reduces the number
of futures markets that have to be operational and brings the economy closer
to what is observed in practice. To formalise this idea it is simplest to return
to the economy with uncertainty but a single period of consumption. Assume
that there exists a set of securities, with S members, such that security s
pays one unit of numeraire commodity if state s occurs and nothing otherwise.
Security s has price qs . Economic activity is then organised in two stages.
Firstly, households allocate their incomes to the purchase of securities. Hence
securities will be purchased by household h to satisfy the budget constraint
Mh =
S
X
q zh ,
(7.3)
=1
where M h is the income of h and zh is the quantity of security purchased by

h. Naturally, this procedure will determine the income of consumer h in state
s to be zh . The second stage of the economy occurs after the state has been
realised and involves the households using their income from security purchases
to buy the commodities that are available in that state.
The central feature of this reinterpretation is that it leads to precisely the
same equilibrium as the economy with a full set of futures markets. To see this,
that in the original economy household h chose to purchase the vector
assume
xh1s , ..., xhns in state s. Given the equilibrium price vector, this choice will cost
199
the household an amount given by

n
X
pis xhis .
(7.4)
i=1
If securities are purchased so that

zh =
n
X
pis xhis , all s = 1, ..., S, = s,
(7.5)
i=1
the household can guarantee the necessary income to purchase the optimal consumption plan. They will, of course, also choose this level of income since it was
defined by the optimality of the consumption vector. The final step is to note
that since the total income to be allocated to securities is defined by the value
of the contingent endowments and the profits of firms, this remains the same
in both interpretations of the economy. The prices of the securities are then
determined on their market so that precisely the correct levels of income are
transferred forward. In summary, if for each state there exists a unique security
that has a unit return in that state and zero return in all others, the equilibrium
of the economy will be the same as that with a full set of futures markets. It
therefore has the same welfare properties and the Two Theorems can be proved
to hold.
Two further points are worth making about this economy. Firstly, the use
of securities has economised upon the number of markets that are required to
achieve optimality. The initial economy employed nS markets whereas the alternative possessed S security markets and n goods markets since the goods
market was only opened after the state was realised. There is therefore greater
likelihood that the optimum will be attained in the latter case. Secondly, although the economy was introduced in terms of the elementary securities
that had returns in only a single state, it will also function when securities have
returns in several states provided the returns are such that the securities can be
combined so that any vector of returns across the states can be achieved; see the
discussion in Due and Sonnenschein (1989). This property of the securities is
generally termed spanning in the literature. The applicability of this economy
is therefore broader than the initial description may have suggested.
7.2.5
Comments
The discussion above has demonstrated that the Arrow-Debreu economy may
be extended to accommodate uncertainty and, that when extended, it retains
the welfare properties of the economy with certainty. However, the extension
required either an expansion of the number of markets or the introduction of
security markets. The first interpretation probably has little to commend it
other than formal elegance since economies with complete sets of markets are
not observed in practice. The second is rather more applicable and does suggest
how economies actually attempt to cope with the existence of uncertainty but,
200
CHAPTER 7. RISK
although securities certainly exist, it is not clear that they actually satisfy the
spanning property. Furthermore, the structure of the economy also places very
strong informational requirements upon the participants. In order to determine
the optimal purchase of securities, it is necessary for the household to know the
prices that will rule in each possible state that may be realised; this is implicit
in the statement of (7.5). It must also be the case that all the agents agree upon
the prices that will be ruling in each state. This position is clearly untenable.
Although the interpretation of general equilibrium with risk shows that there
is no formal dierence between the economy with certainty and that with uncertainty, the structure of the latter makes it apparent that even less emphasis
should be placed upon its welfare properties than was placed upon those of the
certainty economy. As before, the economy provides a benchmark case against
which other, more restricted, economies should be placed. As already noted, the
existence of a market failure is not a sucient condition for supporting government intervention since the government may also be restricted by some of the
features of the economy that prevent the market being ecient. In every case
it must be verified that the government can actually achieve a better outcome
than the market.
7.3
Private and social attitudes to risk
It is typically assumed that households prefer a certain outcome to a risky outcome with the same expected payo. That such risk-aversion exists is supported
by the observation that the purchase of insurance policies is commonplace. The
holding of shares with limited liability and of money, which pays no interest,
can be similarly explained by the desire to avoid risk. Given that society is
the sum of its members, does it follow from this that the government, as the
representative of society, should also act in a risk-averse manner? The answer
to this question has important implications for policy choice since a risk-averse
government would undertake fewer risky projects than a risk-neutral one and
would structure policies to avoid uncertainty in revenues and expenditures.
This section will first recall some aspects of the theory of household attitudes
to risk and of the measurement of risk aversion. This is followed by a summary
of alternative views on social attitudes to risk including the Arrow-Lind theorem
that asserts the risk-neutrality of the government. The section is completed by
an analysis of the valuation of projects in a contingent-commodity framework
with, possibly, incomplete markets.
7.3.1
Private attitudes
Consider household h which derives utility level U h = U h (M ) , U h0 () > 0,

from certain income M . Assuming that the preferences of the household satisfy
the assumptions of von Neumann-Morgernstern expected utility theory, so that
the utility function is unique up to ane transformations, the expected utility
f with probability and M
c with probability 1 is
of receiving income level M
7.3. PRIVATE AND SOCIAL ATTITUDES TO RISK

given by

f + [1 ] U h M
c ,
E U h = U h M
201
(7.6)
f + [1 ] M
c = M then household
where E [] is the expectations operator. If M
h is risk-averse if and only if

f + [1 ] U h M
c .
U h (M ) > U h M
(7.7)
The interpretation of (7.7) is that a risk-averse household would not take on an

actuarially fair bet. If the inequality was reversed the household would be a risk
lover. Equality in (7.7) would imply risk-neutrality.
For (7.7) to hold for all actuarially fair bets the marginal utility of income for
h must be decreasing, that is U h00 < 0. Equivalently, the utility function U h (M )
must be concave. A measure of the degree of risk-aversion of a household must
therefore involve the second derivative of utility. By itself, this cannot act as
a measure since it is not invariant of the particular transformation of utility
chosen. Two measures that are invariant to such transformations are given by
the Arrow-Pratt (Arrow (1965), Pratt (1964)) measures of absolute and relative
risk-aversion defined respectively by
h00
Absolute risk aversion, RA (M ) = UU h0 ,
and
h00
Relative risk aversion, RR (M ) = MU
.
U h0
It is often hypothesised (for instance in Arrow (1965)) that RA (M ) is a decreasing function of M and that RR (M ) is an increasing function. The first of
these hypotheses implies that the willingness to take small bets of fixed size increases with income, an observation that is not implausible. The second implies
that as the level of wealth and the size of bets increase in the same proportion, the willingness to accept the bet should fall. This observation is not as
immediately appealing as the first.
Now consider a risk-averse household assessing whether to undertake a risky
investment project, for example the purchase of securities. The concavity of
the utility function that follows from the assumption of risk-aversion, implies
that the expected return to the household of the investment is less than the
expected value of the returns. This is simply a more general restatement of
(7.7). An alternative view is that the household discounts the returns for the
risk involved at a higher rate than is justified by the probabilities of the outcomes
alone. In any case, a risk-averse household will not undertake a project that has
an expected value of zero.
An alternative perspective can be obtained by using the concept of a risk
premiums. Consider a household facing a random income M and choose the
fixed income level M0 such that
U (M0 ) = E [U (M )] .
Now set M0 = M where M = E [M ]. From (7.8) this implies that
E [U (M )] = U (M0 ) = U M0 M .
(7.8)
(7.9)
202
CHAPTER 7. RISK
The value is termed the risk premium and it can be interpreted as the amount
the household would be willing to pay to avoid the risk associated with the
random income M . It should be clear that as risk-aversion increases, so will
the value of the risk premium that the household will pay. In fact, denoting
the variance of the random variable M by 2 , Pratt (1964) shows that the
risk premium and risk-aversion are related by = 12 2 RA (M0 ) + higher-order
terms.
7.3.2
Social attitudes
The implication of household risk- aversion for the social treatment of risk has
received considerable discussion and several alternative positions have been proposed. These are based upon diering views on the role of the government, its
relation to the agents that compose the society over which it governs and its
relative ability at dissipating risk.
A first view, propounded most forcefully by Hirshleifer (1964, 1966) and
Sandmo (1972) is that the government should really not be given any privileged
role with respect to risk bearing. In a perfect capital market, the decisions
and attitudes to risk of the agents that constitute those markets are reflected
in the equilibrium levels of discounting that are applied to risky projects since
the market functions to allocate risks eciently. As the market captures the
views of agents, it is then argued that the government can do no better than
to adopt the market rate of discounting when making its own project appraisal
decisions. A further development of this argument would be to suggest that if
the government did use a discount rate lower than the market rate, it would
adopt projects that were socially undesirable and which displaced private sector
projects with higher returns.
An alternative view, stated for example in Samuelson (1964) and Vickrey
(1964), is that the government is in an advantageous position with respect to
accepting risk relative to private sector agents. This advantage arises due to
the size of the public sector and the number and variety of projects that it
undertakes. If the returns from its various projects are not perfectly positively
correlated, it is possible for the government to pool the risks arising from the
projects and therefore lessen the overall uncertainty of its returns. The argument
is then completed by noting that an ability to pool risks allows a lower discount
factor to be applied to any single risky project than would be done if the project
were undertaken separately in the private sector.
There are several arguments that can be raised against this view. Firstly,
projects should be evaluated individually by the public sector and not as a package as implied by the risk pooling argument. Packaging projects may result in
some being adopted which are socially undesirable. Secondly, if the government
is able to reduce risk by pooling, it should confer the advantages of doing so on
the private sector rather then simply exploit them in the choice of public sector
project. One method of doing so is by direct subsidy of the private sector in
order to allow previously marginal projects to be adopted.
Rather than focus on the risk pooling ability of government, Arrow and Lind
203
(1970) argue in favour of a lower discount factor for the public sector on the
grounds of the ability of the public sector to spread risk. A simple version of
this argument would go as follows. Although each taxpayer is risk-averse, the
total cost of a project will be divided between taxpayers with each carrying only
a small amount of the cost and therefore only a fraction of the risk. Provided
that the number of taxpayers is suciently great, that the project is not large
relative to the economy as a whole and that the (random) cost to any taxpayer
of the project is not correlated with any existing uncertainty in their income
stream, then as the number of taxpayers increases the risk premium of each
taxpayer and the total risk premium to society tend to zero. Given this, the
public sector should act as if it were risk-neutral.
To prove this theorem, consider an economy with H identical households who
have identically distributed random incomes, M , and concave, bounded, strictly
increasing and dierentiable utility functions U (M ) which satisfy the axioms of
expected utility theory. The government carries all costs of investment in a
project and receives all returns. The payo, less costs, is denoted by . Its
budget is to be balanced in the absence of the project so positive returns from
the project are given equally to the taxpayers (by reducing taxes) and costs are
carried by the taxpayers (by raising taxes).
The first step is to write the payo to the project in the form = + X
where = E [] and E [X] = 0. Now consider an individual household and
let s, 0 s 1, be their share of the returns of the project. This household
would
(s) to avoid the
h pay a risk premium
i
risk where (s) is defined by
E U M + s + sX
= U M + s (s) . The value of (s) is the cost
to the household of bearing the risk. With these definitions, Theorem 7.1 can
now be proved.
1
Theorem 37 (Arrow and Lind) Assume that
) = 0 and that s = H
.
1 cov(M,
The total cost of risk bearing, H (s) = H H = Hk (H), then tends to zero
as H tends to infinity.
Proof. Dierentiating expected utility with respect to s gives

i
h
h
ii
h
E U M + s + sX = E U 0 M + s + sX + X .
s
(7.10)
Setting s = 0 and employing the fact that cov(M, ) = 0, which implies

cov(M, X) = 0, it follows that
h
ii
h
(7.11)
E U 0 (M ) + X = E [U 0 (M )] .
From the equality in (7.11), using the definition of a derivative shows that
h
i
E U M + s + sX U (M )
(7.12)
lim
= E [U 0 (M )] ,
s0
s
204
CHAPTER 7. RISK
1
or, equivalently under the assumption that s = H
,
"
!
#
+X
lim HE U M +
U (M )
H
H
"
!
#
= lim HE U M +
k (H) U (M )
H
H
= E [U 0 (M )] ,
(7.13)
where the second equality follows from the definition of the risk premium.
Noting that limH H

k (H) = 0, the definition of a derivative can be
used to write
h
E U M+H
k (H) U (M )
lim
(7.14)
= E [U 0 (M )] .
H
k
(H)
H
Using (7.14) to divide through (7.13) gives
i
h
lim Hk (H) = ,
H
(7.15)
or
lim Hk (H) = 0.
(7.16)
The limit in (7.16) proves the theorem.

Although a formally attractive theorem, this result is not entirely persuasive.
The most obvious shortcoming is the assumed independence between private
and public risk. These risks may well be highly correlated, for instance in a
recession both public and private sector incomes will fall. More importantly, the
theorem overlooks existing private sector institutions for risk sharing. Foremost
amongst these are insurance and joint stock companies and the theorem does
not prove that the public sector can spread risk any more eectively than these
institutions. Furthermore, if there is an element of self-selection in the adoption
of private sector risk, it may be the case that risk in the private sector is borne
by those who are risk-lovers and that the private sector is, in aggregate, less
than risk-neutral.
To provide further insight into these issues, consider the following analysis
due to Grinols (1985) which is based on the state-preference economy of Section
2 augmented by the existence of a stock market. The economy is assumed here
to last for two periods. There is no uncertainty in the first period, labelled 0, but
any one of S possible states, s = 1, ..., S, may arise in the second period. Trades
must be completed in the first period. It is possible to trade shareholdings in
firms and to enact futures trades for some, or all, goods. A shareholding in a
firm entitles the owner to a share in the production of the firm proportional
to the shareholding and this is the only means by which households can obtain
consumption in period 1. To motivate a social role for government risk taking,
the analysis will focus on the case where there are incomplete futures markets.
205
A consumption
plan for household
h, h = 1, .., H, is written xh = xh0 , xh1 , ..., xhS
of which eh = eh0 , eh1 , ..., ehS is derived from the public sector and xh eh is
from the private sector; the consumption set is X h . Household h owns an initial
endowment h0 of the input; no endowments are held
of the state-contingent
j
commodities. The production plan for firm j, y = xj0 , xj1 , ..., xjS , is chosen
from the production set Y j . The production set is defined so that y0h 0 is the
input and ysh 0 is the output if state s occurs. Government production is denoted e = (e0 , e1 , ..., eS ) with e0 0, es 0, and is chosen from the production
set Y g .
The important feature of the economy is that inecient risk bearing may
arise due to forward markets being incomplete. If the forward market does
not exist for good i, there will be no price quoted in period 0 for delivery of
good i in period 1. In order to determine its optimal production plan a firm
must decide upon a value for the missing prices. This is assumed to be done
by weighting the valuations of the firms shareholders of good i by the level
of final shareholding. This approach, and possible alternatives, are discussed
in Diamond (1967), Dreze (1974) and Grossman and Hart (1979). Formally,
denote the price vector that firm j perceives to be facing by pj = pj0 , ..., pjS
and normalise so that pj0 = 1, all j. The vector
marginal rates of substitution
of
h
h
h
U0h
US
h
for consumer h is written q = U h , ..., U h = 1, q1 , ..., qnh . Letting j be the
0
shareholding of h in firm j before trade in shares and hj be the shareholding

after trade, the perceived prices for firm j are given by
pj =
H
X
hj q h .
(7.17)
h=1
It should be noted that the economy with complete markets is a special case of
(7.17). With complete markets, trade results in the marginal rates of substitution being equalised across households and, since the shareholdings must sum
to 1, (7.17) then states the usual relation that price is equal to the (common)
marginal rate of substitution.
With the share price for firm j represented by rj , an equilibrium of the
economy is defined as follows.
Equilibrium
h n h oi
An equilibrium is an array pbj , rbj , ybj , x
such that:
bh , b
j j
0
j
b y for y Y ;
For all j = 1, ..., m, y
b maximises p
h
h
h b
b
For all h = 1, ..., H, x
b0 , 1 , ..., m maximises U h xh0 , xh1 , ..., xhS subject to
i) xh X h ,
i
Pm
Pm h h
ii) xh0 + j=1 hj rj h0 + j=1 j rj + y0j + eh0 ,
P
h j
iii) xhs ehs m
j=1 j ys 0, s = 1, ..., S.
Conditions (ii) and (iii) capture the first- and second-period budget constraints. In (ii) the initial ownership of a share in a firm implies that the
206
CHAPTER 7. RISK
household must provide a share of the input to the firm proportional to the
shareholding.
It is now possible to consider the evaluation of public sector projects. The
comparison will be made between the valuation that a typical private firm, say
j, would place on a project and that which the government would adopt. To
remove one reason for these to dier, it is assumed the returns to the public
sector project are distributed in precisely the same proportions as the returns
to the private firm. Therefore, the proportion of the public sector project that
is distributed to h, h , satisfies h = hj . To value public projects, the government employs a social welfare function that is a weighted sum of the household
valuations
H
S
X
X
ah
qsh xhs ,
(7.18)
W (x) =
h=1
PH
s=0
where a 0, all h, and h=1 a = H. A project

is interpreted as a dierential
change in consumption dx = dx10 , ..., dxH

.
Such
a project is valued by the
S
social welfare function at
W (dx) =
H
X
h=1
ah
S
X
qsh dxhs =
s=0
H
X
ah h
S
X
qsh dxs ,
(7.19)
s=0
h=1
using dxhs = h dxs . Conversely, a change in production is valued by firm j at

F j (dx) =
H X
S
X
hs qsh dxs .
(7.20)
h=1 s=0
Employing the restriction that h = hj , the valuation in (7.19) and that in

(7.20) are related by
W (dx) = F j (dx) +
H
S
X
X
h
hs qsh dxs .
a 1 h
h=1
(7.21)
s=0
It is immediately apparent from (7.21) that if all households were weighted

equally in the social welfare function, so that ah = 1, private and social valuations would be equal. If (7.18) is viewed as a linear approximation of a
standard Bergson-Samuelson social welfare function, then there is no reason
why the weights should be equal. Consequently, there is no reason to expect
equality of private and social valuations.
To further investigate the dierence in valuations when are not weighted
equally requires a further distinction has to be drawn. A public project is
said to be contained within the private sector if the vector of returns dx can
be achieved by a portfolio of holdings in the existing private sector firms. If
this is the case, then existing trading arrangements ensure that the valuations
of households and firms are equalised for such a project. Conversely, a public
project is outside the private sector if its returns cannot be duplicated by the
207
purchase of shareholding in firms. A project outside the private sector has the
property that it extends the set of securities that are available and permits
further risk-spreading.
PS
For a project contained within the private sector it follows that s=0 qsh dxhs =
F j (dx), all h = 1, ..., H and j = 1, ..., m. Using this relation in (7.21) then gives
"
#
H
X
1
W (dx) = F j (dx) 1 +
ah 1 h
= F j (dx) 1 + cov ah , h .
H
h=1
(7.22)
The interpretation of (7.22) is that a project contained within the private sector
should be valued more highly by the public sector than by the private sector
when its returns accrue disproportionately to those households that have a high
weighting in the social welfare function. If the returns accrue to those with a low
weighting, the public sector should value the project less than the private sector.
The major implication of (7.22) is therefore the demonstration that there is no
a priori reason for the public sector to employ a discount rate that is uniformly
higher
or lower
than
that of the private sector. As regards the choice of project,

if 1 + cov ah , h > 0, any project that satisfies F j (dx) > 0, and hence would
be undertaken by the private sector, also satisfies W (dx) > 0 so that it would
also be undertaken by the public sector. This result also emphasises the role
played by the assumption of the Arrow-Lind theorem that private and public
risks are uncorrelated. The argument of Hirshleifer (1966) that the government
should employ the same discount rate as private firms follows from assumption
of complete futures markets which implies that all projects are within the private sector. Similarly, Sandmo (1972) although assuming incomplete futures
markets, reaches the identical conclusion by considering only projects within
the private sector.
As already noted, a project that is outside the private sector introduces
a set of payos that cannot be duplicated by shareholdings in firms. If such a
project is undertaken, households aected by the project will need to adjust their
portfolios of assets to achieve optimal risk-taking. If the portfolio adjustment
raises utility, then the project can be viewed as providing social insurance and
this will raise the welfare valuation of that project. It is also possible that the
portfolio adjustment may reduce utility in which case the welfare valuation of
the project will be reduced.
To formalise these observations, the project dx (dx0 , dx ), dx <s , is
written in the form dx = de
x + db
x, where de
x is that part of the project that
can be obtained by combining existing private sector returns and db
x is the
uninsurable part. Expressed alternatively, de
x
(dx
,
de
x
)
can
be
written
as a
0
linear combination of basis vectors of the form dx0 , yj where yj are the future
returns of firm j and is therefore the projection of dx onto the space spanned by
the future returns of the firms. The uninsurable part of the project, representing
an addition to the existing set of risky assets, takes the form db
x (dx0 , db
x )
and is orthogonal to the space of existing risks.
Substituting into (7.21) determines the relation between social and private
208
CHAPTER 7. RISK
valuations as
W (dx) = F j (dx) + F j (de
x)
H
H
S
X
X
X
h
h
qsh de
xs
a 1 h +
a 1 h
h=1
s=0
h=1
= F j (db
x) 1 + cov ah , h + F j (de
x) +
H
X
h=1
S
X
h
h
q(7.23)
xs .
a 1 h
s de
s=0
The evaluation of the project can therefore be divided into three components.
The first is the evaluation of the part of the project that lies within the private
sector and can be interpreted in the manner following (7.22). The second and
third components are the firms and the households evaluations of the component of the project that is uninsurable. Since this part of the project is not
traded, these need not be equal. These can be positive or negative; hence the
social evaluation may be above or below the private valuation. There is therefore no a priori reason for believing the social valuation to be above or below
the private valuation. In the Arrow-Lind case for which the public project is
entirely uncorrelated with private returns, de
x would be zero and as the number
of households increased without limit the final term would tend to zero. This
leaves the government evaluation equal to the valuation of the risk-neutral firms.
7.3.3
Summary
This section has considered alternative perspectives on the relation of social

attitudes to risk to private attitudes. There is, of course, some merit in each of
these proposals and it was shown how each could be generated as special cases
of a general state-preference economy. As a final approach to this issue, Glazer
(1989) has considered social attitudes to risk from a political perspective and
has argued that strategic voting behaviour may lead to projects being adopted
by the public sector that could only be rationalised by a discount rate below that
of the private sector. This emphasises that the political outcome may remain
distinct from that which is economically ecient.
7.4
Household choice and taxation
Although the contingent commodity framework used in the previous sections

has the appeal of generality, this is also its shortcoming as a vehicle for developing simple insights into the interaction between taxation and risk-taking. A
literature has therefore developed that focuses upon household portfolio choice
when there are only two assets available; one safe, one risky. If it is further
assumed that there are only two potential future states of the world, the household decision problem can be represented diagrammatically. This approach, and
its extensions, will be the subject matter of this section.
Consider a household with a given initial wealth, , to divide between the
two available assets. One of the assets, termed the safe asset, yields a known
7.4. HOUSEHOLD CHOICE AND TAXATION
209
return r. The other, the risky asset, has a random return which is distributed
with density () and variance 2 . () has support on (1, ). The household
is concerned only with their final level of wealth, Y , and seeks to maximise
E [U (Y )] by choosing the amount, a, of their initial wealth invested in the risky
asset. The household is assumed to be risk-averse; hence U 00 (Y ) < 0. The
eects of two taxes will be analysed: a wealth tax, denoted tw , and an income
tax, tm .
In order to specify the payo to the household from a given portfolio, it is
first necessary to clarify the provisions in the tax system for osetting losses
(negative returns) against taxable income. If all losses can be oset, the tax
system is said to have full loss oset. The other extreme is termed no loss
oset. Between the extremes lie systems with partial loss oset. The relevance
of loss oset provisions is in their eect upon the expected post-tax return from
the risky asset and the distribution of tax revenues. With full loss osets, an
income tax reduces both the potential gains from holding the risky asset and the
potential losses. Conversely, with no loss oset only the gains are reduced. These
distinct eects upon the structure of post-tax returns show why the eects of
taxation will be dependent upon loss oset provisions. In addition, in all but the
no loss oset case, the government will incur some risk in the form of potentially
reduced tax revenues whenever the household holds some of the risky asset. The
extent of loss oset then determines the degree to which the government shares
the households risk.
Although full loss oset may seem at first sight a practically uninteresting
case, it should be noted that the investment behaviour under consideration may
only be a part of the broader economic activity of the household. When the
household earns income in addition to that derived from investment, then any
loss on investments can be set against other income in the determination of
taxable income. If all sources of income are taxed at the same rate, the tax
system would then appear as if it had full loss oset provisions.
Having clarified these distinctions, the analysis will first consider the eect
of a wealth tax and then turn to the income tax. The formal results will be
developed for the general case described above but will be diagrammatically
illustrated for the special two-state case in which the risky asset either yields
a high return of a low return. The majority of the results are due to Mossin
(1968) and Stiglitz (1969a).
7.4.1
Wealth taxation
The wealth tax is levied upon the final wealth of the consumer that is determined
after the realisation of the return on the random asset. Provided that the final
wealth of the household is positive, the loss oset provisions are not relevant for
the determination of the eect of the tax.
When the realised return on the random asset is , the final wealth of the
household with wealth tax tw < 1 is
Y = [[ a] [1 + r] + a [1 + ]] [1 tw ] .
(7.24)
210
CHAPTER 7. RISK
It then follows that the expected utility of the household can be calculated as
Z
E [U (Y )] =
U ([[ a] [1 + r] + a [1 + ]] [1 tw ]) () d.
(7.25)
1
Dierentiating with respect to a and assuming that E [] > r, which is necessary

and sucient for an interior solution with a > 0 (see Arrow (1970)), the optimal
portfolio satisfies
[1 tw ] E [U 0 (Y ) [ r]] = 0,
(7.26)
i
h 0
(7.27)
E U 0 (Y ) [ r]2 < 0.
and
From (7.25) and (7.26) it can be seen that the eect of the wealth tax is
equivalent to a reduction in initial wealth Y . An increase in the rate of tax
will decrease the proportion invested in the risky asset if the wealth elasticity
da
of demand for the risky asset, d
a , is greater than 1, will leave it unchanged if
da
da
=
1
and
increase
it
if
<
1. The eect of the taxation is this case is
d a
d a
straightforward.
To illustrate this result, assume that the risky asset provides a high return
1 with probability 1 in state 1 and a low return 2 with probability 2 in state
2; 1 > r > 2 . Investing all wealth in the risky asset will lead to final wealth
of [1 + 1 ] [1 tw ] in state 1 and [1 + 2 ] [1 tw ] in state 2. Alternatively,
if all income is invested in the safe asset, final wealth will be [1 + r] [1 tw ]
in both states. The expected utility of the household is given by
E [U (Y )] = 1 U ([[ a] [1 + r] + a [1 + 1 ]] [1 tw ])
+2 U ([[ a] [1 + r] + a [1 + 2 ]] [1 tw ]) .
(7.28)
The indierence curves of (7.28) and the budget opportunities facing the household are illustrated in Figure 7.2.
In Figure 7.2 the budget constraint is given by AB and the chosen portfolio
AO
is at O. The proportion of wealth placed in the risky asset is AB
. The eect
of an increase in the wealth tax or, equivalently, of a reduction in wealth is to
move points A and B towards the origin with the new budget line being parallel
to the old. Repeated for a series of changes in wealth, this will trace out the
wealth-portfolio locus of optimal portfolios. This is shown in Figure 7.3.
If the wealth-portfolio locus is a ray through the origin, as in Figure 7.3,
the proportion of wealth invested in the risky asset is constant as wealth or
the rate of wealth taxation change. This occurs when the wealth elasticity of
demand for the risky asset is equal to 1. When the wealth elasticity is greater
than 1 the wealth-portfolio locus will bend downwards away from the ray and
the proportion invested in the risky asset will fall as the rate of tax increases.
Conversely, a wealth elasticity less than 1 will lead to a wealth-portfolio locus
that rises more quickly than the ray and a consequent increase in the proportion
invested in the risky asset as the wealth tax increases.
Figure 7.2: Portfolio Choice
Figure 7.3: Wealth-Portfolio Locus
211
212
7.4.2
CHAPTER 7. RISK
Income tax
As already noted, it is necessary in the case of an income tax to specify the loss
oset provisions in the tax code. The analytically simpler case of full loss oset
will be treated first and then partial loss oset will be considered.
Full loss oset
To initially simplify the analysis, it will first be assumed that r = 0 so the safe
asset has a zero return. Under this assumption, when the realised return on the
random asset is , the final wealth of the household with income tax tm < 1 is
Y = a + a [1 + [1 tm ]] = + a [1 tm ] ,
so that the expected utility of the household becomes
Z
U ( + a [1 tm ]) () d.
E [U (Y )] =
(7.29)
(7.30)
Assuming that solution to the decision problem is interior, the optimal portfolio
satisfies
[1 tm ] E [U 0 (Y ) ] = 0.
(7.31)
The eect of a change in the tax rate upon the portfolio is found by dierentiating (7.31) with respect to tm . This gives
a
E U 00 (Y )
[1 tm ] a [1 tm ] U 0 (Y ) = 0.
(7.32)
tm
Employing the
h first-order
h condition (7.31)i implies that
i equation (7.32) can be
a
00
reduced to E U (Y ) tm [1 tm ] a [1 tm ] = 0 or
a
a
=
.
tm
1 tm
(7.33)
From (7.33) it can be seen that the holding of the risky asset is increased as the
rate of income tax increases. With full loss oset, an income tax encourages the
household to increase the holding of risky assets.
The surprising aspect of the result in (7.33) is that the response of the
household to the change in taxation is independent of the structure of preferences. It should be noted, however, that this is strongly dependent upon the
assumption that the safe asset has zero return. The reasoning lying behind
(7.33) is that the change in a described there is such that it keeps the mean
final wealth of the household, E [Y ] = + aE [] [1 tm ], the variance of final
wealth, var (Y ) = a2 2 [1 tm ]2 , and all higher moments of the final wealth
distribution constant. Since the riskiness of the portfolio does not increase for
the household, social risk taking must have increased due to the government
holding part of an increased level of risky assets.
213
Figure 7.4: Income Tax with Zero Return on the Safe Asset
In terms of the two-state economy, the result is illustrated in Figure 7.4. The
assumption of zero return on the safe asset implies that point A is not aected
by taxation. Point B moves inward to B 0 . Provided that the change in tax
rate is not so great as to push the household to a corner solution, the original
choice of returns, O, is still available but this now implies a greater holding of
the risky asset. The attainment of O requires the increase in investment in the
risky asset given by (7.33).
When the return to the risky asset is non-zero the result is less clear. The
level of expected utility is given by
Z
E [U (Y )] =
U ( [1 + r [1 tm ]] + a [ r] [1 tm ]) () d.
(7.34)
1
Maximising (7.34) by the choice of a and dierentiating the resulting first-order

condition with respect to tm gives
da
E [U 00 (Y ) [ r]] r
a
i
= h
+
.
2
dtm
1
t
1
tm
m
E U 00 (Y ) [ r]
(7.35)
The first term of (7.35) can be viewed as a wealth, or income, eect on asset
demand whilst the second term is in the nature of a substitution eect. Since
these two conflict, no unequivocal conclusion can be given on the eect of the
income tax on portfolio composition.
Further insight can
h be gained byi observing that the concavity of the utility
2
function implies E U 00 (Y ) [ r] . To evaluate the term E [U 00 (Y ) [ r]]
note that the wealth elasticity of demand for the risky asset is equal to
da
1 + r [1 tm ] E [U 00 (Y ) [ r]]
h
i .
=
d a
1 tm
E U 00 (Y ) [ r]2 a
(7.36)
214
CHAPTER 7. RISK
Figure 7.5: Eect of an Income Tax

Therefore the sign of is opposite to that of the wealth elasticity of demand.
Relating this to (7.35), if the wealth elasticity of demand for the risky asset is
negative, this further increases the level of demand for the risky asset caused
by the tax increase. When the elasticity is positive, the increase in demand is
a
reduced below 1t
. Substituting from (7.36) into (7.35) determines the eect
m
of the tax as
r
da
da
a
=
.
(7.37)
+
dtm
1 + r [1 tm ] d 1 tm
Returning to the two-state case, the eect of the income tax is illustrated in
Figure 7.5. The eect of the income tax is to shift the budget constraint from
AB to A0 B 0 and the optimal choice from O to O0 . As drawn, the proportion of
wealth invested in the risky asset remains constant (the move down the dashed
line from O to O0 ). An altered portfolio composition would be shown by a move
above or below the dashed line.
7.4.3
Imperfect loss osets
When loss-oset provisions are imperfect, taxation has an asymmetric eect

upon the post-tax returns to the household. The government takes a share of
positive returns but carries no responsibility, or at most a limited responsibility,
for any losses. It should therefore be expected that this will reduce the demand
for the risky asset relative to a system with full loss-oset. Due to the existence
of competing income and substitution eects, the validity of such a claim depends upon the basis for comparison adopted (for instance, at constant revenue
or at constant tax rate) and the method of specifying an imperfect loss-oset.
An analysis of imperfect loss-osets can be found in both Mossin (1968)
and Stiglitz (1969a) but the approach taken here is based on the more general
analysis of Eeckhoudt and Hansen (1982). Let the income tax on positive returns
be fixed at tm and denote the tax on negative returns by tn with 0 tn tm .
215
If tn = tm the tax system has perfect loss-oset and if tn = 0 there are no oset
provisions. A move away from full loss-oset can be interpreted as a reduction
in tn .
The return to the household investing amount a in the risky asset is negative
the tax rate tn is applicable. For
if [ a] r + a < 0. Hence for < [a]r
a
other values of returns are taxed at tm . The expected utility of the household
is therefore
E [U (Y )] =
[a]r
a
[a]r
a
U ( [1 + r [1 tn ]] + a [1 tn ] [ r]) () d
U ( [1 + r [1 tm ]] + a [1 tm ] [ r]) () d.
(7.38)
The first-order condition for the choice of a to maximise (7.38) follows as
Z
+
E [U (Y )]
=
a
[a]r
a
1
Z
[a]r
a
U 0 ( [1 + r [1 tn ]] + a [1 tn ] [ r]) [1 tn ] [ r] () d
U ( [1 + r [1 tn ]] + a [1 tm ] [ r]) [1 tm ] [ r] () d.
(7.39)
The negativity of the second-order condition implies that sgn.

n 2
o
E[U (Y )]
. Calculating the latter derivative gives
atn
2 E [U (Y )]
=
atn
[a]r
a
a
tn
= sgn.
[U 0 [ r] + U 00 [1 tn ] [ r] [ a] r + a] () d.
(7.40)
Since the integration is conducted over values of for which [ r] < 0 and
2
E[U(Y )]
[ a] r + a < 0, the concavity of the utility function implies that at
>
n
0. An increase the loss-oset provisions (tn increasing) therefore increases the
households investment in the risky asset. Phrased alternatively, the households
purchases of the risky asset will fall as the loss-oset provisions are reduced.
A similar analysis is conducted by Diamond and Stiglitz (1974) but with
the tax rate adjusted as loss-oset is changed in order to keep utility constant.
The same general result is also shown to hold in that case. With respect to the
eect of changes in the tax rate on positive returns, Mossin (1968) shows that
the result is indeterminate due to the conflict between income and substitution
eects even when the return on the safe asset is zero. Since these results are
straightforward to generate, they are not reproduced here.
216
7.4.4
CHAPTER 7. RISK
Generalisations
The results given above are based on a specification that is restrictive in a

number of directions. The analysis has been entirely partial equilibrium and
the use to which tax revenues were put was not considered. It was assumed
that there was a safe asset; such an asset need not exist. The restriction to two
assets is unjustifiable. Some of the consequences of relaxing these restrictions
will now be considered.
With respect to the use of tax revenue, it can always be assumed that this is
used to purchase a public good and that all households preferences are separable
between final wealth and levels of the public good. Similar reasoning was used
in Chapters 4 and 5. It is possible to consider the implications of tax revenue
being distributed between households but, since no particular mechanism for
redistribution suggests itself as most appropriate and the results are mechanismdependent, few general insights can be gained from this. The partial equilibrium
nature of the results will be relaxed somewhat in Section 6 when the design of
the tax system will be addressed.
Although it may appear innocuous, the assumption that there exists a safe
asset is rather a strong one. An obvious candidate for this role is holding cash
which earns a known nominal return of zero. However, if there is inflation the
real return will be less than the nominal return and, if there is uncertainty about
future levels of inflation, the real return will be uncertain. The same comments
apply to any asset whose nominal return is fixed but whose real return may
vary. Fortunately, the analysis is not critically dependent upon the existence of
a safe asset and can be reworked to include two risky assets.
Introducing further risky assets leads to a number of complications. Firstly,
the extent of risk taking cannot be as easily measured as in the one safe-one
risky economy where the degree of risk taking is given by the proportion of
income invested in the risky asset. Secondly, basic comparative statics results
on the relation of the investment in the risky asset to wealth (with decreasing
absolute risk aversion and increasing relative risk aversion the amount invested
increases with wealth whilst the proportion falls, see Arrow (1965)) no longer
apply except in rather special cases, some of which are noted in Cass and Stiglitz
(1972). As shown by Hart (1975b) with more than one risky asset (plus one
safe) it is generally possible to construct the returns to the risky assets in such
a way that converse comparative statics results can be derived.
Despite these diculties, Sandmo (1977) demonstrates that some results still
apply. Consider a household faced with the choice between n risky assets and
a safe asset that pays return r. Let the return on risky asset k, k = 1, ..., n, be
given by k and the amount that asset purchased be denoted ak so that final
wealth is
n
X
Y = [1 + r [1 tm ]] +
ak [ k r] [1 tm ] .
(7.41)
k=1
The eect of changes in the tax rate upon the demand for asset k is then given
7.5. LABOUR SUPPLY AND OCCUPATIONAL CHOICE
217
by
r
ak
ak
ak
=
, k = 1, ..., n.
+
tm
1 + r [1 tm ]
1 tm
(7.42)
This result has precisely the form of (7.37).
7.5
Labour supply and occupational choice
The choice of portfolio is not the only uncertain decision that a typical household will have to take. It may not be unusual for labour supply decisions to be
taken before the actual level of the wage is realised. This can occur in some
forms of self-employment (such as farming) where production takes place prior
to the return being realised and, even if the nominal wage is known, future price
variation can make the real wage random. An alternative source of uncertainty
can arise in the choice of occupation; in particular between entering paid employment with (possibly) known return and some form of entrepreneurship that
has a random return. In both these examples, the tax system interacts with the
uncertainty in determining household choice.
7.5.1
Labour supply
The analysis of labour supply under certainty is restricted in the number of clear
predictions that it provides about the eects of taxation due to the inherent
conflict between income and substitution eects. It does, however, provide two
clear conclusions. Firstly, if taxation is raised on exogenous non-labour income
then labour supply is increased provided that leisure is non-inferior. Secondly,
income-compensated increases in taxation upon earned income reduce labour
supply.
To see the implications of the introduction of uncertainty for these conclusions, consider augmenting the standard analysis of labour supply by assuming,
as in Eaton and Rosen (1980), that the household does not know the wage
that they will receive but instead holds a probability distribution over possible
wages. Denoting labour supply by x0 and consumption by x1 , the taxpayer
chooses labour supply to maximise expected utility, E [U (x0 , x1 )], given the
constraint that
(7.43)
x1 = wx0 [1 t] + I,
where t is the tax rate and I is lump-sum income. Assuming an interior solution,
the necessary condition for maximisation is
E [U0 + U1 [1 t] w] = 0.
(7.44)
From (7.44), the eect of an increase in lump-sum income can be found to be

E [U01 + U11 [1 t] w]
dx0
=
,
dI
S
(7.45)
218
CHAPTER 7. RISK
where S < 0 is the second-order condition for maximisation. If the utility

function is separable, so that U01 = 0, concavity with respect to consumption
0
implies that U11 < 0 and hence dx
dI < 0. An increase in lump-sum income,
which is received with certainty, therefore reduces labour supply. This result is
in agreement with that which would be derived with a certain wage.
Now consider the eect of an increase in the tax rate that is compensated
for by an increase in lump-sum income. The eect of this upon labour supply
is given by
dx0
dx0
dx0
|comp. =
+ wx0
,
(7.46)
dt
dt
dI
in which wx0 dt, with w the expected wage, is the compensation required for the
tax change dt. Calculating (7.46) gives
E [U1 w + [U11 [1 t] + U01 ] wx0 [w w]]
dx0
|comp. =
.
dt
S
(7.47)
With certainty w = w so that the tax increase would unambiguously reduce

labour supply. In contrast, the eect of uncertainty is to make the sign of
E [U1 w + [U11 [1 t] + U01 ] wx0 [w w]] ambiguous and it is possible for a compensated increase in the tax rate to raise labour supply.
Uncertainty can therefore have the eect of reversing the direction of labour
supply responses to changes in taxation and the results of the certainty analysis
are not directly applicable. The results developed above will be employed in the
analysis of tax design in Section 6.
7.5.2
Occupational choice
The analysis of occupational choice involves a dierent decision framework to

that of the choice of labour supply. As emphasised by Kanbur (1980), occupation is not really a variable that can be continuously adjusted and it is more
reasonable to think of a discrete choice being taken between entering alternative
occupations that diered in the certainty of their returns. The discreteness of
the choice prevents the household obtaining insurance against risk by combining
a portfolio of employment in several occupations. Analyses of the interaction
between taxation and occupational choice are given in Kihlstrom and Laont
(1983) and in Kanbur (1980). The presentation given below is based on the
latter.
Consider an economy in which each of the set of identical households has a
choice between either supplying 1 unit of labour and receiving the competitive
wage w with certainty or becoming an entrepreneur and accepting uncertainty
in earnings. Those households who choose to become entrepreneurs choose the
quantity of labour they employ in order to maximise
E [U (R (y, ) wy)] ,
(7.48)
where U () is the concave utility of income function common to all households, y

is the quantity of labour employed and R (y, ) is the revenue from entrepreneurial activity when value of the random variable is realised. It is the dependence
7.5. LABOUR SUPPLY AND OCCUPATIONAL CHOICE
219
of revenue on and the randomness of that is the source of uncertainty in

the return from entrepreneurship. It is assumed that each household holds a
common belief about the distribution of . The maximisation of (7.48) leads
to a labour demand from each entrepreneur of y = y (w). If the proportion of
the population becoming entrepreneurs is denoted by , an equilibrium for the
economy, in the absence of taxation is given by a wage rate, w,
b and a proportion
b
such that
(i) b
y (w)
b = 1b
,
and
(ii) E [U (R (y (w)
b , ) wy
b (w))]
b = U (w)
b .
Condition (i) is the full employment condition that balances the labour demand
from entrepreneurs with the labour supply of those choosing the risk-free occupation. The second condition equates the expected return from the occupations
so that here is no incentive at equilibrium to change occupation.
One course of action open to a government faced with this economy is the
dierential treatment of income from the risky and the safe activity. If such
dierentiation is feasible, it is natural to investigate which of the two forms of
income should carry the lowest rate of tax. To formalise this, let income from
entrepreneurship be taxed at rate t and that from risk-free labour supply be
subsidised at rate s. Post-tax incomes in the two occupations are then given by
and
Mne = [1 t] [R (y (w)
b , ) wy
b (w)]
b = [1 t] M e ,
Mns = [1 + s] w
b = [1 + s] M s ,
(7.49)
(7.50)
in entrepreneurship and safe labour supply respectively. The analysis now determines the direction that a balanced budget reform of the tax system should
take beginning in an initial position with t = s = 0. Assuming a utilitarian
social welfare function, the level of per-capita welfare is
W = E [U ([1 t] [R (y (w)
b , ) wy
b (w)])]+[1
b
] U ([1 + s] w)
b = U ([1 + s] w)
b ,
(7.51)
since, in equilibrium, E [U ([1 t] [R (y (w)
b , ) wy
b (w)])]
b
= U ([1 + s] w).
b In
order that the budget remains balanced, the tax and subsidy must satisfy the
budget constraint
tE [R (y (w) , ) wy (w)] [1 ] sw = 0.
Hence
"
E[M e ]
wy
E [M e ]
ds =
+t
wy
t
E [M e ]
dt =
dt,
wy
(7.52)
(7.53)
220
CHAPTER 7. RISK
where the second equality follows from evaluating at the initial point with t = 0.
From (7.51) the eect of a balanced budget reform upon social welfare is
dw
dW
ds
0
= U [1 + s]
+w
.
(7.54)
dt
dt
dt
Dierentiating the equilibrium condition guaranteeing equivalence of utilities
and evaluating at t = s = 0 provides the relation
ds
dw
[yE [U 0 ] + U 0 ] + E [M e U 0 ] + wU 0
= 0.
(7.55)
dt
dt
Combining (7.53), (7.54) and (7.55) and evaluating at t = s = 0 provides the
final expression
e
0
e 0
dW
0 E [M ] E [U ] E [M U ]
=U
.
(7.56)
dt
yE [U 0 ] + U 0
The concavity of the utility function implies that marginal utility and income
are negatively correlated; hence E [M e ] E [U 0 ] > E [M e U 0 ]. Using this inequality
shows that (7.56) is positive. From this it can be concluded that starting from
the initial position with zero taxes the government can raise welfare by taxing
the risky occupation whilst subsidising the safe occupation. Furthermore, as
shown by Kanbur (1980), with non-increasing absolute risk aversion and nondecreasing relative risk aversion such a tax reform will reduce the proportion of
the population that chooses the risky occupation. The analysis provides a clear
prediction that, given the use of dierentiated income taxes as available policy
tools, the government should attempt to reduce entry into the risky occupation.
A similar analysis can be conducted to characterise the structure of a linear
income tax that is common to earnings from both occupations. Denoting the
parameters of the income tax by and t, post-tax income is given by
Mni = [1 t] M i , i = e, s.
(7.57)
Mi
Since the marginal rate of tax is t and the average rate

+ t, the tax is
progressive if < 0 and regressive if > 0. The direction of reform from an
initial position with no taxation can be found, as above, by dierentiating the
equilibrium conditions and the revenue constraint and evaluating at = 0, t = 1.
Noting that welfare is a monotonically increasing function of the post tax wage,
Mns , the direction of reform can be found by noting that
dMns
E [RU 0 ] E [R] E [U 0 ]
< 0,
|t=0 =
dt
yE [U 0 ] + U 0
(7.58)
where the negative sign follows from observing that R and U 0 are negatively
correlated. Social welfare is therefore raised by the introduction of a progressive
linear income tax with < 0 and t > 0. This form of tax reduces the risk
carried by entrepreneurs since it raises their raises their income in the worst
state relative to that in the best state.
This analysis of occupational choice is applied to the taxation of profit in
Peck (1989) and is extended to include heterogeneous households in the analysis
of optimal linear income taxation in Boadway et al. (1991).
7.6. OPTIMAL TAXATION
7.6
221
Optimal taxation
In designing optimal taxation in the presence of uncertainty, it is necessary to

face several issues that did not arise in the analysis under certainty. There is
first the appropriate definition of a government budget constraint and whether
this should be satisfied in an expected sense, which was implicitly what was
imposed in the previous section, or whether a separate budget constraint should
be enforced in each state of the world. The choice of policy instruments is also
broader since taxes may potentially be dierentiated across states in addition to
being dierentiated across goods. When risks in the economy are purely private
and uncorrelated across individuals, there is also a role for taxation in providing
social insurance in addition to raising revenue and eecting redistribution.
Given this list of extensions beyond the analysis of certainty, it is clear that
the results given below cannot deal comprehensively with all of these. In fact,
the existing literature has made only initial investigations into each of them and
none is yet treated in an exhaustive fashion. The discussion in divided between
an analysis of tax design with private risk and that with aggregate risk.
7.6.1
Private risk
The majority of the literature on optimal income taxation under uncertainty

has focused upon the social insurance role played by taxation. This has been
achieved either by considering identical households who make a single-period
labour supply decision in the face of uncertainty about the wage they will receive
or by considering households with two-period lives, who may all be identical,
but face uncertainty about the second period of their lives. In both cases income taxation, by providing the households with partial insurance against wage
variations, can increase welfare through a channel that does not exist without
uncertainty.
To develop the first result, the labour supply analysis described in (7.43)
to (7.47) is employed and the optimal linear income tax is characterised for
a population of identical households. With a linear income tax, the budget
constraint of a typical household becomes
x1 = [1 t] wx0 + I,
(7.59)
and the government budget constraint, which is required to hold in an expected

sense, expressed in per-capita terms is
R = tE [wx0 ] .
(7.60)
Assuming that the population is suciently large and employing the fact that
the identical households all choose the same level of labour supply, over the
entire population (7.60) is equivalent to
R = tE [wx0 ] = twx0 ,
where w is the expected wage.
(7.61)
222
CHAPTER 7. RISK
The optimal tax structure can be found by solving

max E [U (x0 , x1 )] = E [U (x0 , [1 t] wx0 + I + twx0 R)] ,
{t}
(7.62)
where x0 = x0 (t, ) = x0 (t, twx0 R). The first-order necessary condition for
(7.62) follows from dierentiation as
x0
x0
E [U0 + U1 [[1 t] w + tw]]
+ wx0
+ U1 [wx0 wx0 ] = 0. (7.63)
t
I
From (7.63), the optimal tax is characterised in the following proposition.
Theorem 38 (Eaton and Rosen) The optimal value of t (0, 1) .
Proof. From (7.44) E [U0 + U1 [[1 t] w]] = 0. In addition, (7.46) shows
x0
0
that x
t + wx0 I is formed as an expectation on the distribution of w. Using
these, (7.63) reduces to
x0
x0
E [U1 tw]
(7.64)
+ wx0
+ U1 [wx0 wx0 ] = 0.
t
I
Evaluating at t = 0 gives
E [U1 [wx0 wx0 ]] = x0 cov (U1 , w) > 0,
(7.65)
1
since dU
dw = U11 x0 < 0. Therefore t = 0 is not optimal. When t 1, x0 = 0
and the first-order condition reduces to

x0
x0
wE U1
+ wx0
< 0,
(7.66)
t
I
x0
0
where the inequality follows from noting that (7.47) implies x
t + wx0 I is
negative when x0 = 0. Finally, any value of t < 0 can be shown by a revealed
preference argument to be dominated by t = 0 since a negative value of t
increases the variability of income.
The important aspects of Theorem 7.2 become clear when it is contrasted
with the case of certainty. Under certainty, the optimal way to raise revenue from
a set of identical households is to employ a lump-sum tax which will raise the
required revenue without introducing any distortions. This can be seen in the
derivations above by noting that certainty implies w = w and that t = 0 then
solves the first-order condition. Viewed this way, the proposition states that
lump-sum taxes are not the most ecient way to collect revenue in the presence
of uncertainty. This seemingly surprising result can be explained by repeating
the observation that income taxation provides partial insurance against the risk
due to the uncertain wage. That the optimal rate of tax is non-zero is then
simply a reflection of the fact that the gain from insurance initially outweighs
the loss of welfare arising from the ineciency caused by the distorting nature
of the tax.
223
An alternative perspective on the social insurance role of income taxation has

been provided by Varian (1980), who considers households that face a two-period
decision problem. In the first period the household works for a known wage and
divides income between consumption and saving. Uncertainty is introduced by
assuming that the level of consumption in the second period is equal to savings
plus a random component with zero mean. The interpretation of this framework
is that initially identical households dier in the luck they have in choosing the
correct form of deposit for their savings. For instance, if dierent portfolios of
stocks were chosen by dierent households, the final value of these would be
expected to dier.
Formalising this discussion, each of the identical households receives a fixed
income of w which is divided between savings, x, and consumption, w x.
Consumption in the second period in the absence of taxation is x + , where
is a random variable with zero mean that is independently and identically
distributed across the population. It is assumed that the government cannot
observe x and separately but can only monitor their sum x + . A tax system
can then be seen to take the form of the imposition of a budget constraint such
that second period consumption is given by c (x + ).
Under the assumption of identical households, the optimal tax system is
chosen to solve the following maximisation program
max U (w x) + E [U (c (x + ))] ,
(7.67)
E [c (x + )] = x,
(7.68)
U 0 (w x) E [U 0 c (x + ) c0 (x + )] = 0.
(7.69)
{c()}
subject to
The constraint (7.68) restricts the tax system to break even in expected terms
and (7.69) is the first-order condition for optimal savings choice by each of the
households.
In the case of a linear consumption function with gradient and intercept
, (7.68) can be solved to write = [1 ] x and (7.69) used to write x = x ().
Together these allow the objective function, (7.67), to be written entirely in
terms of as
max U (w x ()) + E [U ( [x () + ] + [1 ] x ())] .
{}
(7.70)
Setting the derivative of (7.70) equal to zero and employing (7.69) the gradient
of the consumption function can be characterised implicitly by
=
U 0 (w x ()) x0 ()
.
U 0 (w x ()) x0 () E [U 0 (x () + ) ]
(7.71)
Since an increase in reduces U 0 (x () + ), E [U 0 (x () + ) ] < 0. From

(7.71) it is then clear that lies between 0 and 1. The reason for this finding
is once again the social insurance role of the income tax. The numerical calculations presented in Varian (1980) also show that as the degree of risk-aversion
224
CHAPTER 7. RISK
rises so does the level of the marginal rate of tax, an observation which reinforces
the insurance interpretation of the tax system.
In the case of a general nonlinear consumption function, Varian establishes
that the marginal rate of tax must be less than 1, as in the standard income tax
analysis, but that the marginal rate of tax may be non-zero on the household
with the highest value of x + . The latter result follows from the absence of
the disincentive eect in this framework; high second period income is due to
luck rather than eort.
Diamond, Helms and Mirrlees (1980) also consider an economy where individuals have a two-period horizon but introduce uncertainty by making the
introducing the possibility that a given individual will be unable to work in the
second period of their life (for instance through ill health). Since the government cannot distinguish between those who are unable to work in the second
period and those who choose not to work, there is an additional moral hazard
problem in the creation of social insurance to cover the risk in the second period. Numerical simulations for economies based on Cobb-Douglas preferences
show that linear taxation, although having some eect, is of limited value in
providing social insurance.
The analyses of optimal income taxation descried above have focused on the
role of income taxation in providing social insurance in economies with uncertainty. The same conclusion arises from each despite the somewhat dierent
formulations of the source of uncertainty: the social insurance value of income
taxation outweighs its disincentive eects. Therefore, even when no revenue is
to be raised, social welfare can be increased by the introduction of an optimal
linear income tax. Even when revenue is to be raised, the linear income tax is
preferred to a uniform lump-sum tax for a population of identical households.
7.6.2
Aggregate risk
In the presence of aggregate risk taxation cannot have a social insurance role.
This being the case, the focus is now shifted to the design of a tax system that
is ecient in its collection of revenue. To make the situation precise, the singlehousehold economy of Christiansen (1993) is considered in which the household
allocates an exogenous income between a safe and a risky asset. The incomes
accruing from the two assets are tax at dierentiated rates and the aim of the
analysis is to characterise the form of the tax structure. Tax revenue is used to
provide a public good. The aggregate risk is captured in the variable return to
the risky asset. In this framework, taxation aects the division of risk between
private and public consumption.
Denoting the exogenous wealth of the household by and investment in the
risky asset by a, the consumption level of the household in state s is given by
xs = + [1 tr ] r [ a] + [1 t ] s a
(7.72)
where s is the return to the risky asset in state s and tr , t are the tax rates
on the income from the safe and risky asset respectively. The tax rates are
225
chosen subject to the constraint that expected government revenue is equal to

expenditure, g, or
(7.73)
E [tr r [ a] + t a] = g.
The form of the government budget constraint is not inconsequential with uncertainty. Expressing it in terms of expected revenue permits variations in the
level of revenue, and hence expenditure, across states of nature. This formulation therefore leaves an element of risk in public consumption. An alternative
would be to fix the level of expenditure, and required revenue, in each state.
This would remove the uncertainty about the level of public consumption. Except when utility is linear in public consumption or when there is a large number
of taxpayers with uncorrelated risks so that expected and actual tax revenue
are equal, the specification of the budget constraint will aect the optimal tax
system. Further discussion of this issue can be found in Christiansen (1993).
Assuming that the households derives utility from private consumption and
government spending and that the utility function is linear in g, social welfare
is given by
W = E [U ( + [1 tr ] r [ a] + [1 t ] s a)] + E [g] .
(7.74)
The optimal tax rates are found by maximising (7.74) subject to (7.73).
To characterise the tax rates, first note from (7.73) that if expected revenue
is to remain constant, tr and t are related by
a
aE [] + [t E [] tr r] t
dtr
=
.
dt
r [ a]
(7.75)
Dierentiating (7.75) with respect to t whilst varying tr so as to keep revenue

constant gives

dW
0 dtr
E U
r [ a] a = 0.
(7.76)
dt
dt
Substituting (7.75) and the first-order condition for household portfolio choice
E [U 0 ] =
E [U 0 ] [1 tr ] r
,
[1 t ]
(7.77)
into (7.76) and rearranging, the optimal tax rates satisfy
[1t ]
t E [] tr r
1
= ,
[1 t ] E [] [1 tr ] r
b
a
(7.78)
a
is the elasticity of demand for the risk asset with respect
where b
a = a t
to [1 t ]. The characterisation of optimal taxes in (7.78) is in the form of an

inverse elasticities rule in which the inverse of the elasticity is inversely related
to the ratio of the net tax on the risky asset and the net return on the risky
asset.
226
7.7
CHAPTER 7. RISK
Conclusions
In a competitive economy with complete markets, the presence of uncertainty

does not destroy the Pareto optimality of the equilibrium and suitably interpreted versions of the Two Theorems still apply. Indeed, eciency is still
achieved even if for each state there is only a single asset that has a positive
return in that state. Despite these results there may still be a role for the public
sector in an economy with complete markets if the public sector is able to bear
risk more eciently than the private sector. Whether the public sector should
discount for risk at a lower rate than the private sector has been the subject of
some dispute and alternative perspectives on this issue were presented in Section 3. In brief, the treatment of risk by the public sector is dependent upon
the form of the risk that is involved, in the terminology used above whether it
is inside or outside the private sector.
The eects of income and wealth taxation upon portfolio choice were considered and can be viewed as an example of how taxation aects risk taking.
They also illustrated the manner in which the use of taxation turns the government into a partner in a households risk-taking activities. This also emphasised
the distinction between private and social risk taking, with social risk entering
through the variability of government revenue. Similar eects were also noted
in the treatment of taxation and labour supply with an uncertain wage and in
the study of occupational choice.
The design of optimal taxes involves issues that do not arise in an economy
with certainty, in particular the specification of the government budget constraint and the range of policy tools. Furthermore, in addition to the revenueraising and redistributional roles of taxation, there is now the possibility that
taxation can also provide social insurance by transferring from those who are
benefiting from high random returns to those who are suering low returns. This
eect is suciently strong that a distortionary income tax can raise welfare in
an economy of individuals who are identical except for random components in
their wealth. More interestingly, the optimal linear income tax is preferable to
a uniform lump-sum tax.
Chapter 8
Corporate Taxation
8.1
Introduction
The corporation is treated as a separate entity for tax purposes in all developed
countries. It has been subject to numerous tax instruments with a variety of different motivations. The transfers between the corporation and its stockholders
result in the behaviour of the corporation also being influenced by the structure
of the personal tax system, most notably through the favourable tax treatment
of capital gains. The intention of this chapter is to describe the relevant tax
instruments and to determine their eects. This will give an insight into the
many issues that arise in the analysis of corporate taxation.
This chapter is distinguished from those that proceed it by its focus upon the
eects of taxation rather than upon optimisation exercises. There are several
reasons for this. Input taxes have often been employed in many countries and
the eects of such taxes are important because of this, but it has already been
shown that they would not form part of an optimal tax system for a competitive economy. Therefore there is no need for a further study of optimisation. In
simple settings where shareholders exercise direct control, the corporation cannot be identified as a entity distinct from its owners. A coherent tax structure
would then involve a comprehensive income tax on owners, covering all sources
of earnings, with no need for separate taxation of the corporation. Although
the eects of corporation taxation are still of interest in such a framework since
they suggest issues that may arise in more complex settings, optimisation is
again of limited interest. When the setting becomes more complex, and the
existence of managers leads to a separation between ownership and control, the
task of clarifying the eects of taxation is dicult enough, without considering
optimisation.
The next section will discuss the various taxes to which the corporation
has been subject and will consider the rationale for treating the corporation
as a distinct taxable entity. The incidence of a range of taxes will then be
considered in a two-sector general equilibrium economy. This form of economy
227
228
CHAPTER 8. CORPORATE TAXATION
was popular in the 1970s following the work of Harberger (1962) and stills
remains instructive. The incidence results will be derived in the simplest setting
but a number of extensions will also be described. As a tool for studying taxation
of the corporation, the two-sector economy is restricted by its static nature and
its lack of integration with the personal tax system. These features prevent the
study of dividends and bonds and the consequences of preferential tax treatment
of some sources of income. Adopting an intertemporal framework, Section 4 will
consider the eect of taxation upon the policy of the corporation under both
certainty and uncertainty.
8.2
Taxation of the firm
As has already been noted, the corporation (and the firm generally) has been
subject to a range of taxes. This section will describe a number of these but
with particular focus placed upon the corporate profits tax. This mirrors the
emphasis upon this tax in the literature. A brief description will also be given
of how the corporate tax system interacts with the personal tax system.
8.2.1
Input and output taxes
The most prevalent form of input tax has been that levied upon the employment
of labour. In the US, the Social Security tax provides a notable example and
the economics of this are discussed further in Chapter 14. National Insurance
payments play a similar role in the UK. Both the Social Security tax and National Insurance raise the cost of labour for the employer relative to the price of
capital and other inputs. Another example of a tax on labour is the Selective
Employment Tax which was levied in the UK between 1966 and 1973. The
rate of Selective Employment Tax was sector-specific: it taxed employment in
service industries and subsidised it in manufacturing. For further discussion of
the eects of this tax see Reddaway (1970).
Factor subsidies have also been used to promote additional investment. Such
subsidies have the eect of lowering the cost of additional units of capital relative to labour. These subsidies are often provided in the form of depreciation
allowances but cash subsidies to some forms of investment in defined geographical areas were available under the 1972 Industry Act in the UK. The corporate
profits tax has often been interpreted as a tax on capital in the corporate sector.
This interpretation is explained further in 2.3.
Viewing the provision of finance as an input to the corporation, there has
also been dierential treatment of payment to providers of finance. Interest
payments to bondholders may be tax deductible for the firm, in contrast to
dividends which are taxed. Provision of finance by equity holders may lead to
capital gains which are taxed under the personal tax system at a dierent rate
to interest received from bondholdings or from dividends. How these various
provisions aect the choice of financial policy for the firm is investigated in
section 4.
8.2. TAXATION OF THE FIRM
229
The Value Added Tax (VAT) levied by the European Union is essentially a
tax on the output of the firm. The tax is based on the value added in production.
Alternative taxes on output include production and turnover taxes. In contrast
to the VAT, these are based on the gross output of the firm rather than its net
output.
8.2.2
Profit tax
A tax on corporate profits is levied in all developed countries and is significant

in the level of revenue it raises. Its economic eects have also been extensively
analysed and this will be the focus of much of the analysis below. The discussion
here will simply set the scene and point-up some of the issues that are addressed
below.
With full allowance for capital expenditure, the firm will optimise, by choice
of capital and labour, the level of after-tax profits given by
= [1 c ] [pF (K, L) wL rK] ,
(8.1)
where c is the rate of profit taxation, p is the product price, r the rental rate on
capital, L the labour employed and K the level of capital. It is clear from (8.1)
that, provided the tax rate is not greater than 100%, the firms optimal choice of
inputs will be unaected by the imposition of the tax. In this circumstance, the
profit tax will not cause any substitution eects in the pattern of input use by
the firm. This should not be taken as a claim that the tax is completely without
distortion. Since the firms net of tax profit is reduced by the tax, its return
to its owners will fall and this may cause substitution in the asset holdings of
households and changes, for example, in their labour supply.
The results are modified if payments to capital cannot be deducted before
tax. In this case the firm seeks to maximise
= [1 c ] [pF (K, L) wL] rK,
(8.2)
= [1 c ] [pF (K, L) wL rK] ,
(8.3)
pF (K, L) = wL + rK + ,
(8.4)
or equivalently
1
where = 1
. If the tax rate is positive, the corporation tax raises the price of
c
capital relative to that of labour. Denoting total tax payments by , assuming
constant returns to scale it follows that
which is the identity that total income must equal total disbursements. Substituting from (8.4) into (8.2) and using the zero profit identity gives
= [1 c ] [rK + ] rK = 0,
(8.5)
= c rK.
(8.6)
or
230
Given (8.6) it can be seen that with constant returns to scale and with the
non-deductibility of capital costs, the corporate profits tax can be seen equally
as a tax on the value of capital use. Indeed, this is the interpretation that has
been adopted in much of the incidence literature that is discussed in section 3.
8.2.3
Personal taxes
The discussion of corporation tax given above provides a starting point for a
more detailed analysis but, because of the form of most tax codes, the corporation tax cannot be considered in isolation from other aspects of the tax system.
To briefly illustrate this consider a firm about to finance an extra unit of investment. This investment can be paid for either from retained earning, from
additional borrowing or from the issue of new equity. The fact that interest on
borrowing may be tax deductible leads to an obvious incentive to borrow rather
than issue equity. Also, the equity holders of the firm may experience capital
gains and these will be taxed but at a lower rate than dividends. The response
of the firm to corporation tax cannot then be seen in isolation from the capital
gains tax. Furthermore, dividends may be taxed twice: once as profit to the
firm and then as income for a shareholder. The decision of the firm in issuing
dividends must then be considered with the corporation tax. Finally, the static nature of the analysis does not permit the discussion of investment. The
corporation tax, by aecting financial policy, will aect the cost of investment
and this needs to be addressed. Each of these points is addressed in turn in the
remaining sections of the chapter.
8.2.4
Why tax the corporation?
Having made these points about the structure and eects of taxes, there remains
a further issue that must be addressed. This is the reason why the corporation is
taxed at all. If the corporation is seen merely as earning income and transmitting
this to its ultimate owners, then there is no reason why the corporation should
be taxed. Instead, the tax liability should be placed upon its owners alone. Kay
and King (1990) provide a forceful exposition of this viewpoint. This arguments
reflects the view that the corporation does not have a personality or existence
of its own other than that given to it in law.
The alternative perspective is that incorporation carries legal and economic
privileges and that the corporation tax is a tax upon the gains enjoyed from
the benefit of these privileges. Foremost among these privileges is the limited
liability that the shareholders in the corporation enjoy in the event of bankruptcy. Another possible view, and one reflected in US Tax Reform Act of 1986
which shifted the tax burden from the personal sector to the corporate sector,
is that corporations can aord to pay taxes and should therefore carry their
share of the burden. There is also the argument, already explored above, that
corporation tax is taxing rent so is a distortion-free way of raising revenue. As
already noted, there are limits to how far this argument can be pushed since it
relies at the very least on the tax being levied on true economic profit.
8.3. TAX INCIDENCE
231
Ultimately, the eect of a tax depends upon how it aects the individuals
in the economy and the correctness, or otherwise, of taxing the corporation
depends upon the final incidence of the tax. If the tax can achieve objectives,
and so raise social welfare, that other taxes cannot then there is a justification
for its existence. In a sense, many of the arguments noted above simply direct
attention away from the main justification for introducing any form of taxation,
which is to achieve specified aims. As is always the case in Second-Best theory,
although a policy instrument may have no role in the First-Best, its use may
still be justified in other circumstances.
8.3
Tax incidence
One aspect that has already been stressed in earlier chapters is that the economic
incidence of a tax is rarely the same as the legal incidence. Legal incidence
relates to who has to formally pay the tax to the tax collection agency whereas
economic incidence is identified by the agents who suer reduced welfare due
to the imposition of the tax. Since there are general equilibrium repercussions
to any tax change, the identification of economic incidence is not always a
straightforward exercise.
In terms of the corporation tax, Harberger (1962) was the first to present
a comprehensive analysis of incidence. The framework employed was that of a
two-sector economy with two factors of production: capital and labour. One
of the sectors of the economy was treated as incorporated and capital in that
sector bore the legal incidence of the corporation tax. This framework permitted
identification of the real eects of the corporation tax in terms of an output
eect, capturing the change in the relative outputs of the two goods, and an
input substitution eect, representing the adjustment of inputs within sectors.
In addition, Harberger also calculated the change in relative factors rewards in
order to determine the economic incidence of the tax in terms of whether the
tax burden fell on capital or labour.
In the period since its publication, the Harberger analysis has been subject to
many extensions and modifications, many of which are detailed in the surveys of
Mieszkowski (1969) and McLure (1975). The analysis of this section will describe
the Harberger economy and show how this can be employed to derive results on
the incidence of the corporation tax and a range of other tax instruments. Some
of the extensions will then be described. The analytical technique used to solve
the incidence question follows closely the original development of Harberger
(1962) and its description in Shoven and Whalley (1972).
8.3.1
Tax incidence in the Harberger economy
The Harberger economy is competitive with two goods, denoted 1 and 2, and
two factors of production, capital and labour. The factors of production are in
fixed supply. The initial equilibrium for the economy is defined in the absence
of taxation and the eect of introducing infinitesimal taxes is then considered.
232
This point must be stressed since it has not always been clarified in some of the
literature. The extension of the analysis to finite taxes is described in 3.2.1.
To simplify the analysis by eliminating income eects, it is assumed that the
government spends the tax revenue it receives in the same way that consumers
would have done in the absence of taxes. In conjunction with the fact that taxes
are infinitesimal, this implies that income does not need to be considered as a
determinant of demand. The demand for good 1 is further assumed to depend
only upon the ratio of consumers prices or

q1
X1 = X1
.
(8.7)
q2
Since demand is determined by the price ratio, the underlying utility function
must be homothetic; see Rapanos (1991). Denoting the compensated elasticity
of demand by , total dierentiation of (8.7) gives

q1
d
q2
dX1
= q1 .
(8.8)
X1
q2
The elasticity appearing in (8.8) is compensated since the marginal propensities to consume of households and government are assumed equal so that the
income eect in government demand exactly osets that in consumer demand.
This leaves only the substitution eect. Units of measurement of the commodities are chosen so that the initial consumer prices are equal to unity. Equation
(8.8) can then be approximated by
dX1
= [dq1 dq2 ] .
X1
(8.9)
Production in sector i is determined by the dierentiable and homogeneous

of degree 1 production function
Xi = F i (Ki , Li ) .
(8.10)
Total dierentiation of the production relation (8.10) gives

dXi =
F i
F i
dKi +
dLi ,
Ki
Li
(8.11)
which can be divided by Xi to read

dXi
i
= fK
dKi + fLi dLi ,
Xi
where
i
fK
F i
Ki Ki
Xi
(8.12)
(8.13)
is the relative share of capital in output (or the partial elasticity of output with
respect to capital) and fLi is defined similarly.
8.3. TAX INCIDENCE
233
Let rbi be the price paid by users of capital in sector i and w

bi be the price paid
for labour. The definition of the elasticity of substitution, si , between inputs
then implies

i
d K
d wrbbii
Li
= si rbi .
(8.14)
Ki
w
bi
Li
The set of input taxes that are to be considered include sector-specific taxes
and general taxes on capital and labour. Incorporating these taxes into factor
prices and taking an approximation linearises (8.14) to
dLi
dKi
= si [dr + dTK1 + dTK dw dTL1 dTL ]

Ki
Li
(8.15)
where r is the price received by owners of capital, TK1 the sector-specific tax on
capital and TK the general tax on capital. The notation is interpreted similarly
for the taxes TL1 and TL on labour. Units of measurement are also chosen for
inputs so that initial prices are unity. Given this, the tax rates can be interpreted
as being either be ad valorem or specific since, for example, the dierential of
r [1 + TK1 ] equals dr + dTK1 when r = 1 and TK1 = 0 which is equal to the
derivative of r + TK1 .
The assumptions of competitive behaviour and homogeneity of the production functions imply that factor payments are equal to revenue in both sectors.
Hence
pi Xi = w
bi Li + rbi Ki ,
(8.16)
which when totally dierentiated gives
bi dLi + Li dw
bi + rbi dKi + Ki db
ri .
pi dXi + Xi dpi = w
(8.17)
Profit-maximising input choice equates the marginal revenue product of the

input to its price so
F i
rbi
= ,
(8.18)
Ki
pi
and
w
bi
F i
=
.
(8.19)
Li
pi
Substituting (8.18) and (8.19) into (8.11) gives
dXi =
Given (8.20), (8.17) reduces to
dpi =
w
bi
rbi
dLi + dKi .
pi
pi
Li
Ki
dw
bi +
db
ri .
Xi
Xi
(8.20)
(8.21)
As initial input prices are unity, (8.13) and (8.18) imply that at the initial
Li
i
i
i
equilibrium fK
=K
Xi and (8.20) that fL = Xi . Hence (8.21) becomes
i
dpi = fLi dw
bi + fK
db
ri .
(8.22)
234
Incorporating taxes (8.22) can be written

i
dpi = fLi [dw + dTLi + dTL ] + fK
[dr + dTKi + dTK ] + dTci + dTc ,
(8.23)
where Tci is the sector-specific consumption tax and Tc the general consumption
tax.
This completes the initial derivations. The solution for the eect of the
introduction of the taxes into the economy is given by equations (8.9), (8.12),
(8.15) and (8.23). These equations will shortly be solved but first a preliminary
result is given that shows how various sets of taxes are equivalent in the sense
that they have the same eect upon the equilibrium. This result is stated as
Theorem ??.
Theorem 39 When levied at the same rate the following equivalencies hold:
(i) TKi and TLi to Tci , i = 1, 2;
(ii) TK and TL to Tc ;
and
(iii) Tc1 and Tc2 to Tc .
Proof. To prove (i) it is first noted that taxes only appear in (8.15) and
(8.23). When introduced at the same rate, dTKi = dTLi , and the taxes cancel
I
from (8.15). From Eulers theorem fLi +fK
= 1 so that from (8.23) the increases
in dTKi and dTLi are equivalent to an equal increase in dTci . This establishes
that the eect of raising dTKi and dTLi is identical to that of raising dTci .
The same argument applies to show the equivalence of TK and TL to Tc ,
where it must be noted that the change in rates aects both sectors. Finally
(iii) is proved by noting that the tax changes involved only aect (8.23).
The explanation lying behind (i) of Theorem 13.1 is that the pair of tax
changes TKi and TLi together raise the price of factors equally for sector i, so
that no substitution between factors occurs. The consequent increase in costs
has the same eect upon the equilibrium as a consumption tax Tci . In the case
of (ii), the pair of taxes raises the price of inputs equally in both sectors and so
is equivalent to a general consumption tax. The same general reasoning applies
to the pair of consumption taxes in (iii).
The equivalence demonstrated in Theorem 8.1 shows that only a restricted
set of taxes need be considered. To maintain the focus upon corporation taxation, the taxes upon labour will now be set at zero. In addition, the general
consumption tax will not be considered. In the present context with no saving
and fixed labour supply, the general consumption tax is equivalent to an income
tax. A change in its value leads only to an income eect and not a substitution
eect so that it is of limited interest as a source of ineciency. Due to the equivalence result, it is also assumed that dTc2 = 0. To understand the reason for the
final restriction, it is best to return to the original motivation for the analysis.
Harberger (1962) was concerned with the distortion that would arise through
the taxation of capital in the corporate sector while capital in the non-corporate
sector remained untaxed. Under this interpretation, only one of the two sectors
should bear a corporation tax. Correspondingly, it is assumed that sector 1 can
8.3. TAX INCIDENCE
235
be treated as incorporated but sector 2 cannot. Hence the capital tax in sector
2, dTK2 , is set at zero. Finally, the wage rate is chosen as numeraire so that
dw = 0 throughout.
Adopting these restrictions and substituting (13.23) for i = 1, 2 into (13.9)
provides
1
dX1
2
= fK
[dr + dTK1 + dTK ] + dTc1 fK
dr .
X1
Equating (8.24) to (8.12) and rearranging
1
2
dL1
1
1 dK1
fK
+ fL1
= fK
[dTK1 + dTK ] + dTc1 .
dr + fK
fK
K1
L1
(8.24)
(8.25)
Since the supply of factors is fixed, it follows that dK2 = dK1 and dL2 = dL1 .
Employing these facts, (8.15) for i = 2 can be transformed to
s2 dr +
K1 dK1 L1 dL1
= 0,
K2 K1
L2 L1
(8.26)
and for i = 1 rearrangement of (8.15) gives

s1 dr +
dK1
dL1
= s1 dTK1 + s1 dTK .
K1
L1
(8.27)
Taken together, (8.25), (8.26) and (8.27) provide a three-equation system determining the dependence of the endogenous variables dr, dK1 and dL1 upon the
tax changes dTK1 , dTK and dTc1 .
Taking the sector-specific capital tax first, it follows from the solution of the
system that
L1
K1
dr
K1
1
1 L1
= fK
+ fL1
+ s1 fK
A,
(8.28)
dTK1
K2
L2
L2
K2
where
2
K1 L1
K1
1
1 L1
A = fK
fK
+ fL1
.
s2 s1 fK
K2 L2
L2
K2
(8.29)
To interpret (8.28) it is first necessary to establish that A is positive. To

see this, note that , s1 and s2 are negative. Now if sector 2 is relatively
L1
2
1
1
capital intensive then fK
> fK
and K
K2 < L2 . The converse inequalities apply
if sector 1 is the most capital intensive. In either case the product of the two
terms is negative thus ensuring A is positive. The next point to note is that
since it has been assumed that dw = 0, a negative value of dr implies that
the imposition of the tax reduces the return to capital relative to that of labour
whilst a positive value indicates an increase in the relative return of capital. The
input substitution and output eects of the tax can clearly be seen in (8.28).
The output eect is determined by the elasticity of demand whilst the input
substitution eect is dependent upon value of the elasticity of substitution.
Given these preliminaries, the next theorem follows from (8.28).
236
Theorem 40 (i) dTdrK can only be positive if sector 1 is more capital intensive
1
than sector 2;
(ii) If |s1 | > || then dTdrK < 0;
1
(iii) As |s2 | increases without limit,
(iv) Assume
K1
L1
K2
L2 .
dr
dTK1
tends to zero;
Then if |s1 | = |s2 |, it follows that
and capital bears the full burden of the tax. If |s1 | > |s2 | , then
dr
dTK1
K1
K1 +K2 ;
dr
dTK1
dr
dTK1
=
>
K1
K1 +K2
K1
K1 +K2
and if |s1 | < |s2 | , then

<
(v) If |s1 | = |s2 | = 0, then the incidence of the tax depends only on factor
proportions.
h
i
L1
1 K1
Proof. (i) In (13.28), only the term fK
K2 L2 can be positive. Hence
L1
1
for dTdrK > 0 it is necessary that K
K2 L2 < 0, which is the condition that sector
1
1 is more labour intensive.
dr
1 L1
1 L1
(ii) If |s1 | > || then s1 fK
L2 fK L2 < 0 so dTK1 < 0.
(iii) This follows from observing that s2 only appears in the denominator of
(8.28) so, all else equal, an increase in s2 reduces the absolute value of dTdrK .
1
(iv) The expressions are obvious after substitution. Their interpretation
follows from noting that the total tax burden is K1 dTK1 so if the price of capital
1
net of tax falls by dr = K1K+K
dTK1 , then capital has borne the entire tax.
2
(v) From substitution.
It is the calculations of (iv) of Theorem 8.2 that describe the means of cal1
culating the economic incidence of the tax. Since a value of dr = K1K+K
dTK1
2
implies that capital bears the entire burden of the tax, incidence can be determined by contrasting this with the value of dTdrK given by (8.28). If the value in
1
(8.28) is higher then labour is also carrying some of the burden of the tax. What
(iv) achieves is to provide the answer to the incidence question under a specific
set of assumptions. Following the same line of enquiry, Harberger substituted
into (8.28) employing values from empirical studies in the US and reached the
conclusion that capital generally bears the entire tax burden. The value of this
conclusion is naturally restricted by the assumptions made about the economy
in which it is derived. Much of the work following Harberger was concerned
with testing the result under alternative assumptions. Some of this work will
be described in Section 3.2.
Returning to the analysis of the present economy, it can easily be calculated
that the introduction of the capital tax in sector 1 reduces employment of capital
in that sector. The eect upon labour use in sector 1 is given by
2
K1
dL1
K1
1
1
= L1 s1 s2 fK
fK
fL1
s2 + s1
s1 fK
A.
(8.30)
dTK1
K2
K2
In general, the change in labour use cannot be signed because of the conflicting
input substitution and output eects but (8.30) permits the eect of various
sets of restrictions on the elasticity of demand and elasticity of substitution
to be tested. As an example, when the elasticity of demand is zero, labour
8.3. TAX INCIDENCE
237
demand in sector 1 increases as labour is substituted for capital. The eect of

the sector-specific output tax upon the return to capital is
K1
L1
dr
=
A,
(8.31)
dTc1
K2 KL2
so that whenever the elasticity of demand is non-zero the change in the price of
capital depends entirely upon the relative capital intensity of sector 1.
These results are intended to be indicative of what can be achieved in the
context of the two-sector Harberger economy. Although only a limited number of
taxes have been considered here, the same methods can determine the incidence
eect of a broad range of tax instruments. However, the assumptions of the
economy are rather restrictive and more is to be gained by relaxing these than
by extending the set of taxes.
8.3.2
Extensions
Since the publication of Harberger (1962), many extensions of the analysis have
been completed. Included among these are studies that have simply addressed
the same issues but under alternative assumptions and others that have applied
the analysis, often under very similar assumptions, to dierent issues. A survey
of the latter, which will not be covered here, can be found in McLure (1975).
In addition, Jones (1965, 1971) and McLure (1974) have developed alternative
analytical techniques for studying the Harberger economy. The extensions that
are now discussed are those that modify the structure of the Harberger economy
in order to relax the restrictiveness of its assumptions.
Income eects
The discussion of the demand function (8.7) described how income eects were
eliminated by the assumptions that infinitesimal taxes were introduced from an
initial position with no taxation and that the government spent the tax revenue
in the same way as consumers. This assumption has been relaxed in two ways.
Mieszkowski (1967) considered demand to be derived from two distinct groups of
consumers whilst Ballentine and Eris (1975) retain the Harberger specification
but incorporate income eects for non-infinitesimal tax changes.
The consumers in Mieszkowski (1967) are comprised of workers, who earn
income from the supply of labour, and capitalists, who receive the return from
capital. In such a framework, a number of anomalies may arise when compared
to the standard Harberger analysis. For instance, when these two groups have
very divergent spending propensities and the elasticities of substitution in production are small, a tax on capital in the capital intensive sector may increase
demand for that sectors commodity and increase the price of capital relative
to labour. Since the same general principals are involved, this analysis also
suggests the possibilities that would arise if the government did not spend its
revenue in the same way as consumers.
238
Ballentine and Eris (1975) write the underlying demand function as

X1 = X1 (q1 , q2 , rK + L + rtK1 K1 ) ,
(8.32)
where rK + L is the income of consumers and rtK1 K1 the revenue of the government from the tax on capital in sector 1. The dependence of demand on
total income implies the assumption that government and consumer spending
patterns are identical. Taking the total derivative of (8.32) gives
dq1 dq2
m
dX1
=
[[K + tK1 K1 ] dr + rK1 dtK1 X1 dq1 X2 dq2 ] ,

+
X1
q1
q2
X1
(8.33)
where m is the marginal propensity to consume good 1. Solving the system
with (8.9) replaced by (8.33) gives the result
1
K1 A + s1 B mq1 CK1 s1 L
dr 1 + tK1
L2
h
i,
=
r dtK1
1
[K2 K1 ] A s1 B s2 + mq1 CK1 s1 L
+
s
2
L2
where Ki =
tK1
1+tK1
r [1+tKi ]Ki
,
qi Xi
A=
K1
K2
L1
L2 ,
L1
1
B = L1 K
K2 + K1 L2 , L1 =
(8.34)
wL1
q1 X1
and
.
C=
The value of (8.34) is that it permits the incorporation of finite taxes into
the analysis. For zero initial taxes, it reduces to (8.28). It follows from (8.34)
that when sector 1 is capital intensive, capital will bear more of the burden of
any finite tax than labour (i.e. dr < 0). This extends (i) of Theorem 8.2 to
finite initial taxes. Similarly (iii) and (v) of Theorem 8.2 extend to finite taxes
but (ii) and (iv) do not. In addition, (8.34) shows that if s1 = s2 = 1 and
the elasticity of substitution in consumption, , also equals - 1, then the fall in
payment to capital due to the tax exactly equals government revenue. To see
this, note that the homotheticity of demand implies
mq1 =
q1 X1
,
q1 X1 + q2 X2
(8.35)
and
q2 X2
.
[q1 X1 + q2 X2 ]
With these restrictions, (8.34) reduces to
=
K1
dr 1 + tK1
=
,
r dtK1
K
(8.36)
(8.37)
so
dr 1 + tK1
rK
r
dr
K=
=
.
(8.38)
dtK1
r dtK1 1 + tK1
1 + tK1
The eect of the tax change upon government revenue, R = rtK1 K1 , is
dR
dtK1
dr
dK
+ tK1 r
(8.39)
dtK1
dtK1
dr 1 + tK1 rtK1 K
s2 L1 s1
rtK1 K L1 s1
1
= rK1 +
(8.40)
.
r dtK1 1 + tK1
A
L2 A
1 + tK1 L2 A
= rK1 + tK1 K1
8.3. TAX INCIDENCE
239
Substituting into (8.39) using (8.37) and the restriction that s1 = s2 = 1 gives
rK
dR
=
,
dtK1
1 + tK1
(8.41)
thus demonstrating that the fall in income to capital is equal to government

revenue.
Imperfect competition
Monopoly was first introduced into the analysis by Harberger in the original 1962
article. This was achieved by including a mark-up, representing the presence
of some monopoly element, in the pricing equation for sector 1. Doing this for
i = 1, (8.23) becomes
i
dqi = fLi [dw + dTLi + dTL ] + fK
[dr + dTKi + dTK ] + dTci + dTc [1 + ] ,
(8.42)
where is the monopoly mark-up. Although this is a straightforward method of
incorporating imperfect competition, it is not entirely persuasive since it omits
the optimisation of firms in the determination of the mark-up. In addition,
it also changes the nature of the sector-specific tax on capital. It was argued
in Section 2 that this could be viewed as a corporation tax in a competitive
economy with constant returns to scale. The introduction of a mark-up in the
pricing equation now implies the existence of pure profits which, in turn, prevent
this interpretation of the sector-specific capital tax.
With the monopoly mark-up included, the eect of the capital tax upon the
return to capital is given by
dr
K1 L1
1
1 L1
1 K1
b
= fK [1 + ]
+ fL
+ s1 fK
A,
(8.43)
dTK1
K2 L2
L2
K2
where
1
2
K1 L1
K1
1
1 L1
b = fK
A
fK
[1 + ]
+ fL1
s2 s1 fK
K2 L2
L2
K2
(8.44)
Comparison of (8.42) - (8.43) with (8.28) - (8.29) shows that the inclusion of
the monopoly mark-up does not aect in any fundamental way the content of
Theorem 8.2.
An alternative formulation of imperfect competition is given in Atkinson
and Stiglitz (1980) but this is based on a particular specification of monopolistic competition with a precise functional form for utility. Although instructive,
it is not able to address the full range of issues associated with imperfect competition. A more general presentation of monopolistic competition is given and
Anderson and Ballentine (1976). They conclude that the existence of imperfect
competition does not much alter the incidence of taxation, concurring with the
findings of Harberger (1962), and that compared to the competitive case, imperfect competition amplifies the welfare loss caused by the introduction of a
distortionary taxation.
240
In contrast to these static analyses, Davidson and Martin (1985) study a

dynamic economy in which collusion can arise in the repeated game played by
the imperfectly competitive firms. The output level in the collusive equilibrium
is partly determined by the need to make deviation from the collusive level an
unprofitable strategy. Deviation is rewarded by increased profits in the period
in which it takes place followed by reduced profits in all future periods as the
firms switch to punishment strategies. Full details of this form of equilibrium
can be found in Friedman (1977). Whether deviation is profitable then depends
on the rate at which future profits are discounted. In fact, as the discount rate
rises so must the collusive output level. The relation of this to the incidence
of capital taxation is due to the fact that the firms will use the net return to
capital, r, as the discount rate. The corporation tax then aects r in the manner
already analysed above and this eect then leads, via changes in the collusive
output level, to changes in other factor and product prices. Consequently, even
though capital may be in fixed supply, a general tax on capital which applies
to all sectors is shifted rather than absorbed by capital because of the eect of
the discount rate on the collusive equilibrium output level.
Intermediate goods
The main eect of the introduction of intermediate goods into the Harberger
economy is to increase the possibilities for substitution between inputs. Unless
the technology is Leontief with fixed input coecients, these substitution possibilities modify the conclusions on tax incidence. In a series of papers, Bhatia
(1981, 1982a, 1982b, 1986) has addressed these issues.
In an economy where there is a single intermediate good which is not used
for final consumption, Bhatia (1981) shows that (i), (iii) and (v) of Theorem
8.2 still apply to the extended economy but the (ii) cannot be proved in the
presence of intermediate goods. In addition to this, it is also proved that when
factor proportions in the final goods industries are the same, an increase in
the sector-specific capital tax will burden capital relatively more than labour
so long as capital and labour, and capital and the intermediate good, are not
complementary to each other in the taxed sector. Bhatia (1982b) considers
the contrast between a corporation tax and a VAT. It is shown that neither of
these tax instruments aects relative prices if input coecients are fixed and
the elasticity of demand is zero. When input coecients are variable, only a
VAT applied at an equal rate to all industries is neutral.
8.3.3
Taxation and finance
An important issue in the study of corporate taxation is the question of how

the tax system aects the financing of a firm and the investment plans of the
firm. To finance investment, a firm has three sources of finance. The firm can
issue new equity, it can issue bonds or it can employ retained earnings. The
issue of new equity makes the firm liable for future dividend payments, but does
not commit the firm to any specific level of payment, whereas bonds involve a
8.3. TAX INCIDENCE
241
fixed commitment to pay interest and, eventually, to redeem the bonds. A profitmaximising firm will naturally wish to choose the mix of these three instruments
that finance the investment at minimum cost. Furthermore, the extent of future
investment will in turn be determined by the cost of finance, usually referred to
as the cost of capital. The focus is therefore upon how the tax system influences
the means of finance and the cost of capital.
Most of the issues involved can be treated adequately under the assumption
of certainty with all agents in the economy fully informed of the future prospects
of the firm. However, the assumption of certainty does imply some restrictions.
With certainty, there can be no possibility of any firm becoming bankrupt since
such a firm would simply not operate. In contrast, in an economy with uncertainty there may be some states of nature in which a firm is unable to meet its
obligations, essentially the contratual payments to bondholders, and therefore
has to go into bankruptcy. The consequences of this will be discussed further
below. A second issue that arises in the presence of uncertainty is that of the
objective of the firm. In an economy with certainty, there will be unanimous
agreement of the shareholders that the firm should maximise its profits. This
need not be the case with uncertainty if there are incomplete markets. Generally, the problem facing the firm is that of aggregating the diverse preferences of
its shareholders into a single objective. One possible resolution of this problem
has already been noted in Section 3 of Chapter 7. The issues that arise with
uncertainty have not been fully resolved and this limits what can be said about
the eects of taxation.
The following section will discuss the essential aspects of the tax system that
will be analysed. The dierential tax treatment of dierent forms of transfer
from the corporate to the private sector imply that the analysis must involve
an integrated treatment of both corporate and personal taxation. The eects of
taxation are then derived under the assumption of certainty with an emphasis
upon the financial decisions of the firm. Uncertainty will then be introduced
and the Modigliani-Miller theorem proved. Some further aspects of taxation
will then be discussed.
8.3.4
Systems of corporate and personal taxation
The tax system that is now described is commonly termed the classical system
and is in use in the US and many other countries. This is to be distinguished
for the imputation system used in the UK and the two-rate system; these are
described later.
The motivation behind the classical system is that the corporate tax is a tax
on the benefits that follow from incorporation. As such, the tax liability of the
corporation is treated as entirely distinct from that of the shareholders of the
company. Consequently, profits are taxed at the rate set for corporation tax,
dividends are taxed at the personal income tax rate applicable to the shareholders who receive them, as is interest received by the bondholders of the firm, and
a separate rate applies to capital gains which is levied on realisation of those
gains. Interest paid by the firm is tax-deductible. Many of the consequences
242
of this tax system with respect to corporate finance follow from the distortions
introduced by the dierential tax treatment of dividends and interest payments.
In what follows, the rate of corporation tax will be denoted c , dividends are
taxed at the personal rate p and capital gains at the rate g . To reflect the
reality of tax codes, it is assumed that g < p .
One of the perceived diculties of the classical system is the double taxation
of dividends: they are taxed once as corporate profit and then again as personal
income. The imputation system represents an attempt to avoid this double
taxation by integrating the corporate and personal tax systems. It does this by
giving each shareholder a credit for the tax paid by the company on the profit
out of which dividends are paid. In essence, any profits distributed as dividends
are deemed to have already been subject to personal tax at what is known as the
rate of imputation. The shareholder receiving the dividend is then only liable
for the dierence between the rate of imputation and their personal tax rate. In
the UK the rate of imputation is equal to the standard rate of tax. A further
alternative system that has been employed in the UK and in West Germany
and Japan is the two-rate system. Under this system dierent tax rates apply
to distributed and undistributed profits with the latter being taxed at a higher
rate. This is designed to partly oset the double taxation of dividends inherent
in the classical system. King (1977) provides further discussion of alternative
systems of corporation tax.
8.3.5
Finance and investment with certainty
The study of the eect of taxation upon the corporation has gradually developed from the initial static analysis of Stiglitz (1973, 1976) and King (1975)
through to fully intertemporal presentations such as Auerbach (1979) and Bradford (1981). Results derived in a static setting can be instructive but are unable
to capture many aspects of the problem. For example, one of the notable features
of the personal tax system is the favourable treatment of capital gains relative
to other sources of income but, since capital gains are essentially intertemporal
in nature, this aspect cannot be capture in a static framework. The same can
be said of the more general fact that the value of the firm is determined by
the flow of funds it provides to the household sector. The approach taken here
will be to focus entirely upon the intertemporal analysis originally developed in
Auerbach (1979) and extended in Auerbach and King (1983).
To motivate the approach taken to the analysis of corporation taxation, it is
worth first looking at the investment decision in a simple two-period economy.
Consider a firm that is entirely owned by a single household. Assume that the
firm can undertake an investment project which requires the input of 1 unit
of the economys single commodity. Under what conditions should the firm
make the investment? If the household faces a price ratio of pp12 = 1 + r for
consumption in the present period (good 1) against consumption in the future
period (good 2) and can lend or borrow as it likes at interest rate r, then the firm
should undertake the investment project if the return in units of consumption
in period 2 is greater than 1 + r. If this inequality is satisfied then the firm
8.3. TAX INCIDENCE
243
is able to provide, via the investment project, units of good 2 more cheaply
than the consumer can purchase them on the market. Eectively, undertaking
the investment expands the budget set of the firms owner. Furthermore, since
the household will be maximising utility subject to the intertemporal budget
constraint, the rate of interest will also be equal to the households personal
rate of time preference.
In this context, the interest rate is termed the cost of capital and plays a
key role in determining the choice of investment projects. The conclusion of
the example illustrates a general truth about the investment decision: the firm
should undertake any investment that gives a rate of return of at least the
rate of interest. In addition, the firm should continue making investments until
the marginal investment just achieves the rate of interest. These observations
underline the role played by the cost of capital and are now applied to the
analysis of corporate and personal taxation. Further discussion of the cost of
capital from this perspective can be found in Auerbach (1983) who also considers
the eect of imperfections in capital markets.
No personal taxes
The problem of firm financial policy and investment strategy in the presence
of a corporation tax is first approached in the absence of personal taxes. The
analysis assumes that the firm can finance future investment through either
issuing new equity, by issuing bonds or through the use of retained earnings.
The timing of transactions is as follows. At the beginning of each period the
firm distributes dividends to the shareholders of the previous period, it pays
interest on its debt and repays the principal (the bonds issued by the firm are
assumed to have a term of 1 period) and sells new shares ex-dividend. Interest
payments are tax deductible for the firm. The actions of the firm in period t
are chosen so as to maximise the wealth of those who hold shares in the firm at
the beginning of period t. The construction that follows determines the level of
wealth and its dependency upon the structure of the tax system.
At the beginning of period t the firm chooses an investment policy, a debt
policy and an equity policy. Together these determine its dividend policy. An
investment policy is characterised by the set of cash flows that arise as a result
of the investment rather than by looking directly at the investment itself. Such
an investment policy is denoted by a vector z = (zt , zt+1 , zt+2 , ...) where zt is the
firms cash flow at the beginning of period t net of corporation tax. Since bonds
have a term of 1 period, a bond policy, B = (Bt , Bt+1 , Bt+2 , ...), describes both
the number of bonds that will be issued in each period and the total stock of debt
n
n
of the firm in that period. An equity policy is denoted V n = Vtn , Vt+1
, Vt+2
, ...
n
with each Vt being the value of new equity sold at the start of t. It follows that
these policies imply the dividend policy through the identity
n
Dt = zt+1 + Bt+1 + Vt+1
[1 + it [1 c ]] Bt ,
(8.45)
where Dt is the value of dividends paid at the beginning of period t + 1 to the

shareholders of period t and it is the interest rate paid to bondholders.
244
After the payments of dividends in period t, the value of the firms equity
is denoted Vto . Following the sale of new equity the value of the total stock of
equity is
Vt = Vto + Vtn .
(8.46)
In period s t, some of the dividends paid will go to equity issued before the
beginning of t whilst the remainder will be received by holders of equity issued
from t onwards. The proportion of equity in period s that is in existence before
the beginning of t is
st = [1 t ] 1 t+1 ... [1 s ] ,
(8.47)
Vn
with t = Vtt . In an ecient market, the value of the firm will equal the
discounted value of future dividends. Hence, letting t be the discount rate
applied by equity holders, the value of the equity at the beginning of t is
" s
#
X
Y
1
o
[1 + u ]
(8.48)
st Ds .
Vt =
s=t
u=t
Now noting that (8.46) and (8.47) can be used to write

Vto =
and from (8.45)
1 t
o
Dt + Vt+1
,
1 + t
1 + t o
V = [1 + t ] Vt ,
1 t t
(8.49)
(8.50)
it follows that
o
t Vt = Dt + Vt+1
Vt .
(8.51)
The firm aims to maximise the wealth of existing shareholders and this
wealth is given by
Wto = Vto + Dt1 ,
(8.52)
which after substituting from (8.44) and (8.45) is equal to
Wto = Vt + zt + Bt [1 + it1 [1 c ]] Bt1 .
(8.53)
As the last term of (8.52) is pre-determined at t, maximising Wto is equivalent

to maximising
cto = Vt + zt + Bt .
W
(8.54)
The firm therefore acts to maximise the sum of security market value (equity
plus bonds) plus current cash flow. From (8.50)
Vt =
1
[Vt+1 + At ] ,
1 + t
(8.55)
8.3. TAX INCIDENCE
245
n
where At = Dt Vt+1
, which on solving for Vt gives
" s
#
X
Y
1
[1 + u ]
As .
Vt =
s=t
(8.56)
u=t
The result in (8.55) shows that the value of the firm is independent of the
level of dividends and new equity issues. From the definition of At it can be seen
that a one unit increase in dividends financed by a unit increase in the value of
new equity has no eect on the value of the firm, nor has an equal reduction in
both. In particular, the firm has no reason to sell new equity in order to pay
increased dividends.
n
Assuming that no new equity is issued, so Vt+1
= 0, rearranging (8.54) and
adding to both sides gives
Vt [1 + t ] + Bt + it Bt [1 c ] = Vt+1 + Dt + Bt+1 Bt+1 + Bt + it Bt [1 c ] .
(8.57)
Using the definition of the dividend in (8.44) to replace Dt , (8.56) becomes
Vt [1 + t ] + Bt + it Bt [1 c ] = Vt+1 + Bt+1 + zt ,
or
Bt + Vt =
Vt+1 + Bt+1 + zt
,
1 + bt it [1 c ] + [1 bt ] t
(8.58)
(8.59)
t
is the leverage of the firm. Combining (8.58) with (8.53)
where bt = BtB+V
t
relates the level of shareholder wealth in two consecutive periods by
o
cto = zt + [1 + rt ]1 W
ct+1
W
,
(8.60)
with rt = bt it [1 c ] + [1 bt ] t . The interpretation of rt is that this is now

the cost of capital for the firm since it is the weighted average rate of return that
has to be paid for finance. The weights are determined by the present financial
structure of the firms in terms of its leverage. When the firm can exercise choice
over its financial structure it will aim to minimise rt ; more will be said about
this shortly. Solving (8.59) under the assumption that there is convergence
"s1
#
X
Y
1
cto = zt +
W
[1 + ru ]
(8.61)
zs .
s=t+1
u=t
From (8.60) it can be seen that the wealth of shareholders is equal to the present
value of future after tax cash flows discounted using the cost of capital derived
above and it is this that the firms policy should be chosen to maximise.
In considering the choice between bond financed investment and investment
via the sale of new equity, it should be noted that both these policies keep the
stream of cash flows constant. Their only eect is felt through the change in the
cost of capital and the firm should choose bt to minimise rt . Assuming that the
firm takes t and it as parametric, when the rates of return on bond and equity
246
dier there is no general presumption about the nature of the choice and the
outcome is dependent upon the corporation tax rate. However, in the central
case in which the rates of return are equal, bond financing is always preferred.
This reasoning is summarised as Theorem 8.3
Theorem 41 (i) If it [1 c ] > t then it is optimal to choose bt = 0 so finance
is by equity alone;
(ii) If it [1 c ] = t then the firm is indierent about the value of bt ;
(iii) If it [1 c ] < t then it is optimal to choose bt = 1 so finance is by
bonds alone.
In particular, if it = t then bond financing is preferred if c > 0 and the
firm is indierent about its financial policy if c = 0.
Proof. All these results follow from dierentiating the definition of rt . Doing
this gives
rt
= it [1 c ] t .
(8.62)
bt
When this is non-zero a corner solutions follows. If it is zero, indierence arises.
The results in Theorem 8.3 encompass several of the most well known findings in the study of corporation taxation and its eects upon financing. The
conclusion that the firm is indierent about its financial policy in the absence
of taxation is the Modigliani-Miller theorem applied to the case of certainty. It
will be shown in Section 4.3 how this can be extended to the case of uncertainty.
The superiority of bond financing over equity financing when the discount rates
are equal is usually attributed to Stiglitz (1973). As the theorem shows, this
preference for bond financing is entirely due to the preferential treatment of
bonds under the corporation tax.
Integration with personal taxes
The analysis developed above can be easily modified to incorporate the provisions of the personal tax system provided that it is assumed that capital gains
are taxed on accrual in each period. This is counter to the usual practice under
which capital gains are taxed upon realisation but is a necessary assumption
since analysing taxation upon realisation would require a comprehensive study
of household choice over time in order to link taxation with timing of asset
sales. Undertaking this would remove the transparency of the analysis. What
the analysis will capture are the consequences of the basic fact that capital gains
are typically taxed at a lower rate than other sources of income.
When dividend income and capital gains are taxed, the net dividend received
by the shareholders is
o
t = [1 p ] Dt g Vt+1
Vt ,
(8.63)
8.3. TAX INCIDENCE
247
and (8.47), giving the value of the firm in period t, becomes

" s
#
X
Y
1
Vto =
[1 + u ]
st t .
s=t
(8.64)
u=t
Following the same method of derivation as in (8.48) and (8.49)

o
t Vt = t + Vt+1
Vt .
(8.65)
The wealth of shareholders is still given by (8.51) but after substitution can
be written as
Wto = [1 g ] Vt + [1 p ] [Bt [1 + it1 [1 c ]] Bt1 + zt ]
[ p g ] Vtn + g Vt1 .
(8.66)
Since terms with subscripts t 1 are pre-determined at time t, maximisation of

Wto is equivalent to maximising
cto = [1 g ] Vt + [1 p ] [Bt + zt ] [ p g ] Vtn .
W
o
Substituting for t and Vt+1
in (8.64) and simplifying gives
1 p
1 + t g
n
+ Vt+1 .
Vt =
Dt Vt+1
1 g
1 g
(8.67)
(8.68)
Employing the definition of At and solving gives the value of the firm as
" s
#
X
X 1 + g 1 1 p
p g
u
n
Vt =
At
Vt+1 .
(8.69)
1 g
1 g
1 g
s=t u=t
The value given in (8.68) implies the following theorem which can be found in
Pye (1972) and King (1974).
Theorem 42 Given an investment policy, z, and a bond policy, B, if p > g
a decrease in Vsn for any s > t increases Vt .
n
Proof. Since At = Dt Vt+1
= zt+1 + Bt+1 [1 + it [1 c ]] Bt , any change
n
in Vs only has a direct eect upon Vt . Given p g > 0, a decrease in Vsn for
any s > t will increase Vt .
The significant implication of Theorem 8.4 is that the increases in value for
the firm achieved by reducing new equity issues need not be exhausted when
Vsn = 0. In fact, provided dividends remain positive, the firm would wish to
repurchase its own equity (i.e. choose Vsn to be negative) rather than increase
dividends. This result arises entirely from the preferential tax treatment given
to capital gains relative to dividends and emphasises how directly the structure
of the personal tax system can aect the financial behaviour of the firm. In
practice, firms are restricted in their freedom to repurchase shares. To do so
248
in the UK requires a court order whilst in the US the Internal Revenue Code
prohibits firms from repurchasing shares in lieu of distribution dividends.
As a consequence of Theorem 8.4 it becomes reasonable to assume that the
firm will not issue new equity. Hence Vtn will now be taken as zero for all t.
Following the derivation in (8.56) to (8.60), the wealth of the shareholders is
given by
"
"s1
# #
X
Y
1
o
ct = [1 p ] zt +
W
[1 + ru ]
zs ,
(8.70)
s=t+1
u=t
where the cost of capital is

rt =
bt it [1 c ] [1 p ] + [1 bt ] t
.
[1 g ] [ p g ] bt
(8.71)
The financial policy of the firm can be inferred from (8.70) through the firm
attempting to minimise the cost of capital. To do this, note that after tax a
bondholder receives [1 p ] it from each bond held. If households are to hold
both debt and equity simultaneously then their post-tax returns must be equal
so it follows that [1 p ] it = t where t is the return on equity defined in
(8.64). This observation provides the basis for Theorem 8.5
Theorem 43 The firm will employ only equity financing if [1 c ] [1 g ] >
[1 p ] and will employ only debt financing if [1 c ] [1 g ] < [1 p ].
Proof. Dierentiating (8.70) with respect to gives
[1 p ] [it [1 c ] [1 g ] t ]
drt
=
,
2
dbt
[[1 g ] [ p g ] bt ]
(8.72)
so the firm will choose only debt if it [1 c ] [1 g ] < t and only equity if the
inequality is reversed.
Now suppose that [1 p ] it = t so both debt and equity are held. If
[1 c ] [1 g ] > [1 p ] then it follows that it [1 c ] [1 g ] > t and the
firm would switch to an equity-only policy. Conversely, if [1 c ] [1 g ] <
[1 p ] then it [1 c ] [1 g ] < t and the firm will issue only debt.
The significant content of Theorem 8.5 is its demonstration that the financial
policy of the firm is determined solely by the structure of the corporate and
personal tax systems. As in Theorem 8.3, higher values of the corporate tax
work in the direction of favouring debt finance as do increases in capital gains
tax. Conversely, an increase in the personal tax rate favours equity finance. It is
also important to notice that the theorem predicts that a firm will never use both
equity and debt financing except in the special case that it [1 c ] [1 g ] = t .
Conclusions
The analysis of the firm under certainty has made a number of strong predictions concerning the eect of the tax system upon the financing of the firm.
8.3. TAX INCIDENCE
249
Without taxation, the certainty version of the Modigliani-Miller theorem shows

that the firm is indierent about its choice of financial policy. The introduction of taxation directly alters this conclusion, with the corporation tax always
working in the direction of favouring bond finance. The results also emphasise
how directly financial policy is aected by the personal tax system.
8.3.6
Uncertainty
The analysis above has been restricted by the assumption that all actions by
the firm lead to known future outcomes. Although a valuable assumption for
generating insights into financial policy, it is a poor reflection of reality especially
where long-term investments are considered. Uncertainty is now introduced
but it must be noted at the outset that the resulting analysis cannot be as
comprehensive as that conducted under certainty. Much of the reason for this
follows from the need to include the diering risk characteristics of equity and
bonds in the portfolio choice problem of consumers. If the firm does not become
bankrupt, bonds always pay a known return whereas the return to equity, in
both dividend and capital gain, will be dependent upon the state of nature. In
addition, if a firm becomes bankrupt the equity of that firm will be worthless
whereas bondholders will receive a share in the remaining value of the firm.
The Modigliani-Miller theorem has already entered the earlier discussion and
this section begins with a formal demonstration of the theorem for the case of
uncertainty. This will provide a sucient set of assumptions under which the
value of the firm is independent of the choice of financial policy. Following this,
the eects of introducing taxation upon financial policy are discussed.
The Modigliani-Miller theorem
The Modigliani-Miller Theorem extends the observation in Theorem 8.3, that in
the absence of taxes the value of the firm is independent of its financial policy, to
an economy with uncertainty. It was first published in Modigliani-Miller (1958)
and extended by Stiglitz (1969b). The limitation of both of these derivations
is that they were basically static and did not capture the intertemporal aspects
of the firms decision. The theorem was extended to a multiperiod setting in
Stiglitz (1974) and it is upon this that the following is based.
Consider an economy with m firms and H households. All firms use one input
to produce an output of the same commodity. A plan for the firm is a choice of
investment levels for each period t and, implicit in what follows, a production
technique contingent on the state of nature that arises at t. The investment
level of firm j at time t in state of nature s (t) is denoted by Ij (t, s (t)) and the
level of gross profit by j (t, s (t)). These are taken as given. The firm can use
two financial instruments: bonds and equity. Both of these are sold on perfect
markets. A bond represents a promise to pay 1 unit of output in period t. The
price of a bond with maturity at time in period in state s is p (t, s (t) , ). It
is assumed that there is no bankruptcy so it must be the case that p (t, s, t) = 1
all s, t.
250
Now suppressing the state of nature, profits can be written

Ij (t) =
=t+1
[p (t, ) [Bj (t, ) Bj (t 1, )]]+qj (t) [ j (t) j (t 1)]+j (t) ,
(8.73)
where Bj (t, ) denotes the number of bonds at the end of period t with maturity
at , qj (t) the price of a share in the jth firm at t, j (t) the number of shares
outstanding at the end of period t and j (t) retained earnings. Hence the level
of investment must be equal to the change in value of outstanding bonds plus
the value of the change in the number of shares plus retained earnings. Define
Vje (t) = qj (t) j (t) ,
(8.74)
to be the value of equity outstanding at the end of period t and

Vjo (t) = qj (t) j (t 1) ,
(8.75)
to be the value outstanding at the beginning of the period. From these definitions it follows that Vje (t) Vjo (t) is the value of the change in the number of
shares and Vjo (t 1) Vje (t) is the capital gain or loss between periods. Since
there are no taxes, the gross profits of the firm must be either held as retained
earnings or distributed as dividends or used to redeem bonds. Thus
j (t) = Bj (t 1, t) + Dj (t) + j (t) ,
(8.76)
where Dj (t) denotes dividend payments in period t.

The value of the jth firm at the beginning of period t before the bonds maturing in that period are redeemed is given by the present value of it outstanding
bonds plus the value of equity
Wjo (t) = Vjo (t) +
X
=t
p (t, ) Bj (t 1, ) ,
(8.77)
and the value at the end of period t is

Wje (t) = Vje (t) +
p (t, ) Bj (t, ) .
(8.78)
=t+1
Hence
Dj (t) = j (t) Ij (t) + Wje (t) Wjo (t) .
ho
(8.79)
Turning now to the household side of the economy, let w (t) be the wealth
of h at beginning of t, whe (t) be wealth at end of t, he
j (t) be ownership of h
in j at the end of t and B h (t, ) be their ownership of bonds with maturity .
The assumption that there is no bankruptcy makes all firms bonds identical so
they must have the same pricing structure. The wealth of h is therefore given
by
m
X
X
whe (t) =
he
(t)
+
p (t, ) B h (t, ) ,
(8.80)
j
j=1
=t+1
8.3. TAX INCIDENCE
251
or
whe (t) =
m
X
hj (t) Wje (t) +
=t+1
j=1
he
j (t)
ej (t)
hj
where
(t) =
Similarly,
w
ho
j=1
hj (t) Bj (t, ) ,
(8.81)
hj (t) Wjo (t + 1)
j=1
p (t, ) B h (t, )
m
X
is the share of h in the equity of firm j at the end of t.
m
X
(t + 1) =
=t+1
p (t + 1, ) B h (t, )
m
X
j=1
hj (t) Bj (t, ) .
(8.82)
Pm
During period t household h receives dividends of j=1 hj Dj and consumes
xh (t) so the wealth levels at the beginning and end of t are related by
whe (t) = who (t) xh (t) +
m
X
j=1
hj (t 1) Dj (t) ,
(8.83)
and the level of consumption is determined by

"
#
m
X
X
h
h
e
x (t) =
j (t 1) j (t) Ij (t) + Wj (t)
p (t, ) Bj (t 1, )
j=1
=1
=1
p (t, ) B h (t 1, ) whe (t) , all h, t.
(8.84)
The description of the economy is completed by noting that equilibrium

requires the value of ownership claims on firm j to equal the value of equity or
H
X
Vjhe (t) =
h=1
H
X
hj (t) V je (t) = V je (t) , all j, t,
(8.85)
h=1
and for the supply of bonds to equal the demand

m
X
j=1
Bj (t, ) =
H
X
B h (t, ) , all t, .
(8.86)
h=1
The following theorem can then be proved.

Theorem 44 (Modigliani-Miller/Stiglitz) Assume that there exists an equilibrium for the economy. Then there is another equilibrium in which any firm (or
group of firms) has changed its financial policy but in which the value of the
252
firm and of all its bonds are unchanged. In addition, the portfolio changes of
investors are given by
B h (t, ) =
m
X
hj (t) Bj (t, ) , all t, , j,
(8.87)
j=1
and
hj (t) = 0, all h, j, t.
(8.88)
Proof. Assume that whe (t), p (t, ) and Wje (t) remain unchanged after the
change in financial policy. Under this assumption, it is clear from (8.81) that
the changes in portfolio described in (8.86) and (8.87) are feasible. From (8.83),
if these changes are undertaken then xj (t + 1) + whe (t + 1) remains constant.
Hence if xj (t + 1) is unchanged, so is whe (t + 1). In addition, for periods before
the change in financial policy, wealth levels will be unchanged. Consequently,
any consumption stream that was feasible before the change in financial policy
remains feasible after the change. Furthermore, the original consumption stream
must still be optimal since the opportunity set has not changed.
It remains to show that the markets for equity and bonds remain in equilibrium. From (8.87) it is clear that the equity market must be unaected. For
the bond market, it follows from (8.86) that
m X
H
X
hj (t) Bj (t, ) =
j=1 h=1
m
X
Bj (t, ) .
(8.89)
j=1
Since all markets are unaected, this justifies the original assumption that
wealth, prices and values remained unchanged.
The basis of Theorem 8.6 is that any change in financial policy by firms
can be neutralised by adjustment of household portfolio composition. In this
way, the equilibrium before the change can be maintained. The theorem does,
however, rely on some strong assumptions. Firstly, it requires that households
are unrestricted in their borrowing and lending and face the same rates on
loans as the firms. Secondly, it was assumed that there was no bankruptcy.
The eect of this assumption is to make all bonds maturing at some given date
perfect substitutes regardless of the firm that issued them. This assumption is
clearly an inappropriate one. The role of the bankruptcy assumption is discussed
further in Hellwig (1981) who shows that the theorem can be extended to include
potential bankruptcy provided that, when only securities issued by firms can
act as collateral on loans, short-selling of securities is permitted. However when
moral hazard is incorporated as a feature of loan contracts, even this theorem
does not apply.
Uncertainty and financial policy
In an economy with a full set of Arrow-Debreu contingent commodities, the
profit level of the firm is certain, as described in Chapter 7, despite the underlying uncertainty. This follows from the contingent production plan having a
8.3. TAX INCIDENCE
253
certain value at the known prices for contingent commodities. Since an increase
in profits of a firm expands the budget set of all its shareholders, they will be
in unanimous agreement that the firm should maximise profit. The implication
of these observations are that with a full set of markets the results on the interaction of taxation and financial policy are the same under uncertainty as for
certainty. Additional content to the problem can therefore only be given in an
economy with incomplete markets.
With incomplete markets, there may no longer be unanimity between the
shareholders about the objective of the firm. King (1977) presents an example where households receive the same utility from consumption but dier in
the probabilities they assign to the occurrence of alternative states. As a consequence, their von Neumann-Morgernstern expected utility functions dier.
Because of these dierences, there may be a conflict of interest between shareholders arises since the expected utility of a policy is a product of the returns,
earned by the firm in the various states, and the probabilities assigned to those
states. In addition to the lack of unanimity, the shareholders may prefer the
firm not to choose the policy that maximises the value of the firm. The explanation for this is that shareholders are also concerned with the prices of the goods
they consume. If the actions of the firm aect both wealth and relative prices,
then wealth maximisation need not be optimal. This aspect is investigated
in Taggart (1980) who shows how the incompleteness of market prevents the
equalisation of shareholders marginal rates of substitution between current and
future consumption. This was also the reasoning behind the valuation adopted
in the analysis of incomplete markets in (7.17).
The requirement of including incomplete markets and analysing the firms
decision problems when there is potentially no unanimity between shareholders
makes further progress dicult to obtain. One contribution that is worth noting
in this context is that of Auerbach and King (1983). Although this is restricted
by being developed within a framework in which households are concerned only
with the mean and variance of returns, so limiting the generality of preferences,
it does develop a number of important conclusions. It is shown that two forms
of equilibrium can arise. The first form of equilibrium arises when there is a firm
whose returns are spanned by the returns of the other firms. When this occurs,
value maximisation is optimal by firms but households will be specialised in their
purchases. Because of taxes, some households will prefer to purchase only equity
whilst others will purchase only debt. The value of each firm will be independent
of its financial policy. In the alternative equilibrium value maximisation is not
optimal and households will hold both debt and equity. However, for either
of the equilibria to exist, constraints upon the activities of households must
be imposed. There must be a limit upon the extent of borrowing that can
be undertaken and no households aggregate portfolio can be short-sold. The
force of these restrictions is to prevent infinite trades to take advantage of tax
arbitrage.
254
8.4
Conclusions
This chapter has considered the eect of corporate taxation upon both the productive activities of the firm and its choice of financial policy. These represent
the two channels through which the corporation interacts with the other agents
in the economy. The structure of the chapter involved a process of moving from
the analysis of the productive decisions of the corporation in the Harberger
economy to the combination of financial and investment decisions in an uncertain environment. The analysis of financial policy emphasised the importance
of both the corporate tax system and the personal tax system for the determination of financial policy. The eect of taxation in determining financial policy
is emphasised most emphatically when contrasted with the conclusion of the
Modigliani-Miller theorem which asserts the irrelevance to the firms value of
the choice between equity and debt.
Part III
Relaxing the Assumptions
255
Chapter 9
Public Goods
9.1
Introduction
When a public good is provided, it can be consumed collectively by all households. Such collective consumption violates the assumption of the private nature
of the goods in the Arrow-Debreu competitive economy. The existence of public
goods then leads to a failure of the competitive equilibrium to be ecient. Such
failure implies a potential role for the state in public good provision to overcome
the failure of the market.
The formal analysis of public goods began with Samuelson (1954) who derived the rule characterising ecient levels of provision and, after defining some
necessary terms, this will also be the starting point of this chapter. Ecient
provision will be considered for pure public goods and for public goods subject
to congestion. The theme of eciency is continued into the study of Lindahl
equilibria with personalised prices. Following this, the analysis of private provision demonstrates the nature of the outcome when prices are uniform and
illustrates why a competitive market fails to attain eciency.
If government provision is to be justified, it must be shown that the government can improve upon the market outcome. Section 6 shows what can
be achieved when policy instruments are restricted to commodity taxation and
uniform lump-sum taxes. In seeking the attainment of an ecient outcome, the
government is faced with informational constraints of which the lack of knowledge of household preferences is the most significant. Section 8 on mechanism
design shows why households may choose to misrepresent their preferences and
how mechanisms can be designed to overcome this. Finally, the chapter is completed by a review of experimental evidence on private provision and preference
revelation, and the use of market data to elicit valuations..
257
258
9.2
CHAPTER 9. PUBLIC GOODS
Definitions
A public good can be distinguished from a private good by the fact that it can
provide benefits to a number of users simultaneously whereas a private good can,
at any time, only benefit a single user. If the public good can accommodate any
number of users then it is said to be pure. It is impure when congestion can
occur. This section defines a public good, clarifies the distinction between pure
and impure and develops its economic implications.
9.2.1
Pure public goods
The pure public good has been the subject of most of the economic analysis
of public goods. In some ways, the pure public good is an abstraction that
is adopted to provide a benchmark case against which other results can be
assessed. Before proceeding, it should be noted that public goods can take the
form of inputs into production in addition to their more commonly-presented
role as objects of consumption. A simple translation of the comments below can
be made in order to allow them to describe the public good as an input.
A pure public good has the following two properties.
Non-excludability
If the public good is supplied, no household can be excluded from consuming
it except, possibly, at infinite cost.
Non-rivalry
Consumption of the public good by one household does not reduce the quantity available for consumption by any other.
The implication of non-excludability is that consumption cannot be controlled eciently by a price system since no household can be prevented from
consuming the public good if it is provided. It is evident that a good satisfying
this condition does not fit into the framework of the competitive economy used
to derive the Two Theorems of Welfare Economics in Chapter 2. In the form
given, those theorems are inapplicable to an economy with public goods.
From the property of non-rivalry it can be deduced that all households can,
if they so desire, simultaneously consume a level of the public good equal to its
total supply. If it is possible for households not to consume the public good,
then some may consume less. In the latter case, the public good may satisfy
free-disposal, so that consumption can be reduced at no cost, or else disposal
can be costly. Further discussion of the modelling of free-disposal is given in
Milleron (1972) and of costly disposal in Oakland (1987). When all households
must consume, or want to consume, to the maximum, the welfare level of each
household is dependent on the total public good supply.
In reality, it is dicult to find any good that satisfies both the conditions of
non-excludability and non-rivalry precisely. For example, the transmission of a
television signal will satisfy non-rivalry but exclusion is possible at finite cost.
Similar comments apply, for example, to defence spending which will eventually
be rivalrous and from which exclusion is possible.
9.3. OPTIMAL PROVISION
259
Figure 9.1: Consumption Possibilities for One Unit of Good
9.2.2
Impure public goods and congestion
In practice, public goods tend to eventually suer from congestion when usage
is suciently great. Obvious examples include parks and roads. Congestion
results in a reduction in the return the public good gives to each user as the
use of a given supply by households increases. Such public goods are termed
impure. The utility derived by each household from an impure public good
is an increasing function of the level of supply and a decreasing function of
its use. There are a number of ways of representing the eect of congestion
upon preferences and some of these will be described when optimal provision is
characterised.
To obtain further insight into these definitions it may be helpful to think of a
continuum of types of good running from purely private goods, for which there
is complete rivalry and exclusion at zero cost, to pure public goods. Figure 9.1
illustrates the possible division of the consumption of 1 unit of a good between
two households for the two extremes and for an impure public good. With a pure
public good it is possible for both to consume a maximum of 1 unit. In contrast,
the private good must be divided between the households. The consumption
possibilities for an impure public good lies between these limits.
9.3
Optimal provision
The characterisation of the ecient provision of a pure public good was first
published in Samuelson (1954) and was followed by a diagrammatic explanation
in Samuelson (1955). For this reason, the rule for ecient provision is typically
called the Samuelson rule. The following analysis will derive the Samuelson rule
for a pure public good, with and without free disposal, and for public inputs
and public goods with congestion.
260
9.3.1
Pure public good
To provide a reasonably simple derivation of the eciency rule it will be assumed

that there is available a single public good and, initially, that disposal is not
possible. The latter assumption implies that all households must consume a
quantity of the public good equal to its supply. The extension to many public
goods is entirely straightforward.
The economy consists of H households, indexed h = 1, ..., H. Each household
has a utility function
U h = U h xh , G ,
(9.1)
where xh is the consumption of household h of the vector of private goods and

G is the supply of the public good. The fact that total supply, G, appears in all
households utility functions indicates that the public good is pure.
It is assumed that the combinations of xh , h = 1, ..., H, and G that the
economy can produce are constrained by production possibilities. The implicit
representation of the production set is written
F (X, G) 0,
(9.2)
where
X=
H
X
xh .
(9.3)
h=1
To characterise the set of first-best, or Pareto ecient, allocations the government chooses xh , h = 1, ..., H, and G to maximise the utility level of the
first household, constrained by the requirement that households 2 to H obtain
given utility levels and by production possibilities. Varying the given utility
levels for households 2 to H traces out the set of Pareto ecient allocations.
The Lagrangean for this maximisation problem can be written
H
h
i
X
h
L = U 1 x1 , G +
h U h xh , G U F (X, G) ,
(9.4)
h==2
where U is the utility level that must be achieved by h = 2, ..., H. Assuming

that the specified utility levels can be reached simultaneously, the necessary
condition describing the choice of a component xhi from xh is
h
F
L
h U
= 0, h = 1, ..., H,
Xi
xhi
xhi
(9.5)
with h 1 for h = 1. At an optimum (9.5) holds for all i = 1, ..., n. For the
choice of the level of public good, forming the Lagrangean and optimising with
respect to G gives
H
X
L
U h
F
h
= 0.
(9.6)
G
G
G
h=1
261
Solving (9.5) for h , substituting into (9.6) and rearranging gives

U h
G
U h
h=1 xh
i
H
X
F
G
F
Xi
, i = 1, ..., n.
(9.7)
To interpret (9.7) note that each term in the summation on the left-hand side
is
h
U
G
U h
xh
i
(9.8)
which is the marginal rate of substitution between the public good and the ith
private good for the hth household. The right-hand side of (9.7) is the marginal
rate of transformation between the public good and private good i. Equation
(9.7) can thus be written as
H
X
h
M RSGi
= M RTGi .
(9.9)
h=1
Equation (9.9) is the Samuelson rule which states that Pareto ecient provision
of the public good occurs when the marginal rate of transformation between the
public good and each private good is equated to the sum, over all households,
of the marginal rates of substitution.
The result in (9.9) should be contrasted to the corresponding rule for ecient
provision of two private goods i and j
h
M RSji
= M RTji , all i, j, h.
(9.10)
The dierence between (9.9) and (9.10) occurs due to the fact that an extra unit
of public good increases the utility of all households so that the social benefit
of this extra unit is found by summing the marginal benefit, measured by the
MRS, to individual households. At an optimum, this is equated to the marginal
cost given by the marginal rate of transformation. In contrast, an extra unit of
private good only increases the welfare of its single recipient and an optimum
occurs when marginal benefits are equalised across households and to marginal
cost.
Two points must be noted in the interpretation of this result. Firstly, although non-excludability has been adopted as a defining characteristic of a pure
public good, it played no role in the derivation of the Samuelson rule. In fact,
the optimal level of provision is not dependent on the degree of excludability.
Instead, excludability is only relevant for determining feasible provision mechanisms. Secondly, although the Samuelson rule is deceptively simple in form, care
should be taken before believing that it can be easily implemented. In order to
derive the rule it was assumed that the government had complete control over
the allocation of resources. Equivalently the government could employ lumpsum taxation to redistribute income and to finance the provision of the public
good, with decentralisation of the provision of private goods. It has already
262
been argued in Chapter 2 that optimal lump-sum taxes can rarely be employed
in practice and the same observation negates the implementation of the Samuelson rule. This conclusion has motivated the investigation of alternative forms of
finance and the contrast of the resulting allocations to the first-best Samuelson
rule.
9.3.2
Free disposal
If free disposal of the public good is possible it is no longer necessary that each
household need consume the total quantity provided. If it is assumed that all
households wish to consume some of thepublic good, the utility level of household h can be written U h = U h xh , g h , where g h is the consumption of the
public good by h, and the constraint g h G, all h, added to the maximisation
in (9.4). The resulting Lagrangean is
H
H
h
i
X
h
L = U 1 x1 , g 1 +
h U h xh , g h U F (X, G) +
h G g h .
h=2
h=1
(9.11)
The version of the Samuelson Rule that applies in this case is given by
U h
g h
U h
h=1 xh
i
H
X
F
G
F
Xi
(9.12)
with the condition U

= 0 if g h < G. The interpretation of (9.12) is that
g h
the marginal benefit of increasing provision of the public good is set equal to
the marginal cost but allowing for the possibility that some households may be
satiated with the public good and receive no benefit from additional provision.
This derivation can be extended to permit some households to consume a
zero quantity of the public good by introducing the constraint g h 0 into the
maximisation. It is also straightforward to remove the free disposal assumption
and replace it with a costly disposal. Oakland (1987) provides a derivation of
the relevant version of the Samuelson rule in these cases.
9.3.3
With congestion
For many public goods congestion is a very real phenomenon. Naturally, congestion reduces the benefit that all households receive from their use of the public
good and therefore modifies the rule for ecient provision.
In the presence of congestion, the welfare of a household is typically written
as dependent upon the total supply of the public good and the usage of the
public good by all households. One approach to this (Oakland (1972)) is to
write
U h = U h xh , g 1 , ..., g H , G ,
(9.13)
h
U
U
with U
G > 0, g h = 0 and for j 6= h, g j < 0. It is the sign of the latter
term that captures the congestion eect. If disposal of the public good is not
263
possible then g h = G all h and (9.13) collapses to
U h = U h xh , G, H .
(9.14)
A specialisation of (9.14) will be employed in Chapter 12. Alternative specifications which employ a household production approach have been suggested
by Sandmo (1973) and Muzondo (1978), who consider production processes using private and public goods, and Ebrill and Slutsky (1982) who focus on the
combination of public goods with scarce time.
If (9.13) is employed and the Lagrangean formed as in (9.11), the resulting
necessary conditions for the maximisation can be combined to give
U j
g h
U j
j=1 xji
H
X
and
U h
gh
U h
h=1 xh
i
H
X
h
F
X
i
, i = 1, ..., n, h = 1, ..., H.
U j
g h
U j
h=1 j=1 xji
H
H X
X
F
G
F
Xi
i = 1, ..., n.
(9.15)
(9.16)
If, at the optimum, g h < G for all h, then from the complementary slackness
conditions h = 0 and (9.15) states that each households use of the public good
should be expanded until the private return is exactly balanced by the sum
of negative externalities that this inflicts on other households. Additionally,
the second term in (9.16) will be identically zero and (9.16) will describe the
standard Samuelson rule.
If some, or all, households are not satiated at the optimum, so that g h = G
for some households, then the second term in (9.16) is positive and the left-hand
side of the equation provides a measure of benefit in excess of that used in the
standard Samuelson rule. This excess arises due to an increase in provision
of the public good aecting utility both directly and through the reduction in
congestion that is brought about.
9.3.4
Public input
The characterisation of eciency conditions for economies with pure public

inputs began with Kaizuka (1965). The subsequent literature, which has also
analysed congestible public inputs, is discussed in Feehan (1989).
To derive the eciency conditions for the supply of a pure public input,
consider an economy with m firms each using labour and the public good to
produce a single form of output. Denoting the labour use of firm j by `j , the
firms production function is given by
yj = f j `j , G .
(9.17)
The public good is produced
by using labour alone according to the production
function G = `G . This is assumed to have an inverse `G = (G) , (G)
264
1 (G). The equilibrium conditions, that supply must equal demand for goods
and labour, are given by
H
X
xh =
yj =
j=1
h=1
and
m
X
H
X
` =
m
X
j=1
m
X
f j `j , G ,
(9.18)
`j + (G) .
(9.19)
j=1
h=1
For this economy, an optimum allocation is found from the Lagrangean
H
H
m
h
i
X
X
X
h
L = U 1 x1 , `1 +
h U h xh , `h U +
xh
f j `j , G
h=1
H
X
h=1
`h
m
X
j=1
h=1
`j (G) .
j=1
(9.20)
Carrying out the maximisation in (9.20) provides the eciency criteria

U h
`h
U h
xh
and
f j
, h = 1, ..., H and j = 1, ..., m,
`j
m
X
f j
j=1
f j 0
, j = 1, ..., m.
`j
(9.21)
(9.22)
The first condition, (9.21), ensures that the marginal rate of substitution between labour and consumption is equated between households and this value
is set equal to (minus) the firms common marginal product of labour. This is
a standard eciency condition. Condition (9.22) is the form of the Samuelson
rule for the public input and requires the sum of marginal products of the public
inputs for the firms to equal the private good foregone in producing marginally
more public good.
This completes the analysis of rules for ecient provision and from this point
the focus will be placed upon pure public goods. The Samuelson rule may characterise the set of Pareto ecient outcomes but, as noted, it cannot in general
be implemented. This motivates the study of feasible allocation mechanisms
and the comparison of their outcomes to those that satisfy the Samuelson Rule.
9.4
Personalised prices and the Lindahl equilibrium
Now that the rule for Pareto ecient provision has been derived the natural
question is whether there is any form of economy in which competitive behaviour
9.4. PERSONALISED PRICES AND THE LINDAHL EQUILIBRIUM
265
will lead to an ecient outcome. The equilibrium in the standard model of

Chapter 2 will not be Pareto ecient in the presence of public goods. This
arises from the fact that consumers dier in the valuation they place upon a
given supply of the public good. Insisting that they all pay an identical price
for the supply cannot therefore be ecient.
Following this reasoning, it would appear likely that Pareto eciency would
result if each consumer could pay an individual or personalised price for the
good. In this way, each will be paying a price that reflects their valuation.
Allowing such personalised prices represents an extension of the Arrow-Debreu
economy which assumed that each commodity had a single price. The equilibrium with personalised prices is often called a Lindahl equilibrium after its
introduction by Lindahl (1919).
Two variants of the Lindahl equilibrium will be described in this section.
The first is for a simple economy with two households in which the price of the
public good is given by the share of the cost of the public good each household
must cover. Following this, it is shown how eciency can be sustained in a
competitive economy with truly personalised prices. The section is completed
by analysing the relationship of the Lindahl equilibrium, and its extensions, to
the core.
9.4.1
Simple model
The first formal analysis of the Lindahl equilibrium can be traced to Johansen
(1963) who provided an analytical interpretation of Lindahls (1919) equilibrium
concept. With a single public good, the central aspect of this formulation is
that each household bases their consumption decision upon the share, h , of
the cost of provision of the public good that they must pay. Assuming that the
h
demand of household h for the public good increases
1without
limit as 0, an
H
equilibrium can be defined as a set of cost shares, , ..., , that sum to 1 and
are such that they lead all households to demand the same quantity of the public
good. The importance of this equilibrium is that it satisfies the Samuelson rule
and is therefore Pareto ecient despite the existence of the public good. If there
are many public goods, a cost share can be introduced for each public good and
optimality again achieved.
To illustrate these ideas in the simplest setting, consider an economy with
2 households who have an endowment of h units, h = 1, 2, of the numeraire
which they supply inelastically to the market. Each household therefore has
a fixed income of h . There is a single private good produced with constant
returns to scale using the numeraire alone and the units of measurement of this
good are chosen so that a unit of output requires one unit of numeraire input.
The price of the private good is therefore also equal to one. Production of the
public good is subject to constant returns to scale and each unit requires pG
units of labour. The marginal rate of transformation in production between the
public good and the private good is therefore constant at pG .
266
Assume that each household has a utility function such that
U h = U h xh , G , h = 1, 2,
(9.23)
where xh is the quantity consumed of the single private good and G is the
quantity of the public good. Utility is non-decreasing in xh and G. Now let
Gh denote the quantity of the public good that household h would like to see
provided when faced with the budget constraint
xh + h pG Gh = h .
(9.24)
In (9.23) pG Gh is the total cost of providing the good and h the fraction of
this paid by h. From (9.22) and (9.23) household h chooses Gh to maximise
U h = U h h h pG Gh , G .
(9.25)
The necessary condition for this maximisation is
h
UG
= h pG .
Uxh
Solving (9.25) for Gh generates the Lindahl reaction function
Gh = Lh h ; h ,
(9.26)
(9.27)
which describes the households demand for the public good as a function of the
cost share it faces and its initial endowment. If the second-order condition for
maximising (9.24) is satisfied and the utility function is strictly concave, then
Lh () is a decreasing function of h .
n
o
A Lindahl equilibrium is a pair of cost shares b
1, b
2 such that
2 = 1,
(i) b
1 + b
and
(ii) Lh b
h ; h = G 0, h = 1, 2.
The first condition guarantees that sucient revenue will be obtained to
finance the equilibrium public good provision and the second-condition that the
households will both be satisfied with the supply. It follows from the fact that
utility is non-decreasing in G that the cost shares will be non-negative.
The nature of the Lindahl equilibrium is illustrated in Figure 9.2. The
Lindahl reaction functions are formed as the loci of the vertical points on the
indierence curves of the utility function in (9.24) and the equilibrium is given
by the intersection of the functions. At this point, the indierence curves for the
two households are tangential and the equilibrium is therefore Pareto ecient.
To demonstrate the latter point, note that (9.25) must hold for both households
at the equilibrium. Summing for the two households then gives
2
X
Uh
G
Uxh
h=1
2
X
h=1
h
M RSGx
=
2
X
h=1
h pG = pG = M RTGx .
(9.28)
267
Figure 9.2: The Lindahl Equilibrium

Since (9.27) is the Samuelson rule for this economy, it demonstrates that the
Lindahl equilibrium is Pareto ecient. This establishes a form of the First
Theorem of Welfare Economics for the Lindahl equilibrium. The relation of
the Lindahl equilibrium to the Second Theorem will be investigated formally
below. For the present, it is sucient to note that by redistributing the initial
endowment it is possible to generate a new Lindahl equilibrium which represents
another point in the set of Pareto ecient outcomes.
9.4.2
A general treatment
The analysis of the Lindahl equilibrium in terms of cost shares clearly indicates
how the reasoning can be generalised. In the budget constraint (9.24) the price of
a unit of public good for household h is given by h pG . Whenever the cost shares
dier, the households face dierent prices for the public good. It is the existence
of these personalised prices that leads to the attainment of eciency. The
asymmetry between private and public goods should also be noted: with private
goods all households face the same prices but demand dierent quantities; with
public goods the households face dierent prices but, at the equilibrium, demand
the same quantity.
The presentation of the Lindahl equilibrium via cost shares and reaction
functions has the benefit of simplicity but, as careful study of the previous
section shows, it does not make it obvious how general theorems on the existence
of equilibrium or the welfare properties of the equilibrium can be obtained. To
develop these it is preferable to exploit the fact that the use of personalised
prices allows the Lindahl equilibrium to be expressed as the equilibrium of a
suitably defined competitive economy. Since standard existence and welfare
theorems apply to the competitive economy, they naturally apply to the Lindahl
equilibrium. This approach to the Lindahl equilibrium was developed by Foley
(1970), Milleron (1972) and Roberts (1973, 1974).
To formalise this idea consider an economy in which there are s non-disposable
public goods available with these public goods used only for final consumption.
268
The aggregate production set of the economy is denoted Y and, with n private
goods available, Y <n+s . It is assumed that Y is a closed, convex cone that
contains the origin and thus that the technology satisfies constant returns to
scale. A production plan is written in the form (g; y) where the first s elements represent public goods and the final n private goods. It is assumed that
no public good is required as an input: if (g; y) Y then (b
g ; y) Y where
gbi = gi if gi > 0 and gbi = 0 if gi < 0. Each household, h, from the set of H
h
n+s
households possesses a consumption
with a consumption choice
h h set X <
for the household denoted g ; x . Household
h
has
continuous preferences
over X h represented by the utility function U h g h ; xh , where U h () is strictly
monotonic and quasi-concave. The endowment of h is denoted by h = 0; h .

There are no endowments of public goods.
A state of this economy is an array g, x1 , ..., xH of public and private good

vectors. A feasible state is now defined.
Feasibility
(i) g h ; xh X h for all h = 1, ..., H,

P
h
h
(ii) g; H
x
Y.
h=1
Condition (i) ensures the consumption allocation is in the consumption set

for each household and (ii) that it is productively feasible.
The relevance of personalised prices has already been discussed and they are
now employed to define the Lindahl equilibrium. Let phG be the price vector for
public goods faced by h and p be the common vector of private goods prices.
The formal definition of a Lindahl equilibrium can now be given.
Lindahl equilibrium
1
A Lindahl equilibrium
with
respect to the endowment
, ...,
H is a fea
sible allocation gb, x

b1 , ..., x
bH and a price system pb1G , ..., pbH
, pb 0 such that
i h P
iG
hP
i hP
H
H
H
bhG ; pb gb; h=1 x
bhG ; pb [g; y] for all [g; y]
(i)
bh h
h=1 p
h=1 p
Y,
(ii) pbhG gb+ pbx
bh pb h,
h
(iii) U gb; x
bh U h g; xh for all g; xh X h such that pbhG g + pbxh pb h .
This definition is a direct extension of that in 4.1.
The first issue concerning this economy is the existence, or otherwise, of an
equilibrium. Although this is not of direct relevance here, it is worth making
some comments. To develop an existence proof, both Foley (1970) and Milleron
(1972) construct a quasi-economy in which the commodity space is expanded
to allow each public good for each household to be treated as a separate good.
Consumption sets, preferences and the production set are redefined to be compatible with this. Equilibrium is then proved to exist, under assumptions closely
related to those in Section 9 of Chapter 2, for the quasi-economy and this equilibrium is shown to be an equilibrium for the original economy. The alternative
approach adopted by Roberts (1973) is to work directly with the Lindahl economy and to construct a mapping from the price space for private goods and the
quantity space for public goods into the same two spaces. A fixed point of this
mapping is shown to exist and to be the equilibrium for the economy. Further
269
discussion of these proofs can be found in Roberts (1974) and an extension to

non-convex productions sets in Bonnisseau (1991). From these results, it can
be concluded that under relatively weak conditions a Lindahl equilibrium exists
and it is justifiable to proceed to an analysis of its welfare properties.
Given the existence of a Lindahl equilibrium and the obvious parallel of this
to the competitive equilibrium, it follows that a version of the First Theorem
must hold. The proof of this exactly mirrors that for the standard competitive
economy: assume the equilibrium is not Pareto ecient and that there exists
a preferred allocation; the preferred allocation can then be shown not to be
feasible.
Theorem 9.1 The First Theorem for Lindahl Economies (Foley)
The Lindahl equilibrium is Pareto ecient.
Proof
See Foley (1970). ||
Since the Samuelson rule describes the set of Pareto ecient allocations, it
follows that the Lindahl equilibrium also satisfies the Samuelson rule.
To prove the equivalent of the Second Theorem, that any Pareto optimum
can be decentralised as a Lindahl equilibrium, requires the application of a
separation theorem as in the proof of Theorem 2.4. The proof proceeds by
constructing two convex sets which have no interior points in common and can
therefore be separated. The normal to the separating hyperplane is then taken
as a set of equilibrium prices which are such that the Pareto ecient output is
profit maximising and each consumption plan is cost minimising over choices
yielding at least as much utility.
The proof given is based on Foley (1970)and requiresthe following additional
assumption:
given the feasible allocation g, x1 , ..., xh there exists, for all h,
h
g, x
e such that x
eh X h . When the price vector is positive
eh < xh and x
this assumption states that, given any consumption plan, there is always an
alternative plan with lower value. Using this assumption, the Second Theorem
for Lindahl economies can now be proved.
Theorem 9.2 The Second
for
Theorem
Lindahl Economies (Foley)
1
H
,
...,
x
b
If a feasible
allocation
g
b
,
x
b
is Pareto ecient, there exists a price
1
H
vector p = pG , ..., pG , p 0 such that
i h P
i
hP
i hPH
H
H h
h
h
h
p
;
p
g
b
;
p
;
p
[g; y] for all [b
g ; y]
(a)
x
b
h=1 G
h=1
h=1 G
Y,
(b) if U h g; xh > U h gb; x

xh .
bh then phG g + pxh phG gb + pb
Proof
Define the set F = g 1 , ..., g H : g 1 = ... = g H = g and (g; y) Y . Since

Y is a convex cone, F isnalso a convex cone. It is also non-empty since 0 F .
o
PH
Next, define the set D = g 1 , ..., g H ; y : y = h=1 y h with U h g h ; y h + h > U h gb; x
bh .
D is convex and nonempty due to the quasi-concavity and monotonicity of utility.
As gb, x
b1 , ..., x
bH is Pareto ecient, D and F have no points
in common.
Therefore there exists a separating hyperplane with normal p1G , ..., pH

G , p 6= 0
270
and a scalar r such that

PH
h h
(i) for all g 1 , ..., g H ; y F ,
h=1 pG g + py r,
and
PH
h h
(ii) for all g 1 , ..., g H ; y D, D the closure of D,
h=1 pG g + py r.
1
H
b
is Pareto ecient, monotonicity of prefSince the allocation gb, x
b , ..., x
erences implies (b
g , ..., gb; yb) D and the feasibility of the allocation implies that
P
P
h
h h
b + pb
y=r H
(b
g , ..., gb; yb) F . Hence H
h=1 pG g
h=1 pG g + py for all (g; y) Y .
This proves (a).
Since F is a convex cone containing zero, r 0 However, if r > 0 then it
could be increased without limit by expanding any plan that gave positive profit.
Hence r = 0. Since preferences are monotonic, p 0 or (ii) would be violated
by increasing the consumption on any good with a negative price. Similarly,
p = 0, phG 0 cannot arise since public goods with positive price could then
be produced using costless inputs (public goods are not needed as inputs) and
profit would be positive. Hence p 0.
P
h
Suppose U k g k ; xk > U k gbk ; x
bk . Then g 1 , ..., g H ; H
where
h=1 x
P
P
H
H
h h
h
h
bh , g h = gbh all h 6= k, is in D so
xh = x
h=1 pG g + p
h=1 x
PH
P
H
h
h
h
p
g
b
+
p
x
b
.
But,
as
only
the
terms
relating
to
k
in
the
inh=1 G
h=1
equality dier, this reduces to pkG g k + pxk pkG gb + pb
xk . Suppose this was an
equality. Then the assumption that for any feasible allocation there is a point
in the consumption set of k with lower value implies that along the line joining
k
k
the point
is, by continuity, a point of lower value
k ofk lower value to g ;x kthere
than g ; x and preferred to gb; x

b . This would imply the existence of a point
in D contradicting (ii). Therefore the inequality is strict and (b) is established.
||
Theorems 9.1 and 9.2 demonstrate how the Lindahl equilibrium replicates
the properties of the competitive equilibrium for economies with public goods.
Implicit in the Second Theorem is the fact that the distribution of resources
is chosen so that the Pareto ecient allocation satisfies the budget constraint
of each household. To achieve this objective, starting from an arbitrary initial
allocation, requires the use of lump-sum transfers or taxation. This makes the
Second Theorem for Lindahl economies subject to the criticisms discussed in
connection with Theorem 3.2.
9.4.3
Core equivalence
One of the observations supporting the use of the competitive equilibrium notion is, that for economies with large numbers of consumers, the set of core
allocations shrinks to the set of competitive equilibrium. This was stated below
as Theorem 2.2. In this way, identical equilibrium allocations are isolated using
two very dierent concepts of equilibrium. If it were also true that the core of
an economy with public goods shrank to the set of Lindahl equilibrium as the
number of consumers increased, this would provide similar support for the use
of the Lindahl equilibrium concept.
271
Unfortunately, this is not the case. It is true that the Lindahl equilibrium is
in the core, and this will be proved below, but Muench (1972) presents an example for which the set of core allocations is larger than the set of Lindahl equilibria
despite the economy possessing a continuum of consumers. The relevance of the
continuum is that Aumann (1964) demonstrated the core and competitive equilibrium are equivalent in this case. Therefore, although the Lindahl equilibrium
has many of the properties of the competitive equilibrium, it does not satisfy
the same core equivalence so that its relation to the core cannot be appealed
to in support of the equilibrium concept. The failure of core equivalence can
be traced to the fact that when a coalition, say S, attempts to improve upon a
given allocation it must be able to provide, from its own resources, quantities
of the public goods. Since there are returns to coalition size in the provision of
public goods, it is dicult for coalitions to improve upon existing allocations.
The result that the Lindahl equilibrium is in the core is given in the following
theorem.
Theorem 9.3 (Foley)
If gb, x
b1 , ..., x
bH; pb1G , ..., pbH
b is a Lindahl equilibrium
G,p
1 forH the
endowment
1
H
vector , ..., , it is in the core with respect to , ..., .
Proof
bH with
Suppose the coalition S
can improve upon the allocation gb, x
b1 , ..., x
the allocation g, x1 , ..., xH . Since the new allocation is preferred by all members of S, it follows from the definition of a Lindahl equilibrium that
P
P
P
P
P
biG g + pb iS xi > iS pbiG gb + pb iS x
bi = pb iS i .
iS p
P
P
P
Since pbiG 0 for all i, H
biG iS pbiG so that gb 0 implies H
bhG g+
i=1 p
h=1 p
P
i
i
pb iS x > 0. The latter condition contradicts profit maximisation in
the Lindahl equilibrium and proves the theorem. ||
9.4.4
Cost-share equilibria
At the heart of the Lindahl equilibrium is the concept of personalised pricing. In

the economy described in 4.2 the cost that household h faces for the public good
provision, phG g, is a linear function of the quantity of provision. In a natural
generalisation of this framework Mas-Colell (1980b), Mas-Colell and Silvestre
(1989) and Weber and Wiesmeth (1991) have investigated the implications of
permitting monotonic cost functions h (G) which are not necessarily linear for
economies with a single public good.
The main result of this literature is the equivalence theorem of Weber and
Wiesmeth which proves the that the set of cost-share equilibria is identical to the
set of core allocations. The additional flexibility oered by the nonlinear costshares is therefore sucient to provide the equivalence that is absent in the linear
Lindahl equilibrium. However, since it is already dicult to envisage linear
personalised pricing being applied in practice, nonlinear personalised pricing
would seem devoid of practical interest.
272
9.4.5
Comments
The Lindahl equilibrium demonstrates how eciency can be attained in an

economy with public goods by the use of personalised prices. The personalised
prices succeed in equating the individual valuations of the supply of public
goods to the cost of production in a way that uniform pricing cannot. These are
important observations that support the relevance of the Lindahl equilibrium
concept.
Unfortunately, the Lindahl equilibrium is not without fault. It is central to
the equilibrium that each household should face a price system that is designed
to capture that households evaluation of the public good supply. When it participates in the public goods market each household is the only purchaser at its
particular price ratio and is not in the position it would be in a competitive
market of being one purchaser among many. In a competitive equilibrium there
is no incentive for the household to act in any other way than just to purchase
its most preferred consumption plan. In contrast, the fact that the household
is in a stronger position in the Lindahl equilibrium raises the very clear possibility that it can gain by false revelation of preferences in an attempt to adjust
equilibrium prices to its advantage. Such strategic behaviour on the part of
households undermines the foundation of the Lindahl equilibrium. If it does
occur, the Lindahl equilibrium with strategic behaviour no longer possesses the
eciency properties set out above. The consequences of strategic behaviour,
and responses to it, will be discussed further in Section 8.
9.5
Private provision of public good
The characterisation of optimal provision in Section 3 was concerned with an

economy in which the government provided the public good and was unrestricted
in its policy instruments. This first-best outcome is now contrasted to the
equilibrium of an economy in which the public good is funded entirely by the
voluntary contributions of individual households. Economies with government
provision alone and those with only private provision should be seen as the two
extreme cases since in practice, as charitable donations indicate, public goods are
usually provided by a combination of both methods. The focus of the analysis
will be upon the welfare properties of the private provision equilibrium and the
level of public good supply relative to ecient levels. In addition, the eect of
the number of households on supply and changes in the income distribution will
also be considered.
In order to analyse private provision, it is necessary to make an assumption
about how each household expects their contribution to the provision of public
goods to aect the contributions of others. The assumption that was made in
the initial literature on private provision (for example Bergstrom, Blume and
Varian (1986), Chamberlin (1974, 1976), Cornes and Sandler (1985), McGuire
(1974) and Young (1982)) was the standard Nash assumption: in planning their
contribution, each household takes the contribution of the others as given. This
9.5. PRIVATE PROVISION OF PUBLIC GOOD
273
is not the only assumption and alternatives have been investigated; some of
these will be discussed briefly in 5.6.
9.5.1
Equilibrium
The equilibrium with private provision will be derived in the economy used to
introduce the Lindahl equilibrium. The economy therefore has H households
who each have an endowment of h units of the numeraire which they supply
inelastically. The income of each household is fixed at h . The single private
good is produced with constant returns to scale using the numeraire alone and
a unit of output requires one unit of numeraire input. The price of the private
good is equal to one. Production of the public good is subject to constant
returns to scale and each unit requires pG units of labour. The price of the
public good is constant at pG .
Each household has a utility function
U h = U h xh , G , h = 1, ..., H,
(9.29)
P
h
h
where xh is the quantity of private good consumed, G = H
h=1 g and g is the
contribution of h. The contribution towards the public good by all households
other than h, Gh , is defined by
h
Gh = G g h .
h
(9.30)
Using the budget constraint x + pG g = , utility can be written in terms of

Gh and g h as follows
U h xh , G = U h h pG g h , g h + Gh = V h g h , Gh , pG .
(9.31)
h
i
h
Household h chooses g h to maximise (9.30) given Gh and subject to g h 0,
pG .
h
h
Indierence curves of V () can be drawn in g , Gh space. Increasing Gh
will always lead to a higher attainable level of V for given g h , g h is limited by
the budget constraint and preferred sets are convex.
Since the household takes the provision of others as given when maximising,
the optimal choice of g h for a given value of Gh occurs at the tangency of the
indierence curve and the horizontal line at Gh . This is shown in Figure 9.3 and
the solid locus, the Nash reaction function, traces out the optimal choices of g h
as Gh varies. The gradient of the reaction function, when the chosen value of
g h is interior, can be derived by noting that for all g h on the reaction function
g h = arg max U h h pG g h , g h + Gh .
(9.32)
Dierentiating this, g h solves
h
Uxh pG + UG
= 0.
(9.33)
Now considering variations in g h and Gh that satisfy the first-order condition

(9.32),
dg h
U h pG U h
= h 2 xG h GG h ,
(9.34)
Uxx pG 2UxG pG + UGG
dGh
274
Figure 9.3: Indirect Preferences over Private Provision

h
> 0.
which is always negative when UxG
Alternatively, taking incomes as fixed, (9.32) can be solved to express g h as
a function of Gh . The reaction function can then be written as

h
h
h
h
g = Gh , Gh 0,
.
(9.35)
pG
This function traces out the optimal response of the household to the supply
Gh of the other households.
The equilibrium of the private provision economy occurs at a set of choices for
the households such that all the reaction functions are simultaneously satisfied.
Private provision equilibrium

is an array of contributions gbh , gbh
h Ah iprivate provision equilibrium

P
0,
bj .
bh = h Gh for all h = 1, ..., H, with Gh = H
j=1,j6=h g
pG , such that g
If gbh > 0 household h is termed a contributor and is a non-contributor if

gb = 0. The set of contributors is denoted C.
The proof that such an equilibrium exists employs the standard argument for
demonstrating the existence of a Nash equilibrium.
Assuming
that preferences
are strictly convex, the composite function R g 1 , ..., g H h 1i, ..., H defines
h
a continuous function from the compact, convex set H

h=1 0, pG to itself. From
1
Brouwers
R g , ..., g H has at least one fixed point and a fixed point
1 theorem
H
of R g , ..., g
is clearly a private provision equilibrium.
It has also been shown that under weak conditions the private provision
equilibrium is also unique. Given
G
h , (9.31) determines the contribution of h as
a function of income, g h = f h h . Bergstrom, Blume and Varian (1986, 1992)
f h (h )
show that the restriction 0 < h < 1 is sucient to prove uniqueness.
This restriction is simply the requirement that both private and public goods
are normal.
For a two-household economy the private provision equilibrium can be presented diagrammatically. The households solve the following maximisations
275
Figure 9.4: Private Provision Equilibrium

Household 1:
Household 2:
The solution
g 1 = 1 g 2 and
max{g1 } U 1 1 pG g 1 , g 1 + g 2 , with g 2 fixed.
max{g2 } U 2 2 pG g 2 , g 1 + g 2 , with g 1 fixed.

to these maximisations
are given by the reaction functions
g 2 = 2 g 1 . The equilibrium is then a pair gb1 , gb2 such that

gb1 = 1 gb2 , gb2 = 2 gb1 .
(9.36)
This equilibrium is illustrated in Figure 9.4 in which the reaction functions are
simultaneously satisfied at their intersection.
9.5.2
Pareto improvements
It is a property of Nash equilibria that they are not, in general, Pareto ecient
and, although no agent can improve their welfare when acting independently,
a simultaneous change in actions can benefit all agents. This observation applies to the private provision equilibrium in which a simultaneous increase in
contributions will raise all households welfare.
To demonstrate this it is first observed, as is always the case, that the set
of Pareto ecient allocations are the points of tangency of the households
indierence curves. In contrast, it follows from the construction of the reaction
function in Figure 9.3 and the structure of equilibrium in Figure 9.4 that for
the two household economy the private provision equilibrium occurs at a point
where two households indierence curves are at right-angles. The set of Pareto
ecient allocations and the private provision equilibrium are shown in Figure
9.5. Since the private provision equilibrium is not Pareto ecient, there exist
Pareto improvements and these are given by the shaded area.
In is also apparent in Figure 9.5 how an increase in both the households
contributions can lead to a Pareto improvement. Therefore, local improvements
in welfare can be achieved by an increase in the provision of public goods.
Compared to Pareto preferred allocations, the private provision equilibrium
leads to an undersupply of public goods.
276
Figure 9.5: Non-Optimality of Private Provision
Figure 9.6: Oversupply
9.5.3
Quantity of provision
The result described above, that the private provision equilibrium is Paretodominated by allocations with a higher level of public good, has often been
interpreted as demonstrating that private provision leads to undersupply relative
to the socially optimal level. However, a global optimum of a Paretian social
welfare function may lie anywhere on the locus of Pareto ecient allocations
and not necessarily on that part of the locus that Pareto-dominates the private
provision equilibrium.
Buchanan and Kafoglis (1963) demonstrated that it was possible for counterexamples to be constructed in which the quantity of a public good is decreased
as the economy moves towards the optimum from the private provision equilibrium. The possibility of such anomalies arising is illustrated in Figure 9.6. The
private provision equilibrium occurs at point N and the set of Pareto ecient
allocations are given by the locus of tangencies of the indierence curves.
The line CC represents an aggregate level of public good supply equal to
that in the private provision equilibrium. For the optimum to have less of the
public good simply requires the locus of Pareto optima to cut the line CC and
277
for social welfare to be maximised at some point below CC. If the optimal point
is P a reduction in the total supply of the public good is required in the move
from the private provision equilibrium to the optimum. It should also be noted
that if the locus does cross CC, a social welfare function can always be found
that places the optimum below CC.
As shown by Diamond and Mirrlees (1973), such anomalies can only be ruled
out by placing restrictions upon the second derivatives of the households utility functions. There are no fundamental reasons why restrictions upon second
derivatives should be satisfied and hence anomalous cases may well occur even
in the two-household model. Therefore, although local results can be established without too much diculty, it is not straightforward to provide global
comparisons.
9.5.4
The number of households
It has already been established that the private provision equilibrium is not
Pareto ecient. A further issue that has been addressed (Chamberlin (1974,
1976), McGuire (1974)) is how the deviation from eciency depends upon the
number of households that may potentially contribute. The natural expectation
would be that an increase in the number of households would lead to greater
divergence as each household expected all others to contribute. However, as
is often the case, the actual result diers somewhat from this expectation. In
addition, the limiting properties of equilibrium, as the number of households
increases without limit, will be analysed following the approach of Andreoni
(1988a).
To consider the consequence of variations in the number of households, assume that all the households are identical in terms of both preferences and
endowments. Assuming both goods are normal, the uniqueness result then justifies the study of symmetric equilibria. For an economy with H households, it
follows that at the symmetric equilibrium
g=
G
,
H 1
(9.37)
where g is the households common contribution. In terms of g, G space,

an allocation satisfying (9.36) must lie somewhere on a ray through the origin
with gradient H 1 and, for each level of H, the equilibrium is given by the
intersection of the appropriate ray with the reaction function. The welfare
optima, if all households are treated equally, are the locus of tangencies between
the rays and the indierence curves. This is shown in Figure 9.7.
Whether the quantity of public good at the private provision equilibrium
increases as
increases is dependent upon the gradient of the reaction func H
tion. If 0 G < 1, then provision, given by Hg, is an increasing function of

H. When 0 G = 1, provision is independent of H and it is a decreasing

function of H if 0 G > 1. These conditions can be related to the structure
278
Figure 9.7: Equilibria and Optima for Various H

of preferences via (9.33). For instance, with the utility function
U (x, G) = x + K (G) ,
(9.38)
h
h
= UxG
= 0 and the total level
the reaction curve has a gradient of - 1 since Uxx
of provision is independent of the number of households.
The eect of increasing the number of households upon the divergence between the private provision equilibrium and the equal-treatment welfare optima
is not so easily determined. It can be seen in Figure 9.7 that this is dependent
upon the curvature of the indierence curves between the two equilibria and,
as shown by Cornes and Sandler (1984a), both income and substitution eects
are involved. What can be said is that examples can be constructed, such as
the case shown in Figure 9.7, in which the divergence decreases as H increases.
The opposite result can also be shown to be possible.
The natural extension of the above analysis is to consider the equilibrium
level of provision as the number of households tends to infinity. An idea of what
will occur with identical households can be seen by considering the consequence
of the ray in Figure 9.7 becoming vertical: provision will tend to the level at
which the reaction function crosses the vertical axis and the provision of each
household will tend to zero. This result is due to Chamberlin (1974). Now
consider an economy where all households have the same preferences but dier
in their endowments. The following theorem is due to Andreoni (1988).
Theorem 9.4(Andreoni)
For an economy of H households with identical preferences and endowments
distributed according to a continuous density function f () with support (0, ),
then as H increases to infinity:
(i) The proportion of the population contributing decreases to zero;
(ii) Only the households with the highest endowment contribute;
(iii) Total contributions increase to a finite value;
(iv) The average contribution decreases to zero.
Proof
279
The maximisation of household h can be written max{xh ,G} U xh , G subh
h
h
ject to the constraints
G and G G . This yields the demand
n x + G =
+h o
h
h
function G = max + G , G . It is assumed that private and public
h
goods are normal so 0 < 0 < 1. If G > G , the demand function can be inverted
to give g h = h (G) where (G) = 1 (G) G and 0 < 0 (G) < . Define
by the level of endowment above which the household becomes a contributor
and note = (G). Then
gh =
h if h > ,
0 if h .
= 1 ( ).
h
1
P
1
Now consider the function MH (s) H(s) = H
h >s s which, for
given endowments for the H households, has solution s = H . Now add to the
vector of endowments by making random draws from the density f (w). By the
R
law of large numbers, MH (s) tends to M (s) = s [ s] f (w) d as H tends
to infinity. Denote the solution to this equation , then limH H = .
Hence G =
PH
h=1
gh =
h >
1 ( )
H
= 0.
Since H is bounded and 1 is finite, limH MH = limH
H
R
It then follows that M ( ) = s [ ] f (w) d = 0. From this, =

and limH H = .
From this result, (i), (ii) and (iv) follow immediately. (iii) follows from noting
that G = 1 (). ||
This theorem has been generalised by Andreoni (1988) to economies with
heterogeneous preferences. In that case, the same conclusions hold with the
addition that the set of contributors will converge to a set of individuals of a
single type.
9.5.5
Invariance results
The private provision equilibrium is characterised by some surprising invariance

properties. Changes in the endowment distribution that satisfy certain conditions will not aect the total level of provision. An increase in exogeneous public
good supply, for instance through the government providing some public good
in addition to the private sector, will not aect total supply in the limit. Public
provision therefore crowds out private provision on a one-for-one basis. These
results are now demonstrated.
In most economic situations a change in the endowment, or income, distribution will aect the equilibrium except when households have identical ane
Engel curves. For the private provision equilibrium, Warr (1982) proved the
much stronger result that, provided all households contribute, the total level
of public good is independent of the endowment distribution. This result was
extended by Bergstrom, Blume and Varian (1986) to allow for the possibility of
non-contributors. This result is given in the following theorem.
Theorem 9.5 (Bergstrom, Blume and Varian)
280
The total provision of the public good is unaected by any income redistribution that leaves the set, C, of contributors unchanged.
Proof
P
h
in the proof of the previous theorem that G = H
h=1 g =
h shown
P It was
and that is independent of the income distribution. From

hC
this equality it follows trivially that any redistribution of income that does not
change the set of contributors will leave G unchanged. ||
To consider the eect of an exogeneous increase
in public good supply,
notice that if h chooses G > Gh , then G = h + Gh implies that g h =
h + Gh Gh . Now let the exogeneous increase in supply be denoted by

d. The response of household h to this change is given by
(9.39)
dg h = 0 dGh + d dGh d,
where dGh is the total response of households other than h. Rearranging and
summing across all households gives
dG =
PH
1+
1 0
h=1 0
PH 1 0
h=1 0
d,
(9.40)
from which the total change in supply can be calculated as

dG + d =
1+
1
PH
h=1
1 0
0
d 0,
(9.41)
assuming normality, so that 0 is strictly less than 1. There exists such that
P
1 0
H H
h=1 0 . Substituting into (9.40) this implies
dG + d =
1
d
1 + H
(9.42)
From (9.41) the invariance result of Andreoni (1988) follows as a consequence

of noting that the total change in supply tends to zero as H tends to infinity.
For large populations, an exogeneous increase in public good supply is exactly
met by a reduction in private provision. Expressed dierently, any government
provision will crowd out an exactly equal amount of private provision.
9.5.6
Alternative formulations
The properties of the private provision equilibrium that have been derived have
been taken as indicative of the failure of the equilibrium concept to accurately
capture reality. The conclusion that only the richest household in a large population will contribute is not an accurate representation of, for example, charitable
contributions in the United States. Nor does the average level of contribution
appear to be close to zero. Crowding out has been estimated to be of the order
of a 5 to 28 cent reduction for each dollar of government spending (Abrams and
281
Schmitz (1984)), not the one-for-one predicted by the analysis. At the private
provision equilibrium an increase in contribution by one household will lead to
a reduction by all others. This feature has also been criticised as an inaccurate
representation of reality. These observations have lead a number of authors to
investigate alternative maximisation procedures for the households and dierent
preference structures with the view to generating equilibria whose properties are
more in accord with empirical observations.
The simplest modification to the private provision equilibrium is to consider
conjectural variations that dier from the Nash conjecture. In the maximisation
in (9.31) each household takes the contributions of others as given when making
their decision. The alternative to this is to assume that the household views the
choices of others as being dependent upon their decision and takes this variation
into account when maximising. Including this variation, the first-order condition
(9.32) would becomes
h
Uxh pG + UG
1+
j
H
X
j=1,j6=h
g j
= 0,
g h
(9.43)
g
where the terms g
h are the conjectural variations. Cornes and Sandler (1984b)
investigate the eects of alternative values for the conjectural variation and show
that if they are positive the equilibrium will have greater total public good
supply than the Nash equilibrium. Despite this modification, Dasgupta and
Itaya (1992) demonstrate that the invariance to the distribution of endowment
still holds for any constant conjectures.
Consistent conjectures are those which agree with the actual responses of the
g j
households involved, that is g
h must be equal to the change in contribution
that j would make if h were to change theirs. Sugden (1985) argues that the
g j
only consistent conjectures are for all the terms g
h to be negative. If this is
the case, equilibrium provision of the public good will be zero under reasonable
assumptions.
Moving to non-Nash conjectures can therefore alter the equilibrium level
of the public good but does not necessarily eliminate the invariance properties.
Overall, this approach must be judged as somewhat arbitrary. There are sensible
game-theoretic motives for focusing upon the Nash equilibrium and these are
not matched by any other set of conjectures. If the Nash equilibrium of the
private provision economy does not agree with observations, it would seem that
the objectives of the households and the social rules they observe should be
reconsidered, not the conjectures they hold when maximising.
One approach to modified preferences has been taken by Andreoni (1989,
1990) who considers the case of impure altruism given by the utility function
U h = U h xh , G, g h . Here the contribution of h provides both a public and
a private return. This representation of utility leads to equilibria that are not
invariant and produce predictions closer to observed behaviour. There remains
the problem of why a pure public good should provide the private benefit.
282
A final modification that has been considered is to remove the individualism

by modifying the rules of social behaviour. Sugden (1984) considers the principle
of reciprocity by which each household considers the contributions of others and
contrasts them to what they feel they should make. If the contributions of others
match, or exceed, what is expected then household h is assumed to feel under
an obligation to make a similar contribution. Formalising this notion, Sugden
(1984) proves that the outcome will be Pareto ecient if all households agree
upon the expected contributions. In all other cases Pareto improvements can
be made by increasing supply. Bordignon (1990) has provided a formalisation
of Kantian behaviour that leads to similar conclusions.
9.5.7
Summary
In the absence of government intervention, public good provision will be left to

the private contributions of households. The basic model of private provision is
built upon the assumption of Nash behaviour and it has been shown that this
leads to an inecient outcome. In addition, the equilibrium level of provision is
invariant to changes in the income distribution and exogeneous changes in public
good supply. These properties, and the limiting behaviour of the equilibrium,
are at variance with empirical observations. Alternative conjectures have been
analysed but these are entirely arbitrary and do not provide a better explanation
of reality. Altering the structure of preferences and the social rules do provide
improved predictions but no alternative has yet received convincing arguments
in its favour.
9.6
Finance by taxation
The rules for ecient provision derived in Section 3 require for their implementation that there are no restrictions upon the tax instruments that can be
employed by the government or, equivalently, that the government has complete
control over resource allocation. When optimal lump-sum taxes are not an available policy instrument the rule for provision, and the resulting level of provision,
must take account of the method of finance. In particular, the gain in welfare
enjoyed due to the provision of the public good has to be oset against any
distortions caused by the method of finance. This section considers the implications of methods of financing in an economy with a set of identical consumers,
which eectively behaves as a single-consumer economy with no distributional
aspects, and in a Diamond-Mirrlees economy with many consumers.
9.6.1
Identical consumers
In an economy with many identical consumers, if a lump-sum tax can be employed at all it must feasible to choose the optimal lump-sum tax. To provide
content to the analysis it is therefore assumed that the only tax instruments
available to the government to finance the public good are commodity taxes.
9.6. FINANCE BY TAXATION
283
This will be relaxed when dierentiated consumers are considered and restricted
lump-sum taxes become meaningful. The aim of the analysis is to determine
how the distortions caused by the commodities taxes aect the Samuelson Rule
and the level of provision. This is undertaken by following the work of Atkinson
and Stern (1974).
Each of the identical consumers maximises their utility U (x, G) subject to
the budget constraint qx = 0 where q is the vector of post-tax prices and x the
vector of net demands. There is one change from the standard commodity tax
model: revenue must now equal expenditure, G, on the public good
H
n
X
ti xi = G.
(9.44)
i=1
Market clearing implies that the revenue constraint and the production constraint may be used interchangeably as argued in Chapter 2 above. The production constraint is used and is written in the form F (X, G) = F (Hx, G) = 0.
It is assumed that
F
F1 =
,
(9.45)
X1
and good 1 is taken to be the numeraire with q1 = p1 = 1. Pre-tax prices are
chosen so that Fk = pk .
For the choice of optimal tax rates, t, and quantity of public good, the
appropriate Lagrangean is
L = HV (q, G) F (X (q, G) , G) .
From this, the first-order condition for the choice of G is
#
" n
X Xi
L
V
Fi
H
+ FG = 0,
G
G
G
i=1
(9.46)
(9.47)
which, using the definition of pre-tax prices, can be written

H
V
G
qk
n
pk FG
X Xi
+
pi
,
qk Fk
qk i=1 G
(9.48)
where is the marginal utility of income for each consumer. From each conU
sumers first-order condition for the utility maximising choice of good k, x
=
k
qk . Therefore the term
H
V
G
qk
=H
U
G
U
xk
(9.49)
is the sum of marginal rates of substitution between the public good and private
good k.
Evaluating the first-order condition (9.47) for k = 1, so that q1 = 1, it can
be rearranged to give
FG
= H
F1
U
G
U
xk
n
X
i=1
[qi ti ]
Xi
,
G
(9.50)
284
Pn
i
but since the consumers budget constraints imply i=1 qi X
G = 0, (9.49) can
be written
P
ni=1 ti Xi
FG
U
= H G
.
(9.51)
U
F1
x
G
k
The expression in (9.50) represents the Samuelson rule for optimal provision
with distorting commodity taxation. It diers from the first-best rule (9.9) in
two ways: the sum of marginal rates of substitution is multiplied by the term
which may not be equal to one and there is an additional term on the right-hand
side. This additional term measures the eect of public good provision upon
tax revenue due to substitutability or complementarity in demand between the
public good and the private goods.
The revenue eect implies that if provision of the public good increases tax
revenue, which will be the case for example if it is a complement to highly taxed
goods, this reduces the cost , measured by the M RT , of providing the public
good. This factor tends to increase provision above the level determined by the
Samuelson rule. The converse holds if provision
reduces tax revenue.
P
n
i=1 ti Xi
To isolate the first eect assume that
= 0 so that the public good
G
is revenue neutral. In this case the departure from the first-best is determined
by
alone. To proceed further, consider the choice of tax rate for good k. From
the Lagrangean
Pn
n
X
i=1 pi Xi
Xi
V
=
Fi
=
.
(9.52)
H
qk
qk
tk
i=1
Using Roys identity and the fact that
Pn
Pn
Pn
i=1 pi Xi
i=1 ti Xi
i=1 qi Xi
+
=
,
tk
tk
tk
(9.51) can be written
Finally, using the Slutsky equation,
Pn
(9.53)
i=1 ti Xi
tk
Xk
n
n
X
Xi X Sik
ti
ti
.
=1
+
I
Xk
i=1
i=1
(9.54)
(9.55)
From (9.54) it can be seen that the divergence of from 1 can be separated
P
i
into two components: (i) a revenue eect given by ni=1 ti X
I ; and (ii) a disPn
tortionary eect i=1 ti SXikk . From the negative semi-definiteness of the Slutsky
matrix it follows that
n
X
Sik
ti
0.
(9.56)
Xk
i=1
The negativity in (9.55) has the eect of tending to reduce belowP

. This
would imply that the true benefit of the public good is less than the
M RS.
285
Pn
i
In contrast, the second eect i=1 ti X
I cannot be unambiguously signed. If it
were positive then would be less than . This would be the case if all taxed
goods were normal, but this may not be the case.
In summary, when the public good has
P to be financed by distortionary taxation there is a divergence of M RT from M RS. This divergence can be broken
down into the eect of provision on tax revenue and two further components:
one of which always P
reduces the benefit measure so that the benefit becomes
something less than
M RS, the other is also likely to reduce it though this
cannot be guaranteed. An alternative perspective upon this issue can be found
in Christiansen (1981) who derives sucient conditions
P for the standard costbenefit calculation of comparing the M RT to the
M RS to be valid even
when there are distributional objectives. Adopting the Mirrlees formulation of
labour supply described in Chapter 5 but with a public good and a second private good incorporated in utility, Christiansen shows that weak separability of
utility between work eort and all other goods combined with the optimal taxation of income is sucient to eliminate the distributional factors and leave only
the basic cost-benefit calculation. Outside of these special cases, the additional
eects described above must be included in the optimality calculation.
The analysis above has been addressed to finding the appropriate benefit
measure in the presence of distortion and has not directly answered the question
of whether more or less public good should be provided when its provision is
financed by distortionary taxation. Unfortunately no complete answer can be
given to this.
Consider an economy where finance is possible either through a lump-sum
tax T or a tax t is levied upon the single factor of production L . The Lagrangean
is
HV (t, T, G) + [HT + tHL G] ,
(9.57)
with first-order necessary conditions
L
Vt + L + t
= 0,
t
(9.58)
HVG = 0,
(9.59)
HT + HtL G = 0.
(9.60)
T is treated as a parameter that can vary from 0 to G.

To provide an answer on the level of provision it would be necessary to
demonstrate that (9.57) for T = 0 and T = G could be drawn as, for example,
in Figure 9.8. In the case shown provision is lower when distortionary taxes are
employed. However, to actually show that GCT < GLS , where subscripts refer
to commodity taxes or lump sum, it is necessary to contrast CT and LS .
The analysis has only been able to determine
which is not sucient for this
purpose. The question of output levels therefore remains unanswered.
In terms of examples, Atkinson and Stern (1974) show that the for the utility
function
U (X, L, G) = a log X + [1 a] log (1 L) + f (G) ,
(9.61)
286
Figure 9.8: Comparison of Lump-Sum and Commodity Tax Finance
the level of provision of public good is lower with commodity taxation than for
the first-best. Wilson (1991a) has extended this conclusion to the function
U (X, L, G) = U ( (X, 1 L) , G) ,
(9.62)
where (X, 1 L) is of the CES form and G is assumed to be a normal good. No

example has been published that has higher provision with commodity taxation
than at the first-best. This lends a degree of support for the conclusion that the
first-best level will exceed the second-best.
9.6.2
Dierentiated households
Allowing the households to dier in their income and preferences provides motivation for considering restricted forms of lump-sum taxation. Even though an
optimal set of lump-sum taxes may not be feasible, it may well remain possible
to levy a uniform lump-sum tax. At the margin, such taxes have the property
of providing a non-distortionary source of finance.
Now consider an economy with an arbitrary, but non-zero, set of commodity
tax rates and implied consumer prices which are taken as fixed. Since the
private sector of the economy is distorted by the existence of the commodity
taxes, the non-distortionary uniform lump-sum tax provides a means by which
resources can be moved from the distorted sector into provision of the public
good. It does not then seem unreasonable that if the private sector is suciently
distorted, more resources would be moved to the public sector than at the firstbest optimum. This will result in the second-best provision of the public good
being above the first-best.
This reasoning has been formalised by Wilson (1991b) who derives a condition that determines the relative levels of public good provision and provides
examples to confirm that second-best provision may be greater than first-best.
287
Figure 9.9: Manipulating the Lindahl Equilibrium
9.6.3
Summary
The results have shown how distortionary financing aects the form of the
Samuelson rule. Although there is a presumption that when finance is entirely
by commodity taxation the second-best level of provision will fall below the
first-best, this has not been formally established and the form of (9.50) suggests
that it cannot be. In contrast, when some financing can be undertaken by lumpsum taxation their non-distortionary nature provides a reason for second-best
provision to rise above first-best and examples have been constructed to confirm
this.Mechanism design
That a household will choose the action that leads to the maximisation of
their welfare is one of the basic assumptions of economic theory. If applied
consistently, this assumption implies that a household will behave dishonestly
if it is in their interests to do so. This simple observation has surprisingly
important implications for the theory of public goods.
The analysis of the Lindahl equilibrium assumed that households were honest in revealing their reactions to the announcement of cost shares. However,
there will be a gain to households who attempt to cheat, or manipulate, the
allocation mechanism. By announcing preferences that do not coincide with
their true preferences, it is possible for a household to modify the outcome in
their favour provided that others do not do likewise. To see this, consider a twohousehold economy in which household 1 acts honestly and household 2 knows
the preferences of 1. In Figure 9.9, honesty on the part of household 2 would
lead to the equilibrium EL . However, by claiming their preferences to be given
by the Lindahl reaction function L20 ( 2 ), rather than the true function L( 2 ),
the equilibrium can be driven to point M which represents the maximisation of
2s utility given the Lindahl reaction function L1 ( 1 ) of 1.
Due to this problem, attention has focused upon the design of allocation
mechanisms that overcome attempted manipulation. As will be shown, the
design of some of these mechanisms leads households to reveal their true preferences. From this property is derived the description of these mechanisms as
288
preference revelation mechanisms.

The general form of allocation mechanism can be described as a game in
which each household has a strategy set and chooses a strategy from this set
in order to maximise their pay-o. The aim of the analysis is to determine
when a game can be constructed such that the equilibrium strategies lead to
the allocation that the policy maker wishes to see implemented. The game
will be one of incomplete information since it is natural to assume that each
household has knowledge only of its own payo function. For such games a
number of equilibrium concepts may be employed. Of these, most attention in
the early literature was upon the dominant strategy equilibrium, where each
household has a dominant strategy regardless of the choices of others, and the
Nash equilibrium in which the chosen strategy must be optimal given that other
households play their equilibrium strategy. In the setting of incomplete information the dominant strategy equilibrium has the appealing property that, since
a dominant strategy is the best response to any choice by opponents, a household does not need to form any expectations about how others will play the
game. Consequently, the diculties caused by the incomplete information are
overcome. More recent work has also considered Bayesian equilibria in which
the information problem is explicitly modelled.
9.6.4
Definitions
The set of potential public projects from which the choice is to be made is
denoted by G, where G is a compact set, and a typical element of G is denoted
G. There are H households who have preferences over the public projects and
over monetary transfers, th , where the monetary transfers arise as part of the
game that is played by the households. For the present, each household is
assumed to have additively separable preferences given by
U h (G, th ) = h (G) + th .
(9.63)
The representation of preferences in (9.62) is interpreted as being net of the

costs of the project imputed to household h. It is then possible to term h (G)
the valuation of project G by household h.
Q
h
The strategy space of household h is given by S h and S = H
h=1 S . A
h
h
strategy choice for household h is denoted
by s S with a set of strategy
choices for the households denoted s = s1 , ..., sH . A game form, , consists of

a strategy space and a payo function, h (s), for each household. If the outcome
of the game is determined by a decision function d : S G, which determines
the project chosen given the announced strategies, and a transfer rule t : S
<h , t (s) = (t1 (s) , ..., tH (s)), determining the transfer to each household, it is
termed a mechanism.
A mechanism is therefore composed of a set of strategy
spaces, S h , and a function f : S G <h , f [d (s) , t1 (s) , ..., tH (s)] where

the accepted project is d (s) and th (s) is the transfer received by household h.
Once determined, the decision d (s) and the transfers th (s) are enacted by the
government or centre.
289
A revelation mechanism is a mechanism for which the strategy space of

each household is the set of possible valuation functions and a strategy is the
announcement of a valuation function. The set of possible valuation functions for h is denoted V h . In the case that the strategy space is the set of
possible valuations, let the reported valuation of h be given by wh (). A direct revelation
sucho that
mechanism is thenn defined as a revelation mechanism
P
h
d (w) = d w1 () , ..., wH () G : G = arg max{G G} H
w
(G)
. A
h=1
direct revelation mechanism therefore has the property that the chosen project
maximises the sum of reported valuations and is therefore optimal given those
valuations. To ensure that the maximum exists, it is assumed the each h (G)
is upper semi-continuous. The reported valuations must then also be restricted
to be upper semi-continuous.
9.6.5
Dominant strategies
If a dominant strategy exists for each participant of a game, a strong argument

can be made that the participants will play those strategies. In addition, the
existence of dominant strategies removes any need to consider how the participants form expectations about the strategy choice of their opponents and
diculties introduced by the existence of incomplete information are overcome.
If the preference revelation problem can be solved by a mechanism involving
dominant strategies, there are persuasive reasons for recommending its use.
These observations provide the motivation for the study of dominant strategy
mechanisms.
A direct revelation mechanism with decision function d (w) and associated
transfers {th (w)} is termed strongly individually incentive compatible (s.i.i.c.)
if truth-telling is a dominant strategy. Announcement of the truth will be a
dominant strategy for all agents if and only if

h arg max h d wh , wh + th wh , wh , wh , h = 1, ..., H,

{wh V h }
(9.64)
where wh = w1 , ..., wh1 , wh+1 , ..., wH , so that the true valuation maximises
the payo. To demonstrate the existence of a mechanism that satisfies (9.63),
first note that from the definition of a direct revelation mechanism it follows
that d (w) has the property that
h arg
max h d wh , wh +
{wh V h }
H
X
j=1,j6=h
wj d wh , wh , h = 1, ..., H.
(9.65)
Hence truthful revelation by h maximises the sum of the payo to h and the
payos, in terms of announced valuations, to the other households. Now write
the transfers in the form
th (w) =
H
X
j=1,j6=h
wj (d (w)) + rh (w) , h = 1, ..., H,
(9.66)
290
which can always be done by suitable choice of the functions rh (w). Using
(9.65), the dominant strategy condition (9.63) can be written
h arg
H
X

max h d wh , wh +
wj d wh , wh +rh wh , wh , h = 1, ..., H.
{wh V h }
j=1,j6=h
(9.67)
Contrasting (9.66) and (9.64) it can be seen
that
social
and
individual
objec
tives will coincide whenever rh wh , wh is independent of wh . Following its
introduction in Groves (1973), a Groves scheme is defined by a set of transfers
such that
th (w) =
H
X
j=1,j6=h
wj (d (w)) + rh wh , h = 1, ..., H,
(9.68)
and a Groves mechanism is a direct revelation mechanism with the transfer rule
given by (9.67). The reasoning above can then be summarised in the following
theorem which was originally proved in Groves and Loeb (1975).
Theorem 45 (Groves and Loeb) A Groves mechanism is strongly individually
incentive compatible.
Proof. Directly from comparing (9.64) and (9.66) after substitution of
(9.67).
The interpretation of the Groves mechanism is that the transfers are designed
so that the only eect the strategy choice of a household can have upon the
size of the transfer is via the eect that the decision on the public project,
based upon that strategy, has upon other households welfare. There is no
direct eect on the transfer. This mechanism can be viewed as internalising
the external consequences of the strategy choice of each household, since the
external consequences are given by the welfare eects on other households of
the public decision. In a Clarke
mechanism, which was introduced by Clarke
(1971), the functions rh wh are given by
H
X
wj dh wh , h = 1, ..., H,
rh wh =
(9.69)
j=1,j6=h
PH
where dh wh is the maximiser of j=1,j6=h wj (G). In this case the transfer is
exactly the change in welfare of other households due to influence of household
h on the public project decision. This is a special case of the Groves mechanism.
Having demonstrated the existence of an s.i.i.c. mechanism it is natural to
consider whether there are any alternative mechanisms that will also serve the
same purpose. The following analysis will show that there are not: the Groves
mechanism is the only s.i.i.c. direct revelation mechanism. To demonstrate this
result, which is due to Green and Laont (1977), it is first necessary to provide
the following definition.
291
Direct revelation mechanism

A direct revelation mechanism satisfies
(i) the transfer independence
property
G i.e.,
h
if th (w)his independent
of h
if
h
h
h0
h
h0
forw , w ,w such that G w , w = G w , w
then th w , wh =
th wh , wh0 ;
PH
(ii) the compensation property if th wh , wh th wh , wh0 = j=1,j6=h wj (G )
PH
j
0
j=1,j6=h w (G ) where G maximises the sum of announced valuations given

h
0
w and G is the maximising project given announcement wh0 .
It is immediately clear from this definition that the transfer independence
and compensation properties uniquely characterise the Groves mechanism. A
direct revelation mechanism is therefore a Groves mechanism if and only if it
satisfies transfer independence and compensation.
The proof that the Groves mechanism is the only s.i.i.c. direct revelation
mechanism can now be undertaken by demonstrating that if transfer independence and compensation are not satisfied by a direct revelation mechanism, it
cannot be s.i.i.c. Therefore it follows from the equivalence of these conditions to
the Groves mechanism that the Groves mechanism is the unique s.i.i.c. direct
revelation mechanism. The relevant theorem is as follows.
Theorem 46 (Green and Laont) An s.i.i.c. direct revelation mechanism must

satisfy the transfer independence and compensation properties.
Proof. First assume that there exists an s.i.i.c. direct revelation mechanism
h
h
h0
for which transfer independence is not
there are
satisfied.
Then
h
w , w , hw
h
h
h0
with the same optimal
G but with th w , w > th w , w . Now let =
wh0 and note that th wh , wh + h (G ) > th wh , h + h (G ). The choice
of h is then not a dominant strategy. An s.i.i.c. direct revelation mechanism
must therefore satisfy transfer independence.
Now assume that the mechanism does not satisfy the compensation
property.
PH
There then exist wh , wh , wh0 such that G maximises j=1,j6=h wj + wh over
PH
G, G0 maximises j=1,j6=h wj + wh0 over G and th wh , wh th wh , wh0 =
PH
PH
j
j
0
j=1,j6=h w (G ) =
j=1,j6=h w (G ) + for some > 0.
PH
PH
0
0
0
eh (G ) = j=1,j6=h wj (G ) , w
eh (G0 ) = j=1,j6=h wj (G0 )+
Let w
eh be defined by w
P
0
H
eh (G) = c for G 6= G or G0 with c > max{GG} j=1,j6=h wj (G) +

2 and w
0
0
eh is upper semi-continuous, it is a permissible valuation anw
eh (G). Since w
nouncement.
P
0
j
Now note that G0 = arg max{GG} H
eh (G), since it has
j=1,j6=h w (G) + w
0
been shown that the mechanism must satisfy transfer independence, th wh , wh =
292
0
th wh , w
eh . Hence
0
0
eh
= th wh , wh th wh , w
th wh , wh th wh , wh
=
H
X
j=1,j6=h
wj (G )
H
X
wj (G0 ) +
j=1,j6=h
= w
e (G ) + w
e (G0 ) + .
2
0
0
0
eh + w
eh (G0 ). Letting
eh (G ) > th wh , w
Rearranging gives th wh , wh + w
h0
h0
h w
eh then contradicts the assumption that the mechanism was s.i.i.c since
truth is not the dominant strategy. A s.i.i.c. direct revelation mechanism must
therefore satisfy the compensation property.
In the form given, Theorem 9.7 only restricts the valuation functions to be in
the class of semi-continuous functions. However, restricting the set from which
valuation functions may be drawn does not extend the set of direct revelation
mechanisms. As proved by Holmstrom (1979), provided the domain of valuation
functions is smoothly connected (that is, there exists a dierentiable deformation
of one valuation function in the domain into any other), the Groves scheme will
be unique.
These results permit the study of s.i.i.c. direct revelation mechanisms to be
undertaken by studying only the properties of Groves mechanisms. Although it
has been shown that a Groves mechanism can implement the correct choice of
project, it should not be presumed that it will necessarily lead to a fully-ecient
equilibrium. To be ecient the mechanism must also have the property that the
sum of transfers must be identically zero, in which case the mechanism is termed
balanced. If it is negative some resources are being taken from the households
and not used to produce welfare elsewhere. It may be felt that these resources
could be redistributed to the households but if this were done the incentives
would be introduced for the households to choose their announced valuation
taking into the eect it would have upon the redistribution. The s.i.i.c. aspect
of the mechanism would then fail. Conversely, if the sum of transfers is positive,
some resources are being transferred from the centre to the households and the
centre is bearing the cost of implementing the mechanism.
To investigate the possibilities, assume that the valuation functions are differentiable and can be given a one-dimensional parametrisation. Let household
h be described by the preference parameter h hwhere h is an open interval
of < and assume the valuation function h G, h is smooth. Finally, assume
there for any array of household preferences = 1 , ..., H , there exists a dif
P
h
h
ferentiable function G () > 0 such that G () = arg max{G>0} H
G,
.
h=1
The following theorem describes the conditions necessary for the existence of a
balanced mechanism.
Theorem 47 (Green and Laont) There exists a balanced mechanism if and

only if
H
X
H1
1
h1
, ...,
, h+1 , ..., , H
h=1
293
h G
0.
G h
Proof. In the case described the transfer functions will depend only on the
h
e
parameter vector . Denoting the true preference parameter
of h byh , truth
h
h
+ th ()
will be the dominant strategy if e
= arg max{h h } G () , e
h
h
th
G
0 when evaluated at e
for all
for all h . This will only apply if
G h + h
R
h
G
h . Integrating with respect to h then gives th h , h =
G h d +
rh h , where rh h is arbitrary.
i
PH hR h G
The mechanism is therefore balanced if h=1
G h d + rh h
0. Dierentiating this condition for a balanced
with respect to
h h mechanism
i
PH
1
H
H1
G
, ..., gives h=1 1 ,...,h1 ,h+1 ,...,,H G h 0 which establishes
necessity. Suciency is proved by integrating the stated condition successively
to obtain the condition for a balanced mechanism.
Since the condition required for the Groves mechanism to be balanced even in
the simple environment of parametrised preferences is restrictive, this theorem
demonstrates that it will not be possible to always find a balanced mechanism.
This result is established formally by Green and Laont (1979) who prove that
there exists no Groves mechanism that is balanced for all possible valuations.
It is therefore unlikely that a Groves mechanism will achieve full eciency. The
lack of eciency is, of course, the price that has to be paid for the revelation of
information.
These negative conclusions are somewhat diminished as the size of the population of households increases. As shown by Green and Laont (1979) and
Rob (1982), the likelihood that any household receives a negative transfer (i.e.
pays a tax) tends to zero as the population increases without bound and the
expected total transfer also tends to zero. These results imply that the limiting
outcome is approximately ecient.
A important assumption that has been employed up to this point has been
the additively separability of preferences described in (9.62). Given the restrictiveness of this assumption, it would be hoped that this was simply a convenient
assumption that could be dropped when necessary. Unfortunately this is not
the case. Bergstrom and Cornes (1983) have shown that the Groves mechanism
will still function successfully if preferences are given by
U h xh , G = A (G) xh + B h (G) , h = 1, ..., H,

(9.70)
where A (G), which is assumed to be known, is common to all individuals and
xh is the quantity of private good consumed by h. This, however, is only a
minor extension.
The possibility of proving a general result on the existence of an s.i.i.c. is
ruled out by the Gibbard-Satterthwaite theorem. To describe the content of this
294
theorem, consider a set of social states and a set of economic agents. Each
of the H economic agents has a preference order, P h , defined over . Let be
the set of possible preference orders. A social choice function (SCF) is defined
H
as a function from H
h=1 into which assigns to any P h=1 a social
state . An SCF, W
, is manipulable at the preference profile P if there
exists P h0 such that W P 1 ,..., P h0 , ..., P H is preferred by h, with preference
P h , to W P 1 , ..., P h , ..., P H . Manipulability means that h can alter the social
decision to one they prefer by announcing a set of preferences, P h0 , which dier
from their true preferences. The SCF is s.i.i.c. if there is no set of preferences,
P , at which it is manipulable. Now let the range of the SCF, W , be 0 .
The SCF is dictatorial when for any 0 , 6= W (P ), W (P ) is preferred by
h to . Agent h is then said to be the dictator.
Given these preliminaries, the Gibbard-Satterthwaite theorem can now be
stated.
Theorem 48 (Gibbard-Satterthwaite) If 0 has at least three alternatives, an
SCF with range 0 satisfying s.i.i.c. and having an unrestricted domain (so no
restrictions on preferences are permitted) is dictatorial.
Proof. See Laont (1987).
The implication of the Gibbard-Satterthwaite theorem is that the search for
a mechanism that will eciently implement social decisions in all circumstances
is a futile one. As has been shown, the Groves mechanism illustrates that
implementation can occur in some restricted circumstances but there is little
hope for proceeding very far beyond this. This negative conclusion provides the
motivation for studying equilibrium concepts that are weaker than dominant
strategies.
9.6.6
Nash equilibrium
The simplest alternative equilibrium concept is that of Nash equilibrium, so that

each households strategy has to be optimal only against the optimal choices of
other households. This weakening of the equilibrium concept does open up new
possibilities and the work of Groves and Ledyard (1977) demonstrates that Nash
equilibrium can lead to optimality. However, there remain diculties with the
interpretation of the Nash equilibrium concept in the context of games with
incomplete information. These will be discussed further below after the results
that can be obtained are described.
Consider a set of social states and an SCF, W , W : H
h=1 . The
game form implements W if, for every P H
h=1 , the set of equilibrium
strategies for , denoted
E (P), is non-empty and (E (P )) W (P ). A set
of strategy choices s1 , ..., sH are a Nash equilibrium for the game form if
and only if for all h
h sh , sh > h sh , sh , sh S h .
(9.71)
295
The first result of this subsection, due to Dasgupta et al. (1979), concerns
the relationship between truthful implementation in Nash equilibrium and in
dominant strategies.
Theorem 49 (Dasgupta, Hammond and Maskin) The social choice function W
can be truthfully implemented in Nash strategies if and only if it can be truthfully
implemented in dominant strategies.
Proof. If W can be implemented in dominant strategies, it can be implemented in Nash since a dominant strategy equilibrium is a Nash equilibrium. Now assume that the game form truthfully implements W in Nash
strategies. For all P H
h=1 , truthful revelation is then a Nash equilibH
h0
rium.
This
implies
that
for
h h
h0 h all h, for all P h=1 , and for all P ,
h
h
P ,P
> P ,P
. Truth is therefore the dominant strategy.
The implication of this theorem is that there is nothing to be gained by
considering Nash equilibria in direct mechanisms since the same result could be
obtained by considering only dominant strategies. If the Nash equilibrium is to
be of value in extending the set of circumstances in which implementation can
take place it will be necessary to consider mechanisms in which individuals do
not announce their true preferences at the equilibrium.
The major contribution of Groves and Ledyard (1977) was to provide a
mechanism that was balanced and which lead to a Pareto optimal allocation.
Assume that there are available K public goods and denote the price vector (to
the public sector) of these public goods by pG . The Groves-Ledyard mechanism
then restricts strategy space of each household to be <K . This can be given the
interpretation that each household announces the incremental adjustment they
would like to be made to the sum of public goods supplies requested by other
households. With this restricted strategy space, the mechanism clearly does not
elicit
functions from the households. For
1 theHvaluation
Pa set hof strategy choices

s , ..., s
the decision function is given by d (s) H
h=1 s and the transfer
functions by
H
X
1
2
2
th (s) h pG
sj +
sh , h = 1, ..., H,
sh sh
2
H
j=1
(9.72)
h PH
h 2
h 2
h
P
PH
1
h
j
where H
=
1,
>
0,
s
s
and
h=1
j=1,j6=h
j=1,j6=h s s
H2
Under this transfer rule the transfer from household h increases as their request
deviates further from the average of other requests and is reduced as the squared
standard error of the others requests increases.
Although the Groves-Ledyard mechanism is balanced and achieves Pareto
optimality, diculties still remain. In the context of the private ownership
economy with government in which they embedded the mechanism, there is no
reason to believe that the mechanism is individually rational. That is, after
participating in the mechanism a household may be left with a utility level
lower than that given by their initial endowment. In these circumstances the
296
rational act for the household would be not to engage in trade but to simply
consume their endowment. Imposing individual rationality, Hurwicz (1979a,
1979b) demonstrates, under mild continuity and convexity restrictions, that the
set of Lindahl allocations is a subset of the Nash equilibria and all interior Nash
equilibria are Lindahl allocations.
These results would seem to suggest that implementation in Nash strategies
provides an attractive means of overcoming the failings of the dominant strategy
mechanisms. Unfortunately, Nash mechanisms are not without their failings.
The major diculty is the level of knowledge implied by the equilibrium concept.
Interpreted literally, it requires each household to know the preferences of the
others and to be able to solve the game for the set of equilibrium strategies.
Such common knowledge amongst the households is in sharp contrast to the
lack of knowledge of the centre. Although it has been argued that there may
be situations when this is appropriate (see Moore and Repullo 1988), the public
goods problem is almost certainly not amongst them. Alternatively, the Nash
equilibrium could be seen as being achieved as the outcome of some iterative
process but, again, this falls outside the scenario envisaged.
9.6.7
Bayesian equilibria
Bayesian equilibria, in which each household holds a probability distribution

over the distribution of preferences in the population and maximise expected
utility subject to this, have also been considered most notably by dAspremont
and Grard-Varet (1979) and Laont and Maskin (1982). Although Bayesian
equilibria can generate fully ecient outcomes, in common with Nash equilibria
they suer from informational diculties. In order that a household can calculate an equilibrium strategy the probability distributions held by the households must be common knowledge. In addition, Bayesian mechanisms suer
from multiplicity of equilibria. These points alone are sucient to cast serious
doubt upon the relevance of Bayesian equilibria.
9.6.8
Conclusions
This review of mechanism design has included some positive results and some
negative ones. It has been shown possible to achieve ecient, though not always fully-ecient, outcomes when preferences are separable using dominant
strategies. Unfortunately, the Gibbard-Satterthwaite theorem shows that this
result cannot be extended to apply to all possible forms of preferences. Eciency is attainable using Nash equilibrium but the informational requirements
of the equilibrium concept appear unsustainable in the public goods context.
The same is also true of Bayesian equilibrium. The mechanism design problem
is therefore far from fully solved.
9.7. EXPERIMENTAL EVIDENCE AND MARKET DATA
9.7
297
Experimental evidence and market data
The previous section has considered the design of mechanisms to elicit households valuations of public goods. The presumption that such a mechanism is
required is based on the view that households act always to maximise welfare
even when this involves dishonesty. Whether this view is justified has been
tested in numerous experiments and the results of a number of these are described in section 9.1. Although households may have an incentive to not reveal
their true preferences in markets involving public goods, this does not apply
in markets for private goods. This suggests the possibility that behaviour in
markets for private goods may reveal information about preferences for public
goods.
9.7.1
Experimental evidence
The analysis of private provision demonstrated that the equilibrium will not be
Pareto ecient and that, compared to Pareto-improving allocations, too little of
the public good will be supplied. A simple explanation of this result can be given
in terms of each household relying on others to contribute and hence deciding to
contribute little themselves. In a sense, each household is free-riding on others
contributions and, since all attempt to free-ride, the total contribution fails
to reach an ecient level. The similar, but distinct, concept of misrevelation
of preferences was analysed in the previous section. If asked to reveal their
valuation of a public good, each household will typically have strategic motives
for revealing a false valuation in order to manipulate the allocation mechanism
in their favour.
The experimental literature on public goods has attempted to test the predictions of these theories and to evaluate methods of overcoming misrevelation.
Unfortunately, some of the literature has not been suciently careful in the distinction between free-riding and misrevelation with the result that some papers
discuss examples of misrevelation and then conduct an experiment that tests
the free-riding hypothesis. With this in mind, the results of a number of these
experiments are now discussed.
Marwell and Ames (1981) report the findings of a series of experiments. The
basic structure of the experiments, which was maintained in many later experiments, was to give participants a number of tokens that could be invested in
either an individual exchange or a group exchange. The individual exchange had
a set repayment per token that was independent of the total investment by the
individual and of the total investment by the group. This should be interpreted
as the purchase of a private good. In contrast, the return from the group exchange was dependent upon the investment into the group exchange of the other
participants and an investment in the group exchange by any participant lead to
a payment to all participants. The group exchange therefore has the properties
of a pure public good. The rates of return were made known to the participants
and the total return from investment was paid to the participant. It was therefore in the interests of each participant to maximise their payo. The structure
298
of payos were such that the private provision equilibrium would result in no
investments in the public good although the ecient outcome would involve
only investment in the public good. Each participant made a single investment
decision.
Although the structure of payos, the distribution of tokens and the nature
of the public good (divisible or lumpy) changed between experiments, the results
were remarkable consistent. In the first 11 experiments, the average investment
in the public good lay between 28% and 87% of tokens with most observations
falling in the 40%-50% range. In the 12th experiment, with a group of first-year
graduates in Economics, the investment in the public good fell to 20%. These
results clearly do not support the predictions of the private provision model.
Isaac, Walker and Thomas (1984) modified the experiment by repeating the
investment decision over ten rounds with the view that this should allow time for
the participants to learn about free-riding and develop the optimal strategy. The
results from this experiment are not as clear as in Marwell and Ames (1981) and
a wider range of investments occurs. Free-riding is not completely supported but
some instances are reported in which it does occur. However, this finding should
be treated with caution since having ten rounds of the game introduces aspects of
repeated game theory. While it remains true that the only credible equilibrium
of the repeated game is the private provision equilibrium of the corresponding
single-period game, it is possible that in the experiments some participants may
have been attempting to establish cooperative equilibria by playing in a fashion
that invited cooperation. Additionally, those not trained in game theory may
have been unable to derive the optimal strategy even though they could solve
the single-period game. Similar comments apply to Kim and Walker (1984) who
also employ a repeated game framework. In their experiments free-riding was
found to occur but some odd responses involving occasional large investments
in the public good also arose.
The repeated game framework was also applied by Isaac, McCue and Plott
(1985) who conducted nine experiments with undergraduate economists. It was
found that the contributions to the public good fell short of the Lindahl level
but remained above the zero level that would occur in the private provision
equilibrium. It is interesting to note that in experiments in which the Lindahl
level was announced after 5 or 6 rounds, the contributions then increased. Allowing communication also raised contributions but less noticeably. Overall, the
experiments show some evidence of free-riding but do not conform exactly to
the predictions of the private provision equilibrium. Isaac and Walker (1988)
use the same format to examine the eect of group size upon free-riding. The
results show that increasing group size leads to increased divergence from the
ecient outcome when accompanied by a decrease in marginal return from the
public good but the results do not support a pure numbers-in-group eect. This
finding is compatible with the theoretical finding that the eect of group size
on the divergence from optimality was in general indeterminate.
The results above indicate that there is little evidence of free-riding in singleperiod, or one-shot, games but in the repeated games the contributions fall towards the private provision level as the game is repeated. Andreoni (1988b)
9.7. EXPERIMENTAL EVIDENCE AND MARKET DATA
299
suggests two reasons for these findings and attempts to choose between them.
The first possibility is that the decay could be due to learning as the participants are initially unable to solve the game and only learn the optimal strategy
by repeated play. Alternatively, the pattern of contributions observed may be a
strategic choice where some cooperation in the early stages is part of that strategy. To distinguish between the two explanations, one set of participants in the
experiment were placed in dierent groups after each round of investment. Since
the groups were randomly chosen, this prevented any benefit existing from playing a long-term strategy. Alongside these varying groups were 3 groups which
remained fixed throughout the 10 rounds of the experiment. If the strategic
explanation is correct, the fixed groups should have higher contribution levels
in the early stages of the game. In addition to the groups playing a fixed number of 10 rounds, an identical number of participants again split into fixed and
varying groups were subject to 10 rounds of the game and then told the game
was to be restarted for another 10 rounds. If learning is the correct explanation,
restarting should not aect the level of contributions.
The results do not confirm either hypothesis. Contributions by the members
of fixed groups were always less than those made by members of the variable
groups and in both cases remained above the level that would be achieved in
the private provision equilibrium After the restart the level of contributions was
temporarily higher for the variable groups. For the fixed groups there was a
lasting increase in contribution after the restart.
Taken together, these experiments do not provide great support for the equilibrium based on the private provision economy with Nash behaviour. In the
single-period games free-riding is unambiguously rejected. Although it appears
after several rounds in repeated games, the explanation for the strategies involved is not entirely apparent. Neither a strategic nor a learning hypothesis
is confirmed. What seems to be occurring is that the participants are initially
guided more by a sense of fairness than by Nash behaviour. When this fairness is
not rewarded, the tendency is then to move towards the Nash equilibrium. The
failure of experimentation to support free-riding lends some encouragement to
the views of Johansen (1977) that although such behaviour may be individually
optimal, it is not actually observed in practice.
In a series of papers Bohm (1971, 1972 1984) has suggested a procedure for
dealing with misrevelation, provided experimental results and has described the
outcome of the procedure in practice. To provide a practical method for determining the valuation of a public good, Bohm suggests running two preference
revelation mechanisms simultaneously. The first mechanism should be designed
to lead to an under-reporting of the true valuation of the project and the second
to the over-reporting. The direction of misrevelation can easily be controlled by
the link between reported valuation and the charge levied for use of the public
good. For instance, if the charge is credibly announced to be zero then overreporting will always take place if provision of the public good is dependent
upon reported valuations being greater than its cost. If those whose valuations
are sought are randomly allocated between the first and second mechanism then
an upper and lower bound are obtained on the valuation of the public good with
300
the true valuation lying somewhere in the interval. The decision to provide the
public good is then taken if all points in the interval are greater than the cost
and it is not provided if all points are less than the cost. A level of cost lying in
the interval indicates potential indierence between provision and non-provision
of the public good at the given level of cost.
Bohm (1972) conducted an experimental implementation of this procedure
on 200 people from Stockholm who had to evaluate the benefit of seeing a previously unshown TV program. The participants were divided into four groups
which faced the following payment mechanisms: (i) pay stated valuation; (ii)
pay a fraction of stated valuation such that costs are covered from all payments;
(iii) pay a low flat fee and (iv) no payment. Although the first two provide an incentive to under-report and the latter two to over-report, the experiment found
that there was no significant dierence in the stated valuations, suggesting that
misrevelation may not be as important as suggested by the theory. In the practical application, Bohm (1984) describes how the choice on whether to collect
statistics on housing is determined by obtaining the local government valuations of these. Two alternative contracts were oered, the first had payment
as a proportion of valuation with the proportion chosen to cover costs and the
second had a fixed payment for any valuation over a given figure. A valuation
below this figure resulted in exclusion from the use of the statistics. The results
indicated that valuations were on average lower (SEK 827) for the oered the
first contract than for those oered the second (SEK 889) with the valuation
interval being small. This work suggests that the misrevelation problem is not
as great as the theory predicts and that simple procedures may well exist that
overcome the actual degree of misrevelation.
In assessing these results, it must be borne in mind that the experimental
procedures are always subject to the criticism that the participants do not treat
them in the way they would act in real situations. The rewards involved are
usually small and the experiment is always somewhat artificial. Furthermore,
the participants may not be a representative sample of the population.
9.7.2
Market data
When acting in a competitive market, a household has no incentive to reveal

false information about their preferences. At the parametric price they face,
their only decision is the quantity to buy and if they do not buy the quantity
that is optimal for them the only eect is to reduce their welfare. This simple
observation can sometimes provide the basis for constructing valuation schedules
for public goods.
To focus the following discussion, consider the case of housing. Viewed in a
characteristics framework (see Lancaster (1966) or Gorman and Myles (1987)),
a house is treated as a collection of attributes and the value to a purchaser of
a house is determined by their assessment of the set of attributes. Equally, the
cost of supplying a house is also dependent upon the attributes supplied. As
shown by Rosen (1974), willingness-to-pay and willingness-to-supply schedules
determine the equilibrium price as a function of characteristics. Now consider
9.8. CONCLUSIONS
301
that part of the attraction of any house is the environmental quality of its
locality, for example whether there is a nearby park. There is obviously no
market for environmental quality as such, but if the prices of a variety of houses
are observed that dier in their environmental quality then a valuation of this
can be derived by obtained by observing the prices at which they trade. The
dierence in price of two houses that are identical in all respects except for
environmental quality captures the value of the environmental dierence.
These observations lead into the theory of hedonic analysis of price which
is the statistical formalisation of the method just described. Examples of the
application of hedonics can be found in Griliches (1971), Ball (1973) and Brookshire et al. (1982). It must be noted that the method does have shortcomings,
particular in that many public goods may have no private good market associated with them.
9.8
Conclusions
This chapter has reviewed the standard analysis of the ecient level of provision
of a public good leading to the Samuelson rule and its generalisations. These
eciency rules generate allocations that can only be achieved if the government
is unrestricted in its policy tools or, as the Lindahl equilibrium demonstrated,
personalised pricing can be employed. If a uniform price must be charged, it was
shown that private provision would not generate a Pareto ecient outcome and
that it would be Pareto-improving for each household to increase their contribution towards provision of the public good. Such results provide a natural role for
the government to participate in financing and provision and the consequences
of financing public goods by distortionary taxation were analysed.
One aspect of public goods that prevents the government making ecient
decisions is the governments lack of knowledge of households preferences and
willingness to pay for public goods. Mechanisms were shown to exist that could
overcome the incentives for households to misreveal their preferences but for
the most relevant equilibrium concept, that of dominant strategies, these were
restricted in the forms of preferences to which they applied. Despite these
rather negative theoretical conclusions, experimental evidence indicates that
household behaviour when confronted with decision problems involving public
goods does not conform with the theoretical prediction and that the private
provision equilibrium may not be as inecient as theory suggests. Furthermore,
misrevelation has not been confirmed as the inevitable outcome.
Chapter 10
Externalities
10.1
Introduction
An externality represents a connection between economic agents which lies outside the price system of the economy. As the level of externality generated is not
controlled directly by price, the standard eciency theorems on market equilibrium cannot be applied. The market failure that can result raises a potential
role for correction through policy intervention.
Externalities and their control are a subject of increasing practical importance. The greenhouse eect is one of the most significant examples of the
consequences of an externality but there are any number of others, from purely
local environmental issues to similarly global ones. Although these may not
appear at first sight to be economic problems, many of the policy responses to
their existence have been based on the economic theory of externalities. The
purpose of this chapter is to demonstrate the consequences of the existence of
externalities and to the review policy responses that have been suggested. In
particular, it will be shown how the unregulated economy generally fails to reach
an ecient outcome and to what degree this can be corrected using standard
tax instruments.
The chapter begins with a discussion of alternative definitions of an externality which dier in whether they focus on eects or consequences. Adopting
an eect-based definition, it is then shown how the market generally fails to
achieve eciency. This lack of eciency is contrasted to the claim of the Coase
theorem that eciency will be eliminated by trade. An emphasis is placed on
the role of missing markets and ineciency in bargaining with incomplete information. The design of the optimal set of corrective, or Pigouvian, taxes is
then addressed under alternative assumptions about the feasible degree of differentiation between dierent households and firms. The chapter is completed
by contrasting the use of taxes with direct control through tradable licences and
value of internalisation.
303
304
10.2
CHAPTER 10. EXTERNALITIES
Externalities
Although the nature of an externality as an eect inflicted by one agent upon

another may seem very clear at an intuitive level, once a formalisation is attempted a number of issues arise that need to be resolved. Of most importance
is the question of whether the existence of an externality should be judged by
its eects or by its consequences. Since both approaches have some merit, but
can lead to dierent classifications, there is no universally agreed definition of
an externality. This section discusses two alternative definitions and describes
the representation of externalities adopted in the following analysis.
10.2.1
Definitions
In the literature there have been a number of alternative definitions of an externality and several attempts at providing classifications of various types of
externality; a survey is presented in Baumol and Oates (1988). There are two
major categories of definition, the first of which defines an externality by its
eects and the second by the reason for its existence and its consequences.
The first definition of an externality is the most commonly adopted and is
based on the eects of externalities.
Externality (1)
An externality is present whenever some economic agents welfare (utility
or profit) includes real variables whose values are chosen are chosen by others
without particular attention to the eect upon the welfare of the other agents
they aect.
This is a very broad definition but does have the advantage of allowing an
externality to be recognised from its eects. The definition also implicitly distinguishes between two broad categories of externality. A production externality
is said to exist when the eect of the externality is upon a profit relationship and
a consumption externality is present whenever a utility level is aected. Clearly,
an externality can be both a consumption and a production externality simultaneously. For a household, an externality can aect either the consumption set
or the utility function. In either case, final welfare will be aected. Similarly,
for a firm, an externality may determine the structure of the production set or
it may enter the profit function directly.
The diculty with this definition is its dependence upon the institutional
context in which it is placed. The following example of Heller and Starret (1976)
illustrates this point. In a barter economy with two households, the utility of
each household is dependent upon the quantity that the other household is
willing to give up in exchange. From the definition above, this must clearly
count as an externality although such a classification seems inappropriate. If
the institutional setting is altered by the introduction of competitive markets
then the externality is removed. Furthermore, it is precisely the nonexistence
of functioning markets for external eects that leads them to be classified as
externalities. Based on this reasoning, Heller and Starret (1976) provide an
10.2. EXTERNALITIES
305
alternative definition of externalities that relates to existence, or otherwise, of

markets and the consequences of the externality.
Externality (2)
An externality is present whenever there is an insucient incentive for a
potential market to be created for some good and the nonexistence of this market
leads to a non-Pareto optimal equilibrium.
Although the conditions of the second definition are stronger than those of
the first so that, for a given institutional framework, the externalities it identifies
will be a subset of those identified by the first, in most cases the two definitions
will delineate precisely the same set of eects as externalities. Moreover, since
the focus below will be placed upon externalities in a competitive economy, the
eects of varying the market institution upon the set of externalities will be of
limited interest. On this basis, the first definition is adopted as the determinant
of what constitutes an externality. The second definition is still important,
however, due to it directing attention to the question of why some markets exist
and some do not.
10.2.2
Representation
Having defined an externality, it is possible to move on to the formalrepresenta

1
H
tion. Denote, as in Chapter 3, the array of
consumption
vectors
x
=
x
,
...,
x
and the array of production vectors y = y 1 , ..., y m . It is assumed that consumption externalities enter the utility functions of the households and that
production externalities enter the production sets of the firms. At the most
general level, this assumption implies that the utility functions take the form
U h = U h (x, y) , h = 1, ..., H,
(10.1)
and the production set is described by

Y j = Y j (x, y) , j = 1, ..., m.
(10.2)
In this formulation the utility functions and the production sets are dependent
upon the entire arrays of consumption and production vectors. The expressions
in (10.1) and (10.2) represent the general form of the externality problem and in
some of the discussion below a number of further restrictions will be employed.
In particular, the case where the total externality eect is due to the sum of
contributions will be used on a number of occasions.
It is immediately apparent from (10.1) and (10.2) that the actions of the
agents in the economy will no longer be independent or determined solely by
prices. The linkages via the externality result in the optimal choice of each agent
being dependent upon the actions of others. Viewed in this light, it becomes
apparent why the ecient functioning of the competitive economy will generally
not be observed in an economy with externalities.
Before proceeding to analyse potential ineciency of the market outcome, it
is worth noting that an economy based on the utility and production functions
defined in (10.1) and (10.2) may still possess a competitive equilibrium even
306
though this equilibrium may not be Pareto optimal. Provided demand and supplies remain continuous functions of their arguments, the proof of the existence
of equilibrium can proceed along the lines sketched in chapter 2. Further details can be found in Arrow and Hahn (1971) and McKenzie (1955). There are
two additional diculties that arise in the proof of equilibrium in an economy
subject to externalities which concern the boundedness of the economy and the
convexity of production sets. Boundedness is a technical point that is not of
great importance here, a detailed discussion can be found in Osana (1973). The
possible non-convexity of production sets is of rather more importance and will
be considered in Section 4.2.2.
10.3
Market ineciency
It has been implicit throughout the discussion above that the presence of externalities will result in the competitive equilibrium failing to be Pareto optimal.
The immediate implication of this fact is that incorrect quantities of goods, and
hence externalities, will be produced. It is also clear that a non-Pareto optimal
outcome will never maximise welfare. This provides scope for economic policy
to raise welfare. The purpose of this section is to demonstrate how ineciency
can arise in a competitive economy. Cases where externalities do not lead to
ineciency will also be described. The results are developed in the context of
a simple two-consumer model since this is sucient for the purpose and also
makes the relevant points as clear as possible.
10.3.1
Equilibrium and ineciency
Consider a two-household economy where the households have utility functions
U 1 = U 1 x11 , x12 , x21 ,

(10.3)
and
U 2 = U 2 x21 , x22 , x11 .
(10.4)
The externality eect in (10.3) and (10.4) is generated by consumption of good

1 by the other household. The externality will be positive if U h is increasing in
xj1 , h 6= j, and negative if decreasing.
To complete the description of the economy, it is assumed that the supply of
good 2 comes from an endowment to the households whereas good 1 is produced
from good 2 by a competitive industry that uses one unit of good 2 to produce
one unit of good 1. Normalising the price of good 1 at one, the structure of
production ensures that the equilibrium price of good 2 must also be one. Given
this, all that needs to be determined for this economy is the division of the initial
endowment into quantities of the two goods.
In specifying demand it is necessary to make an assumption concerning the
treatment of the externality eect by the households in their maximisation decisions. The standard assumption is that both households take the level of
10.3. MARKET INEFFICIENCY
307
externality as given when they maximise. Although this may seem inappropriate for a two-household economy, two points should be stressed. Firstly, this
analysis is illustrative and, in the more relevant case of many households, the effect of each on external eects would be negligible and the assumption is rather
more tenable. Secondly, it is also possible that the households may be aware of
the externality but not of its cause. They therefore take the quantity as given
when maximising.
Incorporating this assumption into the maximisation decision of the households, the competitive equilibrium of the economy is described by the equations
U h
xh
1
U h
xh
2
= 1, h = 1, 2,
(10.5)
xh1 + xh2 = h2 , h = 1, 2,
(10.6)
x11 + x21 + x12 + x22 = 12 + 22 .
(10.7)
and
It is equations (10.5) that are of primary importance at this point. For household
h these state that the ratio of private benefits from each good, determined by
the marginal utilities, is equated to the ratio of private costs. The external eect
does not appear directly in the determination of the equilibrium.
To characterise the set of Pareto optimal allocations, the technique employed
is to maximise the utility of household 1 subject to holding household 2 at a
fixed level of utility and subject to the production possibilities. Varying the
fixed level of households 2s utility will then trace out the set of Pareto optimal
allocations. The resulting equations will then be contrasted to (10.5). In detail,
the Pareto optima are the solution to
2
max U 1 x11 , x12 , x21 subject to U 2 x21 , x22 , x11 U ,
{xhi }
and 12 + 22 x11 x21 x12 x22
0.
(10.8)
Denoting the Lagrange multiplier on the first constraint by , the solution is

characterised by the conditions
U 1
x11
U 1
x12
and
U 2
x21
U 2
x22
U 2
x11
U 1
x12
U 1
2
1 x1
U 2
x22
= 1,
= 1.
(10.9)
(10.10)
In (10.9) and (10.10) the externality eect can be seen to aect the optimal
allocation between the two goods via the derivatives of utility with respect to
308
the externality. If a positive externality, these will raise the value of the lefthand terms. They will decrease them if a negative externality. The value of
can be interpreted as the relative social welfare weight given to the second
2
household and it will change as U changes. For example, if the optimum had
been derived by maximising a utilitarian social welfare function, then would
be identically 1. It can then be seen that at the optimum with a positive
externality the marginal rate of substitution of both households is below its
value in the market outcome. The converse is true with a negative externality.
Since the marginal rates of substitution can be interpreted as valuations, it can
be seen that the externality leads to a divergence between the private valuations
of consumption given by (10.5) and the corresponding social valuations in (10.9)
and (10.10). This observation has the implication that the market outcome is
not Pareto optimal. It is also tempting to conclude that if the externality is
positive then more of good 1 will be consumed at the optimum than under the
market outcome. However this is not always the case, a point that is discussed
in Section 5.
There is one further point worth noting. When a richer production structure
is added to the model, the question of whether it is always optimal for firms
to produce on the frontier of their production sets arises. The answer to this
is given by Murakami and Negishi (1964) who provide an example that has an
optimum with ineciency. However, if it is optimal for at least one firm to be
inecient then the outputs of that firm must cause a negative externality on
other, ecient, firms and the inputs must cause a positive externality. This is
a fairly restrictive condition so that productive ineciency is probably not of
great consequence.
10.3.2
Pareto irrelevant externalities
Although it conveys the message that externalities generally lead to a failure

of the competitive economy to achieve Pareto optimality, the simplicity of the
preceding analysis is somewhat misleading. As shown by Osana (1972), Rader
(1972) and Parks (1991) there are certain classes of preferences for which the
competitive equilibrium is Pareto optimal.
To see how this may arise, assume that the preferences of household h can
be represented by a utility function of the form
h
ih
U h = xh1 xh2 xj1 xj2
,
(10.11)
xh1
= 1, h = 1, 2.
xh2
(10.12)
h
ih
where the term xj1 xj2
captures the externality eect on h arising from the
consumption of j. With these preferences, the competitive equilibrium is then
characterised by
10.4. THE COASE THEOREM
309
The necessary conditions determining a Pareto optimum can be written
and
2 1 1
x1k x21 x22
+ 2 x21 x22 x11 x12
xk = , k = 1, 2,
1 1 2
2
x11 x21 1 x21 x22
xk + x2k x11 x12
= , k = 1, 2.
(10.13)
(10.14)
Dividing the equation for k = 1 in (10.13) by that for k = 2 gives (10.12).

Repeating this for (10.14) also gives (10.12). The existence of the externality
therefore does not aect the eciency of the competitive equilibrium.
Parks (1991) shows that the Pareto irrelevant externalities arise when the
preferences of the households have the general form

U h = U h f 1 x1 , f 2 x2 , ..., f H xH .
(10.15)

In (10.15) the functions f h xh can be interpreted as the private utility that
h derives form their consumption xh . The total utility of each household is
then given as a Bergson-Samuelson function of the vector of private utilities.
Although for each household this Bergson-Samuelson function must be defined
on the true private utilities, it can dier between households as it would in the
example of (10.11) if 1 6= 2 . It is straightforward to show
that
the competitive
equilibrium derived from the private maximisation of f h xh is described by the
same necessary conditions as a Pareto optimum obtained from the maximisation
of (10.15).
The form of (10.15) clearly demonstrates that it is interdependent utilities that give rise to Pareto irrelevant externalities. The result arises because
the externality eects exactly oset each other when the optimal allocation is
determined. Other forms of externality, such as Meades (1952) atmosphere externality which is a function of the sum of consumption of one of the goods,
cannot satisfy this condition and will be Pareto relevant.
10.4
The Coase theorem
The Coase theorem is central to understanding the policy implications of externalities. It does this through indicating those situations in which market activities will eliminate the eects of externalities and suggests new perspectives on
why market solutions to externalities may fail and appropriate policy responses.
This section states the Coase theorem and then explores its interpretations and
implications.
10.4.1
The Theorem
As Coase (1960) never formally stated the theorem but merely described it via
examples, a number of alternative interpretations of the theorem have resulted.
In very general terms, the theorem asserts that if allowed to function freely
the market will achieve an ecient allocation of resources. A generally agreed
version of the Coase theorem is as follows.
310
The Coase Theorem

In a competitive economy with complete information and zero transaction
costs, the allocation of resources will be ecient and invariant with respect to
legal rules of entitlement.
Stated in this form, the theorem makes two distinct claims. Following Regan
(1972), these are termed the eciency thesis and the invariance thesis. In a
given situation it is possible for one of these to be valid whilst the other is false.
No attempt will be made to prove the theorem in the conventional sense. As the
discussion will make clear, it is close to being tautological in some circumstances
and false in others.
The legal rules of entitlement, or property rights, are of central importance
to the Coase theorem. These rules determine the rights of the agents in the
economy, for example the right to unpolluted air or the right to enjoy silence,
and determine the direction in which compensation payments will be made if
the right is violated. If valid, the implication of the Coase theorem is that there
is no need for policy intervention with regard to externalities except to ensure
that property rights are clearly defined. When they are, private agreements over
compensation will generate a Pareto optimal outcome. This conclusion naturally
runs counter to the standard assessment of the consequences of externalities and
explains why the Coase theorem has been of considerable interest.
A natural expectation would be that the assignment of rights will determine
the equilibrium level of an externality, for example that the level of pollution
under a polluter pays system will be less than that under a pollutee pays. The
invariance thesis of the Coase theorem states that this is incorrect and that the
equilibrium level of externality is independent of the assignment of property
rights. To show how this may work, consider the example of a factory that is
polluting the atmosphere of a neighbouring house. When the firm has the right
to pollute the householder can only reduce the pollution by paying the firm a
sucient amount of compensation to make it worthwhile to stop production or
to find an alternative means of production. Let the amount of compensation the
firm requires be C. Then the cost to the householder of the pollution, , will
either be greater than C, in which case they will be willing to compensate the
firm and the externality will cease, or it will be less than C and the externality
will be left to continue. Now consider the outcome with the polluter pays
principle. The cost to the firm stopping the externality now becomes C and
the compensation required by the household is . If C is greater than the
firm will be willing to compensate the household and continue producing the
externality, if it is less than it stops the externality. Considering the two cases,
it can be seen the outcome is determined only by relative to C and not by
the assignment of property rights which is essentially the content of the Coase
theorem.
That the invariance thesis of the Coase theorem is usually false is easily
seen from the example. The change in property rights between the two cases
will cause dierences in the final distribution of income due to the direction of
compensation payments. The invariance thesis can only be correct if there are
no income eects. Since income eects will generally exist, the invariance thesis
311
is false. Attention is now paid only to the eciency thesis.
10.4.2
Markets for externalities
In the study of externalities considerable emphasis has been placed on the value
of markets for externalities. This arises because if the externalities were actually traded, the market outcome would be Pareto optimal. The failure of the
competitive equilibrium to achieve optimality can then be seen as arising from
the necessary markets being missing from the economy. The idea that externalities could be overcome by the introduction of markets for external eects
was first introduced by Arrow (1969) and employed by Meyer (1971). Starret
(1972) provided the formal development of the idea and the proofs of the central
results; the analysis now described is based closely on Starret (1972).
To demonstrate the role of markets, consider an economy with three firms,
labelled j = 1, 2, 3, and two goods. An externality is introduced by assuming
that the production set for each firm depends upon the production plans of
the other firms. The production set, Y 1 , of firm 1 is therefore defined on sixdimensional space with a typical element y11 , y21 , y12 , y22 , y13 , y23 . Clearly, firm
1 only has direct choice upon the first two elements of this set. The same
comments apply to the production sets of firms 2 and 3, where these firms have
direct control over the second and third pairs of entries respectively.
Now let y j be the output vector of firm j. The economys production set is
X
y j , y j Y j , j = 1, 2, 3 .
(10.16)
Y = y:y=
j=1
Provided there is no point interior to the production set which is Pareto optimal,
the ecient production plan will be a point, yb, on the boundary of Y and, if Y is
convex, this will imply a price vector q under which yb maximises qy over all y in
Y . Under this interpretation the price system q captures the true social rates of
transformation between the goods. The optimality question can now be phrased
formally as determining whether there exists any arrangement of trade in this
economy such that the equilibrium is Pareto optimal, so demonstrating the First
Theorem of Welfare Economics, and whether there exists a price system which
will lead, by profit maximisation of the individual firms, to the production plan
yb and hence demonstrating that the Second Theorem of Welfare Economics
applies.
Before proceeding, it may be worth recalling how the Two Theorems of
Welfare Economics were demonstrated in Chapter 2. For the First Theorem, it
is sucient for it to hold that all agents are maximising subject to parametric
prices with no external eects. The proof of the Second Theorem required
that the firms production sets were independent and that the price vector that
ensured the optimal allocation was profit maximising for the economy as a whole
also had the property that it was profit maximising for individual firms to choose
the quantities that lead to the optimal output. These observations indicate that
312
for the Theorems to apply the important point is to define the production sets
so that they are independent between firms.
The required independence can be obtained by delineating goods both by
the firm that produces them and by the firm upon which they have an externality eect. Thus define good yikj to be the net output of good i by firm j as
observed
k. The production
set of firm k is now defined on the variables
k1 k1 byk2firm
y1 , y2 , y1 , y2k2 , y1k3 , y2k3 , k = 1, 2, 3. A count of the commodities shows that

this leads to there being 18 dierent goods. In addition, the production sets of
the firms are formally independent since the goods over which any firm exercises
choice do not appear in the production set of any other firm. To complete the
argument, a set of prices are introduced that correspond to the number of goods
in
the artificial markets. The price vector faced by firm k is therefore written
k1
k2 k2 k3 k3
p1 , pk1
2 , p1 , p2 , p1 , p2 . Since none of the agents now face an external effect, it follows that if agents act competitively and an equilibrium exists for the
economy including the markets for external eects, then the equilibrium must
be Pareto optimal. It is also possible to appreciate that there are now sucient
prices to allow the control of firms decisions that is necessary for decentralisation. In particular, it is possible to choose the prices such that the identity
0
yikj = yik j , all k, j, k0 , i is satisfied.
Starret (1972) proves the following result which extends the Second Theorem
to the economy with markets for externalities.
nm
Theorem 50 (Starret) The price vector p11
can be chosen such that
1 , ..., pn
the profit maximising production plans of the firms have the property that they
sum to the optimal aggregate output yb and the prices satisfy .
Proof. See Starret (1972).
Theorem 10.1 shows that the prices can be chosen so that individual profit
maximisation generates the chosen level of output and that the prices on the
artificial markets are equal to the true social costs. In other words, the equality
states that the total cost of a unit of good i for firm j, including all the externality eects it has upon other firms, sums to the social cost. In this way,
the externality eects are brought into the price system. The motivation for
this result is that each good can be seen as a bundle of commodities traded on
the artificial markets and the price of the bundle is the sum of the prices of its
components. An alternative way of viewing it is to note that the direct price of
a unit of good i for firm j is pij
i . Taking qi to be the consumer price of good, it
follows from the identity
qi = pjj
i +
m
X
pkj
i ,
(10.17)
k=1,k6=j
P
kj
that the sum m
k=1,k6=j pi can be seen as total tax payments per unit of good
i by firm j to cover the externality eects. Equation (10.17) therefore captures
the basic duality between prices on artificial markets and corrective taxes. To
summarise, if artificial markets are created for the externalities, so that they are
313
treated as distinct goods according to the firm that produces them and the firm
that they aect, then a price system defined over these constructed commodities
can support the optimal allocation. In addition, this price system can also be
interpreted as defining a set of optimal taxes to counter the externalities.
This analysis can be given two alternative interpretations. Firstly, it can be
taken as prescriptive of what should be done to overcome ineciency due to
the existence of externalities in the sense that if there is an externality problem
then this can be overcome by the introduction of markets for external eects.
Secondly, it can be treated as a proof of the eciency thesis of the Coase theorem as it shows that if externalities can be traded on competitive markets
then the equilibrium must be ecient. In this context, the Coase theorem becomes almost tautological. If the markets exist then, as in the second definition
of an externality, there must actually be no externalities. As regards policy
prescriptions, since trading in the artificial commodities is equivalent to trading property rights, the analysis can also be interpreted as showing that policy
should be directed to facilitating the exchange of property rights.
10.4.3
Non-existence of markets
The discussion of the previous section has shown that if the economy has sucient markets and if these markets function competitively, then the equilibrium
will be Pareto optimal even in the presence of externalities. From the construction, it is clear that this argument presumes that each externality will have
its own market. The question that is now addressed is, given their obvious
desirability, why a complete set of such markets are not observed in practice.
Property rights
The first diculty in proposing the results above as a solution to the problem
of externalities lies with the assignment of property rights in the market. When
considering trade for commodities defined in the usual sense it is clear who is
the purchaser and who is the supplier and, therefore, the direction in which
payment should be transferred. However, when externalities are interpreted as
commodities, this is no longer the case.
If property rights are not clearly specified, it may not be obvious who should
be seen as the recipient of payment. For example, consider the case of air
pollution. It may be that it is acknowledged that the polluter should pay with
the implicit recognition of the right to clean air. On the other hand, the tradition
may have established a right to pollute, with clean air something that should
have to be paid for. When neither tradition is generally accepted, the direction
in which payment should go remains unclear. As a consequence, without clearly
specified property rights, the markets may not function since none of the parties
involved would be willing to accept the responsibility for payment and costly
litigation would diminish the gain due to the existence of the market. These
observations suggest that the clarification of property rights should constitute
that first step in the construction of a policy towards externalities.
314
Figure 10.1: Non-Convexity
Non-convexities
It has been implicit in the discussion up to this point that the standard assumptions on convexity, particularly of production sets, will still be applicable
to the economy despite the existence of externalities. Without convexity, maximising choices may not be continuous functions of underlying variables and no
equilibrium may exist. When there is no equilibrium, policy proposals based
on its assumed existence, with or without the artificial markets, will be of little
value. Furthermore, even if an equilibrium exists it may not be decentralisable.
Although convexity is not too contentious without externalities, this is not the
case with externalities. As noted by Starret (1972) there are strong reasons for
doubting its applicability.
As an example of the first reason for the possible failure of the convexity
assumption, consider the case of a negative production externality. For a firm
that is aected, if this has the eect of driving its output to zero regardless of
the level of other inputs then the production set will not be convex. Two ways
in which this may occur are illustrated in Figure 10.1 where the output, y, of
the firm is drawn as a function of the level, z, of some externality for a given
level of inputs.
In the left-hand diagram, output becomes identically zero after some level
of the externality is reached. The right-hand diagram does not have the zero
output level but shows output driven to asymptotically to zero. In both these
case the production set of the firm is not convex.
In such cases the economy based on artificial markets will fail to have an
equilibrium. Suppose the firm were to receive a positive price for accepting
externalities. The optimal behaviour for the firm would then be to produce zero
output and oer to accept an arbitrarily large quantity of externalities. Since
its output is already zero the externalities can do it no further harm, so that
this plan will lead to an unbounded level of profits. If the price for accepting
externalities were zero, the same firm would not accept any. The demand for
externalities is therefore discontinuous and an equilibrium need not exist. If
315
the externality were positive, the firm would choose a production plan involving
an arbitrarily large negative quantity of the externality which again leads to
unbounded profit.
The second reason for non-convexity is due to Meade (1952). It is often
assumed that once all inputs are properly accounted for, all firms will have
constant returns to scale since they can always replicate their behaviour. That
is, if a fixed set of inputs (i.e. a factory and sta) produce output y, doubling all
those inputs must produce 2y since they can be split into two identical sub-units
(e.g. two factories and sta) producing an amount y each. Now consider a firm
subject to a negative externality and assume that it has constant returns to
all inputs including the externality. From this view, there are constant returns
from the perspective of society. Now consider the firm doubling all its inputs but
with the externality held at a constant level. Since the externality is a negative
one, it becomes diluted by the increase in other inputs and output must more
than double. The firm therefore faces private increasing returns to scale. With
such increasing returns, the firms profit maximising decision may not have a
well-defined finite solution and market equilibrium may again fail to exist.
Further analysis of these issues is given in Otani and Sicilian (1977). Their
theorem demonstrates that convexity is inconsistent with a set of reasonable
assumptions describing the consequences of external eects. This simply formalises the arguments above. However they also note that by not permitting
firms to demand negative quantities of inputs or supply negative a quantity
of output the problem does not occur with positive externalities. Therefore
non-convexity appears more of an issue for negative externalities.
These comments have been of a fairly negative nature since non-convexities
are the source of severe diculties for conclusions based on competitive behaviour. Although they may not be of critical importance in practice, they do act
as a warning that decentralised solutions may not be appropriate for dealing
with externalities.
Transaction costs
If the exchange of commodities would lead to mutually beneficial gains for two
parties, the commodities will be exchanged unless the cost of doing so outweighs
the benefits. Such transactions costs may arise from the need for the parties
to travel to a point of exchange or from the legal costs involved in formalising
the transactions. They may also arise due to search required to find a trading
partner. Whenever they arise, transactions cost represent a hindrance to trade
and, if suciently great, will lead to no trade at all taking place. The latter
results in the economy having a missing market.
The existence of transaction costs is often seen as the most significant reason
for the nonexistence of markets in externalities. To see how they can arise,
consider the problem of pollution caused by car emissions. If the economy above
is applied literally, then any driver of a car must purchase pollution rights from
all of the agents that are aected by the car emissions each time, and every
time, that the car is used. Obviously, this would take an absurd amount of
316
organisation and, since considerable time and resources would be used in the
process, would certainly not correspond to the frictionless market envisaged in
the Arrow-Debreu economy. In many cases it seems likely that the welfare loss
due to waste of resources in organising the market would outweigh any gains
from having the market.
Despite these negative conclusions, the transactions cost perspective upon
the Coase theorem still has policy value. The prescription arising from it is that
the cost of transferring legal entitlements between agents should be reduced as
far as possible to permit trade.
Missing markets and side-payments
If markets for external eects are missing, it may be possible for direct transfers,
or side-payments, between agents to overcome this and achieve eciency. One
way to formalise this idea has been considered by Bigelow (1993). Assume that
two agents are initially faced with playing a bimatrix nonconstant sum one-shot
game. Denote the payo of this game to 1, if agent 1 chooses strategy i and
agent 2 strategy j, by 1 (i, j) and for 2 by 2 (i, j). Without side-payments,
there is no reason why the Nash equilibrium of the game should be ecient.
Side-payments can be introduced by allowing agent h, h = 1, 2, to oer agent
k an outcome contingent payment of if strategies i, j are chosen for the original
game. Incorporating these side-payments, the payo for 1 if i, j is chosen is
1 (i, j) + 1 (i, j) 2 (i, j) and for 2 is 2 (i, j) + 2 (i, j) 1 (i, j). An
equilibrium for the game with side-payments is then a set of side-payments
which are chosen optimally given that the resulting strategies form a Nash
equilibrium for the game where the payos include side-payments.
A missing market is then represented by an original matrix game in which
the payo of one agent is not aected by the action of the other. Bigelow (1993)
then proves that the existence of side-payments is sucient to ensure that an
ecient equilibrium is achieved. In contrast when two markets are missing,
so any form of payos may arise in the original matrix game, an equilibrium
may not exist in the game with side-payments and when it does, it need not be
ecient. Although they are rather stylised, these results do show the limitations
of side-payments in achieving eciency when there are missing markets.
10.4.4
Bargaining
When external eects are traded, there will generally only be one agent on each
side of the market. This thinness of the market undermines the assumption
of competitive behaviour that can support the eciency hypothesis. In such
circumstances, the Coase theorem has been interpreted as implying that bargaining between the two agents will take place over compensation for external
eects and that this bargaining will lead to an ecient outcome. Such a claim
requires substantiation.
Bargaining can be interpreted as taking the form of either a cooperative game
between agents or as a non-cooperative game. When it is viewed as coopera-
317
tive, the tradition since Nash (1950) has been to adopt a set of axioms which
the bargain must satisfy and to derive the outcomes that satisfy these axioms.
The requirement of Pareto optimality is always adopted as one of the axioms
so that the bargained agreement is necessarily ecient. If all bargains over
compensation payments were placed in front of an external arbitrator, then the
Nash bargaining solution would have some force as descriptive of what such an
arbitrator should try and achieve. However, this is not what is envisaged in the
Coase theorem which focuses on the actions of markets free of any regulation.
Although appealing as a method for achieving an outcome agreeable to both
parties, the fact that Nash bargaining solution is ecient does not demonstrate
the correctness of the Coase theorem.
The literature on bargaining in a non-cooperative context is best divided
between games with complete information and those with incomplete information, since this distinction is of crucial importance for the outcome. One of the
central results of non-cooperative bargaining with complete information is due
to Rubinstein (1982) who considers the division of a single object between two
players. The players take it in turns to announce a division of the object and
each period an oer and an acceptance or rejection are made. Both players
discount the future so are impatient to arrive at an agreed division. Rubinstein
shows that the game has a unique subgame perfect equilibrium with agreement
reached in the first period. The outcome is Pareto optimal.
The important point is the complete information assumed to prevail in this
representation of bargaining. The importance of information for the nature of
outcomes has been already been emphasised on several occasions and its consequences are equally important for bargaining. In the simple bargaining problem
of Rubinstein (1982) the information that must be known are the preferences of
the two agents, captured by their rates of time discount. When these discount
rates are private information the attractive properties of the complete information bargain are lost and, as shown by Rubinstein (1985), there many be
many potential equilibria with the equilibria being dependent upon the precise
specification of the structure of bargaining.
In the context of externalities it seems reasonable to assume that information
will be incomplete since there is no reason why the agents involved in bargaining
an agreement over compensation for an external eect should be aware of the
others valuation of the externality. The formalisation of this situation given
by Myerson and Satterthwaite (1983) considers the sale of an object (which
would be the property right) from one agent to another. Each agent knows
their own valuation at the time of bargaining but is uncertain about that of the
other agent. The characterisation of bargaining is then undertaken by determining the form of Bayesian incentive-compatible and individually rational direct
mechanisms for reaching agreement (see the definitions in Chapter 9 Section
8). Application of the revelation principle, which asserts that for any Bayesian
equilibrium there is an equivalent mechanism that involves truthful revelation,
justifies the restriction to incentive compatible direct mechanisms.
Agent 1 initially owns the object and agent 2 wishes to purchase. The
valuation of the object by 1 is denoted 1 and that of 2 by 2 . Household j,
318

j = 1, 2, believes , i 6= j, to be distributed with density function fi i which
has strictly positive support on [ai , bi ] and
distribution Fi i . The
1 cumulative
outcome is determined by a function p , 2 whichrepresents the probability

that the object is transferred and a function q 1 , 2 that is the payment that
must be made. Given a pair of such functions, the expected payment as viewed
by 1 is given by
Z b2
1

q1 =
q 1 , w2 f2 w2 dw2 ,
(10.18)
a2
and the cost expected by 2 is

q2 2 =
b1
a1

q w1 , 2 f1 w1 dw1 .
(10.19)
Similarly, the probability of transfer expected by 1 is

p1 1 =
and by 2

p2 2 =
b2
a2
b1
a1

p 1 , w2 f2 w2 dw2 ,

p w1 , 2 f1 w1 dw1 .
(10.20)
(10.21)
Employing (10.18) and (10.20), the expected payo of 1 can be written

U 1 1 = q 1 1 1 p1 1 ,
(10.22)
and that of 2

U 2 2 = q 2 2 2 p2 2 .
(10.23)
The characterisation of the mechanism begins by noting that incentive compatibility implies for agent 1 that

U 1 1 = q 1 1 1 p1 1 q 1 w1 1 p1 w1 ,
(10.24)
and

U 1 w1 = q 1 w1 w1 p1 w1 q 1 1 w1 p1 1 ,
1
(10.25)
for any and w . Taken together, (10.24) and (10.25) give

1

w 1 p1 1 U 1 1 U 1 w1 w1 1 p1 w1 .
(10.26)

Choosing w1 > 1 shows that p1 1 p1 w1 so that the probability is
1
decreasing in valuation. Dierentiating
1 (10.26) with respect to and letting

1
1
10
1
w shows that U = p1 . Integrating gives

U 1 1 = U 1 (b1 ) +
b1

p1 w1 dw1 .
(10.27)
319
Repeating the argument for agent 2 shows that p2 () is increasing and

U 2 2 = U 2 (a2 ) +
a2

p2 w2 dw2 .
(10.28)
The central result of Myerson and Satterthwaite (1983) can now be stated.
Theorem 51 (Myerson and Satterthwaite) If the intervals [a1 , , bi ], i = 1, 2,

have an intersection with non-empty interior (so there is a positive probability
of gains from trade) and neither is a subset of the other then there is no ecient
mechanism that satisfies incentive compatibility.

Proof. Substituting for U 1 1 and U 2 2 in (10.27) and (10.28) using
(10.22) and (10.23) and summing the resulting equations gives

q 1 1 q 2 2 = 1 p1 1 2 p2 2 + U 1 (b1 ) + U 2 (a2 )
Z 2
Z b1
1 1

p1 w dw +
p2 w2 dw2 .
(10.29)
+
1
a2
As there is no outside agency involved,

income of 1 is equal to

the expected
the expected payment of 2 or E q 1 1 = E q 2 2 . The expectation of
the right-hand side of (10.29) is therefore zero. Taking the expectation and
integrating by parts gives
"
#

F1 1
U (b1 ) + U (a2 ) =
p1 1 f1 1 d 1
+
1
f1 ( )
a1
#
Z b2 "
2
1 F2 2
2
+
p2 2 f2 2 d(10.30)
.

2
f2 ( )
a2
1
b1

Replacing p1 1 and p2 2 using (10.18) and (10.20) and noting that individual rationality requires U 1 (b1 ) 0, U 2 (a2 ) 0, (10.30) implies
""
# "
##

1 F2 2
F1 1
1

0
+
p 1 , 2 f1 1 f2 2 d 1 d 2 .
2
1
f2 ( )
f1 ( )
a2
a1
(10.31)
Condition (10.31) must be satisfied by any incentive compatible and individually
rational mechanism.
An ecient outcome requires p 1 , 2 = 1 if 1 < 2 and p 1 , 2 = 0 if

Z
b2
b1
320
1 > 2 . Substituting these restrictions into the right-hand side of (10.31) gives
Z
b2
a2
a1
b2
a2
b2
=
=
=
2 2

f2 + F2 2 1 f1 1 d 1 d 2
min{ 2 ,b1 }
a1
1 1

f1 + F1 1 f2 2 d 1 d 2

2 f2 2 + F2 2 1 F1 2 d 2
a2
min{ 2 ,b1 }
b2
a2
b2

1 F2 2 F1 2 d 2 +
a2
b1

1 F2 2 F1 2 d 2
a2
[1 F2 (w)] F1 (w) dw < 0.
b2
b1
b2
b1
b2
a2

min 2 F1 2 , b1 f2 2 d 2
2

b1 f2 2 d 2
2
F2 1 d 2
(10.32)
Contrasting (10.32) and (10.31) shows that an ecient outcome contradicts

individually rationality.
Theorem 10.2 shows how asymmetric leads to ineciency in the outcome
of bargaining. The incentive for each agent to attempt to exploit their private
information prevents all the gains from trade being exhausted. Rob (1989) and
Mailath and Postlewaite (1990) extend this ineciency result to economies in
which the externality caused by one agent aects many agents. As the number of
agents increases the eect of each upon the externality becomes negligible. The
degree of ineciency can therefore increase as the number of agents increases.
The eciency thesis of the Coase theorem relies on agreements being reached
on the compensation required for external eects. The results above suggest that
when information is incomplete, bargaining between agents will not lead to an
ecient outcome.
10.4.5
Conclusions
In summary, the Coase Theorem suggests that externalities can be overcome

by decentralised trading between aected parties. In practice it is dicult to
imagine that its conditions are actually satisfied so that it cannot be given too
much weight as a foundation for the formulation of policy. The arguments above
provide some fairly powerful reasons why the full set of markets required for
optimality may not exist. Furthermore they also show why bargaining between
aected parties is also unlikely to achieve eciency.
10.5
Welfare-improving changes
A natural expectation would be that if the production of a good generated a

negative externality, then too much of that good would be produced at the
10.5. WELFARE-IMPROVING CHANGES
321
competitive equilibrium and that welfare could be raised by reducing output.

The same point would apply to consumption externalities. If this reasoning was
always correct, it would provide a simple rule of thumb for policy to follow.
This section investigates the truth of this statement for a model with a single
externality. The claim is shown to be true for marginal changes but, as will be
discussed, must be treated with care when non-marginal changes are involved.
10.5.1
Local results
Consider an economy that has a single externality that is generated by the firms
in the economy at a level that is related to their output of good 1. The quantity
of externality generated by firm j, z j , is determined by

z j = z j y1j .
(10.33)
The total quantity of externality then amounts to

z=
m
X
zj =
j=1
m
X
j=1

z j y1j .
(10.34)
It is assumed that the externality aects both the consumers, via their utility
functions, and the firms, via their production sets. The utility of household h
is therefore denoted
(10.35)
U h = U h xh1 , ..., xhn , z .
From (10.35) it can be seen that utility is a function only of the total quantity of
the externality and is not related to its distribution across firms. The production
set of each firm is written in implicit form as
n
(10.36)
y1j , ..., ynj : f j y1j , ..., ynj , z 0 , j = 1, ..., m.
The analytical procedure is now to start at a competitive equilibrium and

to calculate the eect of a dierential increase in the output of good 1 by all
firms, with an osetting changes in the consumption and production of all other
goods. This eect is then converted into a social valuation via a utilitarian
aggregation procedure. If the social valuation of the increase has a negative
value, then welfare will be raised by reducing the output of good 1 by all firms.
The competitive equilibrium allocation must satisfy both the definition,
(10.34), of the level of externality and it must be productively feasible for the
firms, that is (10.36) must be met for all j. In addition, an equilibrium allocation
has to meet the market clearance condition
H
X
h=1
xhi i +
m
X
yij , i = 1, ..., n,
j=1
where i is the total endowment of good i.
(10.37)
322
It is now assumed that the constraints (10.34), (10.36) and (10.37) are met
with equality so that the eect of the increase in the output of good 1 can be
calculated by total dierentiation. Carrying out the dierentiation gives
dz =
m
X
z j
dy1j ,
(10.38)
dyij , i = 1, ..., n,
(10.39)
j=1
H
X
dxhi =
j=1
h=1
and
df j =
m
X
y1j
n
X
fij dyij + fzj dz = 0, j = 1, ..., m.
(10.40)
i=1
Equation (10.40) can be rearranged to give the equivalent form

n
X
i=1
fij dyij fzj dz, j = 1, ..., m.
(10.41)
To derive an expression for the social value of this change in the production
plan, it is first noted that utility maximisation of household h leads to the
equalities
Uih = pi , Unh = pn ,
(10.42)
where is the Lagrange multiplier on the households budget constraint. If pn
is normalised at 1, then
Uih
= pi ,
(10.43)
Unh
is a measure of the value of good i to household h relative to the other goods.
To generate a measure of change in social welfare, the eect of rearrangement
in production is defined as
dW =
n X
H h
X
U
i=1 h=1
i
Unh
dxhi
H h
X
U
z
h=1
Unh
dz.
(10.44)
If dW is positive, the modification of production is assumed to be of positive

value for society.
The aim now is to express dW in terms of the output of good 1 so that it
can be ascertained whether a reduction in the externality producing good will
raise welfare. The first step is to note from the resource balance equation that
any change in consumption must be met with an equal change in production.
Hence
H
m
X
X
dxhi =
dyij , i = 1, ..., n.
(10.45)
h=1
j=1
In addition, the fact that the initial point is assumed to be a competitive equilibrium implies that the firms must be equating the marginal rate of transformation
10.5. WELFARE-IMPROVING CHANGES
323
between any two goods to the ratio of their prices. In particular, taking all other
goods relative to good n this gives
j
M RTi,n
=
fij
fnj
= pi , i = 1, ..., n, all j.
Using (10.45) and (10.46), dW can be written

" #
m
n X
H h
X
X
fij
Uz
j
dz.
dW =
j dyi +
Unh
i=1 j=1 fn
(10.46)
(10.47)
h=1
From (10.41) the terms in fij can be eliminated to provide
m j
H h
X
X
f
U
z
z
+
dz.
dW =
j
h
U
f
n
n
j=1
(10.48)
h=1
Finally, using the externality generation process defined in (10.38) gives
m j
H h X
m
X
X
fz
Uz
dW =
z1j dy1j .
(10.49)
+
j
h
U
n
j=1 fn
j=1
h=1
The expression in (10.31) captures the welfare eect of a feasible change in
the production of good 1 by all firms. To interpret this condition, note that
fzj > 0, fnj > 0, z1j > 0 all j, and Uzh < 0, Unh > 0, all h.
(10.50)
Therefore, if dy1j is positive for all firms, dW is negative. Hence a reduction in

the output of good 1 by all firms, that is dy1j < 0 all j, would raise welfare.
In conclusion, a reduction in the output of an externality-producing good
from its level determined at the competitive equilibrium will raise welfare. This
is the intuitively plausible result that agrees with the presumption that it should
always be the intention of policy to reduce the level of a negative externality.
By reversing the signs of fzj , all j, and Uzh , all h, it is also clear that (10.31)
implies that the quantity of a good producing a positive externality should be
increased from its competitive equilibrium level. The methodology employed to
derive this result is clearly analogous to the more general policy reform analysis
of Chapter 6. The methods of that chapter could be employed to extend the
model to several externalities and corresponding changes in several outputs.
10.5.2
Global results
In addition to the usual caveats that must be applied to any result, such as
the appropriateness of the model within which it was derived, the claim of
the previous section is restricted by a further consideration. Since the formal
analysis considered only the consequence of a dierential change in the output
324
of good 1, the result is purely local in that it refers to the correct course of
action only in the locality of the competitive equilibrium. It must therefore not
be read as suggesting that the output of good 1 should be reduced regardless of
the starting point or that the optimal level of the output of good 1 is less than
that at the competitive equilibrium. Both the latter statements are examples
of global results and could only be supported by stronger arguments.
A global result would relate the production and consumption arrays {xc } ,
c
{y } at the competitive equilibrium to the optimal arrays {xo } , {y o }. In fact, as
shown in Baumol and Oates (1988), a comparison between these arrays can only
be made under stronger assumptions than have so far been adopted. In particular, even with a single externality, convexity of the production sets becomes
essential.
Buchanan and Kafoglis (1963) demonstrate that it also possible for counterexamples to be constructed in which, for example, the quantity of a positive
externality is decreased as the economy moves towards the optimum. Consider
a two-consumer, two-good, economy with a single positive consumption externality. For each consumer, the externality arises from the others consumption
of good 2. Under these assumptions the utility functions are of the form
U 1 = U 1 x11 , x21 , x12 , U 2 = U 2 x11 , x21 , x22 .

(10.51)
As each consumer maximises their utility taking the other consumers contribution to the externality as fixed, the equilibrium occurs at a point at which the
indierence curves cross at right-angles. This is precisely the same argument as
for the private provision of public goods in Chapter 9. Therefore, as shown in
Chapter 9 Section 5.3, it is possible for there to be too much consumption of
goods causing positive externalities at the market outcome relative to the social
optimum and too little of goods causing negative externalities.
The conclusion of this section is that although local results can be established without too much diculty it is not straightforward to provide global
comparisons. However, since many policy recommendations are only concerned
with marginal changes, this is not of too great a consequence for the formulation
of practical policies.
10.6
Corrective taxation
In response to the non-optimality of the market equilibrium in the presence of

externalities, a natural policy to adopt is the imposition of a set of taxes to
correct for the distortions. Such taxes, often termed Pigouvian taxes following
the work of Pigou (1920), will be the subject of this section. Before proceeding
to characterise optimal taxes, it is first necessary to clarify several important
issues regarding the extent to which the taxes may be dierentiated.
Taxation has already been implicit in much of the discussion of artificial
markets and was made explicit in (10.17). The content of the first sub-section
will be to derive a set of taxes of the form of those described in (10.17) for a
model of consumption externalities. The perspective chosen emphasises the fact
10.6. CORRECTIVE TAXATION
325
that the rates of tax need to be dierentiated between commodities and between
consumers. This is a far stronger requirement than that normally imposed upon
models of commodity taxation and, arguably, is not administratively feasible.
When the requirement that the taxes must be uniform across consumers is
imposed, the first-best allocation cannot, in general, be sustained except for
some special cases. This leads to the choice of taxes that will generate the
optimal second-best outcome. Uniform taxes are considered in the second subsection.
10.6.1
Non-uniform taxation
It has already been argued in the analysis of artificial markets that the first-best
can be sustained by a system of personalised prices and that the components of
these prices can be given the interpretation of taxes as in (10.17). The analysis
involved asserting the existence of such prices by exploiting the convexity of the
model but did not discuss the means by which such prices could be calculated.
To remedy this, optimal taxes are now reconsidered for a model of consumption
externalities.
Let the utility of household h be given by
hH
h1
hH
U h = U h xh1
,
1 , ..., x1 , x2 , ..., xn
where the externality eects have been included as artificial commodities with
e
xihh denoting the consumption of good i by e
h as viewed by h. The aggregate
quantity, Xi , of good i is defined by
Xi =
H
X
xhh
i ,
(10.52)
h=1
and the production feasibility constraint is expressed as

F (X1 , ..., Xn ) 0.
(10.53)
Following Meyer (1971), the optimal taxes can be characterised by calculating a Pareto optimum for the economy set out above. Such an optimum appears
as the solution to the maximisation problem
max
H
X
HH ,X ,...,X
{x11
1
n}
1 ,...,xn
h=1
hH
h U h xh1
,
1 , ..., xn
(10.54)
subject to
Xi
H
X
h=1
e
hh
hh
e
e
xhh
i , 0 F (X1 , ..., Xn ) , xi = xi , i = 1, ..., n, h, h = 1, ..., H, h 6= h.
(10.55)
The final set of constraints capture the consistency condition that the consumption of h as viewed by e
h must equal the actual consumption of h. Let
326
i , i = 1, ..., n, denote the Lagrange multipliers on the first n constraints, 0

e
be the multiplier on the production feasibility constraint and hh
i , i = 1, ..., n,
h = 1, ..., H be the multipliers on the remaining H(H 1)n constraints.
For the choice of good xhh
i , the first-order condition is
h
U h
i
xhh
i
and for good xhh

i ,
e
U h
e
xhh
i
H
X
hh
i = 0,
e
h=1,e
h6=h
e
+ hh
i = 0.
(10.56)
(10.57)
It can be seen in (10.58) that the multipliers hh

i are proportional to the externality eect. In addition, dividing two of the first-order conditions (10.57) for
goods i and j for a consumer, h, gives
U h
xhh
i
U h
xhh
j
i +
j +
PH
e
hh
e
h=1,e
h6=h i
,
PH
e
hh
e
h=1,e
h6=h j
(10.58)
which would be the first-order condition for choice of the consumer if they faced
P
e
hh
prices given by i + eH
for good i, a fact first noted by Davis and
h=1,e
h6=h i
e
can be treated as the per-unit

Whinston (1965). Therefore the multipliers hh
i
e
hh
tax rate on consumer h for each unit of good xi they generate. The total tax
paid on a unit of consumption, that is for each unit of good xhh
i , is then the
sum of these individual taxes. This is the same interpretation as given after
(10.17). Note that from (10.58), if the externality is negative, the individual
tax component will be positive.
From the above it can be concluded that for the Pareto optimum to be
achieved the taxes must have the potential to be dierentiated between consumers so that each consumer eectively faces a personalised price for each
good. As will be noted below, there are some case where the eects are symmetric across consumer so that they actually face the same optimal prices but
these are the exception rather than the rule. These results have precisely the
same interpretation as the demonstration in Chapter 9 that the Lindahl equilibrium with personalised prices can decentralise the optimum level of public good
provision. Indeed, public goods can be treated entirely as a special case of the
more general externality framework.
The remaining first-order conditions take the form
i 0
F
= 0.
Xi
(10.59)
These equations imply that the marginal rates of transformation between each
pair of goods is equated to the ratio of the s. Since the s can be interpreted
327
as pre-tax prices this has the implication that there is production eciency and
hence the optimal tax system distorts only the consumption side of the economy
in response to a consumption externality.
This discussion has demonstrated that a set of taxes can be derived that
will support a Pareto optimum in the presence of externalities. However, these
taxes will in general need to be dierentiated both across goods and across
consumers. If the model was extended to include production externalities, the
taxes would also need to be dierentiated across firms. Theorem 2.5 investigated
some of the consequences of attempting to implement a tax system with this
degree of dierentiation and showed that such attempts were most likely to fail.
Since they are based on private information of a similar nature, this general
conclusion also applies to the dierentiated Pigouvian taxes that support the
first-best. Consequently, although a first-best outcome can be achieved if the
necessary information were available, the implied tax scheme is unlikely to be
implementable.
10.6.2
Uniform taxation
Uniform taxation refers to the situation in which the taxes can be dierentiated
across goods but not across consumers. As already noted, such taxes will not in
general be sucient to sustain the first-best outcome. This sub-section begins,
however, with a discussion of circumstances in which uniform taxes will sustain
the first-best. Following this, the determinants of the level of taxes are considered and the important distinction between direct and indirect tax instruments
is introduced.
Attainment of first-best
Considering the problem set out in (10.55) and (10.56), uniform taxation will
e
be the solution to this problem when, for a given good i, all the multipliers hh
i
have the same value for all h and e
h. The first case in which this will occur
is the relatively uninteresting one of identical consumers with each given the
same weight in social welfare. Under these conditions the derivatives of utility
in (10.58) will be the same for all h and the claim then follows.
Rather more interesting is the example of Meades (1952) additive atmosphere
externality. The level, , of such an externality is determined by the relation
=
H
X
h=1
xhi
, 0 0.
(10.60)
From (10.60) the level of the externality is determined as an increasing function

of the sum of individual consumption of good i. What is important about the
form of this externality generating process is that the marginal contribution of
each individual to the total is identical. As will now be shown, for this form of
externality uniform taxation generates the first best.
328
To simplify the notation, assume a two-consumer, two-good economy; neither of these restrictions is of any consequence. With the atmospheric externality the utility function are written
U h = U h xh1 , xh2 , , h = 1, 2,
(10.61)
and, assuming that the externality is generated by the consumption of good 1,

is determined by
= x11 + x21 .
(10.62)
The optimum is then characterised as arising from the solution to
max
2
X
{x11 ,...,x22 ,X1 ,X2 , } h=1
hU h,
(10.63)
subject to
Xi
2
X
h=1
xhi , 0 F (X1 , X2 ) , = x11 + x21 .
(10.64)
This problem generates the first-order conditions

h
U h
1 + 1 h = 0, h = 1, 2,
h
x1
x1
(10.65)
U h
2 = 0, h = 1, 2,
xh2
(10.66)
F
= 0, i = 1, 2,
Xi
(10.67)
U 1
U 2
+ 2
+ 1 = 0.
(10.68)
i 0
and
1
As x
1 = x2 , (10.66) and (10.69) have the implication that the consumers
1
1
should face the same shadow prices for the two goods or, equivalently, that the
tax rate on good 1 should not be dierentiated across the consumers. Therefore
uniform taxes can sustain the first-best in the presence of the atmosphere externality. Alternative derivations of this result, for external economies of scale in
production, can be found in Aoki (1971) and Chipman (1970). A more detailed
characterisation of optimal taxes with the atmospheric externality can be found
in Sandmo (1975).
Direct and indirect taxes

When uniform taxation cannot generate the first-best, the range of tax instruments that are available is of great importance. To aid the discussion, it is first
necessary to distinguish between direct taxes, which are levied on the activity
that causes the externality, and indirect taxes that are levied on some activity
329
that is related, via demand or supply functions, to the externality generating

activity. For example, if the use of a car causes air pollution then a tax on
motoring would be a direct tax and a tax on petrol an indirect tax. Although
only direct taxes have been sucient in the example considered above, it is
clear that when direct taxes are restricted to be uniform the extra instrument
of an indirect tax may raise welfare further, a point made clearly by Green and
Sheshinski (1976).
The analysis that is now developed will consist of two parts. The first will
consider only direct taxes and will aim to characterise the factors that determine
their level. This will be followed by the introduction of indirect taxes and the
factors that determine the optimal mix of direct and indirect taxes will then be
of primary importance.
To provide clear conclusions, the first step will be to restrict the individual
utility functions to be linear in all commodities except that causing the externality. Aggregating the non-externality commodities into a single commodity
and labelling this good 2, the utility functions are assumed to take the form
h
U h = U h x11 , ..., xH
(10.69)
1 + x2 .
In addition, the utility functions are assumed to have the concavity and separability properties as below
2U h
2U h
< 0,
= 0, h 6= e
h.
h
h
e
x1 x1
xh1 xh1
(10.70)
The zero cross derivative implies that there are no connections between the
eect on utility of dierent households contributions to the externality.
For consumer h with income M h , solving the utility maximisation problem
h
h
h
h
max U h x11 , ..., xH
(10.71)
1 + x2 s.t. [p1 + t1 ] x1 + x2 = M ,
h
h
{x1 ,x2 }
where the price of good 2 is normalised at 1, generates the demand function
xh1 = xh1 (p1 + t1 ) .
(10.72)
Now assuming, as in Diamond (1973a), that any tax revenue is returned to

the consumers via lump-sum taxes, the social welfare function can be written
in the form
W =
H
X
h=1
H
H
X
X
U h x11 (p1 + t1 ) , ..., xH

xh1 (p1 + t1 ) +
M h.
1 (p1 + t1 ) p1
h=1
h=1
(10.73)
Dierentiating (10.74) with respect to t1 , setting the expression equal to zero
for a maximum, and using the first-order condition from (10.72), the optimal
tax can be written implicitly as
e
h
PH
P
U h x1
H
e
h=1
t1
h=1,e
h6=h xh
1
.
(10.74)
t1 =
PH xh1
h=1 t1
330
From (10.75) it can be seen that the optimal tax is given by the sum of externality eects weighted by the demand derivatives.
The simple characterisation in (10.75) was derived on the basis of a number of strong assumptions. In particular, the separability between the eects
of dierent households contributions to the externality is almost indefensible.
When this is removed however, the possibility is opened for rather more surprising results to emerge. For example, Diamond (1973a) is able to construct
a two-consumer example which has t1 = 0 as the optimal solution despite the
presence of a negative externality. This line of reasoning is developed further
by Green and Sheshinski (1976) who introduce a third good that enters into
the non-linear part of the utility function. With this formulation they permit
indirect taxation of the third commodity and demonstrate that an optimum
may involve a zero direct tax but a non-zero indirect tax.
These results are placed into a common framework by Balcer (1980) whose
analysis is now described. The previous model is extended by assuming a utility
function of the form
h
h
U h xh1 , xh3 + U x11 , ..., xh1
, xh+1
, ..., xH
1 + x2 .
1
1
(10.75)
In (10.76), the externality eect remains separable from the direct level of consumption of good 1 but it is distinguished form (10.72) by the inclusion of good
3 in the first component of the utility function. The focus of the analysis is on
the eect of the substitutability/complementarity relation between good 1 and
3 and the structure of the optimal taxes.
Retaining the assumption that all tax revenues are returned to the consumers
via lump-sum transfers, the welfare function can be written
W
H
X
h=1
h
U h xh1 , xh3 +
U x11 , ..., xh1
, xh+1
, ..., xH
1
1
1
p1
h=1
H
X
h=1
xh1 p3
H
X
xh3 +
h=1
H
X
M h.
(10.76)
h=1
By dierentiating the welfare function with respect to these tax rates and defining
N
i
H
X
xh
h=1
ti
eh
,U
H
X
e
h
e
xh1
e
h=1,e
h6=h
e N
,U
i
H
X
e
h=1
eh xN , N = 1, 3, i = 1, 3,
U
ti
(10.77)
the optimal taxes can be implicitly expressed as
t1 =
e 11
U
11
22 21
22
he
U 12
e 1
U
1
1
21 21
1
(10.78)
10.7. TRADEABLE LICENSES

and
t2 =
e 11
U
11
331
12 11
he
U 12
e 1
U
1
1
22 21 21
1
(10.79)
The first result that can be derived form these expressions is that if
e 11
e 12
U
U
=
,
1
2
11
(10.80)
then t2 = 0 and t1 is given by (10.75). This will occur if the either the externality
eh is the same for all h or if the consumers are
is of the atmospheric kind so that U
identical. As has been noted above, these are the situations for which a uniform
tax can sustain the first-best so the present conclusion is simply an application
of that result.
In other situations the values of the tax rates are determined by two factors:
the degree of aggregate complementarity (21 < 0) or substitutability (21 > 0)
and how those individuals that cause a greater amount of externality at the
margin view the good. When the larger oenders view the goods as complements
and the goods are aggregate complements then t2 < 0 and t1 is less than the
value determined by the Diamond formula. Moving to aggregate substitutability
makes t1 greater than the Diamond value whilst the signs are all reversed when
larger oenders view the goods as substitutes.
10.6.3
Summary
These results conclude the analysis of the determination of tax rates. Although
the economies have been restricted in comparison to some of those in previous
sections, they have illustrated some of the most important determinants of the
tax rates. In particular, it has been shown that uniform taxation cannot sustain
the first-best except in restricted circumstances and that optimality will often
require the use of taxes on related goods. However, even when taxes on related
goods are included the first-best will generally not be achievable. Indirect taxes
can also be motivated in situations where direct taxes cannot be employed, such
as when emissions cannot be monitored directly but the level of production that
generates them can be. Finally, the eect of the related goods depends on their
substitutability or complementarity to the externality producing good and upon
how the larger contributors to the externality view the goods.
10.7
Tradeable licenses
The motivation underlying the use of corrective pricing is that the unregulated
market will produce incorrect quantities of externalities but, by altering the relative cost of generating an externality, taxation can lead to the optimal quantity
of externality. When phrased in this way, it may then actually appear simpler to
directly limit the quantity of externalities rather than to attempt to influence it
332
via the price system. An obvious way to do this, first suggested by Dales (1968),
is to introduce licences that permit the generation of an externality and to allow agents to produce externalities only to the extent of the licences they hold.
Allowing the licences to be traded should see them held by the agents who value
them most highly resulting in ecient generation of externalities. Set against
these observations has to be the fact that the markets on which permits are
traded may be thin so that the competitive outcome will not be achieved.
Administratively, the use of licences has much to recommend it. The calculation of tax rates requires considerable information to correctly forecast their
relationship to externalities generated. In addition, changes in other prices
will aect the optimal tax rates and in a dynamic setting the taxes will need
continuous adjustment. These problems are avoided entirely by licences. In
a spatial economy, the control of the spatial distribution of externalities will
only be achieved through taxation if the tax rates are spatially dierentiated
which raises the information necessary for their design. In contrast, licences can
restrict the right to emit externalities to a given area and control the spatial
allocation directly. Despite these points, when the properties of licences and
taxes are considered in detail, the choice between the two is not as clear-cut as
these administrative advantages may suggest.
10.7.1
Certain costs and benefits
The basis of the argument in favour of licences, which is given in more detail in
Parish (1972), is that a market in pollution quotas would see them purchased by
those who value them most highly and that such purchasers would give the best
return to society for the given level of pollution. The quantity of licences would
determine the level of externality that would be generated, which it is presumed
would be set at the optimal level, whilst the bidding for them would see this
quantity allocated eciently between alternative sources. The tradeable licence
system therefore attains an ecient outcome.
When all costs and benefits are known with certainty by both the government and individual agents, tradeable permits and taxation are equivalent in
their eects up to a redistribution of income. This result is demonstrated for a
very rich spatial economy in Montgomery (1972), for an exchange economy in
Bergstrom (1976) and for a competitive market with entry and exit by Pezzey
(1992). The distribution of income resulting from licences is dependent upon the
method of distribution of licences. If each externality generating agent is sold a
quantity of licences equal to their optimal quantity at the market clearing price
then no further trading will take place and the distribution of income will be
identical to that with taxation. Alternatively, the licences may be distributed
free, possibly in proportion to agents existing level of generation of externality, which will lead to a redistribution of income from the government to the
externality generators relative to the tax solution. Other then these income
dierences, the choice between the two systems under certainty will primarily
depend on administrative convenience.
10.7. TRADEABLE LICENSES
10.7.2
333
Prices versus quantities
When costs and benefits are uncertain, the equivalence argument does not apply.
To see how this can arise, assume that the level of externality generation can be
reduced at some cost but that this cost is uncertain. If a quantity constraint is
chosen based on expected costs, this will lead to the level of externalities being
too high relative to the optimum when a low value of cost is realised. Conversely,
if a tax, or charge, is introduced this will result in excessive reduction when the
cost of reduction is revealed to be low. The consequences of this observation
were first formalised by Weitzman (1974) who derived a condition determining
which of quantity control or pricing would be preferable.
Consider the regulation of a single agent producing an externality and let
denote a reduction in the level of the externality from some initial baseline. The
emission of this externality can be controlled either by announcing a Pigouvian
tax of t per unit of externality or by directly announcing the reduction that must
be achieved. Although the assumption of a single agent prevents any trading
in the licences, the argument can be extended to many agents with the licences
traded between then. The cost of reducing the externality is given by C (, )
where is a random variable whose distribution is known to the regulator but
whose realisation is not observed at the time the policy must be announced. The
benefits are represented by B (, ) where is also random with the distribution
known but the realisation not observed until after the policy is announced. It
is assumed that E [] = E [] = 0 and E [] = 0.
When the regulation policy takes the form of an announcement of the reduction in externality that must be achieved, the optimal reduction, b
, is defined
by
b
= arg max E [B (, ) C (, )] ,
(10.81)
{}
so that it satisfies
E [B1 (b
, )] = E [C1 (b
, )] .
(10.82)
When a charge per unit of externality is employed, a maximising agent will

reduce the externality up to the point where the charge is equal to the marginal
cost of reductions after the realisation of the random variable . This gives the
equality
t = C1 (, ) .
(10.83)
When C11 6= 0, (10.84) can be solved to write
= h (t, ) .
The optimal value of t, denoted e
t, is then defined by
e
t = arg max E [B (h (t, ) , ) C (h (t, ) , )] ,
{t}
so that

t, , h1 e
t, = E C1 h e
t, , h1 e
t, .
E B1 h e
(10.84)
(10.85)
(10.86)
334
Using (10.84) in (10.87) gives

E B1 h e
t, , h1 e
t,
e

t=
,
(10.87)
E h1 e
t,

t, . This level of externality reducand a level of externality reduction e
=h e
tion is known only after is realised.
To permit a comparison of the two methods of regulation, the costs and
benefits are approximated around b
by the quadratic expansions
and
C3
C (, ) = C 1 + C 2 + [ b
] +
[ b
]2 ,
2
(10.88)
B3
B (, ) = B 1 + B 2 + [ b
] +
(10.89)
[ b
]2 .
2
2
From (10.89), dierentiation

C1 (, ) = C + + C 3 [ b
] and from
2
gives
(10.90) B1 (, ) = B + + B 3 [ b
]. The equality in (10.83) then implies
that C 2 = B 2 . Solving (10.84) using (10.89) gives the reaction of the agent as
hence
= h (t, ) = b
+
t C2
,
C3
(10.90)
1
.
(10.91)
C3
Substituting (10.92) into (10.88), cancelling the terms in C 3 and taking the
expectation of B1 (, ) gives
h1 (t, ) =
B3
B3
e
t = B 2 3 C 2 + 3 t.
C
C
(10.92)
t = C 2 . Using this result in

From the equality of B 2 and C 2 , it follows that e
(10.91) yields
e
=b
3.
(10.93)
C
To determine which of fee and quantity regulation is superior, define the
additional gain from using taxation relative to quantity control by
= E [[B (e
, ) C (e
, )] [B (b
, ) C (b
, )]] .
(10.94)
Substituting into (10.94) using the derived terms
2 B 3 + C 3
=
,
(10.95)
2
2 [C 3 ]

where 2 = E 2 . The result in (10.95) is easily interpreted. From the approximations in (10.89) and (10.90) it can be seen that a natural assumption would
10.8. INTERNALISATION
335
be B 3 < 0 < C 3 . Therefore taxation will be preferable if the gradient of the

marginal benefit function is less (in absolute value) than that of marginal cost.
When the opposite holds, quantity restrictions are preferred.
This analysis has been extended to a number of other regulatory schemes by
Laont (1977), Ireland (1977), Yohe (1978), Dasgupta, Hammond and Maskin
(1980) and Chen (1990) (not all of which are directly applicable to the control
of externalities). However, all these analyses are restricted by the employment
of quadratic approximations. Therefore, beyond the observation that charges
and quantities are not equivalent under uncertainty, no general principles have
been derived from this line of analysis.
10.7.3
Nonlinear pricing
The diering eects of fees and licences suggests that, rather than relying on
either one alone, it would be best to combine the instruments. Roberts and
Spence (1976) prove that such a combined system reduces expected social costs
below the level achievable by either fees or licences alone. This, though, does
not exhaust the possibilities.
Spence (1977) observes that control by licences can be interpreted as a nonlinear pricing system in which the price of the right to emit an externality is
zero (or the cost of the licence) until the quantity permitted by the licence is
reached. For any quantity above this level, the price becomes infinite. In contrast, fees represent a linear price system for the externalities. It follows from
these observations that if the set of permissible pricing systems is extended to
the set of continuous functions, both fees and licences, or any combination of
the two, will be dominated by a suitably chosen nonlinear pricing system from
this set. In such a system, the price for emitting externalities is dependent
upon the quantity emitted. These results show that the choice between price
and quantity is therefore irrelevant and all that really matters is the selection
of the optimal nonlinear pricing system. Although the optimal nonlinear price
system can be characterised using methods similar to those of Chapter 5, its
value is limited by the considerable administrative diculties involved with the
implementation of such a system.
10.8
Internalisation
A further method of externality control, as first discussed by Davis and Whinston (1962), is to encourage the internalisation of the externality so that private
and social costs become the same. The essence of the internalisation is that if
firm A causes a negative externality only upon firm B, then the firm formed by
merging both A and B will take account of the externality when choosing its
optimal behaviour. Hence, no ineciency will arise. Such arguments have also
been proposed as providing part of the rationale concerning the existence of the
firm.
336
Internalisation though is not without its diculties. To highlight the first

of these, consider an industry in which the productive activity of each firm in
the industry causes an externality for the other firms in the industry. In this
situation the internalisation argument would suggest that the firms become a
single monopolist. If this were to occur, welfare loss would then arise due to
the monopolistic behaviour and this may actually be greater than the initial
loss due to the externality. Although this is obviously an extreme example,
the internalisation argument always implies the construction of larger economic
agents and a consequent increase in market power. The welfare loss due to
market power then has to be oset against the gain from eliminating the eect
of the externality.
The second diculty is that the economic agents involved may simply not
wish to be amalgamated into a single unit. This objection is particularly true
when applied to consumption externalities since if a household generates an
externality for their neighbour it is not clear that they would wish to form a
single household unit, particularly if the externality is a negative one.
In summary, internalisation will eliminate the consequences of an externality
in very direct manner by ensuring that private and social costs are equated.
However it is unlikely to be a practical solution when many distinct economic
agents contribute separately to the total externality and has the disadvantage
of leading to increased market power.
10.9
Conclusions
Externalities are a prevalent feature of economic life and their existence can
lead to ineciency in an unregulated competitive economy. Although the Coase
theorem suggests that such ineciencies will be eliminated by private trading
in competitive markets, number objections can be raised to this conclusion.
Amongst these are the lack of well-defined property rights, the thinness of markets and the incomplete information of market participants. Each of these
impediments to ecient trading undermines the practical value of the Coase
theorem.
The obvious policy response to the externality problem is the introduction
of a system of corrective Pigouvian taxes with the tax rates being proportional
to the marginal damage inflicted by externality generation. When sucient
dierentiation of these taxes is possible between dierent agents, the first-best
outcome can be sustained but such a system is not practical due to its informational requirements. Restricting the taxes to be uniform across agents allows
the first-best to be achieved in some special cases but, generally, leads to a
second-best outcome. An alternative system of control is to employ marketable
licences. These have administrative advantages over taxes and lead to an identical outcome in conditions of certainty. With uncertainty, licences and taxes
have dierent eects and combining the two can lead to a superior outcome.
Chapter 11
Imperfect Competition
11.1
Introduction
In the previous chapters, the assumption of competitive behaviour has been

maintained throughout. It is often best to view this as a useful restriction
for developing initial ideas and eliminating unnecessary complication. As a
representation of reality it is clearly flawed, an observation easily supported
by casual empiricism. This chapter relaxes the competitive assumption and
reviews some of the major results that have been derived within the framework
of imperfect competition.
The first point to note is that there are numerous forms of imperfect competition which vary with respect to the nature of products, the strategic variables
of the firms, the objectives of the firms and the possibility of entry. Products
may be homogeneous or dierentiated and the strategic variables of the firms
can either be prices or quantities with, possibly, additional instruments such as
investment or advertising. The firms objectives may be individual profit maximisation or, alternatively, joint profit maximisation. Entry may be impossible,
so that an industry is composed of a fixed number of firms, it may be unhindered or incumbent firms may be following a policy of entry deterrence. To
avoid some of this complexity, this chapter will focus primarily upon economies
with quantity setting oligopoly and homogeneous products although at some
points conjectures are introduced to permit flexibility. For a suitable choice
of conjecture, this form of oligopoly has the advantage of being equivalent to
monopoly when the industry has a single firm and, under most circumstances,
to competition as the number of firms increases without limit. This makes the
economy both straightforward and flexible.
A second point of some relevance is that tax incidence is more complex
with imperfect competition. Under the competitive assumption any taxes are
simply passed forward by the firms since price is always set at marginal cost.
In contrast, prices on imperfectly competitive markets are set at a level above
marginal cost and an increase in cost due to taxation need not be reflected in
337
338
CHAPTER 11. IMPERFECT COMPETITION
an identical increase in price. To determine the actual change it is necessary to

work through the comparative statics of the industry in question. In addition
to the price eects, imperfectly competitive firms may also earn non-zero profits
and the eect of taxation on these must also be determined.
The initial sections of this chapter focus on issues related to the construction and analysis of a general equilibrium economy with imperfect competition.
After introducing the economy that forms the basis of the chapter, it is shown
why the equilibrium is not Pareto optimal and measures of the welfare loss due
to imperfect competition are described. This is followed by an analysis of commodity tax incidence and optimal tax rules. The chapter is completed by a
reconsideration of the necessity for production eciency.
11.2
Imperfect competition and general equilibrium
The first formal general equilibrium economy with imperfect competition can be
attributed to Negishi (1961). Also of significance are Arrow and Hahn (1971),
Gabszewicz and Vial (1972), Roberts and Sonnenschein (1977), Cornwall (1977)
and the survey by Hart (1985). Unlike the Arrow-Debreu formulation of the
competitive economy there is not a single, generally accepted framework but a
number of alternative specifications. The economies are dierentiated by the
assumptions made about the form of demand function known by the firms and
on the relation of demand to profit.
11.2.1
Objective and subjective demand
The first distinction to note is between economies that employ objective demands
and those using subjective demands. This distinction arises from the information
that the firms comprising an imperfectly competitive industry must possess in
order to know their profit function. In a competitive economy, a firm need only
observe the set of market prices to determine its profit maximising strategy. In
contrast an imperfectly competitive firm requires the knowledge of the demand
function for its product. Naturally this is a far greater informational requirement
than just the knowledge of prices and, when literally interpreted, implies that
the firms need to be able to solve the economy to generate the excess demands.
Faced with this informational problem, Negishi (1961) assumed that the
firms actually knew only a linear approximation to their true demand functions,
with the restriction that the linear approximation was equal to the true demand
at equilibrium prices. The latter condition ensures that the firms generate
the profit levels that they were expecting. Such demands have been labelled
subjective since they exist only as beliefs held by firms. This approach has been
extended by Gary-Bobo (1989) to consider firms that perceive knowledge of a
kth order Taylor expansion of their true demand. When k = 1 the equilibria
coincide with the equilibrium arising from the true demands.
11.2. IMPERFECT COMPETITION AND GENERAL EQUILIBRIUM 339

In an economy with objective demand, such as Gabszewicz and Vial (1972),
it is assumed that the firms actually know their demand functions. Despite
the informational burden this imposes, it does overcome the problems of the
subjective approach in that it does not require specification of how the firms form
the subjective demands or of how they may seek to revise them. In the analysis
of this chapter only the objective approach is followed. This can be supported
on the grounds of simplicity and the fact that since subjective demands are
typically chosen equal to objective demands at equilibrium, the equilibrium of
the economy should not be aected by the choice.
The assumptions concerning the distribution of profits and their eect upon
demand is another area in which alternative economies dier. If returned to consumers as dividends, whenever income eects are non-zero a firms profits must
appear as an argument of its demand function. This causes some diculties
for the specification of the firms maximisation problem since the quantity to
be maximised (profit) appears as an argument in the objective function. When
this occurs, standard results on maximisation are not applicable.
There have been three responses to this. The first is to assume the structure
of the economy is such that there are no income eects. This approach is exemplified by the work of Hart (1982) in which the economy is divided into separate
sectors and the firms distribute profit generated in one sector to consumers who
purchase in a dierent sector. In this way, although aggregate demand depends
on profit, the demand for each firm is independent of its own profit level. However, assumptions such as this have the disadvantage of being highly artificial.
The second response is more direct and simply involves assuming that all profits
are taxed at a rate of 100% so that profit income accrues to the government
alone. This assumption has the disadvantage of eliminating the motive for the
choice of profit maximising strategy for the firms. The final approach is to meet
the problem directly and to extend the analysis to suit. This has the drawback
of leading to greater complication in the results. The approach taken below will
be to adopt a combination of the latter two possibilities, with the choice made
to suit the purpose.
To complete this discussion of the choice of structure, it should be noted that
intermediate goods are almost never included in general equilibrium economies
with imperfect competition. This exclusion is due to the diculty of formulating
the derived demand for such goods when a number of firms have market power.
For a discussion of this point, see Hart (1985). Although they are similarly
excluded from most of the discussion below, they do occur at two points: in the
discussion of the extent of welfare loss due to imperfect competition and in the
treatment of production eciency.
11.2.2
Price normalisations
Prior to setting out the details of the economy, it is necessary to describe the
method of price normalisation that is to be employed. For competitive general equilibrium economies, the permissible price normalisations are well known
and have been described in Chapter 2: the price vector may be transformed
340
in any way provided the ratios of prices remain unaected. The same claim
cannot be made for economies with imperfect competition and the general class
of permissible normalisation rules is more restricted than for the competitive
economy.
Since the work of Gabszewicz and Vial (1972) it has been recognised that
normalisation rules that would not aect a competitive equilibrium will have
real eects upon equilibrium with imperfect competition as the following example demonstrates. Consider an economy with one single-product monopolist producing at zero cost and whose profit maximising output is finite. Now
imagine a representation of this economy which, for analytical simplicity, uses a
normalisation rule that selects the monopolists good as the numeraire. The monopolists profit maximising output in the representation of the economy then
becomes unbounded and so the normalisation rule has altered the real equilibrium. Since this normalisation rule is known not to aect competitive equilibria,
this example demonstrates that the class of normalisation rules that do not affect equilibrium is smaller for imperfect competition than for the competitive
model. The actual rules employed by Gabszewicz and Vial are essentially arbitrary and were no doubt used for computational convenience. The same general
comment applies to Dierker and Grodal (1986). In contrast, both Negishi (1961)
and Cornwall (1977) restrict prices to the unit simplex in order to exploit the
resulting compactness properties of this set. Cornwall notes that this choice
will aect the equilibrium except in the special case of the Negishi model with
linear subjective demand. Roberts (1980d) employs leisure as the numeraire
and notes that the equilibrium is generally not numeraire free. It is not clear
whether the market for the numeraire is competitive or not. Neither Guesnerie and Laont (1978) nor Roberts and Sonnenschein (1977) use or discuss a
price normalisation. Finally, each economy in the survey paper of Hart (1985)
employs a dierent method of price normalisation.
As the variety of price normalisations noted above illustrates, there is no
common approach but there is general agreement that some price normalisations
will aect the real equilibrium. In Cripps and Myles (1989), it is shown that the
equilibrium is invariant to any normalisation rule that is defined as a function
of competitive goods prices but is independent of the prices of goods traded on
imperfectly competitive markets. For the economies described below, a simple
version of this rule is adopted: labour is assumed to be traded on a competitive
market and the wage rate is taken as numeraire.
11.2.3
The economy
The general equilibrium economy with imperfect competition that will form the
basis for the analysis of commodity taxation and production eciency is now
introduced. The economy is based upon that used for the analysis of commodity
taxation by Myles (1989a). It will be assumed that there is a fixed number of
firms in each industry but the economy can easily be extended to allow for entry.
Consider an economy with n + 1 goods, indexed i = 0, 1, ..., n, where good 0
is chosen to be labour. It is assumed that the labour market is competitive and
11.2. IMPERFECT COMPETITION AND GENERAL EQUILIBRIUM 341

that consumers endowments are of labour alone. The n goods are partitioned
into two subsets. The first subset consists of goods i = 1, ..., K and these are
assumed to be produced by competitive industries, each employing a constant
returns to scale technology that has labour as the only input. The assumptions
of constant returns and a single input imply that for each good there is a constant
ci that describes the labour input per unit of output. With a wage rate p0 , the
competitive assumption implies marginal cost pricing so that
pi = ci p0 , i = 1, ..., K.
(11.1)
Since profits are zero under constant returns to scale, equation (11.1) is a complete description of the competitive sector.
The remaining n K goods are produced by imperfectly competitive industries. In the industry producing good i, there are mi firms, indexed j = 1, ..., mi .
The demand function for good i is denoted by
Xi = Xi (p0 , ..., pn , ) ,
(11.2)
where the aggregate level of profit, , is equal to the sum of profits of all firms
=
mi
n
X
X
ji .
(11.3)
i=K+1 j=1
The demand function in (11.2) is taken to be the true demand for the economy
so it is one of objective demand. Assuming that (11.2) is strictly monotonic in
pi , it can be solved to give the inverse demand function for good i
pi = i (p0 , ..., pi1 , Xi , pi+1 , ..., pn , ) .
(11.4)
To analyse the firms decisions, it is assumed that quantities are the strategic
variable. Hence each firm in industry i chooses its output level to maximise
profits. Denoting a typical firm by j, the firm chooses xji to maximise

(11.5)
ji = i (p0 , ..., pi1 , Xi , pi+1 , ..., pn , ) xji Cij xji ,

where Cij xji is the cost function of the firm. The Cournot assumption is
0
adopted so that when choosing xji firm j takes as fixed the outputs, xji , j 0 =
1, ..., mi , j 0 6= j, of all other firms in its industry. In addition, the Cournot
assumption is extended to the other arguments of (11.4) so the firm also takes
as given the prices on all other markets and the profit levels of all firms other
than itself.
Totally dierentiating (11.5) under these assumptions gives
"
#
Cij
j
j i
j
j i
d i 1 xi
.
(11.6)
= dxi pi + xi
Xi
xji
342
Assuming that the income eect in demand is suciently weak that the lefthand side satisfies
1 xji i > 0,
(11.7)
then the first-order condition that the firms decision must satisfy is given by
pi + xji
Cij
i
= 0.
Xi
xji
(11.8)
In addition to satisfying this condition, at an equilibrium any choice for the firm
must also have the property that it is consistent with the assumptions made on
the profit levels and outputs of other firms. That is, the chosen level of output
must also satisfy the profit identity

(11.9)
ji i (p0 , ..., pi1 , Xi , pi+1 , ..., pn , ) xji + Cij xji = 0.
Simultaneously solving (11.8) and (11.9) for the mi firms comprising industry i,
the output and profit of firm j can be expressed in terms of the prices charged
on other markets and the profits levels of firms in other industries. This gives
the following two equations
mi0
n
X
X
0
ji = ji p0 , ..., pi1 , pi+1 , ..., pn ,
ji0 ,
(11.10)
i0 =K+1,i0 6=i j=1
xji = ji p0 , ..., pi1 , pi+1 , ..., pn ,
n
X
mi0
X
i0 =K+1,i0 6=i j=1
ji0 .
(11.11)
Equation (11.11) can now be used to replace aggregate output, Xi , in (11.4)

by the sum of individual firms outputs and (11.10) to substitute for the profits
levels of firms in industry i. This determines an equilibrium price on market i
as
mi0
n
X
X
0
pi = i p0 , ..., pi1 , pi+1 , ..., pn ,
(11.12)
ji0 .
i0 =K+1,i0 6=i j=1
Using equations (11.10) and (11.12), a general equilibrium can now be defined formally.
Imperfectly Competitive Equilibrium
n
n oo
An imperfectly competitive equilibrium is an array {pi } , ji
such that
(i) pi = ci p0, i = 1, ..., K,
Pmi0 j 0
P
i0 , i = K+1, ..., n,
(ii) pi = i p0 , ..., pi1 , pi+1 , ..., pn , ni0 =K+1,i0 6=i j=1
and
Pmi0 j 0
Pn
i0 , i = K+1, ..., n,
(iii) ji = ji p0 , ..., pi1 , pi+1 , ..., pn , i0 =K+1,i0 6=i j=1
j = 1, ..., mi .
11.3. IMPERFECT COMPETITION AND WELFARE
343
This definition of the equilibrium incorporates both the profit maximisation

of the firms and the existence of positive levels of profit. It should be noted that
the economy takes as given the demand functions of the households, rather than
using their preference relation, and restricts demand to depend only upon total
profits. It is therefore being implicitly assumed in the above statement that
there is some array of shareholdings that determine the allocation of profits to
households and that the households are maximising utility given their income
levels and consumption sets. Since the emphasis here is upon the consequences
of the non-competitive behaviour of the firms, it does not seem necessary to
provide the details of the consumption sector since these remain as for the
competitive economy.
It is considerably more problematic to prove the existence of an equilibrium
for the imperfectly competitive economy than for the competitive case. The
first diculty arises from the maximisation defined in (11.5). If this is to have
a unique solution that is continuous in the variables parametric to the individual firm, then the profit function of the firm must be strictly convex as a
function of its own output. Unfortunately, strict convexity is a strong requirement since it imposes severe restrictions upon the demand function facing the
firm. As shown by Roberts and Sonnenschein (1977) it is possible to construct
economies in which the convexity fails. Moving from the single firm to the economy, the equilibrium requires that the output and profit functions given in the
definition should all be satisfied simultaneously. Although similar in nature to
simultaneously solving the excess demand functions for a competitive economy,
there is less structure in the imperfectly competitive economy and stronger assumptions are necessary to ensure the existence of a solution. Despite these
diculties, given that it is assumed an equilibrium exists, the characterisation
of equilibrium above is straightforward and can be employed in further analysis.
11.3
Imperfect competition and welfare
Imperfect competition is one of the standard examples of market failure which

lead to the non-achievement of Pareto optimality. It is on this basis that economic policy is usually suggested as necessary in the presence of imperfect competition in order to reduce ineciency.
11.3.1
Failure of Pareto optimality
To demonstrate that imperfect competition does not generate a Pareto optimum, it is necessary to provide a suitable characterisation of Pareto optimality.
There are several ways in which this can be done. Firstly, by consideration of
the competitive equilibrium it can be appreciated that competitive firms price
at marginal cost and that this is one of the conditions for Pareto optimality.
Contrasting this to the pricing policy implicit in (11.8), it can be seen that
i
whenever X
is non-zero, price will not be equal to marginal cost in the imi
perfectly competitive industry. Further investigation of the relation of prices to
344
costs, or the mark-up over costs, with imperfect competition can be found in
Cowling and Waterson (1977).
A second method of comparison is to return to the economy summarised in
(9.1) and (9.2) but without the inclusion of the externality eects. Repeating
the analysis given there, the conclusion that follows is that at a Pareto optimum
the ratio of shadow prices for any pair of goods, given by the s, should be should
be equal to the ratio of rates of transformation. In addition, at a competitive
equilibrium the ratio of shadow prices is equal to the ratio of market prices
pi
i
=
=
pk
k
F
Xi
F
Xk
(11.13)
For the economy with imperfect competition, assuming that there is a single
firm in each imperfectly competitive industry in order to simplify the construction, the production possibilities are constrained by the relation that the sum
of labour demand from the competitive and imperfectly competitive sectors
satisfies
K
n
X
X
Ci (Xi )
ai Xi +
0 0,
(11.14)
p0
i=1
i=K+1
where Ci (Xi ) is the cost function of the single firm in industry i and 0 is the
initial endowment of labour. For two goods i and k produced by imperfectly
competitive firms, it follows from (11.14) that
F
Xi
F
Xk
Ci
xi
Ck
xk
(11.15)
(11.16)
Hence using (11.8),

i
pi + xi X
i
k
pk + xk X
k
Equation (11.16) captures the notion of market failure due to the lack of price
taking behaviour. The prices should be proportional to the marginal rates of
transformation which capture the social cost of producing each good. However,
the fact that the imperfectly competitive firms take the eect of their actions
upon prices into account eliminates the direct proportionality.
11.3.2
Measures of welfare loss
It has been shown that the imperfectly competitive equilibrium is not Pareto
optimal. Following from this, the equilibrium cannot maximise the value of any
social welfare function that satisfies the Pareto criterion defined in Chapter 2.
This observation then makes it natural to consider what the degree of welfare
loss may actually be, either for a real economy or for simulated examples. The
assessment of monopoly welfare loss has been a subject of some dispute in which
11.3. IMPERFECT COMPETITION AND WELFARE
345
calculations have provided a range of estimates from the eectively insignificant

to considerable percentages of potential welfare.
Contributions to the literature on monopoly welfare loss can be characterised
according to three criteria: the welfare measure used; whether data or simulations are employed and whether the underlying model is of general or partial
equilibrium. The choice between welfare measures can eectively be reduced
to either calculating welfare loss triangles (in terms of National Income in Harberger (1954) and Gross Corporate Product in Cowling and Mueller (1978)) or
specifying an explicit welfare function and using this to evaluate welfare loss,
a methodology whose case has been argued most forcefully by Bergson (1973).
The earlier contributions were primarily concerned with the use of data to calculate losses for actual economies (Harberger for the USA, Cowling and Mueller
for the UK and USA) but more recent work has concentrated on the use of
simulations to calculate potential losses (Bergson (1973), Kay (1983)). Dickson
and Yu (1989) employs a mix of both data and simulation. With respect to the
form of model, the vast majority of contributions have adopted a partial equilibrium framework. There are some exceptions to this, most notably Ireland
(1978), Kay (1983) and Myles (1994a).
The initial study of monopoly welfare loss is usually attributed to Harberger
(1954) who considered the eect of monopolisation in United States manufacturing industry for the period 1924 to 1928. From the data it is concluded that
welfare loss is equal to 0.08% of national income. Clearly if this figure is accurate,
then monopoly welfare loss was insignificant in the United States. In contrast
to Harberger, Cowling and Mueller (1978) include the cost of advertising in the
measure of welfare loss on the interpretation that advertising is undertaken with
the intention of maintaining a monopoly position. This naturally raises their
estimates. Their analysis of welfare loss in the United States is based on data
for 734 firms between 1963 and 1966 and concludes that welfare loss is between
4% and 13% of Gross Corporate Product. For the United Kingdom, Cowling
and Mueller conclude that the top 103 firms in 1968 to 1969, accounting for
a third of GNP, generated a welfare loss of between 3.9% and 7.2% of Gross
Corporate Product. This contrasts with the loss of 0.2% to 3% using the Harberger measure for the same data set. These two sets of figures clearly provide
conflicting evidence, as do the numerous other contributions that are surveyed
in Sawyer (1981). The actual extent of welfare loss therefore remains an open
question.
Turning now to measures of welfare loss in simulation models, these have
generated far higher figures than analyses of data. Using a constant elasticity
of substitution utility function, Bergson (1973) produces a range of estimates
from 0.06% of national income to 39.03% The drawback to these figures is that
they are calculated on the basis of hypothesised price-cost mark-ups rather than
having the mark-up determined as the equilibrium of a specified economy. Kay
(1983) employs a model with one consumption good that is produced by a
346
monopolist using a single form of labour service. With the utility function
U=
x1
1
x0 ,
(11.17)
a range of estimates were calculated ranging from 3% of GNP when = 0.1

to 134% when = 0.9. In addition, the loss was above 10% for all values of
> 0.3.
The analyses described above have considered monopoly power to be present
only in the markets for final goods and have implicitly taken intermediate goods
markets to be competitive. When the outputs of imperfectly competitive industries are used both as intermediate inputs and for final consumption, welfare loss
can only be increased by the additional distortion of input prices. In a general
equilibrium economy with intermediate goods, Myles (1994a) finds welfare loss
figures ranging from 11% of the attainable utility level to 79%.
It can be appreciated from this discussion that there have been a broad range
of estimates of monopoly welfare loss. The figures generated by simulation
studies are generally greater than those arising from data and indicate that,
theoretically, monopoly may be very damaging in terms of reduced welfare.
11.4
Commodity taxation
The use of commodity taxation when there is imperfect competition has the
additional motivation, beyond those of raising revenue and redistributing welfare
present in the analysis of commodity taxation in Chapter 4, of attempting to
reduce the welfare loss due to non-competitive behaviour. There is consequently
a counter-distortionary role for commodity taxation even in a single consumer
model with zero revenue requirement.
Compared to the competitive model there are several factors that complicate
tax analysis when imperfect competition is introduced. Firstly, the analysis of
tax incidence is more complex in the imperfectly competitive case. With competition, commodity taxes are simply passed forward by the firms since they
always price at marginal cost. In contrast, prices on imperfectly competitive
markets are set at a level above marginal cost and an increase in cost due to
taxation need not be passed directly to consumers. The actual change in price
can only be determined by working through the comparative statics of the industry in question. The analysis of taxation therefore begins with a consideration
of tax incidence. Secondly, in addition to the price eects, imperfectly competitive firms may also earn non-zero profits and the eect of taxation upon these
must also be determined. As with the incidence of taxes upon prices, the profit
eects have to be calculated from the comparative statics of each industry.
With imperfect competition it is also necessary to note that ad valorem and
specific taxation do not have identical eects. That is, for a given level of posttax price, the two methods of taxation may lead to dierent levels of revenue.
To reduce the level of complexity below, the major part of the analysis of this
11.4. COMMODITY TAXATION
347
chapter will the conducted in terms of specific taxes. The interaction of specific
and ad valorem taxes is considered in Section 5.
After a review of tax incidence, two economies will be analysed. The first
economy is has a single consumer and three goods. This is employed to investigate the factors that determine the levels of taxation when the taxes are
used solely to counter the distortions due to non-competitive behaviour. The
second economy characterises optimal taxes using the imperfectly competitive
economy introduced in Section 2.3 and provides a non-competitive variant of
the Diamond-Mirrlees optimal commodity tax rule.
11.4.1
Tax incidence
Tax incidence is concerned with the eect of taxation upon prices and profit
levels. Although tax incidence was of considerable importance in the determination of the optimal tax rules of Chapter 4, the structure of the competitive
economy that was used to derive the Ramsey rule simplifies tax incidence to
such an extent that a formal discussion of incidence was not required. In addition, in the more general economy of Section 6 of Chapter 4, the separation of
producer and consumer sectors again obviates the need for an analysis of tax
incidence. With imperfect competition, an analysis of tax incidence is essential.
To emphasise the points made above, it is worth looking again at the eects
of commodity taxation in the competitive model. Retaining the assumption
that labour is the only input implies that with the wage rate as the untaxed
numeraire that there is a set of fixed producer prices p0 , ..., pK in the competitive
sector. It then follows immediately that
qi = pi + ti , i = 1, ..., K,
(11.18)
qi
= 1,
ti
(11.19)
qi
= 0, i 6= k.
tk
(11.20)
and from this that
and
Equation (11.19) states that the commodity tax is forward-shifted at the rate
of 100%, that is, the entire tax is shifted by the firm onto the consumer. Equation (11.20) illustrates that there is no interaction between the price of a good
produced by a competitive industry and the tax on any other good. It is these
results that give the competitive tax rules their precise structure.
To illustrate the important features of tax incidence with imperfect competition a simple, but flexible, example will now be considered. Consider a single
industry and assume that this industry is composed of mi firms, each with the
same cost function and beliefs about the other firms reactions. These assumptions ensure that there will be a symmetric equilibrium, will all firms producing
the same output level. The inverse demand function facing the industry is
348
denoted
mi
X
qi = i
xji , qk ,
(11.21)
j=1
where xji is the output of firm j and qk is some representative other price. The
second price is introduced in order to allow consideration of the consequences
of its variation. Note that the prices are represented by qs to denote that they
are post-tax consumer prices. Each of the mi firms aims to maximise
mi

X
0
ji = xji i xji +
xji , qk ti xji Ci xji ,
(11.22)
j 0 =1,j 0 6=j

where Ci xji is the cost function that is the same for all firms in industry i
and ti is the specific tax levied on the output of the ith industry.
The flexibility in the example is introduced by assuming that each firm holds
a conjecture about how the other firms will respond to changes in its output
level. By suitable choice of the conjecture, it is possible for the model to generate
all outcomes from competitive pricing to monopoly pricing. The conjecture, ,
can be defined formally as
Pmi
j0
xji + j 0 =1,j
0 6=j xi
=
.
(11.23)
xji
As will be made clear after inspection of the first-order condition (11.24) below,
if = mi monopoly pricing arises, = 1 represents Cournot behaviour and =
0 generates the Bertrand equilibrium with competitive marginal cost pricing.
Employing the definition of the conjecture, the first-order condition for profit
maximisation for firm j is
ji
xji
qi ti + xji
i
Ci
= 0.
Xi
xji
(11.24)
The first question concerning tax incidence is the calculation of the eect
of changing ti upon the equilibrium values of ji , j = 1, ..., mi , and qi . These
eects are derived by first noting that the symmetry assumption implies that
xji = xi , all j and hence that
mi
X
j
xi , qk = i (mi xi , qk ) .
(11.25)
qi = i
j=1
Totally dierentiating the first-order condition (11.24) by varying all outputs

and the tax rate gives
2 i
2 Ci
dxi mi i + i + xi mi
(11.26)
= dti .
Xi
Xi
Xi2
x2i
349
From the inverse demand function

dqi = mi
i
dxi .
Xi
(11.27)
Using (11.27) to eliminate dxi from (11.26) determines the eect of the tax
change upon the equilibrium price as
i
mi X
dqi
i
=
,
2
i
dti
[mi + ] X
+
xi mi X2i
i
(11.28)
where it has been assumed for simplicity that marginal cost is constant. The
i
expression dq
dti is termed the degree of forward shifting of the tax and, except
when the equilibrium is in the sense of Bertrand with = 0, it can be seen from
i
(11.28) that dq
dti is not necessarily equal to 1.
It is standard practice to distinguish between overshifting, in which case the
price rises by more than the increase in tax, and undershifting, for when it rises
by less. This distinction will be seen to be important in the determination of
relative tax rates. To proceed further, it is assumed that the equilibrium is
stable in the sense of Seade (1980), for which a sucient restriction is that the
denominator of (11.28) is negative. Using the stability restriction, it follows from
i
(11.28) that overshifting, which is equivalent to dq
dti > 1, occurs for non-zero
when
i
2 i
mi i < [mi + ]
+ xi mi
.
(11.29)
Xi
Xi
Xi2
Since both the left- and right-hand sides of (11.29) are negative, it can be
rearranged to give the equivalent condition for overshifting
i
2 i
> xi mi
.
Xi
Xi2
(11.30)
Condition (11.30) above can be viewed in two ways. Firstly, it represents a

restriction on the convexity of the inverse demand function since, if overshifting
2
is to occur, it is necessary that X2i is positive. Secondly, it can be phrased in
i
terms of Seades (1985) E by defining
2
Xi X2i
i
i
Xi
(11.31)
where E measures the elasticity of the slope of the inverse demand function.
In terms of E, when = 1, condition (11.30) states that overshifting occurs
when E > 1, that taxes are shifted at a rate of 100% when E = 1 and undershifted when E < 1. For other values of , the critical values of E are adjusted
accordingly.
Turning now to profits, the profit level of firm j can be written as a function
of the tax rate
ji (ti ) = xi (ti ) [qi (ti ) ti ] Ci (xi (ti )) .
(11.32)
350
Dierentiating profits with respect to ti gives
xi
Ci
qi
d ji
= xi
1 +
qi ti
.
dti
ti
ti
xi
Now using (11.24), (11.26) and (11.28) provides the expression
i
h
i
2 i
2
j
+
x
m
2x
2
i
i
i
Xi
d i
Xi
=
.
2
i
dti
[mi + ]
+ xi mi 2i
Xi
(11.33)
(11.34)
Xi
The result in (11.34) provides the interesting conclusion that the firms profit
level may actually increase as the tax rate increases, a result first shown by
Seade (1986). The sucient condition for this to occur, when = 1, can be
found from (11.34) to be E > 2. This possibility arises due to the tax change
moving the oligopolistic equilibrium closer to the monopoly outcome. Further
results along these lines can be found in de Meza (1982), Dierickx, Matutes and
Neven (1988), Myles (1987) and Stern (1987).
The tax incidence analysis above comprises what have been termed the direct
eects of taxation, where direct refers to the fact they relate to the eect of
the tax levied on the industry under consideration. In addition to the direct
eects there may also arise indirect or induced changes in prices and profits in
imperfectly competitive industry i due to changes in the tax rates on industries
other than i. From (11.20) and the fact that profits are zero for the competitive
industries, these induced eects do not arise in the competitive model.
To evaluate the induced eects upon price, (11.24) can be dierentiated with
respect to qk to give
h
i
i
h
i i
2 i
i 2 i
+
x
m
m
2
i
i
i
i
qk Xi
Xi Xi qk
dqi
Xi
=
.
(11.35)
i
2 i
dqk
[mi + ]
+ xi mi 2
Xi
Xi
Provided the inverse demand for good i is not independent of qk , the induced
price eect is almost always non-zero. Inspection of (11.35) shows that the
change in qi may be of either sign. That is, the induced eect may increase or
decrease the price of good i.
To illustrate the importance of this factor assume that good k is produced
by a competitive industry so that the inverse demand function can be written
as
mi
X
qi = i
xji , pk + tk .
(11.36)
j=1
dqi
dqi
Hence dq
calculated above is equal to dt
so, in optimising over tk , these
k
k
induced eects must be taken into account. Similar reasoning also holds when
industry k is imperfectly competitive but the additional, nonlinear, relation of
qk to tk must be included. To complete this analysis, it should also be noted
351
that there will be an induced eect of changes in qk upon the profit levels in
industry i.
To summarise this section, it has been noted that with imperfect competition taxes may be undershifted or overshifted. Overshifting implies that the
final price rises by more than any increase in tax, undershifting the converse.
The rate of shifting is determined by the degree of concavity or convexity of
the demand function facing the industry and the cost functions of the firms
that compose that industry. Concavity of demand leads to undershifting and
sucient convexity to overshifting. When marginal cost is constant, the relevant factor is Seades E, the elasticity of the gradient of the demand function.
The dependence of the market price upon the demand function introduces a
further factor. If a price of another good alters, due perhaps to a tax levied on
it, then this will aect the demand function and hence the market price of the
imperfectly competitive industry in question. These induced price changes are
considered in detail in Myles (1987, 1989a). To the above eects must also be
added the eects of taxation upon the profit levels of the firms which are not
always as straightforward as may be expected.
11.4.2
Optimal taxes
It is now possible to move beyond the study of tax incidence to a consideration

of the factors that determine optimal rates of commodity taxation in imperfectly
competitive economies. The analysis will first focus on tax reform in a two-good
and labour economy in order to highlight the importance of the tax incidence
results. This will then be extended to a construction of optimal tax rules in the
general equilibrium economy detailed at the start of this chapter.
An illustration
To begin the analysis of tax design, consider the following simple economy
taken from Myles (1987) in which the tax analysis consists of characterising
the welfare-improving tax reform starting from an initial position with no commodity taxation. The economy has a single consumer and a zero revenue requirement so the taxes are used merely to correct for the distortion introduced
by the imperfect competition. There are two consumption goods, each produced using labour alone. Good 1 is produced with constant returns to scale by
a competitive industry. It has post-tax price q1 = p1 + t1 . The second good is
produced by an imperfectly competitive industry that faces inverse demand
m2
X
j
x2 , q1 .
(11.37)
q2 = 2
j=1
It is assumed that the preferences of society can be represented by an indirect

utility function
V = V (p0 , q1 , q2 ) ,
(11.38)
352
where p0 is the price of labour. Profits are assumed to be taxed at a rate of

100% and the revenue used to purchase labour.
Rather than repeat the tax incidence analysis, the direct and induced eects
of taxation are understood to have been constructed as above. They are then
denoted
dq2
h1
,
(11.39)
dt1
and
h2
dq2
.
dt2
(11.40)
The expression of these eects at a general level has the advantage that it is
unnecessary to specify the particular model of imperfect competition in order
to derive results. A specific formulation is only needed when the results require
evaluation.
The tax reform problem now involves finding a pair of tax changes dt1 , dt2
that raise welfare whilst collecting zero revenue. If the initial position is taken
to be one with zero commodity taxes, the problem can be phrased succinctly
as finding dt1 , dt2 , from an initial position with t1 = t2 = 0 such that dV > 0,
dR = 0, where tax revenue, R, is defined by
R = t1 X1 + t2 X2 .
(11.41)
This framework ensures that one of the taxes will be negative, the other positive
and the aim is to provide a simple characterisation of the determination of the
relative rates. It should be noted that if both industries were competitive the
solution would be dt1 = dt2 = 0 so that non-zero tax rates emerge due to the
distortion caused by the imperfect competition.
From dierentiating the indirect utility function, it follows that the eect of
the tax change upon welfare is
dV =
V q2
V q2
V q1
dt1 +
dt1 +
dt2 .
q1 t1
q2 tq1
q2 t2
(11.42)
1
Since q1 = p1 + t1 , (11.19) implies q
t1 = 1. Using this fact and the definition of
the tax incidence terms h1 and h2
V
V
V
dV =
+
h1 dt1 +
h2 dt2 .
(11.43)
q1 q2
q2
From the revenue constraint

dR = 0 = X1 dt1 + X2 dt2 ,
(11.44)
where the fact that t1 = t2 = 0 initially has been used. Solving (11.44) for dt1
dt1 =
X2
dt2 .
X1
(11.45)
353
Substituting (11.45) into the welfare expression determines the welfare change
as dependent upon dt2 alone
X2 V
X2 V
V
h2
h1 dt2 .
(11.46)
dV =
q2
X1 q1
X1 q2
Finally, using Roys identity
X2
dV = X2 h2 + X2 +
X2 h1 dt2 .
X1
(11.47)
It then follows that if

X1 [1 h2 ] + X2 h1 < 0,
(11.48)
then dt2 < 0. From (11.48), the output of the imperfectly competitive industry
should be subsidised and the competitive industry taxed when h2 is large, so
that overshifting is occurring, and h1 is negative. These are, of course, sucient
conditions. In general, the greater the degree of tax shifting the more likely is
subsidisation. The explanation for this result is that if firms overshift taxes, they
will also do the same for any subsidy. Hence a negative dt2 will be reflected by an
even greater reduction in price. If h1 is also negative, the tax on the competitive
industry secures a further reduction in the price of good 2.
The conclusion of this analysis is that the rate of tax shifting is important
in the determination of relative rates of taxation. Although the economy is
simplified by abstracting away from profit eects, it does demonstrate that
with imperfect competition commodity taxation can be motivated on eciency
grounds alone.
Optimal taxes
The insights of the example are now employed to provide a characterisation
of optimal commodity taxes when there is also a positive revenue requirement.
This analysis is therefore a development of the commodity tax theory of Chapter 4 to incorporate imperfect competition. The economy to be used is based
on Myles (1989a) and involves introducing commodity taxes into the general
equilibrium economy described at the start of this chapter. To simplify notation, only a single-consumer tax rule will be constructed. The extension to
many-consumers is straightforward.
Treating labour as an untaxed numeraire with a wage rate p0 , the post-tax
price of good i from the competitive sector when the commodity tax is set at ti
is given by
qi = ci p0 + ti , i = 1, ..., K.
(11.49)
Profits remain zero for all firms in the competitive sector.
Introducing taxes into the definition of profit for the imperfectly competitive
firms, each firm in industry i chooses its output level to maximise

ji = [i (q0 , ..., qi1 , Xi , qi+1 , ..., qn , ) ti ] xji Cij xji ,
(11.50)
354
where inverse demand, i (), is now dependent on post-tax prices. Simultaneously solving the first-order condition that results from maximising (11.50) and
the profit identities for the mi firms comprising industry i, the output and profit
of firm j in industry i can be expressed by the following two equations:
mi0
n
X
X
0
(11.51)
ji0 ,
ji = ji q0 , ..., qi1 , ti , qi+1 , ..., qn ,
i0 =K+1,i0 6=i j 0 =1
xji = ji q0 , ..., qi1 , ti , qi+1 , ..., qn ,
n
X
i0 =K+1,i0 6=i
mi0
X
j 0 =1
ji0 .
(11.52)
It should be noted that the arguments of (11.51) and (11.52) are the prices
and profits of other industries and the tax rate on industry i. Aggregating
output and using the inverse demand function determines an equilibrium price
on market i
mi0
n
X
X
0
qi = f i q0 , ..., qi1 , ti , qi+1 , ..., qn ,
ji0 .
(11.53)
i0 =K+1,i0 6=i j 0 =1
The derivatives of f i and ji are precisely the tax incidence terms calculated
in the previous section for which it was assumed that all other variables are
constant.
The general equilibrium of the economy with taxation is then the simultaneous solution to
qi = ci p0 + ti , i = 1, ..., K,
(11.54)
m
0
n
i
X
X
0
ji0 , i = K + 1, ..., n,
qi = f i q0 , ..., qi1 , ti , qi+1 , ..., qn ,
i0 =K+1,i0 6=i j 0 =1
and
ji = ji q0 , ..., qi1 , ti , qi+1 , ..., qn ,
n
X
mi0
X
i0 =K+1,i0 6=i j 0 =1
(11.55)
ji0 , i = K+1, ..., n, j = 1, ..., mi .
(11.56)
Solving equations (11.55) and (11.56) simultaneously for all imperfectly competitive goods prices and firms profits determines the equilibrium level of prices
and profits in the imperfectly competitive industries as functions of the tax rates
and the competitive prices
qK+1 = K+1 (q0 , ..., qK , tK+1 , ..., tn ) ,
..
.
qn = n (q0 , ..., qK , tK+1 , ..., tn ) ,
1
K+1 = 1,K+1 (q0 , ..., qK , tK+1 , ..., tn ) ,
n
m
n
..
.
= 1,K+1 (q0 , ..., qK , tK+1 , ..., tn ) .
(11.57)
355
These equations can be used to determine the eects of changes in the government control variables.
Before proceeding to a derivation of the optimal tax rules, it is helpful to
understand the relation between equations (11.57) and (11.51)-(11.53). To do
this it is necessary to realise that the derivatives of (11.57) capture the eect
of the tax changes taking into account general equilibrium adjustment of all
other prices and profits whilst those of (11.51)-(11.53) capture only the partial
equilibrium eects holding other prices and profits constant.
The choice of optimal commodity taxes are the solution the to maximisation
" n
#
X
max L V (q0 , ..., qn , ) +
ti Xi R ,
(11.58)
{t1 ,...,tn }
i=1
subject to
qi = pi + ti , i = 1, ..., K,
(11.59)
qK+1 = i (q0 , ..., qK , tK+1 , ..., tn ) , i = K + 1, ..., n,
(11.60)
ji = j,i (q0 , ..., qK , tK+1 , ..., tn ) , i = K + 1, ..., n, j = 1, ..., mi .
(11.61)
and
From this system, dierentiation with respect to the tax rate of a typical
good, k, from the competitive sector provides the first-order condition
mi
n
n
X
V s V X X
V
+
k +
j,s
k
qk
qs
s=K+1
s=K+1 j=1
mi
n
n
n
n
n
X
X
X
X
X
X
X
X
X
i
i
i
j,s
s
+ Xk +
ti
+
ti
+
ti
= 0.
qk
qs k i=1
k
i=1
i=1
j=1
s=K+1
s=K+1
In (11.62) the subscripts on s and j,s denote the argument with respect to
which the function is dierentiated.
This first-order condition is distinguished from its equivalent for the competitive model, (4.7), by the inclusion of the eects of the induced price and profit
changes, given respectively by the terms sk , and j,s
k . Using Roys identity and
the Slutsky equation, (11.62) can be written
n
X
i=1
ti Ski

1 k
= Xk +
, k = 1, ..., K,
(11.62)
where
n
X Xi
ti
,
=1
i=1
(11.63)
356
and
k
n
X
s=K+1
n
X
Xs sk
ms
n
X
X
j,s
k +
s=K+1 j=1
n
n
X
X
i=1 s=K+1
mi
n
X
X
Xi j,s
ti
.
k
j=1
ti
Xi s
qs k
(11.64)
i=1 s=K+1

If the term 1 k were not included, (11.65) would be identical to the standard
Ramsey rule so this additional term can be interpreted as the modification
required to incorporate the imperfect competition. As can be seen from its
definition, k is determined by the induced price and profit eects which would
not be present in a competitive model. Retaining the standard interpretation
of the Ramsey rule given in Chapter 4, an analysis of the constituents of k
shows that the reduction in compensated demand for good k is smaller when a
tax on this good increases the prices of imperfectly competitive goods (sk > 0),
reduces profits (j,s
k < 0) and the induced tax changes lower tax revenue.
Repeating the derivation for a typical good k from the imperfectly competitive sector gives the first-order condition for the choice of tax rate
V k
+
qk k
n
X
s=K+1,s6=k
V s V
+
qs k
+ Xk +
n
n
X
X
i=1 s=K+1
mi
n
X
X
j,s
k
s=K+1 j=1
mi
n
n
Xi s X X X
Xi j,s
ti
+
ti
= 0.
qs k i=1
k
j=1
s=K+1
After employing Roys identity and the Slutsky equation, (11.66) becomes
n
X
i=1
ti Ski = k Xk +
where

1 k
, k = K + 1, ..., n,
(11.65)
k =
and
(11.66)
#
P s j,s
P
P
ns=K+1 Xs sk ns=K+1 m
k
j=1
Pn
Pmi Xi j,s .
Pn
Pn Pn
s
i
+ i=1,i6=k s=K+1 ti X
j=1 ti k
i=1
s=K+1
qs k +
(11.67)

In (11.67) the term 1 k again represents a correction to the competitive Ramsey rule for the existence of the induced price and profit eects. The interaction
of its constituent parts with the reduction in compensated demand can be broken down precisely as for (11.65). The distinction between (11.65) and (11.69)
is that for the latter k is dependent upon the good k under good consideration
1
= k
k
k
"
1
X Xi

ti
,
k
i=1
k
11.5. AD VALOREM AND SPECIFIC TAXES
357
due to the appearance of the tax-shifting term in (11.69). Since k is inversely

related to kk , the greater the degree of overshifting the smaller will be the reduction in compensated demand. This result is the natural generalisation of
that derived for the two-good example.
To summarise this discussion, for a good produced by a competitive industry
the standard Ramsey tax rule is adjusted by the incorporation of an additional
term that captures the induced eects of the tax on that good. For goods from
imperfectly competitive industries, the tax rule incorporated the correction term
for the induced eects and the reduction in compensated demand was inversely
related to the degree of tax shifting. The tax incidence results in (11.28), (11.34)
and (11.35) capture the fact that the direct and induced eects are dependent
upon industrial conduct via the value of the conjecture, . Consequently the
tax rules (11.63) and (11.67) show that industrial conduct is as important as
tastes in determining relative rates of taxation.
11.5
Ad valorem and specific taxes
In the competitive model there is no distinction between the eects of specific

taxes, which are an addition to the unit costs of production, and ad valorem
taxes, which represent a proportional reduction in the received price. That
is, for a given increase in consumer price, the two forms of taxation will raise
an identical level of revenue. Although discussions of commodity taxation are
generally phrased in terms of specific taxes, as was Chapter 4, a similar analysis
can be conducted for ad valorem taxes and, as shown by Hatta (1986), the
form of the optimal tax rules are not aected by the choice. When imperfect
competition is under consideration, the equivalence between the two forms of
taxation no longer applies. This distinction has been emphasised by Kay and
Keen (1983) and Delipalla and Keen (1992). Myles (1994b) shows the potential
gains that can be obtained by combining the two forms of taxation. The purpose
of this section is to review these results.
11.5.1
Tax incidence
Returning to the economy used to derive the tax incidence results above and
denoting the specific and ad valorem taxes levied on industry i by tsi and ti
respectively, the profit of firm j is

ji = xji qi [1 ti ] tsi xji Ci xji .
(11.68)
The first-order condition for profit maximisation at a symmetric equilibrium is

then
i
Ci
Xi tsi =
,
(11.69)
[1 ti ] qi +
Xi
xji
358
where =
mi .
Now defining A by
A=
2 Ci
xj2
i
i
[1 ti ] X
i
(11.70)
the procedure used in (11.26) to (11.28) can be repeated to give the eects of
the taxes upon price as
1
dqi
=
,
s
dti
[1 ti ] [1 + [1 + A E]]
(11.71)
where Eis as defined in (11.31), and

Ci
+ tsi
dqi
dqi
xji
= s, =
.
(11.72)
dti
dti
1 ti
Pmi j
i , the eects of the taxes are
With respect to aggregate profits, = j=1
and
Xi [2 + A E]
d
=
,
dtsi
[1 + [1 + A E]]
(11.73)
qi Xi [1 ]
d
d
= qi s d
dti
dti
q [1 + [1 + A E]]
(11.74)
with dq the price elasticity of demand.

These equations indicate that the incidence eects of specific and ad valorem
taxes are not identical. It can therefore be expected that they have dierent
welfare implications. The first step to understanding the welfare eects is to
consider starting from a position with both taxes zero and then varying the
taxes whilst holding revenue at zero. Such a reform satisfies the equation
0 = qi dti + dtsi .
(11.75)
The eect of the reform on the price level is determined by

dqi =
qi qi s
dt + s dti .
ti i
ti
(11.76)
Using (11.77), this reduces to

dqi = [ qi ]
qi s
dt .
tsi i
(11.77)
i
Since (11.71) implies < qi and q
tsi > 0, it can be seen from (11.79) that
consumer price will be lower if the reform is chosen to satisfy dtsi > 0. Therefore
a positive ad valorem tax and a negative specific tax can generate budget-neutral
reductions in consumer price.
11.6. OPTIMAL COMBINATIONS
359
Strengthening this finding, Delipalla and Keen (1992) prove that if the specific tax is restricted to be non-negative, then the solution of the problem
max V (qi ) + subject to [ti qi + tsi ] Xi = R, tsi 0,
{tsi ,ti }
(11.78)
will have zero as the optimal value of tsi . The reason for this result is described
in Venables (1986): raising the ad valorem tax reduces the perceived eect of
changes in output upon prices and leads, in the limit, to the competitive state
in which a change in output has no eect on price. Therefore, the ad valorem
tax is lessening the consequence of the imperfect competition.
11.6
Optimal combinations
Relaxing the restriction that the specific tax be non-negative, the optimal combination of ad valorem and specific taxation can eliminate the welfare loss due
to imperfect competition. In particular, the use of the pair of tax instruments
can lead to the set of prices that maximise welfare given the budget constraint.
Such prices are often termed Ramsey prices and arise as the solution to both the
problem of choosing optimal public sector prices subject to budget constraint
and from analysis of the design of commodity taxation; the latter having already
been discussed in Chapter 4. Bs (1986) clearly elucidates the major principles
of Ramsey pricing and provides a number of alternative characterisations of the
relevant optimality conditions.
To show that Ramsey prices can be generated despite the imperfect competition, consider an economy that has one consumer, a government and an
industry that produces the single consumption good using labour as the only
input. The government levies taxes and uses the revenue to purchase labour.
The labour required by the firms and government is supplied by the consumer
through a competitive market with the wage rate as numeraire. All profit income is paid as a dividend to the consumer. The single industry is oligopolistic
with m firms indexed by j = 1, ..., m. The inverse demand function facing the
industry is given by p = p(X), with first derivative pX < 0. Each firm has
the cost function C = C(xj ) where F C (0) 0 and Cx > 0. The common
dX
conjecture of the firms on the value of dx
is denoted by , with 0 < m. If
i
government receives all profits or production is under state control, the Ramsey
price,p , is the solution to
max V (p, I) subject to R +
{p}
m
X
C(xj ) = pX (p) , I = 0,
(11.79)
j=1
so that it maximises welfare given the budget constraint and the restriction that
lump-sum income be zero. In the single sector economy, the solution to (11.81)
will also satisfy the budget constraint
p X (p ) = R + mC m1 X (p ) ,
(11.80)
360
where the assumption of symmetry between firms has been imposed. The Ramsey price p is optimal given that the existing number of firms is maintained;
changes in the number of firms are not considered.
The first-order condition for profit maximisation of a typical oligopolistic
firm is
(11.81)
[1 t ] p Cx ts + [1 t ] pX X = 0,
. Since the symmetry implies that xj = x = X

where = m
m , all j = 1, ..., m,
(11.83) can be solved to express the equilibrium level of output as a function of
the tax rates. This solution is denoted
X = (t , ts ) .
(11.82)
Substituting (11.84) into the inverse demand function then determines the equilibrium price conditional on the taxes. Hence
p = p ( (t , ts )) = (t , ts ) .
(11.83)
Finally, using (11.83) and the profit identity determines the equilibrium profit
level as
j = = (t , ts ) , all j = 1, ..., m.
(11.84)
Employing (11.84) - (11.86), a combination of ad valorem and specific taxation can generate Ramsey pricing if it is possible to find a pair such that the
following three conditions are satisfied simultaneously
(i) (t , ts ) = p , (ii) (t , ts ) = 0, (iii) [t (t , ts ) + ts ] (t , ts ) = R.
h (11.85)
i
s X
If (11.87i) and (11.87iii) are satisfied, then = [[1 t ] p t ] m C = R m
h
i
1
C, using (11.87iii), R m
[ R mC] = 0 using (11.87i) and (11.81).
Therefore the satisfaction of (11.87i) and (11.87iii) implies (11.87ii) and the optimal tax scheme need solve only two of the three equations. In the analysis
below, the focus is placed upon (11.87i) and (11.87iii).
To give the problem content, it is assumed that the revenue requirement is
suciently small that it can be achieved by use of either of the tax instruments
alone. This ensures that the use of both instruments occurs through choice.
Defining 1 = {t , ts : (t , ts ) = p } and 2 = {t , ts : [t (t , ts ) + ts ] (t , ts ) = R},
the first lemma relates the relative positions of these curves to the profit levels
of the firms.
s
b
52 If,
t , t and e
ts are defined
by
Lemma
fors a given value of t , denoted
s
Cx +ts
s
b
t , t 1 , b
t ,e
t 2 , and for all pairs {t , t } that solve (11.87i), t 1t
<
s
s
s
s
s
s
e
b
e
e
b
e
[1 ] pX X, then t < t if t , t > 0, and t > t if t , t < 0.
s
s
s s
Proof. Using (11.81), define G b

t ,t p b
t ,t .
t ,t
b
t , t RmC m1 b
s
G
The derivative of G b
t , t is given by t
=
[p
C
+
Xp
]
.
Given
that
s
x
X ts
11.6. OPTIMAL COMBINATIONS
361
G
s < 0, ts
t
t Cx +ts
+ [1
1t
G
Since t
s
is positive if p Cx + XpX < 0. From (11.82) p Cx + XpX =

] pX X which is negative by assumption.
s
s
t , t evaluated at b
t ,e
> 0 it is clear that if G b
t is positive then
s
s
s
t , t = 0 is reached by reducing t and hence t < e
ts . The
the solution to G b
s
s
t , t is negative at b
t ,e
t p+e
converse holds if G b
t . Now note that if b
ts X =
s
R, then G = pX R
mC
= pX mC X [t p + ts ] = [[1 t ] pt ]sX
mC
es
b
e
b
b
= m. Therefore G t , t > 0 if m = m t , t > 0 and G t , e
t < 0 if
s
t < 0.
m = m b
t ,e
Cx +ts
The assumption that t 1t
< [1 ] pX X is clearly very weak. The
right-hand side is always positive but, since only values of t satisfying t 1

need be considered and ts < Cx for most points in 1 , the left-hand side is
generally negative.
The next result determines the change in profit on the 1 curve as t increases.
Lemma 53 Profit is monotonically decreasing along the 1 curve as t increases.
Proof. The lemma is directly proved by total dierentiation along and 1 .
See Myles (1995a).
The implication of Lemma 11.1 is that the 1 curve lies outside the 2
curve whenever the firms earn positive profits on the 2 curve. Hence if there
are points on the 2 curve such that the firms earn negative profits then the
continuity implies that the two curves must cross at some point prior to this.
From this observation follows Theorem 11.1.
Theorem 54 If there exist tn , tsn such that tsn < 1, [tn (tn , tsn )] (tn , tsn ) = R
and (tn , tsn ) < 0, then there exists a unique pair (t , ts ) with t < 1 that
generates Ramsey pricing.
Proof. The statements in the proposition imply, by Lemma 11.1, that 1
must lie to the right of 2 at tn , tsn . Therefore,since assumption 1 implies 0
when evaluated at b
ts , where b
ts is defined by b
ts , 0 2 , there must be some
point at which 1 and 2 cross and at this point t < 1. The crossing point
therefore determines the taxes that generate Ramsey pricing. The crossing point
is also unique since if the curves intersected more than once Lemma 11.2 would
be violated.
Theorem 11.1 has demonstrated that Ramsey pricing can be generated when
the firms become unprofitable at some combinations of taxes that satisfy the
revenue requirement. This provides the first possible form of optimal tax policy.
The economic reasoning lying behind this result is that the use of an ad valorem
tax reduces the perceived market power of the imperfectly competitive firms by
lowering marginal revenue whilst the specific tax can be targeted as a subsidy
towards covering fixed costs.
For the case of constant marginal cost, the positivity of F is a necessary
condition that profit become negative at some point on the 2 curve. This can
362
be seen by solving (11.87i) to (11.87iii). Doing this shows that the levels of t
and ts are characterised implicitly by
"
#
C m1 X
Fm
F
s
[1 t ] =
, t = m
,
(11.86)
+
pX X 2
X
pX X 2
where X and pX are evaluated at the Ramsey price and quantity. From (11.88),
t can only be less than 1 when F is positive.
The optimal policy when there is no pair of tax rates on the 2 curve that
lead to negative profits can be found by applying the following lemma.
Lemma 55 Tax revenue is monotonically increasing along the 1 curve as t
increases.
Proof. See Myles (1995a).
Theorem 56 If there does not exist tn , tsn such that tsn < 1, [tn (tn , tsn )] (tn , tsn ) =
R and (tn , tsn ) < 0, then the optimal policy is to let t 1 with ts determined
by the 2 curve. If lim = 0 as t 1 along 2 then Ramsey pricing is
generated in the limit.
Proof. Lemma 11.1 shows tax revenue is less on 1 than 2 for given t .
From Lemma 11.3, the two curves must then converge as t increases. Hence
the optimal policy is always to let t tend to 1 to move as close to 1 as possible
since this takes the price closer to the Ramsey price. The second part of the
theorem follows from Lemma 11.1.
It is therefore possible for the combination of specific and ad valorem taxation to support Ramsey pricing in imperfectly competitive economies and such
an optimal tax system has the eect of eliminating the welfare loss due to
monopoly power. The use of both tax instruments is able to achieve rather
more than the use of a single instrument.
11.7
Production eciency and the taxation of

labour
The important Diamond-Mirrlees production eciency lemma was discussed in

Chapter 4. Since this implied that all producers should face the same shadow
prices, there was no need for taxes upon intermediate goods in an optimal tax
structure. The result was based on several assumptions, amongst them being
those of competition and of constant returns to scale. Although the latter can
be relaxed without harming the essence of the conclusion, the former cannot.
This purpose of this section is investigate why production eciency fails when
there is imperfect competition and to consider the consequences for the taxation
of intermediate goods. It was also assumed in Chapter 4 that labour could act
as an untaxed numeraire. Although valid in a competitive economy, the nontaxation of labour is an additional restriction with imperfect competition. The
consequences of relaxing this are also explored.
11.7. PRODUCTION EFFICIENCY AND THE TAXATION OF LABOUR363
11.7.1
Production eciency
If taxes upon intermediate goods are introduced these will aect the vector of
input prices faced by producers. With competition, a change in input costs is
reflected in final prices only to the extent to which it aects marginal cost. In
contrast, this simple relationship is lost when imperfect competition is introduced. A change in input costs operates through the cost function and then
through the optimisation of the firm. As with commodity taxation, a change
in production costs may be overshifted or undershifted in final price. By manipulating input prices through taxation, tax design can exploit these shifting
eects, and the profit eects, in order to gain welfare increases.
These issues have been analysed by Myles (1989b) and Konishi (1990). Myles
(1989b) considers economies with fixed numbers of firms whereas Konishi (1990)
analyses free-entry Cournot oligopoly in which a number of competitive industries produced goods that were used as inputs by the oligopoly and as final
consumption goods. Both focus only upon the eciency aspects of taxation
and find that, in general, welfare can be raised by the taxation of intermediate goods. The relative rates of tax between dierent industries are dependent
upon returns to scale. Konishi shows that taxation should encourage the use of
intermediate goods that are demanded elastically and reduce the use of those
with inelastic demand. This has been termed the production side Ramsey rule.
To demonstrate the nature of these results, consider an economy in which a
competitive industrys entire output is sold as an intermediate good to a monopolist whose output constitutes the economys final good. Attention is focused
upon eciency arguments by assuming the existence of a single consumer who
consumes the final goods, receives the monopolies profits and supplies labour.
Each firm in the competitive industry is assumed to have a fixed coecient
production function and units are normalised so that each unit of output requires one unit of labour. Writing p0 for the wage rate, the post-tax price of
the competitive industrys good, which is labelled 1, is
q1 = p0 + 1 ,
(11.87)
where 1 is the tax levied upon the intermediate good. Directly from (11.89)
q1
= 1.
1
(11.88)
The monopoly produces with costs given by

C (q1 , p0 , X2 ) + X2 t2 ,
(11.89)
where X2 is the firms output level and t2 the level of commodity tax. The
monopolist chooses its price, q2 , to maximise profit, , where
= [q2 t2 ] X2 (q2 , p0 , ) C (q1 , p0 , X2 ) ,
(11.90)
X2 () being the demand function. The presence of on both sides of (11.92)

captures the income eects that occur in a general equilibrium model. Assuming
364
the condition in (11.7) is satisfied, the profit maximising choice is characterised

by
C X2
X2 + q2 t2
= 0,
(11.91)
X2 q2
and the second-order condition is
2
C 2 X2
X2 2 C
X2
+ q2 t2
< 0.
2
q2
X2 q22
q2
X22
(11.92)
In addition, the equilibrium must also satisfy the profit identity (11.92).
Treating (11.92) and (11.93) as a two-equation system, they can be dierentiated and solved to give the tax incidence terms
C
=
1
1
=
t2
1
q2
1
X2
X
h 2
q2 t2
C
X2
i,
(11.93)
C
X2
i,
(11.94)
X2 2 C
q2 q1 X2
h
2
2 X
q2 + q2 t2
h
+h
1
X2
h q1
q2 t2
X2
h
i2
2C
2
X
q2
X22
h
i
i
2 X2
X2 X2 2 C
C X2
C
+
2
2
2
q1
q2
X2
q2 X2
,
h
h
h
ii
i 2
i2(11.95)
X2
X2
X2
C
C
2C
q2 t2 X2
2 q2 + q2 t2 X2 q2 q2
X 2
C
X2
2 X2
q22
and
q2
t2
X2
q2
i 2
i2
h
h
X2
X2
C
2C
2
2 X
+
q
2
2
2
q2
X2
q2
q2
X22
h
h
i
i
2 X2
X2 X2 2 C
C
2
X2 X
+ q2 q2 t2 X2 q2 X22
.
+h
h
h
h
ii
i 2
i2(11.96)
X2
X2
X2
2C
C
C
2
+
q
1 X
q
2
2
2
2
2
X2
q2
X2
q2
q 2
X 2
2
The first term of (11.98) is the direct eect of the change in tax upon price
2 X2
> 0, the second
and, as shown above, it may be greater than 1. Assuming q
2
term will certainly be negative if marginal cost decreases with output. This
negativity captures the fact that the tax lowers profits, which then leads to a
reduction in demand. The same interpretation applies to (11.97).
To demonstrate that the intermediate good should be taxed, it is worth first
considering the tax reform problem of finding the direction of change in taxation, from an initial position with both taxes zero, that raise welfare whilst
11.7. PRODUCTION EFFICIENCY AND THE TAXATION OF LABOUR365

maintaining a balanced budget. Formally stated, starting from an initial position with 1 = t2 = 0, the sign of d 1 , dt2 such that dV > 0 and dR = 0 will
be determined where V = V (p0 , q2 , ) is the single consumers indirect utility
function, which acts as the measure of social welfare, and R = 1 X1 + t2 X2 .
To characterise the solution, first dierentiate the utility function and the
revenue constraint to give
V
V q2
V
V q2
+
+
(11.97)
d 1 +
dt2 ,
dV =
q2 1
1
q2 t2
t2
and
dR = X1 d 1 + X2 dt2 = 0.
From Shephards lemma, X1 =
C
q1
(11.98)
so (11.100) gives
dt2 =
C
q1
X2
d 1 .
(11.99)
Substituting (11.101) and (11.95) - (11.98) into (11.99)
dV =
V
q2
X2 2 C
q2 q1 X2
h
2
2 X
q2 + q2 t2
C
X2
C X2
q1 q2
X2
2 X2
q22
X2
q2
i2
2C
X22
d 1 .
(11.100)
Using (11.102), the direction of reform can be summarised in Theorem 11.3.

2
Theorem 57 d 1 6= 0 if X2 q1 X
2
2
C
X2 q1 X
2
C
q1
< 0 and d 1 < 0 if
C
q1 6= 0.
2
C
C
X2 q1 X
q
2
1
In particular, d 1 > 0 if
> 0.
V X2
, q2 and the denominator of the bracketed term in (11.102)
Proof. Since q
2
are negative, the resultnfollows from inspection.
o
n
o
2
C
C
C
2C
Assuming that sgn. X2 q1 X
=
sgn.
X
2
q1
p0 X2
p0 , so the pro2
duction function is well-behaved, Theorem 11.3 can be stated in terms of returns
to scale.
Theorem 58 If the monopolist produces with constant returns to scale, d 1 = 0.

Increasing returns imply d 1 > 0 and decreasing returns that d 1 < 0.
Proof. If the monopolist produces with constant returns to scale, C (q2 , p0 , X2 ) =
2
C
C
q
= 0 and hence d 1 = 0. The rest follows
c (q2 , p0 ) X2 . From this, X2 q1 X
2
1
2
C
from noting that increasing returns imply X2 q1 X
2
2
C
q1
< 0 and decreasing
C
C
returns that X2 q1 X
q
> 0.
2
1
Theorem 11.3 demonstrates clearly that with imperfect competition there
will typically be a case for the taxation of intermediate goods providing the
monopolist does not produce with constant returns to scale. The basis for these
taxes are the dierent rates at which the intermediate tax and the final tax
366
aect the consumer price. In terms of Theorem 11.4, when the monopolist
has decreasing returns the input should be subsidised in an attempt to induce
a larger level of output by cost reductions. Although the precise structure of
these statements is due to the simple model used, the same features would also
be of importance in any larger model.
As discussed in Chapter 4, the analysis of perfect competition with decreasing returns has demonstrated that production eciency is desirable when profits
are correctly taxed. This can be investigated for imperfect competition by introducing a profits tax into the analysis above. Writing (< 1) for the rate of
profit taxation, the demand function now becomes
X2 = X2 (q2 , p0 , [1 ] ) .
(11.101)
Including the revenue from the profits tax, the new revenue constraint is
+ 1 X1 + t2 X2 = 0,
and, for a given value of , the dierential of this is
+ X2 dt2 +
+ X1 d 1 = 0.
t2
1
(11.102)
(11.103)
Repeating the tax incidence analysis, dierentiating the utility function and
substituting from (11.105) again generates (11.102). Consequently, the characterisation of policy given in Theorem 11.3 is still valid independently of the
value of the profit tax. This is in contrast to the Dasgupta and Stiglitz (1972)
result for competitive economies.
These results demonstrate clearly that non-taxation of intermediate goods
is not theoretically justified in economies with imperfect competition. Although
they do not match the generality of the competitive economy of Diamond and
Mirrlees, they are suciently rich in possibilities to highlight the major factors.
Furthermore, the conclusion that intermediate goods should be taxed is merely
strengthened by increasing the complexity of the economies.
11.7.2
Taxation of labour
Turning now to the taxation of labour, it should first be noted that in analyses
of commodity taxation labour is treated as an untaxed numeraire in economies
with constant returns and as taxed at a uniform rate when there are decreasing
returns. The arguments of Munk (1980) justify these choices of normalisation
in competitive economies and their optimality is assured via the Production
Eciency lemma. With imperfect competition however, treating labour as an
untaxed numeraire places an additional restriction upon the tax system.
The taxation of labour in the presence of imperfect competition is addressed
in Myles (1995). As in the case of the taxation of intermediate goods, it is the
nonlinear relation of final price to labour cost that is the key factor. Providing
the response of final price to the wage rate diers across industries, then it will
11.8. OTHER FORMS OF REGULATION
367
be justifiable on welfare grounds to use a labour tax that is dierentiated across

industries. By subsidising labour in industries where price is very responsive
to labour cost and taxing those were it is not, a welfare-improving reduction in
the price level can be achieved. The rate of response of price to tax is again
determined by demand conditions, summarised by Seades E, and returns to
scale in the production process. The determinants of relative tax rates are the
returns to scale of each industry and industrial conduct within the industry.
Optimal taxes relate the reduction in supply from each industry due to the tax
system to the returns to scale of the industry, the eect of the tax upon price
and the interaction of the industry with the economy.
11.8
Other forms of regulation
The focus of this chapter has been upon the eects of commodity taxation in
imperfectly economies and the optimisation of such taxes. In practice, a number
of forms of regulation have been employed to reduce the welfare loss arising
from the existence of imperfect competition. The most important distinction
between these is whether the firm is taken into public ownership or remains in
the private sector. The forms of regulation used in these cases may dier but
the aim remains the same: to encourage the firm to act in a socially ecient
manner.
When faced with imperfect competition, the most obvious policy response
would be the encouragement of competition. This could be done by removing
any barriers to entry and by providing initial inducements for competitors to
become active. The limits to the operation of this policy are confronted when
natural monopoly arises. The essence of natural monopoly is that the production
technology has the property that output can always be produced more cheaply
by one firm than it can by two or more. This property, termed sub-additivity
by Baumol et al. (1977), is a necessary condition for natural monopoly to arise.
Stated formally, a sub-additive cost function satisfies
C (y) <
k
X
j=1
C (yk ) , y =
k
X
j=1
yk , k 2.
(11.104)
With natural monopoly, the market will not sustain more than one firm nor can
it be socially optimal to have more than one firm otherwise costs will not be
minimised. It is therefore necessary to devise some policy other than introducing
competition.
The two policy options that have been most widely employed are public
ownership and private ownership with a regulatory body controlling behaviour.
The most significant dierence between these is that under public control the
government is as informed as the firm about demand and cost conditions. It
can therefore determine the behaviour of the firm using the best available information. Although this information may not be complete, so policy can only
maximise the objective function in an expected sense, the best that is possible
368
will be achieved. In contrast, when the firm is in private ownership, the government, via the regulatory body, may well be far less informed about information
pertinent to the operating conditions of the firm than the firm itself. Information about cost structures and market conditions are likely to remain private
and the firm may have strategic reasons for not revealing this accurately.
When the firm is run under public ownership, the level of price should be
chosen to maximise social welfare subject to the budget constraint placed upon
the firm. The firm may be required to break-even or to generate income above
its cost. Alternatively, it may be allowed to run a deficit which is financed from
other tax revenues. In any case, as already noted in Chapter 4, the structure
of the decision problem for the government mirrors that in determining optimal
commodity taxes and will generate the Ramsey price for the firm.
Assume all other markets in the economy are competitive. The Ramsey
price for public firm subject to a break-even constraint will then be equal to
marginal cost if this satisfies the constraint. If losses arise at marginal cost,
then the Ramsey price will be equal to average cost. The literature on public
sector pricing has extended this reasoning to situations in which marginal cost
and demand vary over time such as in the supply of electricity. Doing this leads
into the theory of peak-load pricing (see Brown and Sibley (1986) and Dreze
(1964)). When other markets are not competitive, the Ramsey prices will reflect
the distortions elsewhere (see Bs (1986)). The principles involved are the same
in all cases.
As an alternative to public ownership, a firm may remain under private ownership but be made subject to the control of a regulatory body. This introduces
possible asymmetries in information between the firm and the regulator. Faced
with limited information, one approach considered in the theoretical literature
is for the regulator to design an incentive mechanism that achieves a desirable
outcome. An example of such a regulatory scheme is the two-part tari studied
in Baron (1989) and Lockwood (1995) in which the payment for the commodity involves a fixed fee to permit consumption followed by a price per unit for
consumption, with these values being set by the regulator. Alternatively, the
regulator may impose a constraint on some observable measure of the firms activities. Adopting the framework characteristic of regulation in the US, Averch
and Johnson (1962) study the optimisation of the firm under the constraint that
it must not exceed a given rate of return upon the capital employed. Even more
simple are the regulatory schemes in the UK which involve restricting prices to
rise at a slower rate than an index of the general price level.
11.9
Conclusions
This chapter has shown how imperfect competition leads to a failure to attain
Pareto optimality. As with all such failures, this opens a potential role for
government intervention to promote eciency. Estimates of the welfare loss
due to imperfect competition have been constructed from both observed data
and from numerical simulations. These vary widely from the almost insignificant
11.9. CONCLUSIONS
369
to considerable proportions of attained welfare. On balance, it is likely, however,

that imperfect competition is of significance.
There are numerous ways in which imperfect competition can be represented.
The subjective and objective demand approaches were discussed and a number of unresolved issues were highlighted. An economy incorporating quantitysetting oligopoly using objective demand, and conjectural variations when interesting, was then adopted and it was shown how this could be employed to
determine the structure of optimal commodity taxation. The tax rules derived
were direct generalisations of those for the competitive economy and their implementation would require information additional to the competitive rules.
In contrast to the competitive case, specific and ad valorem taxation are not
equivalent with imperfect competition. In a choice between the instruments, ad
valorem taxation is more eective since it has the eect of reducing perceived
monopoly power. Combining the instruments can lead to further gains and, if
correctly chosen, the welfare loss due to the monopoly power can be eliminated
entirely Finally, imperfect competition was also shown to invalidate the general
argument for production eciency, so that taxes on intermediate goods could
be justified.
Chapter 12
Tax Evasion
12.1
Introduction
An implicit assumption that supported the analysis of taxation in the previous

chapters was that firms and consumers honestly report their taxable activities.
Although acceptable for providing simplified insights into the underlying issues,
this assumption is patently unacceptable when confronted with reality. The
purpose of this chapter can therefore be seen as the introduction of practical
constraints upon the free choice of tax policy. Tax evasion, the intentional
failure to declare taxable economic activity, is pervasive in many economies as
the evidence given in the following section makes clear and is therefore a subject
of practical as well as theoretical interest.
After reviewing evidence on the extent of tax evasion, the chapter considers
the tax evasion decision of consumers. This decision is represented as a choice
under uncertainty and naturally employs the techniques of Chapter 7. Within
this framework, the optimal degree of auditing and of punishment is considered.
This is then extended to include decisions over labour supply, since the choice of
occupation can determine opportunities for evasion, and the role of public goods
and social norms. The analysis predicts the relationship between the level of
evasion, tax rates and punishments. The results of experiments that investigate
these are discussed. A more developed analysis of the optimal choice of audit
is then given. The analysis of tax evasion is then completed by consideration of
evasion by firms.
12.2
The extent of tax evasion
The importance of developing a theoretical understanding of tax evasion can

only be assessed by estimating the actual extent of evasion. If such evasion
constitutes a significant activity within the economy, then a theory of evasion is
of potential use in designing structures that minimise evasion at least cost and
ensuring that policies are optimal given that evasion occurs.
371
372
CHAPTER 12. TAX EVASION
Due to its very nature, the measurement of tax evasion and unreported
economic activity is fraught with diculty and uncertainty. Tax evasion should
be distinguished from tax avoidance, which is the reorganisation of economic
activity, possibly at some cost, to lower tax payment. Tax avoidance is legal,
tax evasion is not. This illegality makes surveys prone to error if the fear of
prosecution remains and, by definition, tax evasion is not measured in ocial
statistics. The estimates reported below therefore rely on a number of methods
of inference which naturally leaves them open to error. They should be regarded
primarily as rough approximations. In addition to measurement errors, there is
also the issue of what should be included. Illegal activities, such as the supply
of drugs or smuggling generally, would not be included in measured GDP even
if they were known. It is open to debate as to whether they should be included
in measures of the hidden economy.
One of the earliest published studies of tax evasion is the analysis by Rey
(1965) of the Italian General Sales Tax. This tax is levied on all exchanges
of goods and services, with some exceptions, and in 1961 raised revenue equal
to approximately 4% of GNP. The tax had several methods of collection and
of the largest of these, which raised two thirds of revenue, Rey estimated that
evasion was equal to 52.46% of actual yield. This is clearly a significant degree
of evasion.
In an article that proved the starting point for many studies, Gutmann
(1977) attempted to measure the extent of unobserved economic activity, or the
hidden economy, in the USA. Based on the observation that transactions in the
hidden economy are invariably financed by cash rather than cheque or credit,
Gutmann used the growth of currency in circulation relative to demand deposits
as an indirect measure of unobserved activity. This procedure resulted in an
estimate of $176 bn. for illegal GNP in 1976 which was approximately 10% of
legal, measured activity. This figure is in accordance with that reported by the
Internal Revenue Service in 1979 which estimated unreported income in 1976
to be between $75 bn. and $100 bn. or 7% to 9% of reported income.
Feige (1979) attempted to measure the same activity as Gutmann but employing a dierent methodology. The method of Feige was to work from the
observation that total economic activity, including both measured and unmeasured sectors, is equal to the price level times transactions. An estimate of the
unmeasured sector is then provided by the ratio of the value of measured income
to that of transactions. The major diculty with this approach is determining
the number of transactions that actually occur. Feige achieved this by using
data on the life-span in months of bank notes in circulation relative to the number of times it is expected each note can be used. This analysis provides an
estimate of the unmeasured sector in 1976 of $369 bn., which is 22% of GNP,
and $704 bn., 33%, in 1978. Given the size of these estimates, Feige concludes
that ocial statistics must be very misleading.
Applying the same methodology to the UK, Feige (1981) calculates that the
unobserved sector produced a GNP of 28 bn. in 1979 which was equal to 15%
of measured GNP. This estimate must be set against the reported comments of
an ex-Chairman of the Inland Revenue that undeclared income could amount to
12.3. EVASION AS A DECISION WITH RISK
373
7.5% of GNP. An alternative approach to measurement of the UK unobserved

sector has been taken by Pissarides and Weber (1989) who employ data from the
1982 Family Expenditure Survey. By assuming that income and expenditure
are reported accurately by employees whose employer filed their income report,
an estimate of the expenditure function for these households then gives the true
relation between income and expenditure. Observing the expenditure of other
households then permits an estimate to be made of their income and, consequently, their tax evasion. The final estimate is that the unobserved economy
is 5.5% of GDP. Finally, Feige and McGee (1983) report an estimate of 10%
for unobserved activity in Sweden and Smith (1981) presents a useful summary
table of results.
Even when the possible degrees of error are taken into account, the impression that these estimates give is that undeclared economic activity, and hence
tax evasion, is a significant part of total economic activity in many western
economies. Although the methods employed are imperfect, they cannot be dismissed entirely. Such an observation clearly justifies further study of the causes
and consequences of tax evasion.
To close this section, it is worth noting that implicit in many discussions
is the assumption that tax evasion reduces tax revenue. However, as shown
by Peacock and Shaw (1982), if unreported activity has a multiplier eect and
would take place at a lower level if it were subject to taxation, then estimates
of revenue loss will be overstated, even to the extent that evasion may lead to
no revenue loss at all. This eect is enhanced by the possibility that evasion
may encourage participation in taxed activities. As the analysis of Peacock and
Shaw is based on a simple Keynesian model, this is a point that could bear
further investigation.
12.3
Evasion as a decision with risk
The decision to evade taxation fits naturally into the framework of choice under
risk. Since not all tax evaders are caught by the tax authorities, risk arises since
an individual who evades tax stands a chance of succeeding with the evasion, and
hence having increased wealth, or a chance of being caught and punished. As an
initial approximation, the individual can be viewed as choosing the extent of tax
to evade, subject to the probability of being caught and punished, to maximise
expected utility. The earliest formal analyses of this decision were given by
Allingham and Sandmo (1972), Srinivasan (1973) and Yitzhaki (1974). These
dier only in the structure of the punishments and that Srinivasan imposes
risk neutrality by assuming the individuals objection is the maximisation of
expected income. The derivation given below will be based primarily upon
Yitzhaki. A diagrammatic presentation can be found in Cowell (1985a).
The tax-payer receives an exogeneous income M which is known to the taxpayer but not to the tax collector. The analysis is simplified by assuming that
declared income, X, is taxed at a constant rate t. If the taxpayer is caught
evading, which occurs with probability p, a fine F > 1 is placed upon evaded
374
tax. The taxpayers aim is to choose X so as to maximise a von NeumannMorgernstern utility function. This decision problem can be written as
max E [U (X)] = [1 p] U (M tX) + pU (M tX F t [M X]) ,
{X}
(12.1)
where E is the expectation operator and F t [M X] is the total fine paid when
caught evading. Defining Y = M tX and Z = M tX F t [M X], the
first- and second-order conditions for maximising (12.1) are
and
p [F 1] U 0 (Z) [1 p] U 0 (Y ) = 0,
(12.2)
h
i
2
S t [1 p] U 00 (Y ) + [F 1] pU 00 (Z) 0.
(12.3)
dX
[F 1] U 0 (Z) + U 0 (Y )
=
> 0,
dp
S
(12.4)
It is assumed that U 0 > 0 and U 00 < 0.

For tax evasion to take place, the solution to (12.2) must have X < M . When
]
(12.3) holds for all X, this will occur when the derivative E[U
X is negative when
evaluated at X = M implying that X must be reduced to arrive at the optimum.
Calculating the derivative at X = M , tax evasion will occur whenever pF < 1.
It is assumed from this point that this condition is satisfied.
The aim of the analysis is to determine how the level of tax evasion is aected
by changes in the models variables. There are four such variables that are of
interest: the income level M , the tax rate t, the probability of detection p and
the fine rate F . The computation of these eects is a straightforward exercise in
comparative statics using (12.2) and (12.3). Totally dierentiating (12.2) with
respect to X and p and rearranging gives
so that an increase in the probability of detection raises the level of income

declared and reduces evasion. This is a clearly expected result since an increase
in the likelihood of detection lowers the payo from evading and makes evasion
a less attractive proposition. Repeating the procedure for a change in the fine
rate upon evaded tax provides the expression
dX
pU 0 (Z) p [F 1] U 00 (Z) t [M X]
=
> 0.
dF
S
(12.5)
As expected, an increase in the fine leads to a reduction in the level of tax

evasion. There is therefore no ambiguity about the eects of the two punishment
variables upon the level of evasion.
The next variable to consider is the tax rate upon declared income. Dierentiating (12.2) with respect to X and t gives
[S] dX+[[1 p] U 00 (Y ) X p [F 1] U 00 (Z) X p [F 1] U 00 (Z) [M X]] dt = 0.
(12.6)
12.3. EVASION AS A DECISION WITH RISK
375
The first-order condition (12.2) implies that p [F 1] U 0 (Z) = [1 p] U 0 (Y )

and this can be used to write the second-bracketed term in (12.6) as
[[1 p] U 00 (Y ) X p [F 1] U 00 (Z) X p [F 1] U 00 (Z) [M X]] =
[1 p] U 00 (Y ) X
p [F 1] U 00 (Z) X
p [F 1] U 00 (Z) [M X]
0
[1 p] U (Y )
.
[1 p] U 0 (Y )
p [F 1] U 0 (Z)
p [F 1] U 0 (Z)
(12.7)
00
(I)
Using the Arrow-Pratt measure of absolute risk aversion RA (I) = UU 0 (I)
to
simplify (12.7), the eect of the tax rate upon tax evasion is given by
dX
[1 p] U 0 (Y ) [X [RA (Z) RA (Y )] + F [M X] RA (Z)]
=
.
dt
S
(12.8)
Since RA is positive, for an increase in the tax rate to increase the level of income
declared it is sucient that RA (Z) RA (Y ) > 0. Thus if absolute risk aversion decreases as income increases, higher tax rates will lead to greater income
declarations and a reduction in evasion. This result has provoked considerable
discussion since it runs counter to the intuitive expectation that an increase in
tax rates should provide a greater incentive to evade. Such a relationship is
supported by the observation that estimates of the unocial economy suggest
that it has grown in periods when the average tax burden has increased.
Dierentiating (12.2) with respect to X and M and repeating the substitution used in (12.7) determines the eect of an increase in income upon evasion
as
dX
F tRA (Z) [RA (Z) RA (Y )]
=
.
(12.9)
dM
F tRA (Z) t [RA (Z) RA (Y )]
dX
< 1
The condition RA (Z) RA (Y ) > 0 then implies, since t < 1, that dM
and hence declared income rises less fast than actual income implying that tax
evasion rises with income.
In assessing the latter two results it should be stressed that the restriction
that absolute risk aversion decreases with income cannot be regarded as universally acceptable so there remains some degree of uncertainty as to the validity
of the presumption that higher tax rates and higher income lead to greater tax
evasion. In addition, these results are also sensitive to the precise form of the
punishment for evasion. If the fine is determined as in Allingham and Sandmo
(1972) by f [M X], f > t, rather than F t[M X], then neither the eect of
the tax rate or of income can be unambiguously signed even with decreasing
absolute risk aversion.
Before proceeding to a discussion of the choice problem of the tax authorities, it is worth noting two directions in which the analysis has been extended.
Cross and Shaw (1982) have conducted a joint determination of evasion and
avoidance activities, where avoidance refers to legal but costly means of altering activity in order to reduce total tax payments. They highlight the possible
complementarities between the two activities so that an increase in the level of
one of the activities leads to a reduction in the cost of the other. For instance,
376
if tax can be avoided by claiming expenses, there is little additional cost involved in inflating an expenses claim and therefore evading taxation. A second
direction taken has been to embed the evasion decision within a more general
choice problem by introducing additional assets with risky returns. This then
makes tax evasion only one of a portfolio of activities that have risky returns.
As shown by Landskroner, Paroush and Swary (1990) even in this generalised
setting the comparative statics results remain much as given above with decreasing absolute risk aversion still being the critical factor. A further extension
of the analysis, the choice of labour supply, will be discussed in more detail in
Section 5.
12.4
Optimal auditing and punishment
In developing the comparative statics properties of the tax evasion decision,

it has been assumed that the probability of detection, or of auditing, and the
fine levied when detected are constant. This is correct from the viewpoint of
the tax evader but from the perspective of the tax collector they are variables
which can be chosen to attain specified objectives. The present section will consider the choice of these variables within the simplest framework; an alternative
perspective is given in Section 9.
From (12.4) and (12.5) it can be seen that both an increase in the probability
of detection and of the fine will reduce evasion. Therefore, as noted by Allingham
and Sandmo (1972), the two instruments are substitutes with respect to reducing
evasion since a reduction in one can be compensated for by an increase in the
other. For this to be strictly true it must also be the case that increases in the
instruments also raise additional revenue, a property that is now demonstrated
to be satisfied.
Following Kolm (1973), if it is assumed that the taxpayer of previous the
section is a representative of a large population of identical taxpayers, then the
average tax revenue from a taxpayer is equal to the expected revenue from each.
The level of average revenue is then given by
R = tX + pF t [M X] .
(12.10)
Dierentiating with respect to p determines the eect of an increase in the

probability of detection upon revenue as
R
X
= F t [M X] + t [1 pF ]
> 0,
p
p
(12.11)
where the positivity follows from (12.4) and the interior-solution assumption
pF < 1. Repeating for the fine rate gives
X
R
= pt [M X] + t [1 pF ]
> 0,
F
F
(12.12)
which is positive by (12.5). These results confirm that the two instruments are
indeed substitutes.
12.5. TAX EVASION AND LABOUR SUPPLY
377
It is now possible to consider the determination of the optimal values of p and

F . A natural assumption is that detection is costly in the sense that resources
are used in the auditing procedure. Thus increases in p require additional expenditure. In contrast, there are no dierences in the cost of alternative levels
of F and, eectively, increases in F are costless to produce. Given this, it is
clear what the optimal, revenue-maximising combination of p and F should be:
p should be set to zero and F increased without limit. This structure provides
maximum deterrence at zero cost. In the words of Kolm (1973), the optimal
policy should be to hang tax evaders with probability zero. This result is also
supported by the analysis of Christiansen (1980) which shows, with a slightly
dierent specification, that when the detection probability and the fine are adjusted to keep the expected gain from tax evasion constant, tax evasion will be
reduced by an increase in the fine so that the fine is the more ecient deterrent.
Further development of this result, introducing non-proportional taxes, can be
found in Koskela (1983).
There are several comments that can be made in respect of this result.
Firstly, it has been assumed that the aim of the tax collector was to choose
the probability and the fine in order to maximise tax revenue. This runs in contrast to the position of previous chapters in which social welfare is maximised
subject to a revenue constraint. If this latter viewpoint is accepted in the present
context and the set of choice variables extended to also include the tax rate,
then an interior solution may exist. This will be considered when public good
are introduced in section 6. Secondly, the level of the fine may not be under the
direct control of the tax collector but may be determined by the courts relative
to punishments for other crimes. In this case, the only choice variable is the
probability of detection and if detection is costly an interior solution is again
likely to exist. Finally, if a majority of the population are evading taxation,
there is little public support for strong enforcement since each consumer may
perceive the threat of punishment to outweigh any gains that may accrue from
additional tax revenue.
This section has considered the determination of the optimal values of detection probability and fine within a tax-revenue maximisation context and in this
particular case the optimum involved zero detection and maximal punishment.
This result should be contrasted to those of later sections in which the structure
is somewhat modified.
12.5
Tax evasion and labour supply
The labour supply decision with evasion has two important components. Firstly,
there is the question of how the possibility for evasion and non-declaration of
income aects the labour supply decision in terms of the comparative statics of
labour supply. Secondly, there is also the issue of occupational choice. If some
forms of employment have greater possibilities for evasion than others, then each
household must reach a decision on the quantity of each form of employment
to be undertaken. The tax system and the punishments for evasion will clearly
378
have an influence upon this decision.
12.5.1
Labour supply
Allowing labour supply to be variable introduces a additional degree of flexibility. However, this is not without cost. As shown by Pencavel (1979) virtually
none of the comparative statics eects of parameter changes can be unambiguously signed when labour supply is variable. The only exception to this is that
an increase in lump-sum income reduces declared income when the marginal
rate of tax is non-decreasing. This failure to provide definite results arises from
the conflict between income and substitution eects. If the substitution eects
alone are considered, then it is possible to derive specific results.
If hours of labour supply are denoted by x1 and utility is additively separable
with the form
U = R (x1 ) + S (wx1 ) ,
(12.13)
where wx1 is the income level achieved, then Andersen (1977) shows that
x1
|U
t
const.
< 0,
[wx1 X]
|U
t
const.
> 0,
(12.14)
where wx1 X is the level of evaded income. Therefore an increase in the tax
rate, holding utility constant, reduces labour supply but increases the level of
evaded income. If evaded income is taxed at a penalty rate f when discovered,
then the eects of increases in f are
x1
|U
f
const.
< 0,
[wx1 X]
|U
f
const.
> 0,
(12.15)
so that an increase in the penalty rate, with utility constant, reduces labour
supply and evaded income. Finally, if relative risk aversion increases with income
and is greater than 1, then an increase in the probability of detection reduces
evaded income.
These results do indicate that some conclusions can be derived when the
labour supply decision is introduced. However, these are almost entirely restricted to substitution eects for a separable utility function. Once such restrictions are invoked, it is not surprising that the conclusions do not dier
greatly from those with exogeneous income.
12.5.2
Allocation of hours
It has been assumed above that tax evasion takes the form of a simple failure
to declare some of the income earned. In practice tax evasion is also linked to
occupational choice, with some occupations providing greater opportunities for
evasion than others. Furthermore, within an occupation it may also possible to
divide labour time between an ocial market, the income from which must be
declared, and an unocial market from which the income earned is not declared.
It is upon this latter aspect that the focus will now be placed.
12.5. TAX EVASION AND LABOUR SUPPLY
379
Two wages are now defined: wr , the wage in the registered or ocial market
and wu , the unocial market wage. Since the recipient of the wages does not
pay tax when working on the unocial market, it is assumed that wu < wr .
This has the implication that the gain from evading tax is split between the
supplier and employer of unocial labour. Compatible with these wage rates
are the hours of labour xr1 , xu1 . The income level when evasion is not detected
is given by
wu xu1 + wr xr1 [1 t] ,
(12.16)
and when detected
wu xu1 [1 f ] + wr xr1 [1 t] .
(12.17)
U = U (M, 1 xr1 xu1 ) ,
(12.18)
xr1 ,
xu1
The consumer chooses

to maximise expected utility subject to the constraints (12.16) and (12.17).
The aim is to determine how the allocation of labour between the two markets is aected by changes in the tax and punishment parameters. However, it
has already been noted that, in general, few of these eects are actually signed.
In order to derive concrete results, it is necessary to place restrictions upon the
utility function. As shown by Cowell (1981), with a utility function of the form
where M is total income and the time endowment is normalised at 1, the important restriction is that
h i
1
U
U2
= 0,
(12.19)
M
U
, U2 U
where U1 M
L , with L, the leisure consumed, defined by L =
r
u
1 x1 x1 . The consequence of this restriction is that it determines the total
labour supply, xr1 + xu1 , as a function of the post-tax wage on the ocial market
and any lump-sum income.
To produce easily interpretable results, a precise functional form that satisfies (12.19) is now employed. Following Isachen and Strom (1980), the utility
function is defined by
U = log (M ) + log (L) .
(12.20)
The maximisation facing the consumer can then be written
max [1 p] [log (wu xu1 + wr xr1 [1 t]) + log (L)]
{xr1 ,xu1 }
+ p [log (wu xu1 [1 f ] + wr xr1 [1 t]) + log (L)] .
(12.21)
To guarantee an interior maximum, the assumption

wu [1 p] > wr [1 t]
(12.22)
is adopted. Solving the first-order conditions for a maximum of (12.21) it can

be found that
xr1
wu [[1 p] [1 f ] wr [1 t] + pwr [1 t] [1 f ] wu ]
=
.
u
x1
wr [1 t] [[1 p] wu + p [1 f ] wu wr [1 t]]
(12.23)
380
Dierentiating (12.23) and using (12.22) shows that

h ri
h ri
x
x
xu1
xu1
1
1
> 0,
> 0,
f
p
(12.24)
so that an increase in either the fine rate on undeclared income or the probability
of detection reduces the proportion of time spent in the unocial labour market.
In addition
h ri
x
xu1
1
< 0,
(12.25)
t
so an increase in the tax rate reduces the proportion of time spent in the ocial
labour market and increases tax evasion. It should be noted that this result is
in contrast to that for the exogeneous income model.
It is the last of these results that is probably the most important since it
supports the general presumption that the growth in unocial economic activity
is due to increases in taxation. However, it should be recalled that it was derived
on the basis of a particular form of utility function and does not constitute a
general finding.
This section has studied two particular assumptions about the method of
reporting income and the flexibility of labour supply. The first involved the
household having complete freedom in choice of hours and reporting whatever
level of income they felt to be optimal. In the second, labour supply was still
variable but evasion was only possible on unocial income. These are not the
only possible sets of assumptions, for instance it is likely that hours of work are
fixed in the ocial market and that changes in supply can only be made in the
unocial market. Cowell (1985b) presents a summary of results to be obtained
from the dierent sets of assumptions; the results dier only in minor details.
12.6
Public goods
When the objective of the policy maker is revenue maximisation, it was shown
above that the optimal choice of detection probability and fine was to have
infinite punishment with zero probability of detection. The reason for the optimality of this extreme strategy was that the objective function did not take
account of the welfare of the taxpayers nor the use to which the revenue would
be put. Once these are incorporated, the choice problem has more content.
The simplest means by which to close the system is to assume that tax
revenue is used to supply a public good from which all households derive welfare.
This provides the motivation for the existence of taxation. In addition to this, it
is also necessary to include a cost function for the detection probability, reflecting
the fact that catching tax evaders requires the input of resources. Combining
these features with a suitable welfare function then provides a well-specified
maximisation problem.
The specification of the welfare function raises some interesting issues in
this context. Since the government, whose preferences are captured by the social
12.6. PUBLIC GOODS
381
welfare function, would prefer all taxpayers to act honestly and does not hesitate
to punish those who do not, should it take account of the welfare of evaders when
formulating policy? Since tax evasion is only a minor crime it may seem that
tax evaders should not be excluded from consideration by society. However,
this cannot be claimed to constitute a general proposition about all crimes since
there probably reaches a point where a crime is so heinous that its perpetrator
does not merit attention in the formulation of societys preferences. Although
there is no clear set of guidelines to answer this question, the tradition in the
literature, for instance Kolm (1973) and Sandmo (1981), has been to adopt
a utilitarian framework in which the utilities of evaders are included in social
welfare. This is probably for analytical simplicity rather than for philosophical
reasons.
12.6.1
The valuation of public funds
Following Kolm (1973), let there be a population of n identical consumers,

where n is intended to be a large number. If p is the true probability of being
detected when evading tax and all the population evade tax, then np consumers
are caught evading and [1 p] n are not. With a cost function C(p), C(0) = 0,
C 0 > 0, representing the cost of achieving the probability p of detection and tax
revenue R = tX + pf [m X], where f is the fine rate, the level of public good
supply is given by
G = R C = tX + pf [m X] C(p).
(12.26)
The expected welfare of a typical consumer is then defined by a utility function

that is additively separable in private consumption and public good consumption
E [U ] = [1 p] U (Y ) + pU (Z) + V (G) ,
(12.27)
where Y = M tX and Z = M tX f [M X]. The level of G is certain for

the individual consumer due to the assumption that n is large. Given that G
is parametric for the individual, the analysis of (12.1) is appropriate here and
the action of each consumer is to choose the optimal value of X. Due to the
separability assumption, the optimal X depends only upon p, t and f , hence
X = X (t, p, f ) ,
(12.28)
with partial derivatives denoted Xt , Xp and Xf .

Invoking the large number assumption again, social welfare is given by
W = n [[1 p] U (Y ) + pU (Z) + V (G)] ,
(12.29)
since exactly np of the population are caught evading. Defining the per-capita
C(p)
R
variables g = G
n , r = n and c(p) =
n , maximising (12.29) is equivalent to
maximising

G
W = [1 p] U (Y ) + pU (Z) + V
n
= [1 p] U (Y ) + pU (Z) + v (g) ,
(12.30)
382
with
g = r c (p) = tX + pf [M X] c (p) .
(12.31)
Dierentiating (12.30) with respect to the choice variables t, p and f and

noting that the optimality of X implies that [1 p] U 0 (Y ) t+pU 0 (Z) [f t] = 0,
the first-order conditions are
dW
= X [v 0 E [U 0 ]] + v 0 [t pf ] Xt = 0,
dt
(12.32)
dW
= p [M X] [v0 U 0 (Z)] + v 0 [t pf ] Xf = 0,
df
(12.33)
and
dW
= U 0 (Z) U 0 (Y ) + v 0 p [M X] v 0 c0 + v0 [t pf ] Xp = 0,
dp
(12.34)
where
E [U 0 ] = [1 p] U 0 (Y ) + pU 0 (Z) .
(12.35)
It has already been demonstrated in (12.4) and (12.5) that Xp and Xf are both
positive. In addition, if it is assumed that Xt < 0, then the first-order conditions
imply that the marginal utilities can be ranked as U 0 (Y ) < E [U 0 ] < U 0 (Z). The
central inequality is reversed if Xt > 0. In particular it follows that
v0
1
h
i
,
=
t X
E [U 0 ]
1 1 pf
t
X t
(12.36)
so that the ratio of the marginal utility of public income to the consumer, v 0 , to
the marginal utility of private income, E [U 0 ], is aected by the eect of changes
in the tax rate upon the declared level income. Since this valuation of public
funds would be employed in any cost-benefit calculations, it can be seen that
tax evasion should be taken into account when determining all public decisions.
The main conclusion to be drawn from endogenising the use of revenue and
formulating the objective function in terms of social welfare is that it is possible
to generate an interior solution for the detection probability and the level of
the fine for evasion. As such, these variables are subject to the same basic
considerations as other instruments of public policy.
12.6.2
Congestible public goods
The second eect that public goods may have is to modify the response of the
household to changes in tax rates. It has been noted that the standard prediction
is that an increase in the tax rate will lower evasion; a result that runs counter
to expectations. If households feel that public goods are over-provided then an
increase in the tax rate will lead to even greater over-provision. As the greater
provision exacerbates the initial feeling of over-provision, the households may
well alter the extent of their tax evasion; in particular circumstances, decreasing
12.6. PUBLIC GOODS
383
absolute risk aversion will lead to increased evasion. This aspect could not arise
in the previous model where the additive separability of the utility function
ensured that the evasion decision was independent of the level of public goods
provided. This intuitive argument has been formalised by Cowell and Gordon
(1988).
Define the level of evaded income by E M X so that the gain from
evasion is tE is evasion in undetected and tEf if detected. The household can
then be viewed as choosing a quantity tE of a random asset which has return
r = 1 with probability 1 p and return r = f with probability p. Hence
consumption can be defined as the random quantity
x = [1 t] M + rtE.
(12.37)
The revenue raised from taxation is

R = n [tM + rtE] nc (p) ,
(12.38)
where the large-population has been used to replace the random variable r by
its mean r = 1 p pf . In contrast to Section 6.1 the quantity of public good
available to each individual is also dependent upon the size of the population;
an assumption that is intended to capture congestion eects. In detail,
G=
R
,
(n)
lim
1
= 0,
(n)
lim
1
n
= > 0.
(n)
(12.39)
If the n 1 other households each evade an amount E, then the evasion E of

the nth household determines the level of public good as
G=
n [tM c (p)] [n 1] rtE

rtE
rtE
= G (n, t)
.
(n)
(n)
(n)
(12.40)
In determining their consumption, it is assumed that the household takes G (n, t)

as fixed. This is the Nash assumption as employed in the analysis of the private
provision of public goods in Chapter 9. It is important to note that in writing
(12.40) it is assumed that the household takes account of how their choice of
evasion aects the level of public good provision. Eectively, the population is
suciently large that the mean can be substituted into (12.38) but not so large
that the eect of each household is negligible. Without this assumption, the
essential link between evasion and provision is lost. The best interpretation to
give this is that a small number economy is considered and the limit assumption
is imposed in (12.42) and (12.47) to provide a simplification.
By choosing their level of evasion, E, each household seeks to maximise the
von Neumann-Morgernstern utility function
U = E [U (x, G)] ,
(12.41)
with G defined by (12.40). The first-order condition for this maximisation is
UG
E Ux r
r = 0.
(12.42)
(n)
384
Imposing the limit assumption in (12.39), the first-order condition reduces to

E [Ux r] = 0.
(12.43)
As noted, the aim of the analysis is to determine the eect of changes in the
tax rate, t, upon the level of tax evasion. To do this, the first-order condition
(12.43) is dierentiated with respect to t and tE to give
E [Uxx r] M E [UxG r] G
d (tE)
t
=
.
dt
E [Uxx r2 ]
(12.44)
At the symmetric Nash equilibrium, the level of the public good is given by
G=
n [tM c (p)] nrtE

.
(n)
(12.45)
Hence dierentiating this

nM
nr tE
G
=
.
t
(12.46)
Substituting (12.46) into (12.44) provides the final characterisation

E [Uxx r] M E [UxG r] M
d (tE)
,
=
r
2
dt
E [Uxx r ] E [UxG r]
(12.47)
where the limit assumption has again be used on the terms n .

To determine the sign of (12.47), it is first noted that stability of the Nash
equilibrium implies that the denominator is negative. Therefore the sign is the
opposite of the denominator. In general, the sign of the numerator cannot be
restricted so the model can therefore generate a positive relationship between
evaded tax and the tax rate, in contrast to the model without public goods. In
order to be more precise, consider the special case where UUGx is independent of
x. This assumption implies that
Ux UxG Uxx UG = 0,
(12.48)
and hence E [UxG r] = E [Uxx r] m, where m = UUGx . Using this restriction, (12.47)
can be written
h
i
m
r]
M
1
E
[U
xG
d (tE)
.
(12.49)
=
2
dt
E [Uxx r ] E [UxG r] mr
To complete the analysis, note that the eect on utility of an increase in t is

given by
m
E [UG ]
dE [U ]
=M
E [Ux ] = M
1 .
(12.50)
dt
If this expression is positive, it suggests public goods are underprovided since

the consumer would welcome an expansion of the public sector. Returning to
12.7. EMPIRICAL EVIDENCE
385
(12.49),
h
i since the denominator is negative, tax evasion will increase whenever
m
1 is positive. Therefore an increase in tax rate will increase evasion when
public goods are over-provided.
The introduction of public goods represents one way of closing the system
so that tax revenue raised has some beneficial eect. It has been demonstrated
how such closure can lead to an interior solution for the punishment variables
and thus places these within the standard framework of public economics. In
addition, it was also demonstrated that tax evasion alters the cost of raising
public funds and that this should be taken into account in any cost-benefit
calculations. Finally, the introduction of public goods may lead to the degree
of tax evasion increasing as the tax rate increases, a result more in conformity
with a priori expectations.
12.7
Empirical evidence
The theoretical analysis of tax evasion has predicted the eect that changes in
various parameters will have upon the level of tax evasion. In some cases, such as
the eect of the probability of detection and the fine, these are unambiguous. In
others, particularly the eect of changes in the tax rate, the eects depend upon
the precise specification of the tax system and upon assumptions concerning
attitudes towards risk. Given these uncertainties, it is valuable to investigate
empirical evidence in order to see how the ambiguities are resolved in practice.
Furthermore, analysis of empirical evidence also allows the investigation of the
relevance of other parameters, such as source of income, and other hypotheses
on tax evasion, for example the importance of social norms.
There have been three basic approaches taken in studying tax evasion. The
first has been to collect survey or interview data and from this to infer the
extent of evasion and some qualitative aspects of its relationship to various parameters. Secondly, econometric analysis has been applied to both survey data
and to standard economic statistics. Such analysis provides a more quantitative
determination of the relationships. Finally, tax evasion experiments have been
conducted which provide an opportunity of designing the environment to permit
the investigation of particular hypotheses.
An early example of the use of interview data can be found in the study
of Norwegian taxpayers by Mork (1975). The methodology was to interview
individuals in order to ascertain their actual income levels. This information
was then contrasted to that given on the tax returns of the same individuals
and indicated a steady decline of declared income as a proportion of reported
income as income rose. This result is in agreement with that of the comparative
statics analysis.
Combining econometrics and survey methods, Spicer and Lundstedt (1976)
sought to investigate the importance of attitudes and social norms in the evasion
decision; the data were taken from a 1974 survey in the United States. Econometric analysis revealed that the propensity to evade taxation was reduced by
an increased probability of detection and an increase in age. Surprisingly, an
386
increase in income reduced the propensity to evade. With respect to the attitude and social variables, an increase in the perceived inequity of taxation and
of the number of other tax evaders known to individual both made evasion more
likely. The extent of tax evasion was also increased by the attitude and social
variables but was also increased by the experience of the tax payer with previous
tax audits. This study clearly demonstrated the importance of social variables
in addition to the economic variables.
Clotfleter (1983) estimated tax evasion equations using data from the Internal Revenue Services Taxpayer Compliance Measurement Program survey of
1969. These indicated that tax evasion did, in fact, increase as marginal tax
rates increased and was decreased when wages where a significant proportion of
income. This result was supported by Crane and Nourzad (1986) employing the
dierence between income and expenditure figures in National Accounts from
the United States for the period 1947 - 1981. Although their major focus was
the fact that inflation raised evasion, the results also showed that increased marginal tax rates also raised evasion. In contrast to this, the study of Geeroms and
Wilmots (1985) of Belgian data found precisely the converse conclusion with tax
increases leading to lower evasion. Therefore these econometric studies do not
resolve the ambiguity about the relation between marginal tax rates and tax
evasion.
Turning now to experimental studies, Friedland, Maital and Rutenberg (1978)
employed a tax evasion game in which participants were given a monthly income and a set of tax and punishment parameters. Given these, they were
requested to make tax declarations. The major findings of this study were that
evasion increased with the tax rate and, that keeping the net gain from evasion
constant, evasion fell as the fine was increased and the detection probability reduced. This result is in agreement with the theoretical analysis of Christiansen
(1980). Further results showed that women evaded more often than men but
evaded lower amounts and that purchasers of lottery tickets, presumed to be
less risk averse, were no more likely to evade than non-purchasers but evaded
greater amounts when they did evade. A similar experiment was conducted
by Becker, Bchner and Sleeking (1987) but with the inclusion of endogenous
transfers of tax revenue back to the taxpayers and with income being earned
by the participants. With respect to the propensity to evade, a high transfer
had a negative eect as did the probability of detection and, surprisingly, the
perceived level of tax. Income level had a positive eect and hence raised the
propensity to evade. Only the audit probability had a significant eect on the
level of evasion.
Finally, Baldry (1986) contrasted the findings of two sets of experiments.
The first was framed as a tax evasion decision and this determined that some
participants never evaded tax and that the decision to evade was influenced by
the tax schedule. This experiment was then repeated as a simple gamble with
the same payos. The finding was then that all participants betted and each
made the maximum bet. From these contrasting findings, it can be concluded
that there is more to tax evasion than gambling and that the moral and social
dimensions are of importance.
12.8. HONESTY AND SOCIAL NORMS
387
The important lessons to be drawn from these results are that the theoretical
predictions are generally supported, with the exception of the eect of the tax
rate which remains uncertain, and that tax evasion is rather more than a simple
gamble; there are attitudinal and social aspects to the evasion decision. This
latter observation naturally carries implications for further theoretical analysis
of the evasion decision. In particular, the fact that some taxpayers never evade
requires explanation.
12.8
Honesty and social norms
The feature that distinguishes tax evasion from a simple gamble is that taxpayers submitting incorrect returns feel varying degrees of anxiety and regret.
To some, being caught would represent a traumatic experience which would do
immense damage to their self-image. To others, it would be only a slight inconvenience. The innate belief in honesty of some taxpayers is not captured by the
representation of tax evasion as a gamble nor are the non-monetary costs of detection and punishment captured by preferences defined on income alone. The
first intention of this section is to incorporate these features into the analysis
and to study their consequences.
The empirical results show a positive connection between the number of tax
evaders known to a taxpayer and the level of that taxpayers own evasion. This
observation suggests that the evasion decision is not made in isolation by each
taxpayer but is made with reference to the norms and behaviour of the general
society of the taxpayer. Given the empirical significance of such norms, the
second part of this section focuses on their implications.
To introduce the concept of a preference for honesty into the utility function,
the simplest approach is to adopt a function of the form
U = U (x) E,
(12.51)
where is the measure of the taxpayers honesty and, with E the extent of
evasion, E is the utility cost of deviating from complete honesty. It is assumed
that taxpayers are characterised by their value of but are identical in all other
respects.
Combining (12.51) with the budget constraint (12.37), individual maximisation leads to the first-order condition for choice of E
E [U 0 (x) rt] = 0.
(12.52)
E will only be positive when the marginal utility of evasion is greater than
zero at a zero level of evasion. Formally, evasion will occur when V0 > 0,
where V0 [1 p pf ] tU 0 (M [1 t]). Hence those taxpayers characterised by
values of that satisfy V0 > 0 will evade and those with higher values of
will not. The population is therefore separated into two parts with honest
taxpayers not evading whilst dishonest will evade. The term honest does not
have its standard meaning in this context since all taxpayers with finite will
evade if the expected gain is suciently great.
388
The comparative statics with respect to the punishment and income parameters are not too dierent from those of the model described by (12.1). The
major dierence is that the aggregate eect is composed of the changes in the
level of evasion by existing evaders and the marginal changes as either more or
less taxpayers evade. Since these eects work in the same direction, there is no
change in the predicted eects on aggregate or individual evasion.
The eect of an increase in the tax rate is given in the following theorem
from Gordon (1989).
Theorem 59 (Gordon) With decreasing absolute risk aversion, there exists
E
some < V0 such that E

t < 0 if < and t > 0 if V0 > > .
Proof. By dierentiation of (12.52), the eect of the tax change can be
found to be
E [U 00 (x) rt [M rE]] t
E
=
.
(12.53)
t
E [U 00 (x) rt [M rE]]
For = 0, (12.53) reduces to (12.8) which has already been shown to be negative. In the limit as tends to V0 (12.53) is clearly positive since E is tending
to 0 from a positive value. As the decreasing absolute risk aversion implies the
numerator is monotonic in , the sign of E
t must change once as stated in the
theorem.
This analysis predicts that there will be some evaders who increase the extent
of their evasion as the tax rate increases. In addition to this eect, an increase
in t also raises V0 so that previously honest taxpayers will begin to evade. It
is therefore possible, though not unambiguous, that the introduction of utility
cost of evasion can generate a positive relation between the tax rate and the
extent of evasion.
Social norms have been incorporated into the evasion decision in two distinct
ways. Gordon (1989) introduces the social norm as an additional element of
the utility cost to evasion. The additional utility cost is assumed to be an
increasing function of the proportion of taxpayers who do not evade. This
formulation is intended to capture the fact that more utility will be lost, in
terms of reputation, the more out of step the taxpayer is with the remainder of
society. The consequence of this modification is to reinforce the separation of
the population into evaders and non-evaders. This approach has been developed
further by Myles and Naylor (1995a) who show that reputation eects can lead
to multiple equilibria and epidemics of evasion.
An alternative approach has been taken by Bordignon (1993) who explicitly
imposes a social norm upon behaviour. The social norm is based on the concept of Kantian morality and, eectively, has each individual assessing their fair
contribution in tax payments towards the provision of public goods. This calculation then provides an upper bound on the extent of tax evasion. To calculate
the actual degree of tax evasion each taxpayer then performs the expected utility
maximisation calculation, as in (12.1), and evades whichever is the smaller out
of this quantity and the previously determined upper bound. This formulation
is also able to provide a positive relation between the tax rate and evaded tax
12.9. OPTIMAL AUDITING WITH AN INDEPENDENT REVENUE SERVICE389

for some range of taxes and to divide the population into those who evade tax
and those who do not.
The introduction of psychic costs and of social norms is capable of explaining
some of the empirically observed features of tax evasion which are not explained
by the standard expected utility maximisation hypothesis. It does this modifying the form of preferences but the basic nature of the approach is unchanged.
The obvious diculty with these changes is that there is little to suggest precisely how social norms and utility costs of dishonesty should be formalised.
12.9
Optimal auditing with an independent revenue service
In the analysis of the optimal fine and detection probability conducted in Section
4, it was assumed that both instruments were under the control of the authority
that also determined the rate of tax. In practice it is more likely that there is a
distinction between the agencies that determine each of these variables. Taxes
are set by the exchequer, punishments by the courts in relation to those for other
oences, and detection rates by a revenue service. If this view is adopted, it is
more appropriate to analyse the determination of the optimal rate of detection
in isolation taking as given the rate of tax and the fines that will be levied.
Having identified the revenue service as an agency in its own right, it follows
that its objective can be identified as the maximisation of revenue collected less
the cost of enforcing its verification policy.
As noted by Reinganum and Wilde (1985), tax evasion is distinguished from
many other forms of crime in that the tax evader is required to make a report of
their income to the revenue service and it is on the basis of this report that the
revenue service founds the decision whether to investigate or not. Consequently
the reports can be viewed as a signal from the tax payer about their true income
and the revenue service can attempt to extract information from this signal.
Formalising this, given their observed income, which is private information,
each taxpayer files an income report and the revenue service decides to audit,
or not, on the basis of this report. The reported income is chosen to maximise
the welfare of the taxpayer and the audit probabilities are chosen to maximise
the net revenue of the revenue service.
To complete the description, the distinction must also be drawn between an
equilibrium with pre-commitment and that without. With pre-commitment the
revenue service is bound by a pre-announced audit policy which it cannot alter in
the light of taxpayers responses even if the audit policy is then non-optimal. The
assumption of pre-commitment reduces the model to a principal-agent problem
in which the revenue service (the principal) designs the contract for the taxpayer
(the agent). Without pre-commitment the audit policy must remain optimal
given the choices of the taxpayers. There is hence a degree of simultaneity in
the choices and the Nash equilibrium concept becomes appropriate.
Taking the no commitment model first, Reinganum and Wilde (1986) con-
390
sider a verification policy as a function = p(X), where X is the taxpayers

reported income. Given a report of income X, the function p(X) determines
the probability with which evasion will be detected. Given a true income level
of M the taxpayer chooses X to maximise
E [U (X)] = p (X) [M [1 t] tf [M X]] + [1 p (X)] [M tX] .
(12.54)
The specification in (12.54) embodies the assumption of risk neutrality so the

consumer is attempting to maximise expected income. Given p(X), the maximisation in (12.54) determines a report as a function of true income. This
function is denoted r(M ), hence the report is given by X = r(M ).
When it receives an income report the revenue service must form an expectation of the true income level of the person filing that report. Let these beliefs be
captured by a function (X), that is if income X is reported the revenue service
believes the taxpayer to be of true income (X). Defining the cost of achieving a
probability of detection as c(), the expected revenue from a taxpayer making
report X is given by
R (X, ; ) = [t (X) + f t [ (X) X]] + [1 ] tX c().
(12.55)
Given its beliefs, the revenue service chooses to maximise R (X, ; ).

The equilibrium that is to be defined is assumed to be a separating equilibrium in which each income type files a dierent report. This implies that the
report function r(M ) is monotonic in M ; in particular it is natural to assume
that r(M ) is monotonically increasing in M . A separating equilibrium can then
be defined.
Separating equilibrium
A set of functions { (X), p(X), r(M )} is a separating equilibrium if r(M ) is
monotonically increasing and
(i) Given (X), p(X) maximises R (X, ; );
(ii) Given p(X), r(M ) maximises E [U (X)];
and
(iii) (r(M )X) = M .
Condition (iii) is the consistency condition for the beliefs of the revenue
service and simple asserts that they cannot be consistently wrong in their assessment of the income levels from the reports that they are given.
For each value of X the optimal audit probability can be found by dierentiating (12.55). This provides the necessary condition
t [1 + f ] [ (X) X] c0 () = 0.
(12.56)
Assuming that c00 is positive, this is also a sucient condition for the choice of
audit probability. From (12.54), the taxpayers optimal report must satisfy
p (r (M )) tf p0 (r (M )) t [1 + f ] [M r (M )] t [1 p (r (M ))] = 0. (12.57)
The second-order condition is
2p0 (r (M )) t [1 + f ] p00 (r (M )) t [1 + f ] [M r (M )] 0.
(12.58)
12.9. OPTIMAL AUDITING WITH AN INDEPENDENT REVENUE SERVICE391

The consistency condition (r(M )X) = M can be given the alternative formulation (r(M )) = r1 (X), where r1 (X) is the inverse of the function r(M ).
Using this identity (12.56) - (12.58) can rewritten
(12.59)
t [1 + f ] r1 (X) X c0 () = 0,
and
p (X) tf p0 (X) t [1 + f ] r1 (X) X t [1 p (X)] = 0,
2p0 (X) t [1 + f ] p00 (X) t [1 + f ] r1 (X) X 0.
(12.60)
(12.61)
The advantage of these transformed equations is that they are now in terms of
X and can be used to characterise the reporting and auditing strategies.
Since (12.59) and (12.60) must hold for all X they can be dierentiated to
give
t [1 + f ] r10 (X) 1 c00 (p (X)) p0 (X) = 0,

(12.62)
p0 (X) t [1 + f ] 2 r10 (X) p00 (X) t [1 + f ] r1 (X) X = 0. (12.63)

The form of the optimal audit function is then given in the following theorem.
Theorem 60 (Reinganum/Wilde) p0 (X) < 0 and r10 (X) (0, 1).
Proof. For (12.62) to hold either p0 > 0 and r10 > 1 or p0 < 0 and r10 < 1.
Similarly, for (12.63) to hold either p0 > 0 and r10 < 0 or p0 < 0 and r10 > 0.
These requirements can only be consistent when p0 < 0 and 0 < r10 < 1. This
proves the theorem.
Theorem 12.2 has the implication that the equilibrium eort at verification
diminishes the higher is the level of reported income. In addition the identity
r10 = r10 and the result 0 < r10 < 1 imply that r0 (M ) > 1 and hence that tax
evasion declines with income since M r(M ) is decreasing in M . The second of
these two conclusions may seem surprising but if evasion increased with income
the revenue service would have an incentive to raise the audit probability on
high incomes.
This analysis has been extended to a social custom model of tax evasion in
Myles and Naylor (1995b). The existence of the social custom results in the
equilibrium being characterised by some taxpayers evading whilst others pay
their taxes honestly. This leads to there being pooling of honest and dishonest
taxpayers at some income report levels. The optimal audit schedule is still
decreasing over the reports by evaders, with the probability of audit reaching
zero at the highest income report of a tax evader. An increase in the fine rate
raises the audit probability (where positive), in direct contrast to the analysis
of Section 4 where the two instruments are substitutes.
Turning now to the analysis with pre-commitment, somewhat dierent results are obtained. Allowing pre-commitment permits the adoption of strategies
by the revenue service which involve a zero probability of auditing some income
levels. In fact, as shown by Scotchmer (1987), the optimal policy takes the form
of a cut-o rule in which income reports above a certain level are not audited at
392
all. Since this leads all taxpayers with an income above the cut-o level to file
a report at the cut-o, this can only be an equilibrium when pre-commitment
is allowed because if there were no commitment the revenue service would gain
by reneging on the zero audit probability and catching all the evaders at the
cut-o point. This optimal strategy is clearly an exaggeration of the decreasing
audit probability shown to be optimal without commitment.
The consequence of the cut-o rule is that the ocial tax which is proportional at rate t becomes regressive since actual tax payment is constant on
incomes above the cut-o point and the eective tax function is regressive. This
observation illustrates the general point that with tax evasion the properties of
the eective tax function may well be very dierent from those of the ocial tax
function. This is a point pursued further by Cramer, Marchand and Pestieau
(1990) who also note the diculties in determining optimal taxes due to the
nonconvexity of the social welfare function.
12.10
Tax evasion by firms
The analysis of the previous sections has been concerned with the tax evasion
decisions of households. To complement this, and to provide the results needed
for considering optimal taxation with evasion, it is also necessary to consider
tax evasion by firms. Firms can evade taxation either by misreporting sales or
profit or by making false declarations about input use. It is possible that all three
methods may be required simultaneously to disguise evasion if the information
gathering process of the revenue service is suciently thorough. The analysis
given below is simplified however by considering only the under-reporting of
sales. Tax evasion will be analysed in both competitive and imperfectly competitive markets.
12.10.1
Competitive firms
The tax evasion decision for competitive firms has been analysed by Virmani
(1989), Yamada (1990) and Cremer and Gahvari (1993), with each employing a
slightly dierent structure; the analysis given here follows Cremer and Gahvari.
Consider a competitive industry producing with a constant marginal cost c
whose output is subject to a specific tax t. Each firm in the industry can choose
to reveal only a fraction of its sales to the revenue service. However, to
conceal output requires the use of resources by the firms. The resource cost of
concealing each unit of output is determined by a convex function G(1 ) of
the proportion of sales concealed. The probability that evasion is detected is
given by . The penalty rate on evaded tax is given by 1.
Denoting the market price of output as q, a typical firm in the industry
maximises expected profit given by
e = [q c [1 ] G(1 ) [1 ] t [t + [ 1] [1 ] t]] y, (12.64)
12.10. TAX EVASION BY FIRMS
393
where y is its output. Given that y is positive, is chosen to maximise

q c [1 ] G(1 ) [1 ] t [t + [ 1] [1 ] t] .
(12.65)
Defining g(1 ) [1 ] G(1 ), the necessary condition for choice of is

g 0 (1 ) = [1 ] t.
(12.66)
The second-order condition is naturally satisfied by the assumption that G(1)

is a convex function. Equation (12.66) therefore characterises the optimal .
Defining the expected tax rate, te , by
te = [ + [1 ] ] t,
(12.67)
the competitive assumption implies that market price must be equal to expected
marginal cost or
q = c + g + te ,
(12.68)
where g and te are evaluated at the optimal value of .
These equations allow the comparative statics eects of changes in the underlying parameters to be calculated. Taking changes in the tax rate first,
dierentiation of (12.66) gives
d
[1 t]
< 0,
=
dt
g 00
(12.69)
so that reported sales decrease as the tax rate increases. The eect upon the
expected tax rate follows from (12.67) as
dte
[1 t]2 t
.
= [ + [1 ] ]
dt
g 00
(12.70)
This eect may be positive or negative since an increase in t directly raises the
expected tax but causes an indirect decrease as evasion rises. Finally, using
(12.69) and (12.70), the eect upon price is
dq
dq
= [ + [1 ] ] , 0 <
< 1.
dt
dt
(12.71)
The post-tax prices rises but by less than the amount of the tax since some of
the tax increase is absorbed in increased evasion.
The eect of changes in the probability of detection are also derived from
(12.66) to (12.68). The results are given by
t
d
= 00 > 0,
d
g
(12.72)
dte
[1 t] t2
>0
= [1 ] t +
d
g 00
(12.73)
394
and
dq
= [1 ] t > 0.
(12.74)
dt
Hence an increased probability of evasion raises the proportion of sales declared,
the expected tax and the market price. It can be seen from the last result that
an increase in the probability has an ambiguous eect upon welfare since it
raises the price level to the detriment of consumers.
These results are generally without surprises but they indicate that tax evasion can be incorporated onto the analysis of the competitive firm. They will
also be employed in the characterisation of optimal taxes below. Before considering imperfect competition, it should be noted that Virmani (1989) studies
firms with U-shaped average costs and establishes the result that with tax evasion production will not take place at minimum average cost. This raises a
channel through which tax evasion can lead to production ineciency.
12.10.2
Imperfect competition
Tax evasion by monopolistic firms has been studied by Marrelli (1984) and
evasion by oligopolisitc firms by Marrelli and Martina (1988). Both authors
assumed that the firms were risk averse. Since this is not standard practice
the analysis below will assume risk neutrality. The case of monopoly will be
reviewed here; the extension to oligopoly is straightforward.
The structure remains as that for the competitive firm but with the addition that the price is chosen by a profit maximising monopolist. Denoting the
demand function by X(q), the price and level of evasion are chosen to maximise
e = X (q) [q c g(1 ) [1 ] t [t + [ 1] [1 ] t]] .
(12.75)
For the choice of , the first-order condition is

X (q) [g 0 (1 ) [1 ] t] = 0.
(12.76)
Assuming that X(q) is positive, this is simply the condition given in (12.66)
again. In addition, since X(q) can be cancelled from (12.76), q does not appear
in the condition determining . Hence the evasion decision is independent of the
pricing decision. The converse is not true since the evasion decision determines
the expected level of costs, inclusive of tax, and therefore the level of price. To
see this formally, note that the first-order condition for the choice of q is given
by
X 0 (q) [q c g(1 ) [1 ] t [t + [ 1] [1 ] t]] + X (q) = 0,
(12.77)
where (12.77) is evaluated at the optimal .
The qualitative nature of the comparative statics are suciently similar to
those of the competitive model not to require close study. However, it is worth
noting that tax overshifting will occur when
E > 2 [ + [1 ] ] .
(12.78)
12.11. OPTIMAL TAXATION WITH EVASION
395
Contrasting this to (11.31), tax evasion reduces the possibility of overshifting

whenever < 1 and increases it if > 1. Therefore, the higher the rate of
punishment, the more likely is overshifting of taxation.
12.11
Optimal taxation with evasion
The existence of tax evasion clearly has implications for the determination of
optimal taxes. It has been noted above that optimal auditing policies will not
generally result in complete elimination of evasion. Therefore the design of taxes
should take this into explicit account. With respect to commodity taxation, the
evasion of firms implies that the relationship between tax and price will be
modified by the existence of evasion. For income taxation, tax evasion has the
eect of altering the elasticity of labour supply due to the possibility of working
in the shadow economy. These factors are now incorporated into the relevant
optimal tax problems.
12.11.1
Commodity taxation
Consider a single consumer competitive economy with n industries as described

in Section 4 of Chapter 4. Normalising the wage rate to unity, industry i will
have a constant marginal cost of ci . Given a tax tk on good k, the result in
(12.68) shows that the post-tax price will be
qk = ck + gk + tek ,
(12.79)
where tek = [k + [1 k ] k ] tk is the expected tax payment per unit of output

of a firm in industry k and the tax evasion cost, gk, the evasion choice, f k, and
the detection rate, k , are all industry-specific. Employing the argument that
each industry is composed of a large number of firms permits the claim that
actual and expected tax revenue are equal so that
R=
n
X
tei Xi .
(12.80)
i=1
In the optimal tax problem, the choice variables of the government are the
set of tax rates (t1 , .., tn ) and the detection probabilities (1 , .., n ). The problem
can be written
max
{t1 ,..,tn ,1 ,..,n }
V (q1 , ..., qn ) subject to
n
X
i=1
tei Xi C(1 , .., n ) = R,
(12.81)
where C(1 , .., n ) is the cost of implementing the chosen set of detection probabilities. This problem has the associated Lagrangean
" n
#
X
e
L = V (q1 , ..., qn ) +
ti Xi C(1 , .., n ) R .
(12.82)
i=1
396
Dierentiating with respect to tk gives the first-order condition

n
h
X
Xi
i
Ak
tei
= 0,
Xk +
qk
i=1
where
Ak
tek
tk
qk
tk
(12.83)
(12.84)
For the optimal choice of audit probability, the first-order condition is

h
Bk
n
X
Xi
i
Ck
tei
=
Xk +
,
qk
[1 k ] tk
i=1
(12.85)
where the right-hand side follows from (12.74) and

Bk
tek
k
qk
k
(12.86)
The fundamental result of Cremer and Gahvari (1993) is now found by subqk
and then solving
tracting (12.85) from (12.83). Replacing [1 k ] tk by
k
provides the conditions
qk
tk
qk
k
tek
tk
tek
k
Ck
Xk
, k = 1, ..., n.
(12.87)
The interpretation of these equations are that the taxes and detection probabilities should be adjusted until the rate of substitution between tax and probability
holding the price of good k constant, and hence welfare constant, given by the
term on the left equals the rate of substitution holding tax revenue constant
which is the right-hand term. This equality of rates of substitution provides the
obvious balance between the eects of the alternative instruments.
To provide a direct contrast between the tax rule in (12.83) and its counterpart with no evasion given by (5.16), the Slutsky equation can be used to write
(12.83) as
#
" n
n
X
X Xi
e
e
ti Ski =
ti
(12.88)
+ Ak Xk .
I
i=1
i=1
This diers from the standard Ramsey rule in two ways. Firstly, it is in terms
of the expected rather than the actual taxes. Secondly, the term Ak enters and
hence the right-hand side is raised or lowered depending on whether Ak is less
than or greater than zero. Ak measures the rate at which the expected tax rate
increases relative to price as the nominal tax is raised. It is therefore preferable
to tax those goods where Ak is relatively high. This is reflected in (12.88) where
a high value of Ak leads to a greater reduction in compensated demand.
12.12. SUMMARY
12.11.2
397
Income taxation
Sandmo (1981) considers the determination of an optimal linear income tax

in the presence of tax evasion. Taxpayers are divided into two groups. The
first group consists of taxpayers who have a choice of allocating some, or all,
of their labour to an unobserved sector and hence avoiding income tax. The
second group of taxpayers do not have this option open to them and must pay
tax upon all their earned income. An optimal income tax is then derived by
maximising a utilitarian social welfare function. This resulting tax rule provides
an implicit characterisation of the optimal marginal tax and can be partitioned
into two parts: the first being the standard formula for the optimal marginal
tax and the second being a correction term for the existence of tax evasion. If
a higher tax rate leads to substitution towards labour in the unobserved sector
then this makes the correction term positive and implies a tendency for the
marginal rate of tax to be increased. This result is in marked contrast to the
view that tax evasion should be oset by lower marginal rates of tax.
12.12
Summary
Tax evasion is an important and significant phenomenon that aects both developed and developing economies. Although there is residual uncertainty surrounding the accuracy of measurements, even the most conservative estimates
suggest the hidden economy in the UK and US to be at least ten per cent of
the measured economy. The substantial size of the hidden economy, and the
tax evasion that accompanies it, requires understanding so that the eects of
policies that interact with it can be correctly forecast.
The predictions of the standard Allingham-Sandmo representation of tax
evasion as a choice with risk were derived and contrasted with empirical and
experimental evidence. This showed that although it is valuable as a starting
point for a theory of evasion, the Allingham-Sandmo representation some key
aspects of the evasion decision, most notably the eects of morals and the social
interactions between taxpayers. In addition, tax evasion also impinges upon the
broader issues of labour supply and the allocation of hours between markets and
occupations. It was shown how each of these issues could be incorporated into
the evasion decision.
Part IV
Introducing Real Time
399
Chapter 13
Overlapping Generations
Economies
13.1
Introduction
The overlapping generations economy, so called because of its assumed demographic structure, was introduced by Samuelson (1958). It has since proved
useful in many areas of economics including macroeconomic growth theory, public economics and monetary economics. One of the economys major points of
interest is the welfare properties of its equilibrium. Even when the standard
competitive assumption are imposed, the equilibrium of the overlapping generations economy may not be Pareto optimal. This is in marked contrast to the
Arrow-Debreu competitive economy.
Despite its value in many areas, as demonstrated by the previous chapters,
there are several shortcomings of the Arrow-Debreu economy when applied to
intertemporal issues. The first is that it is essentially static and, although it can
be interpreted as intertemporal, this is not completely satisfactory as noted in
Chapter 2. It would seem to be stretching the interpretation too far to accept
trading in a single period for all goods into the indefinite future. Trades in the
economy are carried out by barter and there is simply no role for money. This is
a consequence of the assumptions that agents are assumed to know universally
the terms of trade between commodities and that any sequence of transactions
can be completed without cost or hindrance. The equilibrium of the economy
is also Pareto optimal, so there can be no ineciency in investment or in the
choice of production techniques. Finally, the economy implicitly assumes the
lives of each agent to be at least as long as the length of the economy itself. In
many ways, the economy is simply too rich: it can cover all possibilities but can
never describe anything in detail.
These observations provide the motivation for the study of overlapping generations economies. By structuring the evolution of the population and introducing time in a very real sense, the overlapping generations economy is able
401
402
CHAPTER 13. OVERLAPPING GENERATIONS ECONOMIES
to address many issues of interest in public economics. The potential failure

of its competitive equilibrium to be Pareto optimal provides an eciency-based
justification for assessing the benefits of government intervention. In addition
to possessing inecient equilibria, overlapping generations economies can also
generate cyclical equilibria without any requirement for exogenous shocks. Furthermore it is possible for fiat money to be valuable and for a continuum of
equilibria to exist. All these features will be discussed below.
This chapter sets out the structure of both the pure exchange overlapping
generations economies and an aggregate economy with production due to Diamond (1965). A version of the economy proposed by Samuelson is introduced
first, and the failure of eciency demonstrated. The economy is then generalised and placed in an Arrow-Debreu format in order to make the comparison
with the finite economy as sharp as possible. This generalised economy is employed to characterise ecient equilibria. Money, dynamics and indeterminacy
are then considered. For the aggregate production economy, the focus is placed
on characterising its steady state and the welfare properties of the steady state
equilibrium.
13.2
Overlapping generation exchange economies
The features of the overlapping generations economy that have been noted above
are most clearly identified in economies without production and it is these that
have been most extensively analysed. For exchange economies it is straightforward to reinterpret an overlapping generations economy as a special case
of the Arrow-Debreu economy described in Chapter 2 in which the lifetime of
each household is finite but, over the lifetime of the economy, there are an infinite number of households and goods. It is this double infinity that gives an
overlapping generations economy its unique structure.
Following a general description of a typical overlapping generations economy, this section will demonstrate the failure of Pareto optimality in the simple
economy first described by Samuelson (1958). Although instructive in itself,
this style of presentation of the economy does not make clear the link between
overlapping generations economies and the Arrow-Debreu economy. A more
general economy is therefore introduced which is cast in a form that emphasises
the parallels between it and the Arrow-Debreu economy. This general form
of overlapping generations economy is then employed to demonstrate the most
important features of the equilibria of such economies.
An overlapping generations economy is explicitly intertemporal. Time is
divided into discrete periods with the basic interval of time being equal to the
length of time that elapses between the birth of one generation and that of the
following generation. There is no final period for the economy. The population
of households alive at any point in time in a typical overlapping generations
economy consists of a set of finitely-lived consumers. At each date is born a
cohort of young consumers and, if the rate of growth is positive, each cohort is
larger than the previous. In this and the following chapters, the term household
13.2. OVERLAPPING GENERATION EXCHANGE ECONOMIES
403
Figure 13.1: Structure of Generations

is reserved for a sequence of consumers linked by family ties. The lifespan of
each consumer is assumed to be two periods; it will be shown below that this
is not a significant restriction. The structure of the population is illustrated in
Figure 13.1 where the solid lines represent the lifespan of the generations. Each
generation is identified with the period in which its members are born. The
figure shows the motivation for the name applied to these economies.
An equilibrium for the economy is a sequence of prices that equate demand
to supply in each time period. In this section it is assumed that there is no
production and no storage of commodities. Since the absence of storage possibilities prevents any of the endowment of one period being carried over to the
next period, the supply in each period is equal to the fixed endowment. The
existence of the infinite population makes the definition of a Pareto optimum
non-trivial; alternative concepts will be defined later.
13.2.1
The Samuelson economy
The Samuelson economy is defined as one in which all consumers are identical,
except for their date of birth, and in which there is a single good available. Each
consumer receives their endowment of this good entirely in the first period of
their life. Since there is no storage, the only means by which a consumer can
enjoy consumption in the second period of their life is by organising a series
of trades in which they deliver some quantity of the good in the first period
of their life and receive a delivery in the second period. It is the possibility,
or otherwise, of organising such trades that will determine the eciency of the
competitive equilibrium.
s
The consumption in period s of a consumer
in
tborn
period t is denoted xt
t+1
so that their utility level is given as U = U xt , xt
. Such a consumer will
be referred to as belonging to generation t. It is assumed that the utility funcU
U
tion satisfies limxtt 0 x
= so that the consumer strictly
t = limxt+1 0
xt+1
t
t
t
desires consumption in both periods of their life. The consumers endowment,
which for a consumer of generation t is received entirely in period t, comprises
tt units of the good. If prices pt and pt+1 hold in periods t and t + 1, the budget
404
constraint of the consumer is

= pt tt .
pt xtt + pt+1 xt+1
t
Defining 1+ rt+1 =
form
pt
pt+1 ,
(13.1)
the budget constraint can be written in the alternative
= 0,
[1 + rt+1 ] tt xtt xt+1
t
(13.2)
which can be interpreted as saying that any savings out of the endowment earn
interest at rate rt+1 .
Generation t is defined as the set of consumers who are born in period
t. The population grows at rate n, so that if generation t is of size Ht then
Ht+1 = [1 + n] Ht . An allocation is feasible for the economy if consumption
by the two generations alive at each point in time is no greater than the total
endowment. At time t the consumption levels of the generations must satisfy
Ht1 xtt1 + Ht xtt = Ht tt ,
(13.3)
where equal treatment of the identical households in each generation has been
assumed. Using the assumed growth path of the population, (13.3) can be
written as
= 0.
(13.4)
[1 + n] tt xtt xt+1
t
From (13.4) a steady state equilibrium for the economy, in which the endowment
and the consumption levels are independent of t, is restricted to satisfy
(13.5)
[1 + n] 1 x1 x2 = 0,
where x1 and x2 denote the steady state consumption levels in first and second
period of life and 1 is the constant first period endowment. From these equalities can be proved the following theorem due to Samuelson (1958) and Gale
(1973).
Theorem 61 (Gale, Samuelson) There are at most two possible steady state
equilibria. Either (i) rt+1 = n all t or (ii) x1 = 1 all t.
Proof. In the steady state (13.2) reduces to [1 + r] 1 x1 x2 = 0.

Contrasting with (13.5) proves the theorem.
It is interesting to note that there is actually no economic behaviour behind
Theorem 13.1 and that it is simply a consequence of the accounting constraints.
The economic implications follow from studying the welfare properties of the
two steady state equilibria.
Consider case (i) first. Assume that the steady state consumption levels
x1 , x2 chosen by the consumers when rt+1 = n all t yield less utility than an
alternative pair of stationary consumption levels x1 , x2 . Revealed preference
implies that since x1 , x2 were chosen, x1 , x2 must be more costly so that, from
the budget constraint (13.2),
(13.6)
[1 + n] 1 x1 x2 < 0.
405
However, (13.6) implies that the pair x1 , x2 cannot satisfy (13.5) and hence must
be infeasible. It follows that rt+1 = n and the associated choices x1 , x2 represent
a Pareto optimal outcome for the economy. The optimality of this equilibrium
has lead to it being termed the Golden rule path for the economy. Conversely,
the equilibrium in case (ii) with x1 = 1 is clearly not Pareto optimal. Given the
assumptions made about the form of the utility function, each consumer would
be willing to trade consumption in the first period of their life for consumption
in the second period at any finite price.
Since there are two steady state equilibria with one strictly dominating the
other, it is now necessary to identify which will arise as an equilibrium of a competitive economy. To sustain a market equilibrium with rt+1 = n it is necessary
that consumers are able to trade some of their endowment for consumption in
the second period of their life since rt+1 = n implies that the price of second
period consumption in terms of first period consumption is finite. Such a trade
would require the household to form a contract which involved transferring consumption in period t to a second party and receiving, from the same second
party, consumption in t + 1. Unfortunately, the only economic agents alive in
both period t and t + 1 are all members of the same generation and they are
all trying to enact the same trade. This absence of trading partners therefore
prevents the equilibrium rt+1 = n being sustained as a competitive outcome.
In contrast, the autarkic outcome x1 = 1 can be sustained as a competitive
equilibrium since, by default, it involves no trade. Implicitly, this equilibrium
requires the price of consumption in the second period of life relative to that in
the first period being infinite or, equivalently, rt+1 = 1.
The Samuelson economy therefore leads to an outcome in which there exists
a single steady state equilibrium that can be supported by competitive behaviour and this equilibrium is not Pareto optimal. The existence of a competitive
equilibrium that is not Pareto optimal, in contradiction to the First Theorem of
Welfare Economics, is just one of a number of surprising features that overlapping generations economies possess. Although the Samuelson economy forces
this equilibrium by preventing any possibility of all consumers simultaneously
transferring consumption from the first to the second period of their lives, the
same possibility will be shown to hold in more general economies below. In addition, only the steady state equilibria have so far been described. More attention
will be paid to non-steady state equilibria below.
13.2.2
Overlapping generations as Arrow-Debreu economies
It would not be unreasonable to suspect that the source of the failure of Pareto
optimality lay in the fact that the Samuelson economy in particular, and overlapping generations economies in general, were of an entirely dierent nature to the
competitive Arrow-Debreu economy described in Chapter 2. Such dierences
could possibly stem from trades being enacted sequentially in the overlapping
generations in contrast to the completion of trades prior to economic activity
in the Arrow-Debreu economy. Alternatively, the restriction that consumers
conduct trades only during their own life-span may be felt to account for the
406
failure of optimality so that if consumers could trade in any period the inefficiency would perhaps be removed. Two final possibilities may be that the
endowment pattern was responsible for the ineciency or that the assumption
of identical consumers was responsible.
It will now be shown that none of these claims is justified. This will be
achieved by setting out the structure of a general form of overlapping generations economy in an Arrow-Debreu format as in Balasko and Shell (1980).
Several features will be made more precise including specifying the time periods
over which the economy operates. Once this is done, it will be seen that the
overlapping generations economy is simply a special case of the Arrow-Debreu
economy in which there are an infinite number of consumers and goods. Hence,
although the structure of the Samuelson economy was certainly responsible for
producing such stark findings, none of the assumptions was strictly necessary
for producing the ineciency result.
The economy
In each time period t = 1, 2, 3, ... of the economy there is a finite number n of
goods, none of which can be stored. The consumers are indexed by their date
of birth. Each consumer lives for two periods so those born in period t are alive
during periods t and t + 1. At the start of the economy in period 1 there is a
generation, labelled generation 0, whose lives finish at the end of period 1. To
provide some simplification, each generation consists of a single consumer.
The consumption of commodity i by the consumer born in t during period s
is denoted by xs,i
t . The lifespans of the consumers are reflected in their preferences by assuming that the consumers only derive utility from consuming during
the periods they are alive. The utility of theconsumer born
at t is thus given
1,n
by Ut = Ut (xt ), t = 0, 1, ... with x0 = x10 = x1,1
,
...,
x
for t = 0 and xt =
0
0
t t+1 t,1
t+1,1
xt , xt
, ..., xt+1,n
for t = 1, 2, .... The utility function
= xt , ..., xt,n
t , xt
t
is assumed to be strictly quasiconcave, dierentiable and all its first-order partial
derivatives are strictly positive. The indierence curves are also assumed not to
cross any of the coordinate axes. Each consumer receives an endowment
in each
period they are alive. These endowments are denoted 0 = 10 = 1,1

,
...,
1,n
0
0
t,1
t,n
t+1,1
t+1,n
for t = 0 and t = tt , t+1
,
...,
,
...,
for t =
=
t
t
t
t
t
x10 and the sequence

.... The
1, 2,
the vector
notation x0 is used to refer tot both
t+1
1
x0 , 0, 0, ... . Similarly, xt represents xt , xt
and 0, ..., 0, xtt , xt+1
t , 0, 0, ... .
The same convention also applies to the endowments and to the demands defined below.
The price
by pt,i so that pt = pt,1 , ..., pt,n
i in period t is represented 1,1
1 of2 good
and p = p , p , ... . The price normalisation p = 1 is employed. Using these
definitions of the prices, the demands of the consumer comprising generation 0
solve

max U0 x10 subject to p1 x10 p1 10 = M0 ,
(13.7)
1
{x0 }
407
while those of generation t = 1, 2, ... solve
max Ut xtt , xt+1

pt tt + pt+1 t+1
= Mt .
subject to pt xtt + pt+1 xt+1
t
t
t
{xtt ,xt+1
}
t
(13.8)
1
These
maximisations
result
in
demand
functions
x
=
x
,
M
p
and
xt =
0
0
0
t t+1
xt p , p , Mt or, alternatively, x0 = x0 (p, M0 ) and xt = xt (p, Mt ). These

demand functions can be treated equally as mapping into either a vector or
a sequence. For instance, for t = 0 the demand maps either to x10 or into
a sequence x10 , x20 , ... with a positive entry for x10 alone. Given these demand
functions, a competitive equilibrium for the overlapping generations economy
can be defined.
Competitive equilibrium
1 2
A competitive
equilibrium
P
P is a sequence p , p , ... of strictly positive prices
such that t xt (p, Mt ) = t t .
The proof of the existence of an equilibrium is sketched in the next section.
13.2.3
Existence of equilibrium
The proof of the existence of an equilibrium is of interest both in its own right
but also because of the insight it gives into the non-uniqueness of equilibrium.
A formal proof of existence will not be given here but instead a fairly detailed
description of the steps involved is provided.
The first step in the proof is to truncate the economy at period t and to
consider the existence of a solution to the equations
x10 () + x11 () = 10 + 11 ,
(13.9)
x21 () + x22 () = 21 + 22 ,
(13.10)
..
.
(13.11)
xtt1
() +
xtt
() =
tt1
tt1 ,
(13.12)
It should be clear that the solutions of (13.9) to (13.11), which is termed

the truncated economy Et , will determine a set of prices Pb with typical member
pb1 , pb2 , ..., pbt+1 . The equations must have a solution since adding the additional
equation to (13.9) - (13.11) gives a set of equilibrium conditions for a t+1 period
Arrow-Debreu economy. As this extended set of equations must have a solution,
the reduced set certainly does. Thus any sequence of prices p = p1 , p2 , ... can be
treated as an equilibrium for the t-period truncated economy provided it agrees
with some pb1 , pb2 , ..., pbt+1 from Pb in its first t + 1 elements. Alternatively, the
truncated economy determines the first t + 1 elements of a price sequence with
the remaining elements being indeterminate.
The next step is to provide upper and lower limits for possible equilibrium
prices. Assume that the endowment of each consumer is strictly positive. In
equilibrium each consumer must reach a level of utility at least equal to that of
their endowment and their maximum consumption level is limited by the total
408
endowment of the economy. The set of possible equilibrium consumptions for a

consumer is therefore given by the intersection of the set above the indierence
curve through the endowment and below the total endowment. Since the indifference curves of each consumers utility function do not cross the coordinate
axes, this set is compact and the set of possible gradients of the indierence
curves through points in this set is also compact. Measuring the gradient relative to that of good 1, the desirability assumption implies that the gradients
are bounded above zero but are also finite. Hence for each time period t there
is an lower bound t and an upper bound t on the gradient of the indierence
curves and hence on the possible equilibrium prices.
The above reasoning shows that the equilibrium prices for the t-period truncated economy Et belong to a compact set which is a subset of the possible
equilibrium prices for Et1 . As t tends to infinity, a sequence of nonempty
compact sets is generated with each being a subset of the previous. Such an
infinite sequence of sets has a nonempty intersection. An element of the infinite
intersection is an equilibrium price system for the economy. This proves the
existence of an equilibrium.
Welfare properties
It has already been shown that the equilibrium of an overlapping generations
economy need not be Pareto optimal, even though it is characterised by competitive behaviour and an absence of externalities or any other standard source of
market failure. The purpose of the present section is to formalise definitions of
Pareto optimality that are appropriate for infinite economies and to characterise
Pareto optimal equilibria via the structure of the supporting price sequences.
Pareto optimality is usually tested by considering whether there exists any
reallocation of resources that can raise the welfare of one consumer without
reducing that of any other. In a finite economy this is a sucient description. However, in an infinite economy it is necessary to consider the number of
consumers that can be involved in any reallocation of resources. Searching for
improving allocations where only a finite number of consumers can have their
allocation altered will clearly provide dierent results than allowing allocations
to be altered for an infinite number of consumers. Accordingly, the distinction
is made between weak Pareto optimality and Pareto optimality as follows.
Pareto optimality
0
0
The
0 , x1 , ...} is Pareto optimal if there is no allocation {x0 , x1 , ...}
P allocation
P {x
0
0
0
with t xt = t xt , Ut (xt ) Ut (xt ) all t and Ut0 (xt0 ) > Ut0 (xt0 ) for some t0 .
Weak Pareto optimality
, ...} is weakly Pareto optimal if there is no allocation
The allocation
P{x0 , x1P
{x00 , x01 , ...} with t xt = t x0t , xt = x0t for all but a finite number of t, Ut (x0t )
Ut (xt ) all t and Ut0 (x0t0 ) > Ut0 (xt0 ) for some t0 .
To understand the distinction between these forms of optimality, consider
the following example from Shell (1971). Each consumer has a linear utility
function, so that utility is the sum of consumption levels in the two periods,
and each consumer is endowed with 1 unit of the single available good in each
409
period of life. The competitive equilibrium in this case must have consumption equal to the endowment. Now consider a series of transfers in which the
consumer who is born in period 1 transfers their first period endowment to
the consumer who was born in period 0 and, in each successive period, the
consumer born in that period transfers their endowment to the consumer born
in the previous period. Compared to the competitive outcome, all consumers
born from period 1 onwards still attain a utility level of 2 but the consumer
born in period 0 has their utility raised from 1 to 2. The transfer has therefore achieved a Pareto improvement and the competitive equilibrium was not
Pareto optimal. In making this Pareto improvement, the allocation of an infinite number of consumers has been changed. More importantly, it can easily
be seen that a Pareto improvement could not be made if only a finite number
of consumers were involved in the reallocation. The competitive equilibrium is
therefore weakly Pareto optimal but not Pareto optimal.
The conclusions of this example will now be shown to be generally applicable.
That is, competitive equilibria will be shown to be weakly Pareto optimal but
not necessarily Pareto optimal. The first of these statements is proved in the
following theorem.
Theorem 62 (Balasko and Shell) Every competitive equilibrium is weakly Pareto
optimal and every weakly Pareto optimal allocation is a competitive equilibrium
allocation for a suitable assignment of endowments.
Proof. Given an allocation x = {x0 , x1 , ...}, the price sequence p = p1 , p2 , ...

is said to support the allocation x if and only if for all t = 0, 1, 2, ... xt maximises
Ut () subject to pxt Mt for some sequence of incomes Mt .
It is easy to see that every competitive equilibrium allocation x, with its
associated equilibrium price sequence p, is supported by p when the income
sequence is given by {Mt } = {p t }. Conversely, if an allocation x is supported
by a price sequence p, there exists a sequence of endowments { t } such that x
is a competitive equilibrium and the equilibrium prices are given by p. To see
this, note that if p supports x then, from utility maximisation xt = xt (p, pxt ),
for t = 0, 1, 2, ... . If the endowment sequence is chosen so that t = xt the
claim follows. Given these observations connecting competitive equilibria and
supporting price sequences, the proof is completed by showing that an allocation
is weakly Pareto optimal if and only if there exists a price sequence which
supports it.
To show that any allocation supported by a price sequence is weakly Pareto
optimal, let x be supported by p and assume that x is not weakly Pareto optimal.
Therefore, since weak Pareto optimality is concerned with allocations that dier
in only a finite number of components, the assumption that x P
is not weakly
P
Pareto optimal implies that there exists x0 with x0t = xt for t t0 , t x0t = t xt
and Ut (x0t ) Ut (xt ) with at least one strict inequality for t < t0 . From revealed
preference px0t = pxt with strict inequality for at least one t < t0 . Hence
P 0 P
Pt0 1 0 P
t0 1
0
0
t xt =
t xt and xt = xt for t t .
t=0 pxt =
t=0 xt , contradicting
410
Finally, it is necessary to show that an allocation x is weakly Pareto optimal only if there exists a price sequence that supports it. It is obvious that an
allocation is weakly Pareto optimal if and only if the first t elements of the allocation sequence are Pareto optimal for any truncated economy Et , t = 0, 1, 2, ...
. The allocation x0 is uniquely supported in E0 by p1 = U0 (x0 ) with
chosen so that p1,1 = 1. The argument is now completed by induction by showing that, given unique supporting prices p1 , p2 , ..., pt for x0 , ..., xt1 in Et1 ,
there exists a unique pt+1 such that p1 , p2 , ..., pt+1 supports x0 , ..., xt in Et . Assume that p0t , p0t+1 support xt so that p0t , p0t+1 = Ut (xt ) for some
> 0.
Now since x0 , ..., xt is Pareto optimal in Et , p0t = Ut , xt+1

with xt+1
t
t
t1
fixed and p0t = Ut1 xt1
,
with
x
fixed.
But,
by
the
assumption
that
t1
t1
t
0t
p1 , p2 , ..., pt supports x0 , ..., xt1 , pt = Ut1 xt1
t1 , . Hence p = p . Now
define pt+1 = p0t+1 . p1 , p2 , ..., pt+1 clearly supports x0 , ..., xt in Et and the
proof is completed.
The characterisation of Pareto optimal allocations is somewhat more complex than that of weakly Pareto optimal. To provide necessary and sucient
conditions requires considerable work. As a consequence, the theorem that is
proved below describes only sucient conditions. An informal description of
the necessary and sucient conditions is also given. To permit proof of the
theorem, a description of Pareto-improving sequences of feasible transfers is required. Let x be a given feasible allocation. The sequence h = (h0 , h1 , ...) is
aPfeasible sequence of commodity transfers if x + h is a feasible allocation and
t ht = 0. Since commodities can only be transferred between consumers who
are alive
follows
that h0 = h10 = h11 and, for t 6= 0,
t att+1the
same
t time, it
t+1
ht = ht , ht
= ht1 , ht+1 . A sequence of transfers is Pareto improving
on the allocation x if Ut (xt + ht ) Ut (xt ) for all t and with strict inequality for
at least one t. The following theorem gives sucient conditions for an allocation
to be Pareto optimal.
Theorem 63 (Balasko and Shell) A weakly Pareto optimal allocation x supported by prices p is Pareto optimal if the sequence (x0 , x1 , ...) is bounded from
above and lim inf t kpt k = 0.
Proof. The first step in the proof is to show that if an element of a sequence
of Pareto improving transfers, from an initial weakly Pareto optimal position, is
positive at t = t0 , it is positive for all t t0 . To show this, assume
P the converse.
Hence there exists t1 > t0 such that ht1 = 0. Now, since
t ht = 0, the
Pt1
P
structure of transfers implies that t=t0 ht = 0 and t>t1 ht = 0. Therefore the
sequence ht = (0, ..., 0, ht0 , ..., ht1 , 0, ..., 0) would be feasible and Ut (xt + ht )
Ut (xt ) for all t = 0, 1, ... . Due to strict quasiconcavity of utility, the sequence
(0, ..., 0, ht0 , ..., ht1 , 0, ..., 0), 0 < < 1, would be Pareto improving. This
sequence has only a finite number of non-zero elements; a contradiction to the
supposition that the initial state was weakly Pareto optimal which establishes
the claim.
The second step is to show that the transfers involved in a Pareto improvement can be ranked in terms of their values. In particular, it is now shown
411
that
t t
1 1
1 1
1 1
pt+1 ht+1
t+1 p ht ... p h1 = p h1 = p h0 0,
with strict inequality for t t0 , t0 as defined above. To prove this, first note
that for generation
0, U
0 (x0 + h0 ) U0 (x0 ) which, by revealed preference,

implies that p1 x10 + h10 p1 x10 or p1 h10 0. For generation 1, p1 h11 + p2 h21 0
so that p2 h21 p1 h11 . From the structure of transfers, h11 = h10 and h21 = h22
hence p2 h22 p1 h11 = p1 h10 0. This construction can be continued for all t
with strict inequality holding for t t0 .
The proof is completed by noting that (x0 x1 , ...) is bounded from above
and that lim inf t kpt k = 0. Assume that there exists a Pareto improving
sequence of transfers. For such a set of transfers, it follows that for t t0 ,
t0 t0
pt htt < pt0 +1 htt00 +1
+1 < p ht0 . The assumed boundedness of (x0 x1 , ...) implies that
t
ht is also bounded and the assumption that lim inf t kpt k = 0 implies that
lim inf t kpt htt k = 0. This contradicts the preceding inequality and establishes
that under the conditions of the theorem the allocation x is Pareto optimal.
It should be noted how this result provides a characterisation of Pareto
optimality in terms of the price sequence that supports an allocation rather than
in terms of the allocation itself. The restriction that the sequence of allocations
is bounded above prevents the economy enjoying unlimited growth.
The result described in Theorem 13.3 is not the most general that is available. Since lim inf t kpt htt k = 0 is a sucient condition for Pareto optimality,
it leaves open the possibility that there may be important cases in which this is
not satisfied but which are Pareto optimal. After placing further regularity assumptions upon the curvature and gradient of indierence curves and bounding
feasible allocation sequences above and below, Balasko and Shell (1980) prove
the following.
Theorem 64 (Balasko
P and Shell) A weakly Pareto optimal allocation x is Pareto
optimal if and only if t kp1t k = +, where p is the price sequence that supports
x.
Proof. See Balasko and Shell (1980).
P
It should be noted that lim inf t kpt htt k = 0 implies t kp1t k = + but
not the converse. For instance, in the Samuelson economy with no population
growth, so that n = 0, if trades could be organised the Pareto optimal equilibrium would be supported by the price sequence p = (1, 1, ....) which satisfies
P
1
t t
t kpt k = + but not lim inf t kp ht k = 0. This characterisation of Pareto
optimality will be employed further in the discussion of dynamics for overlapping
generations economies.
General demographic structure
The demographic structure of two-period lifetimes may seem unduly restrictive
and potentially responsible for some of the more surprising conclusions. However, this is not the case and, as shown by Balasko, Cass and Shell (1980), it
is possible to transform economies with more general demographic structures
412
into two-period economies by suitable relabelling. The restriction to two period

lives therefore involves no loss of generality.
The basis for this procedure is to distinguish between real calendar time
and (artificial) time used for labelling lifespans. Let calendar time be labelled
s = 0, 1, 2, ... . At s = 0, out of the set of consumers alive at that time, the
consumer with the greatest future longevity is identified. It is then assumed
that s = 0 represents the end of the first half of their life. The second half of
their life ends at their date of death, say s1 . Economic time is then constructed
so that t = 1 is associated with s1 . The same process is repeated at s1 : the
consumer with the greatest longevity is identified and viewed as being at the
end of their first period of life. Their date of death, s2 , is then associated with
t = 2.
Proceeding in this way, the economic periods are constructed so the consumers live at most two periods. It is possible that some may be born, for
instance, at a date s > 0 and die at s < s1 . Such consumers are regarded as
having single-period lifespans. The construction is completed by treating all
commodities available between s = 0 and s = s1 as being traded at t = 1. This
is repeated for t = 2, ... . Preferences are then defined over vectors of goods at
t and t + 1.
13.2.4
Money, dynamics and indeterminacy
As noted in the introduction, overlapping generations economies dier from

finite economies in more ways than simply that their competitive equilibria
may be inecient. Three of these further dierences are now briefly discussed.
Money
In a standard competitive economy that exists over a finite number of periods,
fiat money whose only use is as a store of value would be worthless. This
conclusion can be established by a simple backward induction argument. Money
is clearly worthless in the final period since it has no further use as a store of
value. Seeing that it is worthless in the last period, no consumer would wish to
purchase money in the second to last period so that it is also worthless in that
period. This argument can be continued backward until the beginning of the
economy. Money is therefore worthless unless it is arbitrarily assumed that it
must be used as a means of exchange.
This backward induction argument cannot be applied to the infinite overlapping generations economy and the possibility then arises that fiat money may
have value. This can be shown most clearly by returning to the Samuelson
economy. The ineciency in that economy arose because consumers have no
opportunity for turning their first-period endowment into second-period consumption. Money provides such a possibility. As first shown by Samuelson
(1958), the introduction of money can permit the attainment of Pareto optimality provided that all generations believe money to be valuable and are
willing to accept money in exchange for goods.
413
Figure 13.2: Ecient Allocation

This argument can be illustrated simply as follows. Assume that each consumer is endowed with one unit of the single consumption good when young
and none when old and that if generation t has Ht identical members then
Ht+1 = [1 + n] Ht . In each period the feasible allocations are given by
Ht xtt + Ht1 xtt1 = Ht ,
(13.13)
or, in per capita terms,

xtt +
xtt1
= 1.
[1 + n]
(13.14)
From (13.13) the structure of the population ensures that the feasible allocation
to any individual must be constrained by
xtt +
xt+1
t
= 1.
[1 + n]
(13.15)
This constraint is shown in Figure 13.2. As already argued, the only competitive
equilibrium
for
this economy is the autarkic outcome with consumption allocation xtt , xt+1
= (1, 0) whereas the ecient outcome is shown as the tangency
t
between indierence curve and constraint (13.14) at point o.
Introducing money increases the set of potential trades consumers may make.
Suppose at time 0 a quantity of M units of money is divided equally between
those already alive (the generation who will be old in period 1) and that they
and all following generations believe this money to have value. In period t a
consumer then solves the maximisation
max U xtt , xt+1

= Mtd , (13.16)
subject to pt = Mtd + pt xtt , pt+1 xt+1
t
t
t+1
{xtt ,xt }
implies a demand function for
where Mtd is money demand. This maximisation
Mtd
pt
real balances of the form pt = M pt+1 so that equilibrium on the money
414
market in period t requires

Ht M
pt
pt+1
pt = M.
(13.17)
Equating this to money demand in period t + 1, which must equal the same
stock, provides the equilibrium condition
pt+1
pt
pt
= [1 + n] M
M
.
(13.18)
pt+1 pt+1
pt+2
pt
= ppt+1
1 + , this equation is solved
At a steady state solution with pt+1
t+2
when 1 + = 1 + n. Combining the budget constraints in (13.15) and employing
the solution to (13.17) shows that at the steady state with 1 + = 1 + n, each
consumer faces the budget constraint
xtt +
xt+1
xt+1
t
= xtt + t
= 1.
[1 + ]
[1 + n]
(13.19)
The constraint in (13.18) is identical to (13.14). Faced with this budget constraint, each consumer will choose the ecient consumption point o. This observation demonstrates how the introduction of money that is valued allows
the decentralisation of the Pareto ecient outcome by extending the range of
allocations that can be sustained through competitive behaviour. It should be
noted that the discussion has been restricted to the steady state.
The fact that money can have value in an overlapping generations economy
is another surprising aspect of these economies. Equally surprising are the
potential eciency gains from the introduction of a commodity with no actual
consumption value. Although money will not receive further analysis in the
chapters that follow, these observations provide an insight into the results that
can emerge. Further properties of monetary equilibria are discussed in Balasko
and Shell (1981) and Hahn (1982).
Dynamics
The analysis to this point has characterised the steady state solutions of the
Samuelson economy and investigated the existence and welfare properties of
equilibria for the generalised economy. The intention now is to consider the
possible dynamics of non-steady state solutions. The dynamics of overlapping
generations economies were first investigated by Gale (1973), who provided an
example of an economy with two period cycles, and then in detail by Grandmont (1985). The work of Grandmont revealed the potential complexity of the
dynamics that can arise.
The structure of dynamics can most easily be seen by considering a slightly
modified version of the Samuelson economy. It is now assumed that there is a
single consumer in each generation and that an endowment is received in both
periods of life. In addition, the preferences of all consumers are identical and
415
can be represented by a utility function that is separable between consumption

in the first and second periods of life.
The budget constraint for each consumer is given by
[1 + rt+1 ] tt xtt + t+1

xt+1
= 0,
(13.20)
t
t
so that maximisation of the utility function U xtt , xt+1

= U 1 (xtt ) + U 2 xt+1
t
t
results in the necessary condition

U 10 xtt [1 + rt+1 ] U 20 xt+1

= 0.
(13.21)
t
Eliminating [1 + rt+1 ] by using the budget constraint (13.19) reduces (13.20) to

20 t+1
t
xt+1
U xt
= 0.
(13.22)
t xtt U 10 xtt + t+1
t
t
The aggregate feasibility condition for the economy requires that total consumption does not exceed the total endowment or
tt xtt + tt1 xtt1 = 0.
(13.23)
Taken together, (13.21) and (13.22) fully describe the dynamic evolution
of the economy from any feasible initial starting value of x10 . Given x10 , (13.22)
determines x11 and then, using the value of x11 , (13.21) determines x21 . Returning
to (13.22), x22 can then be found. Repeating this process, whilst taking account
of nonnegativity constraints, provides the entire sequence of consumption levels
for the economy. Although some of the consumption patterns that will arise in
this dynamic process appear not to be feasible trades given what has been said
above, two modifications can overcome this objection. As noted by Gale (1973),
a fictitious central clearing house could be introduced that can organise trades
that would not otherwise be possible. Alternatively, money could be introduced
with only minor modification to the analysis.
The above argument provides a simple constructive approach to the dynamics but it provides limited insight into the dynamic processes that may emerge;
for example it neither confirms or refutes the possible existence of periodic cycles. To proceed further in this direction, (13.22) can be used to substitute for
xtt in (13.21) giving

20 t+1
t
xt+1
U xt
= 0. (13.24)
xt1 tt1 U 10 tt + tt1 xtt1 + t+1
t
t
The structure of (13.23) is (implicitly) that of a nonlinear first-order dierence

to xtt1 . Hence given the consumption in the second
equation relating xt+1
t
period of life for members of generation t 1, (13.23) determines the second
period consumption of generation t. First period consumption levels can then
be read from (13.22). If it is further assumed that all generations receive an
identical pattern of endowments, so that tt = 1 , t+1
= 2 all t, (13.23)
t
reduces to
t

20 t+1
xt1 2 U 10 1 + 2 xtt1 + 2 xt+1
U xt
= 0.
(13.25)
t
416
The relevance of this assumption is that if xs+1

= xt+1
then xss = xtt so that
t
s
if the consumer of generation s has the same second period consumption as
that of t, they also have identical first period consumption. The structure
of the economy then ensures that generations s and t face identical nominal
variables and choose identical real variables. If such equivalence occurs every k
generations, the economy can be said to possess a cycle of period k.
The next step is to show that such cycles can actuallyoccur. Gale
(1973)
t
demonstrated that if the utility function has the form U xtt , xt+1
=
10x
t
t
t t+1
t+1
t+12
t2
and the endowment t , t = (0, 2) then there exists
a
4xt + 4xt xt
5 5
5+ 5
t+2
t+1
t
cycle of period 2 with consumption levels xt = 6 = xt+2 , xt+1 = 6 =
xt+3
t+3 . One of the interesting features of such a cycle is that it is driven entirely
by equilibrium behaviour under certainty and is not the consequence of random
shocks hitting the economy or of false expectations.
Now return to (13.24). Theorem 13.1 shows that (13.24) has two steady state
solutions, that is there are two values
of xtt1 such that xtt1 = xt+1
t . Hence,
t+1
t
if the locus of pairs of xt1 , xt
that solve (13.24) are drawn in xtt1 , xt+1
t
space, the locus must cross the 45o line twice. The gradient of the locus can be
calculated as
U 10 1 A1 2 xtt1
dxt+1
t
,
(13.26)
= 20
dxtt1
U 1 A2 xt+1
1
t
i00
where Ai = UU i0 is the coecient of absolute risk aversion in period i of life.

At the autarkic steady state with consumption equal to the endowment the
10
o
gradient is equal to U
U 20 > 0. The locus therefore cuts the 45 line from below
at the autarkic steady state and hence must cut it from above at the Pareto
ecient steady state.
If the second period endowment is suciently low that the ecient steady
state level of second period consumption is greater than the endowment, then
the locus of solutions to (13.24) will appear as in Figure 13.3. Point a denotes
the autarkic equilibrium and b the ecient steady state. In particular, if the
second period endowment is 0 then the autarkic steady state will be the origin.
Starting with a value of x10 on the horizontal axis, x21 can be found on the
vertical axis by reading o the graph of the curve. Translating this value through
the 45o line back onto the horizontal axis then allows x32 to be plotted on
the vertical. In this way, the dynamics of consumption in second-period of
life for each generation of consumers can be traced on the diagram. There
is a significant mathematical literature on systems that have the form shown
in Figure 13.3 and only a sketch of the most important will be given here.
Further details of the mathematics can be found in Collet and Eckmann (1980)
and simple surveys are given in Baumol and Benhabib (1989) and Boldrin and
Woodford (1990).
The possible forms of dynamic behaviour that the system may display can
be placed into four general categories. These are:
1. Convergence
The system may converge to a steady state.
417
Figure 13.3: Dynamic Solution

2. Convergence to a cycle
After an initial period the system settles down to a cycle in which the same
points are visited after specific intervals. An example of an economy with a
cycle of period 2 has already been described.
3. Non-cyclical
Here the system never returns to the same point twice. The system may
simply be erratic or it may be drawn to a strange attractor in which case it
becomes arbitrarily close to some cycle.
4. Chaotic
In the case of chaos the system never returns to the same point and has
sensitive dependence on initial conditions. That is, two solution paths that start
at two initial points an apart will, after a finite time, be a significant distant
apart. This has the implication that such a system cannot be investigated by
numerical simulation since rounding errors will be sucient to invalidate the
analysis.
To delineate conditions under which each of these would arise requires considerable further formalisation which is not worth undertaking here. However,
two further points are worth noting. Firstly, if for some starting point the system has a cycle of period 3, then starting points can be found that will give a
cycle of any desired order. This is an implication of Sarkovskiis Theorem: if a
system has a cycle of period 3, it has cycles of all orders. Grandmont (1985)
actually provides conditions under which such a cycle exists in an overlapping
generations economy. Secondly, it is dicult to determine if an observed chaotic
times series is generated by a deterministic system or is actually random.
There are two important conclusions that follow from this discussion. The
first is economies may be constructed whose dynamics that reproduce the properties of observed economic time series. Observed series generate cyclical behaviour of fairly stable periodicity. Such cycles can be generated by few economies
other than the overlapping generations economy, except for exceptional sets of
parameter values. This gives the overlapping generations economy an important
role in the analysis of business cycles. Secondly, the fact that the economy can
418
generate endogenous cycles provides a natural motivation for counter-cyclical

policy. The policy implications of these dynamics have been investigated in
detail by Grandmont (1985).
In the context of policy reactions to cycles it may seem surprising to note
that the cycles, with the exception of the autarkic 1-cycle, are actually Pareto
optimal. The argument behind this conclusion is the observation that if the
economy is cyclical equilibrium prices must also be cyclical. During the time
span of the economy the samePprices must therefore be repeated an infinite
number of times. This implies t kp1t k = +, hence the cycle must be Pareto
optimal.
Indeterminacy
A theorem of Debreu (1970) shows that almost all competitive economies of the
form described in Chapter 2 have a finite number of isolated equilibria so that
each equilibrium is locally unique. That a similar conclusion does not apply to
overlapping generations economies has been shown by Gale (1973), Geanakoplos and Polemarchakis (1984) and Kehoe and Levine (1985). For overlapping
generations economies robust examples can be constructed that have a continuum of equilibria so that both finiteness and local uniqueness do not apply.
Such indeterminacy imposes considerable problems for analysis. Comparative
statics exercises are not possible and the concept of perfect foresight along such
indeterminate paths is implausible. The introduction of money compounds the
problem by introducing a further dimension to the indeterminacy.
The problem of indeterminacy is inherent in all applications of overlapping
generations economies. It is sidestepped in the chapters that follow, as well as
in much of the literature, by focusing only upon steady states. As has already
been shown, there are just two steady states for the Samuelson economy and in
the aggregate production economy discussed next it is possible for there to be
a unique steady state.
13.2.5
Summary
This section has described the overlapping generations economy and has illustrated a number of the important properties that such economies possess. In
contrast to standard competitive economies, the equilibria of an overlapping
generations economy need not be Pareto optimal, though all are weakly Pareto
optimal, and there may be an uncountable infinity of equilibria. Furthermore,
fiat money can play a socially useful role in leading to the attainment of a Pareto
optimum and, consequently, may be valued.
These features of overlapping generations economies undermine many of the
presumptions developed from analysis of standard competitive economies. Due
to this, it has proved an important tool in the study of public economics. The
version of the overlapping generations economy that will be employed in the
following chapters is less general than that described in this section but the
results here provide the foundation of the analysis.
13.3. AN AGGREGATE PRODUCTION ECONOMY
13.3
419
An aggregate production economy
The overlapping generations economies described above involved no production

but only exchange of commodities. Each generations of consumers received
their endowment and then traded. In contrast, the overlapping generations
economy described in Diamond (1965) introduced a production process that
involved both capital and labour. This extension permits the analysis of many
interesting topics, including the assessment of the eciency of choice of capitallabour ratio.
In the Diamond economy each consumer again lives for just two periods.
They work only during the first period of their life and inelastically supply
one unit of labour. This unit of labour is their entire endowment. In their
second period of life they are retired and supply no labour. Income earned by
a consumer during the first period of their life is divided between consumption
and savings. Second period consumption is equal to savings plus accumulated
interest. With the exception of their date of birth, all consumers are identical.
At each date is born a cohort of consumers and the population grows at a
constant rate.
The economy has a single consumption good which is produced using capital
and labour. Available capital consists of a stock of the single good and is generated by the savings of the previous period. Capital does not depreciate during
the production process. The production function exhibits constant returns to
scale. The level of production is chosen so as to maximise profits. Finally, all
markets are competitive. The existence of capital as a store of value allows
consumers to carry purchasing power from one period to the next.
13.3.1
Consumers
All consumers have identical preferences which, for the consumer born in period
t, are represented by the utility function
,
U = U xtt , xt+1
t
(13.27)
where xtt is the consumption level of the single good when the consumer is young
and xt+1
consumption when old. There is no disutility from labour supply.
t
To construct the budget constraint of a typical consumer, note that labour
income must be divided between consumption and savings and that each consumer supplies one unit of labour. With the price of the consumption good in
period t denoted pt , consumption and savings, st , must satisfy
Wt = pt xtt + pt st ,
(13.28)
where Wt is the wage received for the single unit of labour. The value of second
period consumption must be equal to the value of savings, hence
+ pt st .
pt+1 xt+1
t
(13.29)
420
Combining (13.27) and (13.28) and defining the interest rate by

rt+1 =
pt pt+1
,
pt+1
(13.30)
the budget constraint becomes

Wt
xt+1
t
wt = xtt +
,
pt
1 + rt+1
(13.31)
where wt is the real wage. Employing the definition of the interest rate also gives
xt+1
= [1 + rt+1 ] st . Note that the relevant interest rate is rt+1 since interest is
t
notionally paid in period t + 1.
From (13.26) and (13.29) the utility-maximising consumption plan satisfies
the first-order condition
U1 xtt , xt+1
t
= [1 + rt+1 ] .
(13.32)
U2 xtt , xt+1
t
Equation (13.30) relates the intertemporal marginal rate of substitution to the
rate of transformation and is a simple extension of the standard condition for optimal choice. The simultaneous solution of (13.30) and (13.29) provides demand
functions of the form
xt+i
= xt+i
(wt , rt+1 ) , i = 0, 1.
t
t
13.3.2
(13.33)
Production
The productive sector of the economy is described by an aggregate production

function
Yt = F (Kt , Lt ) ,
(13.34)
where Kt is the capital stock in period t and Lt aggregate labour supply. Since
each young consumer inelastically supplies one unit of labour, Lt = Ht . The
production function can either be interpreted literally as representing the technology of a single firm or as the aggregate production function of a set of identical firms. A formal demonstration of such an aggregation argument is given in
Sargent (1979). As capital does not depreciate, consumption and saving must
satisfy
F (Kt , Lt ) + Kt = Ht xtt + Ht1 xtt1 + Ht st .
(13.35)
The production function therefore gives net output not including the nondepreciating capital.
It is assumed that F (Kt , Lt ) satisfies constant returns to scale and is therefore homogeneous of degree 1 so that F (Kt , Lt ) = F (Kt , Lt ) for all positive
. As the size of the population is growing at rate n, what matters as a measure
of the output of the economy is not total output but output per head. The homogeneity of production permits the analysis to be phrased in these terms. To

see this, set =
of . This gives
Lt
Lt
Yt = Lt F
421
in (13.33) and use the homogeneity to extract the numerator

1
1
Kt , Lt
Lt
Lt
= Lt F
Now define
yt =

Kt
Kt
, 1 = Lt f
.
Lt
Lt
Yt
Kt
, kt =
.
Lt
Lt
(13.36)
(13.37)
Then output per unit of labour is determined by a function that has the capitallabour ratio as its sole argument
yt = f (kt ) .
(13.38)
As the labour market is competitive, profit maximisation in the choice of

labour by firms implies that the marginal product of labour must be equated to
the real wage. The marginal product of labour is derived by noting that

Kt
Yt = Lt f
,
(13.39)
Lt
so that
yt
=f
Lt
Kt
Lt
Lt f 0
Kt
Lt
Kt
= f (kt ) kt f 0 (kt ) .
L2t
(13.40)
The optimum choice of labour therefore satisfies

wt = f (kt ) kt f 0 (kt ) .
(13.41)
Similarly, the optimum choice of capital equates the rate of interest to the
marginal product
rt = f 0 (kt ) .
(13.42)
13.3.3
Equilibrium
At the equilibrium it is necessary that consumers maximise utility, firms maximise profit and all markets clear. For capital market equilibrium, the relevant
condition is that capital used must be equal to the level of savings, since capital
is the only store of wealth. To derive the capital market equilibrium condition,
first note that the population of young consumers is of size Ht in period t, so
the equality of total savings with capital available in period t + 1 implies that
(13.43)
Ht wt xtt = Kt+1 .
Dividing through (13.41) by Ht , recalling that Ht+1 = Ht [1 + n] and Ht = Lt ,

gives
wt xtt = kt+1 [1 + n] .
(13.44)
422
When equation (13.42) is satisfied, there is equilibrium in the capital market.

Collecting equations, the equilibrium evolution through time of the economy
is determined as the simultaneous solution to
U1 xtt , xt+1
t
= [1 + rt+1 ] ,
(13.45)
U2 xtt , xt+1
t
wt = xtt +
xt+1
t
,
1 + rt+1
(13.46)
wt = f (kt ) kt f 0 (kt ) ,
(13.47)
rt = f 0 (kt ) ,
(13.48)
wt xtt = kt+1 [1 + n] .
(13.49)
and
These equilibrium equations can be used to provide a simple description
of the evolution of the capital stock through time which can then be used to
generate the time paths of the other endogenous variables. To do this recall that
the simultaneous solution of (13.43) and (13.44) provides demand functions of
the form
xt+i
= xt+i
(wt , rt+1 ) , i = 0, 1.
(13.50)
t
t
From the determination of wt described in (13.45), and from stepping (13.46)
one period forward, it follows from substitution into the demand function that
xtt = xtt (f (kt ) kt f 0 (kt ) , f 0 (kt )) .
(13.51)
Now substituting (13.49) into the capital market equilibrium equation (13.47)
gives the final expression
f (kt ) kt f 0 (kt ) xtt (f (kt ) kt f 0 (kt ) , f 0 (kt+1 )) = kt+1 [1 + n] .
(13.52)
Equation (13.50) has as its only arguments the capital-labour ratio in periods
t and t + 1 and it represents the basis for studying the dynamic equilibrium of
the economy. Assuming that (13.50) can be solved for kt+1 in terms of kt , the
relation can be written
kt+1 = (kt ) .
(13.53)
The intertemporal evolution of the capital-labour ratio is summarised in (13.51)
in the sense that it determines kt+1 for given kt . As with (13.23), (13.51) is a
nonlinear first-order dierence equation and the potential dynamics that were
discussed in connection with that equation are again relevant here.
To analyse (13.51) first assume that, for all kt ,
kt+1
(13.54)
kt < 1,
423
Figure 13.4: Stable Convergence to the Steady State

which, after implicitly dierentiating (13.50), will hold if
i
h t
xt 1 k f 00
t
wt
< 1.
xtt
00
1 + n + rt+1 f
(13.55)
In this case, if the system given by (13.50) has a stationary solution with kt =
kt+1 = k all t, it will be globally stable. That is, if there is a fixed point such
that
f (k) kf 0 (k) xtt (f (k) kf 0 (k) , f 0 (k)) = k [1 + n] ,
(13.56)
then the system will always converge to k regardless of the initial level of the
capital stock. A stable system of this form is illustrated Figure 13.4.
The system in Figure 13.4 also has the property that the (kt ) function is
monotonically increasing. This may be a reasonable restriction and it certainly
rules out many forms of badly-behaved solution. Sucient conditions for the
function (kt ) to be increasing in kt are that
xt
xtt +1
> 0,
a) wtt > 0, w
t
and t
xt
b) rt+1
< 0.
Condition (a) is the requirement that both goods are normal. The imporxt
tance of (a) is that it implies wtt < 1. Together conditions (a) and (b) guarantee
that the numerator and denominator of (13.53) are positive. Of course further
restrictions are needed to ensure that the derivative is less than 1.
Alternatively assume that for some values of kt
kt+1
(13.57)
kt > 1,
and that (kt ) is first increasing and then decreasing with a unique maximum
which can occur if second period consumption is inferior. This provides the
424
hump-shaped relationship of Figure 13.3 and permits the possibility of the complex dynamics discussed following that figure. The Diamond economy can therefore potentially generate convergent, cyclical, non-cyclical and chaotic behaviour
in the capital-labour ratio and hence in output per head.
13.3.4
Steady State Solution
Although the dynamic solution is of much potential interest, the focus of most
analyses of overlapping generations economies in public economics has been
placed upon steady state equilibria. A steady state is the situation in which the
economy repeats itself period after period in the sense that the capital-labour
ratio, the output-labour ratio and the interest rate are constant over time. The
steady state is typically interpreted as constituting the long-run equilibrium
for the economy. The typical policy analysis characterises the steady state
and performs comparative statics exercises as the policy variables are modified.
Such exercises will be the subject matter of Chapters 14 and 15. The present
concern is the characterisation of the steady state and the analysis of its welfare
properties.
Since all nominal variables and per capita variables are constant in the steady
state, the time subscripts are now dropped. Denoting the steady state value of
the capital-labour ratio by k and the interest rate by r, the budget constraint
can be written
x2
x1 +
= w,
(13.58)
1+r
where x1 and x2 are first- and second-period consumption levels. Using the facts
that w = f (k) kf 0 (k) and r = f 0 (k), the budget constraint can be written
x1 +
x2
= f (k) kf 0 (k).
1 + f 0 (k)
(13.59)
From capital market equilibrium

w x1 = [1 + n] k,
(13.60)
f (k) kf 0 (k) x1 = [1 + n] k.
(13.61)
or
The points to be made about the properties of the steady state can be derived
from using (13.59) to eliminate x1 from (13.57). Dierentiating the equation
obtained gives
1
dk
=
> 0,
(13.62)
2
dx
[1 + n] [1 + f 0 + kf 00 ]
under the assumption that 1 + f 0 + kf 00 > 0. It is therefore
possible to solve for
k as a function of x2 . This solution is denoted k = k x2 . The next step is to
substitute the solution for k into (13.59) to give

(13.63)
f (k x2 ) k x2 f 0 (k x2 ) x1 = [1 + n] k x2 .
425
Figure 13.5: Consumption Possibility Locus

1 2
The solutions to equation (13.61) generates a locus of pairs
x , x termed
the consumption possibility frontier and, using k = k x2 , an implied value

of k. Each point on this frontier is potentially a steady state with the actual
steady state that arises as the competitive equilibrium being determined by
consumer preferences. There are basic economic reasons for expecting the locus
to describe a non-monotonic relationship between x1 and x2 . For low values of
x2 the capital-labour ratio is also low so that total production, and hence x1 ,
must also be small. As x2 increases the capital-labour ratio grows and permits
x1 to rise. Eventually, the capital-labour ratio will become too large and x1 will
again fall. As an example, for the Cobb-Douglas production function
f (k) = k , 0 < < 1,
(13.64)
the locus has the shape shown in Figure 13.5.

The importance of this construction is the following interpretation. The
long-run steady state will generate some
value of k as the equilibrium of the
economy. This value of k implies a pair x1 , x2 on the consumption possibility
locus. The question then arises as to whether
the
k determined in this manner,

or equivalently the consumption pair x1 , x2 , has any eciency properties.
That is, are some values of k preferable to others and, if so, will the economy
have an optimal value of k as its equilibrium? The answers to these questions
are considered next.
13.3.5
Golden rules
It is possible to define the optimal value of k in a number of ways. The simplest

is to view the optimum as the level of the capital-labour ratio that maximises
per capita consumption in the steady state. The relation that this level of
capital satisfies is termed the Golden rule and the resulting capital-labour ratio
is the Golden rule level. The analysis of the Golden rule is undertaken in 3.5.1.
Alternatively, optimality can be defined in terms of the maximisation of a social
426
welfare function. When the social welfare function is chosen as a discounted

sum of utilities, the modified Golden rule is obtained. This is derived in 3.5.2.
Such rules are important throughout the economic theory of economic growth.
The Golden rule
This section considers the maximisation of consumption per period or, equivalently, consumption per capita, a procedure that leads to the Golden rule level
of the capital-labour ratio. The following section will relate the Golden rule to
Pareto optimality and dynamic eciency.
The objective of the central planner is
max xtt Ht + xtt1 Ht1 ,
which, since Ht1 =
Ht
1+n ,
(13.65)
can be expressed in per capita terms as

max xtt +
xtt1
.
1+n
Since the maximisation will be restricted to the choice between steady states,
(13.64) reduces to
x2
max x1 +
.
(13.66)
1+n
The constraint facing the central planner is that consumption in any period
must be equal to total output less additions to the capital stock. Hence
xtt Ht + xtt1 Ht1 = Ht f (kt ) Ht [kt+1 [1 + n] kt ] .
At a steady state equilibrium (13.66) reduces to
x1 +
x2
= f (k) nk.
1+n
(13.67)
The optimisation in (13.65) is therefore equivalent to max{k} f (k) nk. From

this maximisation the first-order condition
f 0 (k) n = 0.
(13.68)
is obtained. Hence the optimal k, k , should be chosen such that

f 0 (k ) = n.
(13.69)
The capital-labour ratio k is termed the Golden rule capital-labour ratio and
it is the optimal ratio in the sense that it maximises consumption per head.
Returning to the competitive economy, since f 0 = r, if the competitive
economy reaches a steady state equilibrium with r = n, this equilibrium will
satisfy the Golden rule. Since no other equilibrium will, this identifies r = n
as the Golden rule rate of interest. To understand the structure of the Golden
427
Figure 13.6: Golden Rule Competitive Equilibrium

rule competitive equilibrium, (13.57) and (13.59) can be used to show that the
gradient of the consumption possibility locus is given by
kf 00 + 1 + n
dx1
=
.
2
dx
[1 + n] [1 + f 0 + kf 00 ]
When f 0 = n, (13.70) reduces to
dx1
dx2
(13.70)
1
1+n . When facing a rate of interest
U1 (x1 ,x2 )
that U2 (x1 ,x2 ) = 1 + n. Therefore at the
r = n consumer maximisation implies

Golden Rule the consumers budget constraints, maximal indierence curves and
the consumption possibility locus have identical gradients. This is illustrated in
Figure 13.6.
The modified Golden rule
The modified Golden rule is derived by choosing the growth path of the economy
to maximise a social welfare function defined as the discounted sum of future
utilities. This social welfare function is denoted by
X
t=0
t U xtt , xt+1
.
t
(13.71)
Although the size of population in each generation does not enter explicitly into
(13.71), it can be incorporated via the definition of the discount factor .
The objective of the social planner is to maximise (13.71) subject to the production constraint upon the economy which is summarised in (13.66). Dividing
through by Ht , (13.66) becomes
kt + f (kt ) = [1 + n] kt+1 + xtt +
xtt1
.
1+n
(13.72)
s
Substituting into (13.71) for xs1
s1 and xs using (13.72), social welfare can be
428
written
X
t=0
s2
X
t U xtt , xt+1
t U xtt , xt+1
=
t
t
t=0
!
s1
x
s2
+ s1 U ks1 + f (ks1 ) [1 + n] ks
, xs
1 + n s1
xs
+ s U ks + f (ks ) [1 + n] ks+1 s1 , xs+1
1+n s
+
t U xtt , xt+1
.
(13.73)
t
t=s+1
From (13.73) the optimal choice of xss1 and ks is determined by
s1 s
s U xss , xs+1
s
s1 U xs1 , xs1
= 0,
xss1
1+n
xss
(13.74)
and
s1 U
s1 s
xs1 , xs1
xs1
s1
[1 + n] +
s U
s s+1
xs , xs
[1 + f 0 ] = 0.
xss
(13.75)
Along the optimal growth path, (13.74) and (13.75) must hold for all values of
s.
Moving to the steady state, (13.74) and (13.75) become
U1 = 0,
1+n
(13.76)
1 + n = [1 + f 0 ] .
(13.77)
U2
and
Equation (15.77) is the modified Golden rule which relates the optimal
capital-labour ratio to the rate of population growth and the discount factor.
Since social welfare defined by (13.71) is only properly defined if < 1, the
modified Golden rule results in a lower capital-labour ratio than the standard
Golden rule. This is due to the impatience implied by the discounting leading
to higher consumption in the present and less accumulation. The interpretation
of (13.76) is that the steady state must involve optimal intertemporal allocation of consumption for each consumer when faced with an implied interest rate
satisfying 1 + r = 1+n
.
13.3.6
Pareto optimality
Having now characterised the Golden rule capital-labour ratio and corresponding rate of interest, it is possible to relate this to the question of Pareto optimality. To do this, first note that if k > k then r < n. The converse is
13.4. CONCLUSIONS
429
true if k < k . This is a simple consequence of the concavity of the production

function. These two cases will now be discussed in turn, beginning with k > k .
A capital-labour ratio above k represents an economy that has over-accumulated
along the growth path. In such an economy it would be feasible for the consumers alive in any period with k > k to consume some of the existing capital
stock so as to reduce the stock to the level k . Such consumption would have
two consequences. It would raise the welfare of the existing generations because
it increases their present consumption at no cost. Secondly, it would raise the
welfare of all following generations because it would place the economy on the
Golden rule path and would consequently maximise their consumption. Thus,
such consumption of the excess of the capital stock above the Golden rule level
would lead to a Pareto improvement. Therefore, any steady state with k > k
and r < n is not Pareto optimal. Such states are called dynamically inecient.
When k k no such Pareto improvements can be found. Any act of investment that raises the welfare of the following generations must be undertaken at
a cost to the existing generations in foregone consumption. Therefore all such
states, including the Golden rule and modified Golden rule states, are Pareto
optimal and are termed dynamically ecient.
The importance of this discussion is that when r < n the equilibrium is
not Pareto optimal despite the economy satisfying all the standard competitive
assumptions. To illustrate that the failure of Pareto optimality does not require
an unusual structure to be placed on the economy, consider the following example from Diamond (1965). The utility function of the single consumer in each
generation is given by
U xtt , xt+1
= log xtt + [1 ] log xt+1
,
(13.78)
t
t
and the production function is y = Ak . The steady state interest then can be
calculated to be
[1 + n]
r=
.
(13.79)
[1 ] [1 ]
This will only be equal to the Golden rule rate when

n=
.
[1 ] [1 ]
(13.80)
If preferences and production do not satisfy this condition, and there is no

reason why they should, the economy will not grow on the Golden rule growth
path. This example illustrates that a Golden rule economy will be the exception
rather than the norm.
13.4
Conclusions
Overlapping generations economies have a number of interesting features that

are not shared with standard Arrow-Debreu economies. The competitive equilibria can be inecient, complex endogenous dynamics can be generated, fiat
430
money may be valuable as a store of value and its existence can raise the level
of welfare and indeterminacy in equilibria may exist. In the context of public
economics, it is the first of these that is arguably the most important since it
provides a role for corrective policies in the absence of any additional market
failure.
More importantly, the structure of the economy permits the analysis of the
eects of policies that are essentially intertemporal in nature. This will be
utilised in the analysis of social security in Chapter 14 and debt and taxation in
Chapter 15. In both cases, the nature of ineciency in overlapping generations
economies and the structure of Golden rules will be central in explaining the
results of policy analysis.
Chapter 14
Social Security
14.1
Introduction
The provision of social security to provide cover against disability and the inability to work due to old age is a feature of all developed economies. Such programs
are large, both in terms of the proportion of population receiving benefits and in
terms of the total payments as a proportion of national income. The programs
are not without their diculties. Thompson (1983) describes the adjustments
made to the US program following overly optimistic forecasts of real earnings
growth. The expected increase in the ratio of retired to employed due to greater
life expectancy will also place the system under pressure. There is also evidence,
see Kotliko (1989), that social security programs are required due to the inadequate savings and insurance purchases of the elderly which would not support
them through retirement. These observations show that the analysis of social
security and its economic impact is a subject of practical importance.
The first issue in the analysis of social security is its eect upon the equilibrium of the economy and, particularly, upon the level of the capital stock.
If a social security program has the form of a forced saving program, so that
consumers are provided with greater second period earnings than they would
naturally choose, then the program may raise the capital stock. This outcome
will be beneficial in an undercapitalised economy. Conversely, if the program
simply transfers earnings from those who are working to those who are retired,
savings and hence the level of capital may fall. It can be judged from the dierence in outcomes of these simple scenarios that the consequence of the existence
of social security is closely dependent upon the programs structure. In addition to its eects on savings, the interaction between social security and the
retirement decision may also be significant.
A second major issue that arises is the eect of demographic change upon
the social security program. Present trends are for the proportion of retired
consumers to increase and for the retired to live longer. At a practical level,
this raises the question of whether the working population can continue to fund
431
432
CHAPTER 14. SOCIAL SECURITY
social security. A related, but more theoretical issue, is the question of whether
there exists an optimal rate of population growth. This issue arises from the observation that if the rate of population growth increases, there are more workers
to support each retired consumer but the level of capital per worker is reduced.
This trade-o suggests there may be an optimal growth rate.
The introduction of a social security program results in a transfer of resources towards the generation that benefits on the introduction of the program
and away from later generations. This raises the question of how such a program
receives the support that is required for it to be introduced at all. The mechanism by which the level of benefits in the program are selected also needs to
be addressed. Furthermore, the reasons why the private sector cannot provide
insurance cover on terms at least as attractive as those oered by social security
must also be addressed.
This chapter begins by setting out the important distinction between fully
funded and pay-as-you-go social security. The economic eects of these two polar
forms of program are markedly dierent. An optimal social security program is
then characterised under the strong assumptions of certainty and fixed labour
supply. A number of extensions of the basic result are considered including
an analysis of optimal population growth. Determination of the level of social
security by majority voting and various justifications for social security are
then considered including altruism, myopia and aggregate uncertainty. The
final section analyses the eect of introducing individual uncertainty about the
length of life.
14.2
Fully funded and pay-as-you-go
The financing of social security can have important implications for the economic eects of the program and for its sustainability in the face of demographic change. The purpose of this section is to define alternative structures
of financing and to broadly sketch their diering eects.
To make the definitions as precise as possible, assume that the economy
is one with overlapping generations and that each consumer lives for just two
periods. Each consumer supplies labour during their first period of life and
is retired in the second period. Finally, there is one capital good available,
purchases of which provide a repository for savings.
In a fully-funded system each consumer when young make contributions
towards social security via a social security tax and the contributions are used
to purchase capital by the social security program. Total capital in the economy
is then given by the sum of private capital and the publicly owned capital of the
social security program. Total pension benefits received by a consumer when
retired are then equal to their contribution to the program plus interest received.
Such a program satisfies the equalities
pensions = social security tax plus interest = capital plus return.
A fully-funded social security system eectively forces each consumer to save
an amount at least equal to the tax they pay. Consumers may, of course, choose
14.3. AN OPTIMAL PROGRAM
433
to save more. If, in the absence of social security, all consumers chose to save an
amount in excess of the taxed levied by the program then, holding all else constant, a fully-funded system will simply replace private saving by an equivalent
amount of public saving. If these conditions are met, a fully-funded system will
have no eect upon the equilibrium outcome. In more general settings with a
variety of investment opportunities, the possibility must be considered that the
rate of return on private savings may dier from that on public savings. When
it does, a fully-funded system may aect the equilibrium.
In contrast to the fully-funded system, a pay-as-you-go social security program does not own any capital. Instead, a pay-as-you-go system relies on the
contributions of the young of each generation to provide the pensions of the old
of the previous generation. Such a program therefore satisfies the equality
total benefits received by generation t 1 = contributions of generation t.
The system presently in operation in the US is of this form since the capital
it owns would only fund approximately two months of benefit payments (see
Thompson (1983)).
A pay-as-you-go system leads to an intergenerational reallocation of resources whereas a fully-funded system can at most cause an intertemporal reallocation for each generation. From this observation it can be seen immediately
that the two systems will have rather dierent welfare implications, some of
which will be investigated in the following sections.
Systems that fall between these two extremes will be termed non-fullyfunded. Such systems own some of the capital stock but the payments made in
a period may be greater than or less than the revenue, composed of tax payments and interest, received in that period. The dierence between the two will
comprise investment, or disinvestment, in capital.
14.3
An optimal program
The analysis of this section presents Samuelsons (1975a) characterisation of

an optimal social security program. The assumptions under which this will
be undertaken are strong. In particular, it is assumed that labour supply is
completely inelastic and that the date of retirement cannot be varied. The
relaxation of these restrictions, and others, in later sections will place the results
of this section in context.
Given that the competitive equilibrium may be non-optimal in an overlapping generations economy, there is potentially a role for a social security to
enhance eciency. In fact, in the discussion of Pareto optimality in chapter
13, a simple form of social security program was shown to be Pareto-improving
over the competitive outcome. That example involved each young consumer
transferring a unit of the good to the old consumer of the previous generation.
This series of transfers, which is essentially a pay-as-you-go social security program, raised the welfare of the generation that were old when the program was
introduced and left that of all later generations unaected. This argument is
further formalised in Aaron (1966).
434
To proceed beyond this result, consider a Diamond economy in which there

exists a social security program. The program taxes each worker an amount
and pays each retired person a pension . The programme also owns a quantity
Ks
Kts of capital. Equivalently, it can be said to own kts , kts = Ltt , of capital per
unit of labour. A social security program will be optimal if the combination of
, and kts is feasible for the program and leads the economy to the Golden
rule.
A feasible social security program must satisfy the budget identity
s
Lt1 = Lt + rt kts Lt kt+1

Lt+1 kts Lt ,
(14.1)
which states that pension payments must be equal to tax revenue plus return
on capital holdings less investment in capital. Since the population grows at
Lt
rate n, in a steady state the identities Lt1 = 1+n
, Lt+1 = [1 + n] Lt and
s
s
s
kt+1 = kt k can be used in (14.1) to generate the steady state budget
identity
(14.2)
= + [1 + r] ks [1 + n] ks .
1+n
Employing the equilibrium conditions r = f 0 (k), where k denotes the total
capital stock, the budget constraint of a consumer under the programme can be
written
x2
x1 +
= f (k) rk +
.
(14.3)
1+r
1+r
Note that the pension, , which is received in the second period, is discounted
2
. The budget identity

since x1 +s = w and [1 + r] s+ = x2 ; hence s = x1+r
of the program can be used to eliminate in the individuals budget constraint
to give
x2
1+n
(14.4)
x1 +
= f (k) rk +
[ + [r n] ks ] .
1+r
1+r
Equilibrium in the capital market requires that private savings be equal to
total capital less capital owned by the social security program. This condition
can be expressed as
w x1 = [1 + n] [k ks ] .
(14.5)
Substitution for x1 from the individuals budget constraint (14.4) gives

1+n
x2
[ + [r n] ks ] = [1 + n] [k ks ] .
1+r
1+r
(14.6)
The equilibrium of the economy is then the simultaneous solution to

x1 +
x2
1+n
= f (k) rk +
[ + [r n] ks ] ,
1+r
1+r
(14.7)
f 0 (k) = r,
(14.8)
U1
= 1 + r,
U2
(14.9)
14.3. AN OPTIMAL PROGRAM
435
and
x2
1+n
[ + [r n] ks ] = [1 + n] [k ks ] .
1+r
1+r
(14.10)
These equations represent the private budget constraint, the choice of production technique, individual choice and the capital market equilibrium respectively.
The aim now is to investigate the eect that the social security policy can
have upon the equilibrium. In particular, is it possible to design a policy that
will generate the Golden rule? To see why this may be possible it should be
noted that the failure of the competitive equilibrium without intervention to
be ecient results from the savings behaviour of individuals which may lead to
over- or under-accumulation of capital. With the correct choice of social security
program the government can eectively force-save for individuals. This alters
the steady state level of the capital stock and hence the growth path of output.
Equations (14.7) to (14.10) determine the endogenous variables k, x1 , x2 , w
and r conditionally upon the exogenous variables and ks describing the social
security programme. The solution of (14.7) to (14.10) provides the following
system
k = k ( , ks ) , x1 = x1 ( , ks ) , x2 = x2 ( , ks ) , r = r ( , ks ) , w = w ( , ks ) .
(14.11)
For a social security program to achieves the Golden rule, there must exist a
pair { , ks } that satisfies the equality
r ( , ks ) = n.
(14.12)
To see the values that should be chosen, set r ( , ks ) in the capital market
equilibrium condition (14.10) and employ the functional relationships (14.11) to
give
x2 ( , ks )
(14.13)
= + [1 + n] [k ( , ks ) ks ] .
1+n
From (14.13) can be determined the set of pairs of { , ks } that will give the
Golden rule rate of growth. Since this is one equation in two variables, there
will in general be a continuum of solutions rather than a single unique solution.
If there exists a solution for when ks = 0, then the optimum can be sustained
by a pay-as-you-go system.
The structure of this social security program is that the young in each generation give up some consumption, in the form of tax payments, to the old on the
understanding that they will receive a similar a gift when old. There is consequently an element of trust involved in the transactions that support the social
security policy. This should be contrasted to the failure of Pareto optimality in
the Samuelson economy due to the lack of intergenerational trades.
To show that the optimal program will not be fully-funded, note that a
fully-funded program must satisfy the identity
Lt1 = Lt1 [1 + r] = ks Lt [1 + r] .
(14.14)
436
The substitution of (14.14) into the equilibrium conditions (14.7) - (14.10) shows
that they reduce to the original market equilibrium conditions described in
(13.43) - (13.47). The fully-funded system therefore replaces private capital
by public capital and does not aect the consumption choices of individual
consumers. It can therefore have no real eect on the equilibrium and, if the
initial steady state were not at the Golden rule, a fully-funded social security
program cannot restore eciency.
This analysis has demonstrated how a correctly designed social security program can generate the Golden rule equilibrium, provided that it is not of the
fully-funded kind. A fully-funded system simply replaces private savings by
public savings and does not aect the growth path. In contrast a non-fully
funded system can aect the aggregate levels of savings and hence the steady
state capital-labour ratio. The results have been concerned only with the comparison between steady states. Burbidge (1983a) discusses the stability of the
steady states and presents some simulations of the adjustment paths that are
followed. The optimality result can also be extended, as in Gigliotti (1984),
to show that a social security program can be designed such that the implied
steady state maximises the discounted sum of future utilities and achieves the
modified Golden rule.
14.4
Some extensions
The optimality result of the previous section was derived under strong assumptions. Foremost amongst these was the inelastic supply of labour. One obvious
consequence of the provision of a pension is to encourage retirement and through
this mechanism to reduce labour supply. An induced increase in retirement
raises the proportion of retired to working consumers and reduces the welfare
gains obtained from the implementation of social security. The provision of a
pension will also aect the savings decision. There will be an incentive to reduce
saving since a pension is simply a substitute for private saving. Conversely, earlier retirement suggests the need to raise savings to cover the longer retirement
period. The resolution of these eects will have important implications for the
level of the capital stock. Variable labour supply, and other extensions to the
basic analysis, are now considered.
14.4.1
Labour supply and retirement
The interaction between social security provision and the retirement decision
has been analysed from both partial equilibrium and general equilibrium perspectives. Although the important results of the former will be noted, it is the
latter that is of primary interest here.
Under the assumptions of perfect capital markets, actuarial fairness and
known lifespan, Kotliko (1979) shows that the provision of social security will
not aect the retirement decision. This is simply a result of pensions being
equivalent to private savings in that the provision of pensions does not alter the
14.4. SOME EXTENSIONS
437
opportunity set of a consumer. An increase in pension simply replaces private

savings on a one-for-one basis. Relaxing each assumption in turn, Crawford
and Lilien (1981) show that the eect on the date of retirement is in general
ambiguous but a progressive system tends to advance retirement for low-income
workers. Diamond and Mirrlees (1986) focus on the problems raised by the
governments inability to distinguish those unable to work and those who choose
not to work. When consumers are forced to retire due to an inability to work,
but with no prior warning of this, it is shown that the optimal social security
program will have benefits rising with the age of retirement.
Although suggestive, these analyses do not address the interaction between
the retirement decision and the equilibrium of the economy. There are clearly
important connections between these since early retirement reduces labour supply while reduced savings lower the equilibrium capital stock. A simple extension of the Samuelson analysis by Hu (1979) that incorporates endogenous
retirement is now discussed.
The production side of the economy and the structure of the population remains as in Section 3. What distinguishes the present economy is that although
all consumers must supply one unit of labour in the first period of their life, they
can choose what proportion of the second period of life they spend working. To
provide a motive for retirement, utility is increasing in the length of the period
of retirement.
Expressed formally, the utility function is assumed to take the form
U = U xtt , xt+1
(14.15)
t , t ,
where t , 0 t 1 is the proportion of the second period of life that is spent
in retirement. Again denoting the tax paid toward the social security program
whilst working by and the pension by , the consumers budget constraint is
xtt +
xt+1
[1 t ] [wt+1 ]
t
t
= wt +
+
,
1 + rt+1
1 + rt+1
1 + rt+1
(14.16)
or, in the steady state,

x1 + x2 = w + [1 ] [w ] + .
(14.17)
Maximising utility leads to the necessary conditions

U
U
= 1,
x2
x
and
(14.18)
U
U
(14.19)
= [w ] 1 .
x
From (14.18) and (14.19) can be derived consumption demands and a retirement
decision that are dependent upon w, , and . It is assumed that the optimal
satisfies 0 < < 1. If the upper bound were attained the model would be
identical to that already analysed.
438
The total population at time t is denoted by Pt . With growth rate n, the

proportion of the total population that is in the second period of life, , is given
by
Ht1
Ht1
1
=
=
=
.
(14.20)
Ht + Ht1
[1 + n] Ht1 + Ht1
2+n
The social security program is assumed to operate on the pay-as-you-go basis.
Noting that benefits are only paid once a consumer has retired, the program
satisfies the budget identity
Pt = Lt = Pt l ,
(14.21)
where L is total labour supply and l is average labour supply given by [1 + [1 ]]

= [1 ] since proportion 1 of the population work full-time and proportion
work for fraction 1 of the time. Equation (14.21) reduces to
= [1 ] .
(14.22)
The decision problem of firms is unaected by the existence of the social

security program. In the steady state capital and labour demands satisfy r =
f 0 (k) and w = f (k) kf 0 (k), where the capital-labour ratio k = K
L . These
equations can be solved to express the interest rate as a function of the wage
rate or w = (r). This relationship is termed the factor price frontier and is
the locus of pairs of w, r that are consistent with equilibrium for firms. Since
dr = f 00 dk and dw = kf 00 dk, dw
dr = k.
The optimal social security program can be derived by maximising (14.15)
with respect to the level of the pension, , taking into account the dependence
of the choice variables on and the budget constraint of the program. Dierentiating (14.15) and using (14.18) and (14.19) gives
dU
dU dx1
dx2
d
= 1
+
+ [w ]
,
(14.23)
d
dx
d
d
d
i
d
where the total derivatives dx
d , i = 1, 2 and d take account of the induced
change in via (14.22). The derivative of the budget constraint (14.17) can
then be substituted into (14.23) to give
d
d [w ]
dU
dU
2
= 1 [1 + ]
+ w x [w ]
+ .
d
dx
d
d
(14.24)
Using the second-period budget constraint w +[1 + r] s = x2 + [w ],
where s is first-period saving, (14.24) becomes
dU
d [w ]
dU
d
= 1 [1 + ]
[1 + r] s
+ .
(14.25)
d
dx
d
d
dr
1
2
By definition, d
dr = and, from the factor price frontier dw = k . Since
the level of capital is given by K = P s and total labour supply L = P l =

P [1 ], k =
d dr dw
dr dw d
s
[1] .
Substituting these definitions into (14.24), using
439
d
d
and collecting terms
i dw
dU
dU h
d
= 1 1+
[1 + ]
+ .
(14.26)
d
dx
d
d
1
1
Finally, since = 2+n
and = 1+r
, 1 + = [r n] , the optimal pension
solves
dw
dU
d
dU
= 1 [r n]
[1 + ]
+ = 0.
(14.27)
d
dx
d
d
The contrast between this result and the conclusion drawn from the economy
in which no labour can be supplied in the second period of life can be seen by
d
setting = 1 in (14.27). From (14.22), when = 1, d

= 1
. Substituting
into (14.27) then gives
rn
dU
dU
dw
|=1 = 1 [r n]
= 0.
(14.28)
d
dx
d
1
1+r
dU
Since dx
1 > 0, the solution to (14.28) must have r = n so that the program
achieves the Golden rule.
When 0 < < 1, it then follows that r = n will not be achieved at the
d
solution to (14.27). Assuming that d
> 0, if dw
d < 0 then the program will
lead to a rate of interest less than the growth rate of population. The converse
holds if dw
d > 0.
The conclusion to be drawn from this analysis is that the decentralisation
of the Golden rule by the use of a social security program will not be desirable
when the retirement decision is endogenous. This is because the tax used to pay
for pensions introduces a distortion into the choice problem of each consumer.
Although the program can aid the dynamic eciency of the economy by aecting
the level of the capital stock, this has to set against the ineciency caused by
the tax. Continuous-time versions of this analysis are presented in Hu (1978)
and Sheshinski (1978).
In addition to aecting the retirement decision, social security will also have
an impact upon the labour supply decision itself. This impact will be particularly pronounced when a household framework is adopted and the distinction
is made between male and female labour supply. Since female labour supply is
typically more elastic than male supply, it will be female labour supply that is
proportionately more aected by the social security program. Simulation evidence on the strength of these eects is presented by Craig and Batina (1991).
They consider an economy in which each generation is composed of an equal
number of males and females. Pairs of males and females then form households and jointly choose their labour supplies to maximise a household utility
function. Except for the fact that they now supply two forms of labour, male
and female, these households are equivalent to the consumers of the economies
above. It is assumed that the two forms of labour are close, but not perfect,
440
substitutes in a CES production function. Since they are not perfect substitutes,
the equilibrium wage rates may dier.
The social security program considered by Craig and Batina contains provisions other than a simple retirement benefit. The retirement benefit is determined by the value of earnings prior to retirement and may be dierent between
males and females. Furthermore, each female has the option of choosing either
a pension based on their own earnings history or one based on their spouses
earnings. The social security program is financed by a tax on labour earnings, a
lump-sum tax and an additional tax upon labour earnings when, in the second
period of life, these rise above a cut-o level. Incorporating these provisions, the
budget constraint of a household whose members are born in period t is given
by
xtt
xt+1
t
+
1 + rt+1
0
wt+1 Lt+1 + wt+1
L0t+1
0 0
T
= [1 t] wt Lt + wt Lt +
1 + rt+1
"
#
0
[wt Lt + wt+1 Lt+1 ] + 0 [1 ] wt0 L0t + wt+1
L0t+1 + B
+
1 + rt+1
#
"
0
L0t+1 a
wt+1 Lt+1 + wt+1
[[wt Lt + wt+1 Lt+1 ]]
+
(14.29).
1 + rt+1
1 + rt+1
In (14.29) the primes denote variables relating to the female, is the benefit
rate on own earnings, = 0 if the wife chooses a pension based on her earnings
and = 1 if they choose that based on their spouses. is the benefit rate on
spouses earnings. B is the lump-sum benefit, T the lump-sum tax, t the tax
rate on earnings and t the tax on second-period earnings above the level a.
For a program with t = 0.15, = 0 = 0.2, s = 0.1 and = 0.5, the results
obtained are summarised in Table 14.1. The primary eects of the program
are to shift labour supply toward the beginning of the life-cycle for both males
and females and to increase the level of household consumption in the second
period of life relative to that in the first period. Lifetime labour supply of both
male and female falls. The program also causes a slight shift in relative wages
in favour of males.
w
w0
Lt
Lt+1
L0t
Without programme 1.790 0.478 0.232 0.434

With programme
1.804 0.503 0.095 0.462
Table 14.1 Eect of Social Security
L0t+1
xtt
xtt+1
0.084
0
0.561
0.470
These results demonstrate how the introduction of a social security program

aects the allocation of labour and consumption over the lifecycle. In particular,
it acts as a disincentive to supply labour in the later stages of life. The reduction
of total labour supply that is caused will also have consequences for the level of
output that is produced and the capital-labour ratio.
14.4.2
441
Eect on savings and capital
With a pay-as-you-go social security program in operation, any eect that the
existence of the program has upon private savings is reflected directly in the
level of the capital stock since, by definition, the program owns no capital. Social security has two conflicting eects upon the level of private saving. The
first eect is the substitution of social security for private savings which naturally reduces the level of saving. Osetting this eect is the likelihood that
social security will bring forward retirement. If this does occur, private saving
should rise in order to cover the increased length of retirement. At this level of
generality, the net eect is indeterminate.
In contrast to the earlier evidence of Cagan (1965) and Katona (1964) which
showed that consumers covered by private pensions did not save less than those
not covered, Feldstein (1974) estimated that the existence of the US social security program reduced private savings by 30-50%. This evidence was based on
the estimation of a consumption function that included social security wealth as
one of the explanatory variables. The central estimate suggested that, during
the 1960s, the capital stock was 38% lower with the social security program
than it would have been without. Although widely cited, these results have not
always been replicated in later studies. Danziger, Haveman and Plotnick (1981)
suggest that the true figure should be somewhere in the range of 0-20% whilst
Aaron (1982) concludes that there is simply a lack of agreement amongst the
studies. As an example of conflicting findings, work by Lee and Chao (1988)
estimates labour force participation and personal savings simultaneously taking into account private pensions. Although social security wealth is found to
encourage retirement, the payment of contributions to social security has an
insignificant eect on private savings.
Whether it is possible for a theoretical economy to exhibit a similar responsiveness to the introduction of social security as that suggested by Feldstein has
been investigated by Kotliko (1979) in an analysis involving a continuous-time
formulation of endogenous retirement. Consider an economy of identical individuals with has a population growth rate of n and endogenous labour-augmenting
technical progress at rate g. The latter assumption implies that the wage rate
is also growing at rate g. At each point in time, the fraction of the existing
population that have lived out their lifespan of D years will be replaced by
new consumers. Each consumer chooses their consumption stream and date of
retirement to maximise discounted utility.
Denoting instantaneous utility by U (xt (t)), discounted utility is given by
U=
U (xt (t)) et dt,
(14.30)
where is the rate of time preference. Utility (14.31) is maximised subject to

the budget constraint
Z
xt (t) ert dt =
R
0
W [1 ] e[rg]t dt +
e[rg]t dt.
(14.31)
442
In writing (14.32), it is assumed that both the wage, W , and the pension, , are
growing at rate g. The pay-as-you-go identity for the social security program is
given by
W [1 ]
=
,
(14.32)
with 1
the ratio of those working to those retired defined by
R R nt
e dt
1
(14.33)
= R 0D
ent dt
R
At each point in time, additions to the capital stock are equal to the savings of
those working plus the savings of those who are retired. For a working consumer
born at date t, their savings are
dKt (t)
= W [1 ] egt xt (t) + rKt (t) ,
dt
(14.34)
whilst for a retired consumer

dKt (t)
= egt xt (t) + rKt (t) .
dt
(14.35)
Solving (14.35) provides the capital owned by a consumer born in t and solving
(14.36) determines the capital of a retired consumer. Total capital can then be
found be integrating over the population. The total savings provided by consumers can then be equated with the capital demand by firms and equilibrium
computed.
(t)1
Adopting an instantaneous utility function of the form U = xt 1
and a
Cobb-Douglas production function, the simulation results of Kotliko for the
eect of the introduction of a social security program with a value of of 0.1
are summarised in Table 14.2. Inspection of Table 14.2 demonstrates that a
social security program can have eects with the order of magnitude identified
by Feldstein. The reductions in the capital stock range from 10% to 21% which,
although somewhat less than 38%, are still significant. Even in very simple
economies it is therefore possible for the introduction of pay-as-you-go social
security to substantially reduce the capital stock and with it the output of the
economy.
Net rate of interest
Age of retirement Rate of time preference Reduction in capital stock

42
0.0160
-0.213
0.05
45
0.0120
-0.190
50
0.0058
-0.104
42
0.0232
-0.212
0.06
45
0.0196
-0.189
50
0.0148
-0.102
Parameters: = 1, D = 55, r = 0.05, = 0.1, r = n + g, n = g
Table 14.2: Eect on Capital Stock
14.4.3
443
Ricardian equivalence
Ricardian equivalence originally referred to the proposition that the method of

financing government expenditure, whether through taxes or borrowing, was
irrelevant. To illustrate this, consider the following example. To reduce the
level of taxation by D in period t, the government sells quantity D of bonds. To
repay the bonds the following period, the government must levy additional taxes
in t + 1 of [1 + r] D. Since the discounted value of this increase in taxation is
exactly equal to the value of the original tax cut, the net wealth of the economy
is unchanged when viewed from period t. The bond-financed plan is therefore
equivalent to keeping the initial level of taxation unchanged, illustrating the
principle of Ricardian equivalence.
Ricardian equivalence will be discussed in detail in Chapter 15 but a further
point needs to be made here in order to relate the argument to the analysis of
social security. Returning to the example above, it would seem that if some of
the households alive at time t were no longer alive at t + 1 then the equivalence
would fail since they would benefit from the reduced tax payment but would
avoid the increased tax. To make this point stronger, assume the bonds mature N periods after issue. The argument of the example would still apply if
the population remained unchanged over the N period but it now seems less
reasonable to expect this.
There is, as first noted by Barro (1974), a mechanism that will maintain
equivalence even if the population changes between issue and redemption of
bonds. Suppose that each consumer has a single identified descendent and that
they care about their own level of consumption and about the utility level of their
descendent. Such intergenerational altruism then links finitely-lived consumers
into a household whose lifespan is as long as that of the economy. The altruism
will manifest itself in consumers choosing to leave bequests to their descendants
and it is via the bequest motive that Ricardian equivalence arises. To see this
point, assume that the bonds are purchased by generation t but the tax liability
is borne by generation t + 1. If generation t raises its bequest to t + 1 by exactly
the amount necessary for t + 1 to meet their increased tax liability then the
consumption plans of both generations will remain unaected by the switch
from tax-finance to bonds thus maintaining the equivalence result. Although
there are, of course, limitations to this result which are described in Chapter
15, it does have implications for the analysis of social security.
One implication of Ricardian equivalence has, in fact already been described.
The discussion of a fully-funded social security program in Section 2 noted that
such a program would leave the equilibrium of the economy unaected. The
fully-funded system requires that the pensions received by each generation are
equal to the taxes that they paid. In this case an increase in tax when young
is accompanied by the receipt of a pension when old which is equal in value
to the tax plus interest. This is simply the opposite of the initial example of
equivalence and a slight modification of that argument shows why it has no
net eect. The households reduce their private savings by the amount of the
tax, maintain their consumption levels and replace private saving by saving in
444
the pension plan. Since private and public savings have the same return, they
are indierent to this rearrangement. It is worth noting that this argument
presumes that private savings are initially greater than the tax used to finance
the pension. If they are not, then equivalence will not apply.
Introducing intergenerational altruism allows the equivalence argument to
be extended to pay-as-you-go social security. A pay-as-you-go system can be
interpreted as a forced transfer from the young generation to the old. If all
members of the old generation were making a positive bequest to their descendants prior to the introduction of the program, the eects of the program can
be entirely neutralised by the old simply increasing their bequest by exactly
the pension they receive. By the definition of a pay-as-you-go system, this increased bequest will exactly match the taxes paid by the young. Therefore,
with intergenerational altruism, pay-as-you-go social security will have no eect
upon the equilibrium provided the bequest motive is operational prior to the
commencement of the program.
In this context, it should be stressed that intergenerational altruism must
manifest itself through each member of generation t having a utility function of
the form introduced by Barro (1974)
U t = U xtt , xt+1
, U t+1 ,
(14.36)
t
where U t+1 is the utility level of their descendent. The dependence of U t upon
U t+1 provides the linkage that eectively turns the separate generations of the
family into a single household. Furthermore, although there may be a bequest
motive, bequests may still be zero if the consumer
a corner solution. For
is 2at t+1
the separable utility function U (xtt ) + 1 U xt+1
U
, Weil (1987) shows
+
t
that the bequest will only be positive if 2 > 1+n
where
r
e
is
the rate of interest
1+e
r
in the absence of bequests. If the no-bequest economy is dynamically inecient
then re < n and with discount rate 2 < 1 bequests will never occur.
An alternative specification (used, for example, in Hu (1979)) is to assume
that each consumer cares about the size of the bequest, bt , that they leave to
their descendent, rather than their descendants welfare, so that utility takes the
form
U t = U xtt , xt+1
, bt .
(14.37)
t
With the specification of utility in (14.38), the equivalence result no longer

applies. This follows from observing that the introduction of a social security program that has an initial transfer to generation t will result in standard
changes in xtt , xt+1
, bt via an income eect rather than an increase in bt exactly
t
equal to the level of the pension. If it is a normal good, the level of the bequest
will increase but not to the point that maintains the welfare of generation t + 1.
In the presence of intergenerational altruism it is therefore possible for social
security programs to have no eect upon the equilibrium of the economy. Indeed,
the discussion above may suggest that this is the typical outcome. However,
there are many circumstances it which the Ricardian equivalence result does
not apply and, although this pre-empts some of the material of Chapter 15, it
is worth noting some of these. As already stated. the utility function must take
445
the form of (14.37) and the bequest motive must be active for all consumers
prior to the introduction of the program. The existence of bequests in itself is
not sucient; these may simply be left due to uncertainty about the time of
death or poor planning. Taxes and benefits must be equal for each household
so that if there is a redistributive element to the program then equivalence will
not apply. In addition the empirical evidence already discussed suggests that it
does not apply in practice.
Although equivalence may apply in the simple economies used so far if altruistic preferences are introduced, the limits to the result are soon reached when
obvious extensions to these economies are made. With respect to the analysis of
social security, Ricardian equivalence should therefore be viewed as a theoretical
curiosity rather than a result of practical relevance.
14.4.4
Demographics
The growth rate of population is an important determinant of the success of

a social security program. A rapidly increasing population will make it easier
for the young to provide pensions for the (relatively) smaller number of old. It
will also lead to a dilution of the capital stock which makes it more dicult
for the pensions commitments to be met. This trade-o suggests that there
may be a rate of population growth which will maximise the welfare of the
representative consumer. Indeed, Samuelson (1975b) claimed to have provided a
characterisation of this optimal rate. However, it was shown by Deardor (1976)
that in many reasonable cases Samuelsons conditions actually described the
growth rate that minimised welfare. This fact is further clarified in Samuelson
(1976).
The connection between the rate of population growth and the optimal level
of social security can be investigated by optimising the choice of program for
given n and then varying n; this is the method adopted by Lopez-Garcia (1991).
Assuming that each consumer can choose the amount of labour they supply in
each period. Denoting this level of labour supply by `tt , a consumer born in t
maximises
t
U t = U xtt , xt+1
(14.38)
t , 1 `t .
The social security tax is levied proportionately upon labour income at rate
so, in the steady state with identical consumers, the budget constraint for the
program is given by
= [1 + n] w`.
(14.39)
The pension is assumed to be a constant proportion, , of the net wage so that

= [1 ] w`. As a function of the replacement rate , the tax supporting
the program is equal to
=
.
(14.40)
1+n+
The individual budget constraint is
x1 +
x2
[1 + n] [1 + r + ]
=
w` = w`,
b
1+r
[1 + r] [1 + n + ]
(14.41)
446
so that maximisation of (14.39) gives consumption demands xi = xi (w,

b r),
i = 1, 2, and labour supply ` = ` (w,
b r). Employing the factor price frontier,
w = (r), the equilibrium interest rate is the solution to
i
h
1+n
1
b r)
(r) ` (w,
b r) x1 (w,
1+n+
[1
]
w`
x
. (14.42)
r = f0
= f0
[1 + n] `
[1 + n] ` (w,
b r)
The solution to (14.43) is denoted r = r (, n).

Substituting into (14.39), the attainable level of utility can be written U =
V (, n). Dierentiating utility with respect to for fixed n gives
1+n
U
r
(r) `
V
=
[n r (, n)] k`
+
,
(14.43)
x2 1 + n +
1 + n +
where use has been made of the conditions describing consumer choice. The
sign of (14.44) is the same as that of n r since it can be shown, employing
r
the sucient condition for stability, that
> 0. Hence only if r (0, n) < n will
there be an interior, positive solution for . In that case, since the second-order
condition can be shown to be negative at r = n, the maximising value of
satisfies r (, n) = n.
Assuming now that an interior solution exists for , with > 0, the fact that
optimal social security program can be employed to maintain the equality of r
with n allows (14.43) to be written as
i
h
1+n
(n) ` ( (n) , n) x1 ( (n) , n)
1+n+
.
n = f0
(14.44)
[1 + n] ` ( (n) , n)
Solving (14.45) for , the solution is written = (n). This allows utility to be
expressed as U = V ( (n) , n) and hence
V ( (n) , n) V ( (n) , n)
U
=
+
.
n
n
n
(14.45)
However, since is chosen optimally, (14.46) reduces to

V ( (n) , n)
U
`
U
=
=
.
2
n
n
x 1 + n +
(14.46)
Since it has been assumed that > 0, (14.47) is always positive so that increasing
the rate of population growth will increase welfare. In this case, there is no
optimal rate of population growth. The ability of an increasing population to
provide a larger pension dominates in this case.
Alternatively, if the optimal replacement rate is negative, utility is always
decreasing in the rate of population growth. Finally, if a replacement rate of
0 is optimal, so that no social security program is required for the Golden rule
to be obtained, then (14.47) is always zero. It can easily be shown, however,
14.5. DETERMINATION AND JUSTIFICATION
447
that this may either be minimum, as in Deardor (1976), or a maximum. The

competitive outcome that achieves the Golden rule may therefore be sustained
by a rate of population growth that can either maximise or minimise utility,
thus being either the best or worst of all worlds.
In conclusion, the relation between the growth rate of population, the choice
of social security program and the level of utility may take many forms. If
there is an optimal social security program that has a positive pension, then an
increase in population growth always raises welfare. The converse holds if the
optimal pension is negative. More surprisingly, when it is optimal to have no
pension system the rate of population growth that supports such an equilibrium
will either maximise or minimise the level of utility. From a policy perspective,
there may be benefits to be obtained from rapid population growth as the increasing population can support a more generous social security program.
14.5
Determination and justification
There are several justifications for the existence of social security programs.
The optimal program of Section 3 would raise the welfare of all generations that
benefited from the attainment of the Golden rule steady state although there
may be some losers along the adjustment path (see Flemming (1977)). This
need not be true of pay-as-you-go systems with their implied intergenerational
transfers nor of fully-funded systems which may simply replace private saving
by public saving. Given these observations, the purpose of this section is to look
at various methods of determining the structure of social security programs and
of justifying their existence.
14.5.1
Voting equilibria
The most obvious mechanism by which the level of social security can be determined is through a system of voting. Although systems of voting dier widely,
the literature focuses only upon simple majority voting and the median voter
outcome. The rationale for this position is that these are the most tractable
analytically and capture the essence of more sophisticated alternatives.
The earliest analysis of these issues is given in Aaron (1966) who shows
that since a social security program with positive benefits always provides a
transfer to the old, a median-age voter may find it in their favour to vote for
the continued existence of a program despite its total eect being to reduce the
welfare of all households. Browning (1975) considers an economy in which there
is no capital market so that the only method of saving is via social security. In
such circumstances, majority voting leads to a level of social security in excess
of that which maximises lifetime welfare.
A simple result on the outcome of majority voting in an economy with capital
markets is obtained by Hu (1979) for the economy of Subsection 4.1. Since
population growth is positive, at any point in time the young generation is
larger then the old. Combining this observation with the assumption that all
448
consumers are identical, it follows that the program chosen by majority voting
will be that which maximises the lifetime utility of a young consumer. The
program therefore solves
max U x1 , x2 , ,
(14.47)
1 2
{x ,x ,,}
subject to the consumers budget constraint (14.17) and the budget identity of
the social security program (14.22). Assuming that the consumer treats the
wage as independent of when performing the optimisation, the social security
program chosen by majority voting will satisfy
[1 + ]
= 0.
(14.48)
In general, there is no reason to expect that the program determined by (14.49)

will be anymore, or any less, generous than the program arising from (14.27).
However, in the special case of labour supply only in the first period of life, so
= 1, (14.49) reduces to
rn
= 0,
(14.49)
1 1+r
so that majority voting will attain the Golden rule.
This analysis is extended by Hu (1982) who considers an economy in which
consumers live for three periods. They work in the first two and are retired in
the third. The important distinction is that with population growth satisfying
the restriction
2
2 + n > [1 + n] ,
(14.50)
so that, at any given time, the number of consumers in either the second or
third period of their life is greater than the number in their first period, majority voting will always lead to an outcome that is supported by two of the
three generations. Furthermore, if there is a possibility that the social security
program determined by voting at time t may be overturned by a re-vote at t + 1
or later, the possibility of strategic voting arises. That is, a consumer in the
early stages of their life may vote for a program with a low level of tax with the
intention later in life of voting for one with higher provision.
Employing this framework Hu (1982) shows that the equilibrium arising
from majority voting when consumers are uncertain about the impact of present
voting intentions on later outcomes may have an interest rate either above or
below that of the Golden rule. This finding indicates that although majority
voting need not be ecient, the possibility arises that it may be possible to
design a mechanism that implements the Golden rule. The weakness of this
framework is that the uncertainty is necessary for a consumer to have an interior
solution to their choice of preferred social security tax rate. In the certain case,
when no re-voting can occur, they will always choose a tax rate of 1 when the net
gain from the program is positive and zero when it is negative. The uncertainty,
however, is introduced in a fairly arbitrary fashion: it is simply assumed that
future provision, though uncertain, is increasing in the present level.
449
An alternative approach to this issue is provided by Boadway and Wildasin

(1989) who employ the continuous time framework described in (14.31) - (14.34)
but with g = 0. The majority voting assumption results in the equilibrium
level of social security being determined by the median voter. To overcome the
problem that in the absence of uncertainty this choice would either be a tax
rate of 0 or 1, the capital market is assumed imperfect so that consumers face
constraints upon borrowing. In addition, it is assumed that voting is infrequent
so that when determining their voting intentions each consumer expects the
level of social security adopted to remain constant for the remainder of their
lifespan. The major findings are that, given an existing level of social security
benefits, preferences are single-peaked over alternative levels so that the median
voter outcome applies and that when consumers cannot borrow against future
benefits, if the existing level of benefits are below the optimal level for the
median-age voter, the median-age voter will also be the median voter.
14.5.2
Altruism
Voluntary intergenerational transfers can arise when consumers have altruistic

feelings about other generations. These feelings can extend both to previous
generations and to following generations. Some of the consequences of bequests
have been discussed in Subsection 4.3; the focus here is primarily upon voluntary
gifts from young to old since this is the essence of a pay-as-you-go program.
Altruism has been represented as a consumption externality by Veall (1986)
with the utility of generation t being of the form
U t = U xtt , xt+1
, xtt1 .
(14.51)
t
In the formulation of (14.52) the consumption of generation t 1 in the second

period of life aects the utility of generation t. When the externality eect is
positive, it can be interpreted as a form of altruism. The alternative representation of Hansson and Stuart (1989) develops the Buiter (1980) and Carmichael
(1982) formulation of altruism and assumes that each consumer alive at time t
has the utility function
"
t #
2

xt1
U 1 xt1
t1 + U
t
U =
+ U 1 xtt + U 2 xt+1
t
[1 + n] [1 + ]
X
1 z
z+1 1 + n zt
2
+
,
(14.52)
U (xz ) + U xz
1+
z=t+1
where the first term captures concern for the older generation, with the relevant discount rate, and the third term that for following generations with
discount rate . Although analytical details are rather dierent, the two formulations provide an essentially identical motive for the introduction of a payas-you-go social security program that, on its introduction, involves an initial
transfer to the older generation. Here the approach of Hansson and Stuart will
be followed.
450
Assume that production is subject to a constant coecient production function so that the wage rate, normalised at 1, and the interest rate, r, are both
constant. Two restrictions upon the utility function are adopted. Firstly, > n
so that the infinite sum in (14.53) converges and > r so that, as noted in 4.3,
no consumer desires to leave a bequest. Denoting the gift given by generation t
to generation t 1 by at , the budget constraints in the first- and second-periods
of life are
xtt = 1 st at , xt+1
= [1 + r] st + [1 + n] at+1 .
(14.53)
t
The gift at satisfies the necessary condition

U 20 xtt1
U 10 xtt 0, = 0 if at > 0,
1+
(14.54)
where is the rate of discounting applied to the older generations utility in

(14.53).
Hansson and Stuart (1989) note that this economy has two steady state
equilibria. The first arises when the first part of (14.55) is satisfied with equality
and positive gifts are passed between generations.
In such a steady state the fact

that > 0 implies, from (14.55), that U 20 xtt1 > U 10 (xtt ) so that the marginal
utility of consumption when old is greater than whenyoung. This is in contrast
to the outcome without altruism in which U 20 xtt1 = U 10 (xtt ), with < 1,
see (14.18). This equilibrium is therefore not Pareto optimal since
a marginal
increase in st would raise the utility of generation t by U 20 xtt1 [1 + r]

U 10 (xtt ) > 0. The second steady state equilibrium occurs when (14.55) is an
equality and gifts are zero. If n r, then the outcome is not Pareto optimal
since an introduction of gifts can raise welfare of all generations. This is not
possible if n < r.
The structure of a social security program that is acceptable to the two
generations alive at the time of its introduction, say time 0, is constructed by
first deriving the optimal program for the old, generation -1, and the then for the
young, generation 0, where the program takes the form of gifts from young to
old {a0 , a1 , ...} and savings {s0 , s1 , ...}. For generation -1 the optimal program
solves

U 10 x00
U 20 x01
(14.55)
0, = 0 if a0 > 0,
1+
and
U 10 (xtt )
U 20 xtt1
0, = 0 if at > 0, t > 0,
1+
U 10 xtt + U 20 xt+1
[1 + r] 0, = 0 if st > 0, t > 0.
t
(14.56)
(14.57)
The optimal program for generation 0 also satisfies (14.57) and (14.58) but
replaces (14.56) by

U 20 x01
(14.58)
U 10 x00 0, = 0 if a0 > 0.
1+
451
The first point to note is that savings must tend to zero. Assume otherwise.
Then (14.57) and (14.58) imply that
t+1 U 20 xt+1
U 20 xtt1 [1 + ]2
U 10 (xtt ) [1 + ]
t
10
U xt+1 =
=
=
.
(14.59)
1+
1+r
1+r
From the assumption

that > r, (14.60) implies that as t , U 10 (xtt )
20
t
and U xt1 or xtt 0 and xt+1
0. The budget constraints (14.53)
t
and (14.54) can then be combined to give
st = 1 xtt
xtt1
1+r
+ st1
.
1+n
1+n
(14.60)
0 imply st contradicting st 1. Thus for some

Hence xtt 0 and xt+1
t
t, st = 0. Now let b
t be the first period with zero savings. Because savings are
t
positive for all earlier periods, (14.57) and (14.58) imply that xt+1
t+1 < xt and
t+1
t
xt < xt1 . From (14.61) it then follows that
1+r
st1 st > [st st+1 ]
, for 0 t < b
t.
(14.61)
1+n
As b
t is the first period with st = 0, sbt1 > sbt which implies from (14.62) that
t. If there were any period after b
t with st > 0 then the
st > st+1 for all 0 t < b
same backward induction argument would contradict sbt = 0.
Contrasting (14.56) and (14.59), it can be seen that generation -1 would
prefer that a0 be larger than the level that would be chosen by generation
0. More importantly, since (14.57) are common to the characterisation of the
optimal program for both generations, given a value of a0 both generations
would agree upon the values of {a1 , a2 , ...} and {s0 , s1 , ...} that should constitute
the remainder of the program. Furthermore, (14.57) and (14.58) also capture
optimal choices for the older generation in every future time period so that they
would always support the continuation of the program.
That the program raises the welfare of the introducing generations can be
seen by setting a0 at the level, say a0 , determined in the equilibrium without
social security with further transfers set optimally. Comparison of (14.55) and
(14.57) then shows that the equilibrium set of transfers without social security is
not optimal for generations -1 and 0 so that they must gain by the introduction
of the program. Their welfare could potentially be raised further by choosing
a value of a0 6= a0 . If the solutions of (14.56) and (14.59) both determine a
value of a0 lower than a0 then the maximum of the two would from part of an
equilibrium program. The converse holds when a0 is less then the solutions to
(14.56) and (14.59). In other cases, the level a0 will lie in the acceptable range.
The move to such an equilibrium social security program moves the economy to
a Pareto optimum. Although the program raises the welfare of the generations
that introduce it, this need not apply to any of the later generations although,
as already noted, no generation will seek to change the program. In particular,
if the initial steady state was one without transfers, the introduction of the
452
program reduces the utility o all generations after a given date. To see this,
note that without social security, total lifetime consumption
would
be 1 + rst
with consumption levels chosen to maximise U 1 (xtt ) + U 2 xt+1
. With social
t
security savings fall to zero by date b
t so, if n = 0, total lifetime consumption
b must equal 1 and, from (14.58), this is not chosen to maximise U 1 (xtt ) +
after
tt+1
2
U xt
. Generations after b
t must clearly be worse o.
14.5.3
Myopia
An obvious motivation for the introduction of social security is that consumers

do not make appropriate provision for their retirement. This may be because
they are myopic and fail to appreciate their later needs, either discounting the
future completely or placing a lower weight upon than would capture their true
preferences. Alternatively, consumers may make mistakes in their planning, lack
information or simply be irrational. Whatever the reason, empirical evidence
(Diamond (1977), Kotliko, Spivak and Summers (1982)) does suggest that
consumers accrue insucient savings to cover retirement.
Although the behavioural foundations of myopia dier significantly from
those of consistent utility maximisation, the formulation of myopic behaviour
of Feldstein (1985) can be incorporated into the analysis developed above by
making only minor adjustments. Assume that the preferences of each consumer
can be represented by the utility function

U t = U 1 xtt + U 2 xt+1
,
(14.62)
t
but, due to myopia, chooses their saving decision to maximise

U t = U 1 xtt + U 2 xt+1
,
t
(14.63)
where 0 1 captures the degree of myopia. Complete myopia is given by

= 0. Each consumer works only during the first period of life and divides posttax wages between consumption and savings subject to the budget constraint
xtt = [1 ] wt st .
(14.64)
A second aspect of consumers myopia is that they underestimate the level of

benefits received from the social security program in their retirement. The
perceived budget constraint is therefore
xt+1
= [1 + r] st + t+1 ,
t
(14.65)
with < 1. It should be clear that the two aspects of myopia have conflicting
1
t
eects
upon
t+1
the level of savings. Adopting the assumption that U (xt ) and
2
U xt
are logarithmic, the optimal level of saving is
st =
t+1
[1 ] wt
,
1+
[1 + ] [1 + r]
(14.66)
453
where it should be noted that low values of reduce saving while low values
of increase it. Using the budget constraint for the social security program,
t = wt [1 + n], and assuming that the wage grows at rate g, so that wt+1 =
wt [1 + g], (14.67) can be written
st =
[1 + ] wt
[1 ] wt
,
1+
[1 + ] [1 + r]
(14.67)
where [1 + ] = [1 + n] [1 + g].
In contrast to the myopia in consumer behaviour, the social security program
is chosen to maximise the social welfare on the basis of true preference. In year
t the total utility of those alive is
t
Wt = [1 + n] ln ([1 ] wt st ) + [1 + n] ln ([1 + r] st1 + t ) .
(14.68)
The optimal social security program is defined as that which maximises the
steady-state value of (14.69). When = 0, so that the future pension is entirely
discounted, the optimal tax rate is
=
[1 + ] [1 + ] [1 + r] [2 + n]
.
[1 + ] [1 + ] [2 + n] [1 + r] [2 + n]
(14.69)
1
With complete myopia, h = 0i and (14.70) reduces to = [2 + n] with corresponding pension = 1+n
2+n w. If population growth is zero, the pension is
exactly half the wage. It rises above half the wage when population growth is
positive. More generally, it can be seen from (14.70) that the tax rate falls, and
with it the pension, as rises. Hence the less myopic the consumers, the lower
is the pension. If the pension is restricted to be non-zero, then = 0 if
[1 + ]
.
[1 + r] [2 + n] [1 + ]
(14.70)
Therefore, even though myopia may in some cases justify social security, even
in this simple framework there is a range of partial myopia described in (14.71)
for which social security is not justified. If 6= 0 broadly similar conclusions
apply and there remains a significant range of parameter values for which social
security is not justified. In addition, it can be shown that the optimal level of
pension is decreasing in .
The main message of this analysis is that myopia is not in itself justification
for the introduction of social security. Although myopic consumers will not
make sucient provision for their retirement this is not a sucient reason for
providing a state pension.
This myopia formulation is also used by Feldstein (1987) to assess the means
testing of social security. In an economy with a mixture of myopic consumers
and fully-rational consumers, the there is a natural trade-o between providing
protection for the myopic who fail to make adequate provision for retirement
and reducing the incentive to save of the rational. The argument in favour of
454
means-testing is that it allows an overall reduction in the size of the program

and therefore of the distortion it causes. The reduction in distortion is especially
relevant in this situation since those who would not receive benefits after meanstesting would not have their savings behaviour aected at all by the existence
of the program.
In an economy that is composed of a mixture of completely myopic ( = 0)
consumers and fully-rational ( = 1) who, except of the value of , have identical
preferences and earnings, Feldstein (1987) shows that there is a strong case for
means-testing but it is not always superior to a program of universal benefits.
The reason why the universal system may be superior is that means-testing
may result in a low level of benefits since, if the benefit is not to be universal by
default, it is necessary to provide an incentive for some consumers to continue
saving. An upper bound is therefore placed upon means-tested benefits which
may be below the level attainable in a universal program.
14.5.4
Uncertainty
The motivation for social security in the presence of uncertainty has already
been addressed for static economies in Chapter 7. It was shown there how social
insurance could raise welfare by enabling intragenerational transfers of income.
Essentially, the insurance contract reallocated income from the fortunate to the
unfortunate. In a dynamic setting a social security program can also provide
intergenerational income transfers as insurance against unfavourable outcomes.
This intergenerational insurance aspect of social security clearly has important
implications for the design of programs.
Two forms of uncertainty will be considered in this section. The economies
considered up to this point have been characterised by populations that have
grown at a steady rate so that the size of each generation has been known
with certainty. The first source of uncertainty to be considered arises from
randomness in the size of the generation that is born at each date. In this case,
social security can play a role in insuring against the risk of being born in a large
generation. The second source of uncertainty is introduced via randomness in
the production technology. That is, given known inputs, output is determined by
a random process. Social security can then act so as to insure those generations
faced with unfavourable realisations of the productive shock.
Random population growth and random output both represent forms of
aggregate uncertainty that cannot be aected by the actions of individual consumers. In contrast, Section 6 will consider individual uncertainty about the
length of life. Since the length of life may potentially be aected by decisions
taken by consumers, lifetime uncertainty is of a dierent nature to aggregate
uncertainty and requires a separate treatment.
Population uncertainty
To be born into a generation that is large relative to its predecessor and its
successor has two disadvantages. When working, the wage received will be low
455
since the labour-capital ratio is high and when retired the return on savings
will be low since the capital-labour ratio will be high in the next period. In the
absence of social security, large generations will therefore have a relatively lower
level of welfare than small generations. These observations are incorporated
into the analysis of social security by Smith (1982).
Population uncertainty can be introduced by taking the economy of Section
3 but treating the growth rate of population as a random variable. The size of
the generation born at t is given by Ht = [1 + nt ] Ht1 where nt is drawn from
some known distribution. To simplify the analysis, it is assumed that consumers
born at t 1 know the value of nt when making their savings assumption.
Two alternative structures for the pay-as-you-go social security program are
considered. In the fixed tax system, the level of tax remains constant for all
generations but the level of pension received is dependent upon generation size.
Alternatively, in a fixed pension system it is the level of the tax that is variable.
Under the fixed tax system, a tax rate of in period t is equal to a pension
of [1 + nt ] for each member of generation t 1. The budget constraint of a
consumer born in t is therefore
xtt +
xt+1
rt+1 nt+1
t
= wt
.
1 + rt+1
1 + rt+1
(14.71)
Hence, in a dynamically ecient equilibrium with rt+1 nt+1 if the interest rate
and the wage are unaected by the introduction of social security, no generation
(with the exception of the generation that benefits when the system is first
introduced) can be made better-o by the introduction of the fixed tax program.
The same result can also be shown to apply if the wage rate and interest rate
do adjust provided consumption is never an inferior good. The reason for this
conclusion is simple. With the fixed tax large generations will receive smaller
pensions than small generations despite being the poorest.
Under a fixed pension system with the pension is fixed at , a member of
generation t will pay a tax of 1+n

so that their budget constraint is
t
xtt +
xt+1
t
= wt
+
.
1 + rt+1
1 + nt
1 + rt+1
(14.72)
If the economy is dynamically ecient with rt+1 nt+1 , this does not rule
out the possibility that rt+1 < nt which, when the interest rate and the wage
rate are constant, will occur when generation t is large. For large generations
(14.73) therefore shows that the introduction of a social security program with
a fixed pension may benefit large generations although it will reduce the welfare
of small generations. When the wage rate and interest rate are not fixed, the
osetting eects of the reduction in wage rate due to reduced saving and the
receipt of the pension leave the outcome ambiguous. However Smith (1982) does
provide an example for which expected utility, where the expectation is taken
with respect to the distribution of generation size, increases if risk aversion is
suciently great.
456
An alternative approach to the insurance role of social security with an

uncertain population is discussed by Brandts and de Bartolome (1992). As
already noted, if the generation born at t is large, the wage will be low and
the return on capital high. Retired consumers therefore gain by the birth of a
large generation whilst the large generation loses. The converse occurs when the
generation is small. Since, at any time t, the incomes of the young and the old
are negatively correlated, there are potentially gains to be obtained from the
provision of insurance. To fully exploit these, it must be necessary for transfers
to be made from young to old or vice versa.
The analysis of Brandts and de Bartolome determines whether the introduction of a small program can be Pareto-improving for the existing members of the
generation born at 0 and potential members of generation 1. The existence and
structure of a Pareto-improving contract is then dependent upon the process
by which potential consumers become actual consumers. The interpretation is
that there exists a pool of potential consumers who may become members of
generation 1 with some probability and the size of generation 1 is determined
according to a known distribution function.
Three cases are examined. In the first, the welfare of an unborn worker
is taken to be the expected utility in first period of life if they were born. A
Pareto-improving contract will almost always exist in this case although the
structure of payments may run in the opposite direction to those that would
provide insurance. The second case corresponds to that of Green (1988) and
assumes that unborn workers are arranged in a queue with those are the front
certain to be born and others further down the line only born if the generation is
large. Since those who will only be born if the generation is large care only about
the payment they will receive in that state, no contract exists that is Paretoimproving. In the final case, each potential consumer in the pool of unborn
consumers is treated symmetrically and has the same probability of being born.
In this case, a social insurance contract is always Pareto-improving.
Output uncertainty
The interaction between output uncertainty and social security has been analysed
Enders and Lapan (1982) in an economy in which labour is the sole input into
production and production is subject to a multiplicative disturbance. Since the
form of technology leads to an absence of capital as a store of value, fiat money
is introduced to serve this purpose. In consequence, the underlying structure of
the economy is distinct from those studied in previous sections.
Given labour supply of `t from a consumer of generation t, the output of
that consumer is given by
yt = t `t ,
(14.73)
where
t is the random disturbance. It is assumed that E (t ) = 1, that
cov t , t+i = 0 all i and that the same value of t is realised by all consumers in generation t. With product price pt , the budget constraint in the first
period of life is
pt xtt + Mt = pt t `t [1 ] ,
(14.74)
457
and in the second period
pt+1 xt+1
= Mt + t+1 = pt t `t [1 ] xtt + t+1 .
t
(14.75)
It is assumed that the pension is determined according to the relation

`t
,
(14.76)
t+1 = pt+1 t+1 Lt+1
Lt
where Lt denotes aggregate labour supply. This formulation relates the level
of pension to work-eort in period t. Alternative specifications could equally
well be adopted. If each generation is of identical size, the market for the
consumption good will clear in period t if
M
+ xtt = t Lt [1 ] ,
pt
(14.77)
with M total money supply. From (14.78) it can be seen that given xtt the
market clearing price level is dependent upon the realisation of t .
If there were no uncertainty in the economy, so that t = 1, all t, the
stationarity of the economy implies that the price level would be constant over
time, and the identical consumers of each generation would each supply the
same quantity of labour. Eliminating Mt between (14.75) and (14.76), and
using (14.77), it can be seen that the resulting lifetime budget constraint is
independent of the parameters of the social security system so that the social
security system described has no eect in a certain environment. This conclusion
arises because under the assumption of a constant population, the return on
social insurance taxes is equal to the return of private saving.
When uncertainty is present, each consumer chooses their labour supply
prior to the realisation of t but chooses consumption and savings after t
has been realised. Their preferences are represented by the utility function
xt
, `t . Denoting the propensity to consumer out of gross income, t`t ,
U xtt , xt+1
t
t
by , a stationary rational expectations equilibrium for this economy is now
defined.
Stationary rational expectations equilibrium
A stationary rational expectations equilibrium is a pair of functions p (),
() and a level of labour supply ` that satisfy
(i)
(ii)
[1 ()] H` =
() = arg max U
{s}
`s, E+1
M
+ H`,
p ()

p () ` [1 s]
+1
`
,` ,
+
p (+1 )
and
(iii)

p () s [1 ()]
+1
s,
s
,
` = arg max E,+1 U ` () ,
+
p (+1 )
{s}
458
where is the productivity level in the first period of a consumers life, +1

the level in the second period and Ex denotes the expectation taken over the
random variable x. Condition (i) captures market clearing, (ii) the choice of
consumption to maximise utility taken after is realised and (iii) the choice of
labour supply made before is realised.
To permit an explicit solution a utility function of the CES form
h
i 1
1 t t+1
U=
+ [1 `t ] , < 1, 6= 0,
(14.78)
xt + xt
is adopted where 1 is the rate of time preference. For = 0 this collapses

to a log-linear function. The following theorem can then be proved.
Theorem 65 (Enders and Lapan) Two equilibrium solutions to (ii) exist: (a)
1
1 () = 1 , (b) 2 () < 1 . If 1
< 1, (a) is the unique solution.
Furthermore, for equilibria characterised by (b) there is a level of the tax rate
1
1+ 1
, such that a social security program with tax rate maximises
the expected utility of each consumer.

Proof. The solution of the maximisation in (ii) is characterised by
"
+1 +1 #1
`
p () ` [1 ] + p
p ()
1
,
= E+1
[`]
p (+1 )
p (+1 )
(14.79)
which can be reduced to

1
,
1 +1 1 +1
(14.80)
by employing (i) and the assumption that all consumers are identical. Solutions
(a) and (b) can then be seen by inspection.
Noting that and +1 have the same distribution, for solution (b) (14.81)
can be further reduced to
1 !
1 ()
= 1.
(14.81)
E
()
()1 [1 ()] = E+1
+1
i1
1
must hold
< 1
1 1
in such an equilibrium. Hence for tax rates such that
< 1, (b) cannot
be an equilibrium.
h
i
1
Now let satisfy 1
= 1. Then as from below, 1
h
1
h
i1
i1
1
1
1
with 1 = 1 + 1
and ` ( ) = 1 + 1
[E ( )] 1 + 1
.
2
Since () < 1 , the inequality E
1()
()
14.6. LIFETIME UNCERTAINTY
459
Again employing the observation that and +1 have the same distribution,
expected utility can be written

1
1
(14.82)
E (U ) =
[[1 `] + ` E ( [ () + [1 ()] ])]
It is clear that this is maximised by ( ) and ` ( ).

This analysis demonstrates how the existence of a social security program can
raise the level of welfare when the economy is faced by production uncertainty.
Since no social security was required in the equivalent economy with certainty,
production uncertainty provides an argument for the provision of social security
that does not depend on any form of economic irrationality, such as myopia, or
market failure.
14.6
Lifetime uncertainty
The distinguishing feature of uncertainty about the length of life is that it constitutes individual, rather than aggregate, risk. Since the risk is individual,
information about the nature of the risk may be held by the individual and
be unobservable to the government. Furthermore, individuals may take unobservable actions that aect the nature of the risk. Each of these aspects of
individual risk leads to a position of asymmetric information between agents in
the economy which can result in the failure of markets to achieve a Pareto ecient outcome. Such failure provides a further motivation for the introduction
of social security.
Consider an economy in which each consumer lives with certainty for one
period but with some positive probability less than 1 may die before they enter
their second period of life. The payment of tax towards a social security program
can then be viewed as the purchase of an annuity which has a return equal to
the pension if the consumer survives into the second period and a return of zero
if they do not. The analysis of Sheshinski and Weiss (1981), which is described
in more detail below, introduced this perspective of social security and showed
how the introduction of such annuities with an actuarially fair return, where
none previously existed, could raise welfare.
If all information were public, there is clearly no reason why such annuities
could not be provided privately at actuarially fair rates of return. If they were,
the introduction of a social security program with the same structure of returns
would have no eect on welfare. To avoid this conclusion, Sheshinski and Weiss
(1981) rule out such private annuities while Karni and Zilcha (1986) allow private annuities but assume that they can only be provided at actuarially unfair
rates for reasons of resource cost in provision. The individual nature of lifetime
uncertainty however provides sucient reasons for not expecting actuarially fair
private annuities to be available. If the probability of not surviving into the second period of life diers between consumers and is private information, then
the problem of adverse selection may arise. That is, an annuity designed for
those who have a low probability of survival will also appeal to those with a
460
high probability. Hence any actuarially fair annuity will entail a loss for its
supplier. If individuals can take actions that aect the probability of survival
which cannot be observed by the suppliers of annuities then moral hazard will
lead to no actuarially fair annuities being oered.
The basic model of uncertain lifetime and the annuity approach to social
security are now described. This is followed by an analysis of the consequences
of adverse selection and moral hazard for economies with both private annuities
and social security.
14.6.1
Symmetric information
Uncertainty in the length of lifetime is captured by assuming that each consumer

lives with certainty for one period of life during which they work and that they
then live for a fraction , 0 1, of the second period during which they are
retired. The retirement period is distributed randomly across each generation
and the distribution is the same for all generations. The value of is known
neither to a consumer or to the government but its distribution is known to
both.
In addition to the standard form of saving that pays a return of r, a fullyfunded social security program is represented as an annuity that pays a return
throughout a consumers retirement. The social security program is assumed to
be actuarially fair so that its return is given by 1+r
where = E (). For later
comparison with a pay-as-you-go program, the optimum social security program

is interpreted as the level of annuity that is chosen by one of the (ex ante)
identical consumers. Denoting the purchase of annuities by a representative
consumer born in t by at , a consumer solves the maximisation

max
U = U 1 xtt + E U 1 xt+1
, + V (bt ) ,
(14.83)
t
t+1
t
{xt ,xt ,at ,st }
subject to the budget constraints
xtt = w + bt1 at st ,
and
1+r
t+1
bt = [1 + r] st +
,
at xt
(14.84)
(14.85)
where bt1 is the bequest received, bt the bequest left and V (bt ) the utility
derived from the bequest. The first-order conditions for this maximisation with
respect to st , at and xt+1
are respectively
t
U 10 + [1 + r] E (V 0 ) = 0,
1+r
10
U +
E (V 0 ) = 0,
and
E Ux2 V 0 = 0.
(14.86)
(14.87)
(14.88)
461
From (14.87) and (14.88) it follows that E (V 0 )E (V 0 ), so that cov (, V 0 ) =

0 and hence, when V 00 < 0, bt is constant for all . From the budget constraints
(14.85) and (14.86) it then follows that xt+1
= 1+r
at and bt = [1 + r] st . Since
t
the consumer chooses the purchase of annuities to maximise welfare, a fullyfunded system with tax rate equal to the value of at that arises from the
solution to conditions (14.87) to (14.89) will be optimal.
Turning now to a pay-as-you-go system, let the tax paid by generation t be
denoted by t so that the benefit received by a member of generation t 1 is
t
. With this notation, the budget constraints facing a consumer are
and
xtt = w + bt1 t st ,
(14.89)
(14.90)
t+1
t+1
bt = [1 + r] st +
.
xt
To characterise the optimal policy, it is first noted that level of welfare of generation t is now determined by the net bequest, bt t+1 , left to the following
generation. The optimal program then solves
max
xtt ,xt+1
, t ,st
t

U = U 1 xtt + E U 1 xt+1
, + V (bt t+1 ) ,
t
(14.91)
subject to (14.90) and (14.91). The first-order conditions for this maximisation
are respectively
with respect to st , t and xt+1
t
U 10 + [1 + r] E (V 0 (bt t+1 )) = 0,
and
E V 0 (bt t+1 ) = 0,
E Ux2 V 0 (bt t+1 ) = 0.
(14.92)
(14.93)
(14.94)
and bt = [1 + r] st .
The solution to (14.93) to (14.95) satisfies t+1 = xt+1
t
More importantly, it is clear from inspection of (14.87) to (14.89) and (14.93) to
(14.95) that the outcomes under the two social security programs will involve the
same levels of consumption and the bequests will dier by an amount equal to
the social security tax. The optimal levels of the two program therefore achieve
the same real equilibrium and the method of financing, being it fully-funded or
pay-as-you-go, is irrelevant. In addition, it is clear that the introduction of the
social security system raises expected utility.
Sheshinski and Weiss (1981) extend this analysis to show that a decrease
in the birth rate increases the optimal level of social security whilst reducing
bequests and savings. As to be expected, an increase in life expectancy also
increases the optimal level of social security.
462
14.6.2
Adverse selection
In the economy described in section 6.1 the optimal allocation achieved by the
fully-funded social security program could also be achieved by an actuarially
fair privately provided annuity. If such a private annuity were provided, no further welfare improvements could be achieved by the provision of social security.
However, the existence of private annuities is not, in itself, sucient to make
the provision of social security redundant. Myopia on the part of consumers,
which was discussed in 5.3 and in this context by Feldstein (1990), will result
in an inecient equilibrium as consumers will not purchase sucient annuities.
Enforced purchases through social security could then raise welfare. The same
would apply to mistaken assessments of survival probabilities.
A social security program may also be justified if private annuities are not
actuarially fair. The lack of actuarial fairness may arise simply as a result of
the resources that are employed in operating the annuities and in the provision
of risk premia to underwriters. This aspect of private annuities provides the
motivation for the analysis of Karni and Zilcha (1986) which shows that the
introduction of social security will improve welfare provided that it oers better
terms than private annuities. There is no reason though why the costs of running
a public annuity system should be any less than those of a private system so
that this motivation for social security is not entirely persuasive.
A second reason why private annuities should not be actuarially fair arises
from private information about survival probabilities and the adverse selection
problem that arises. Unlike a statutory social security program, the suppliers
of private annuities are subject to competition in the market. As first shown
by Rothschild and Stiglitz (1976) in the context of insurance markets, with
adverse selection it is possible for the competitive equilibrium to be Pareto
inecient with the equilibrium contract not oering full insurance. The same
reasoning was shown by Eckstein, Eichenbaum and Peled (1985) to apply to
private annuity markets.
It is now assumed that each generation can be divided into two groups, A
and B, with type B consumers for each type A. All consumers are endowed
with units of the single, non-produced consumption good and live for a least
one period. Consumers in group i have a probability i of surviving into the
second period of life with 0 < A < B < 1. Assuming that each generation
is large, these probabilities are also the proportions of each type that live for a
second period. Focusing attention on steady-state equilibria only, the utility of
a type i consumer is given by

U i = U xi1 + i U xi2 ,
(14.95)
where the survival probability is used to discount second-period utility.

Pareto ecient allocations for this economy are characterised by solving the
optimisation
X
X
max
i U i , i 0,
i = 1,
(14.96)
{xi1 ,xi2 ,i=A,B } i=A,B
463
subject to
A
B
B
xA
1 + A x2 + x1 + B x2 = [1 + ] .
(14.97)
When both > 0, the solution to the optimisation is characterised by

U 0 xA
U 0 xB
1
=
1 .
(14.98)
0 xB
U 0 xA
U
2
2
For later comparison, it should be noted that this optimum can be sustained by
the provision of two private annuities that have returns of 1 and 1 . The first
A
B
will be purchased by type A consumers and the second by type B. This results
in a separating equilibrium with both types of consumers purchasing annuities
that are actuarially fair for their group.
Asymmetric information is introduced by assuming that each consumer knows
their probability of survival and the population proportions. The government
knows the probabilities of survival and the population proportion but cannot
infer the survival probability of an individual consumer. An annuity contract is
defined as a pair {av , v } where av is the number of units of the annuity that
must be purchased and v the return. A consumer purchasing contract will
achieve utility level
V i (a, ) = U ( a) + i U (a) .
(14.99)
From (14.100) the gradient of an indierence curve is

dxi1
U 0 (a)
= i 0
,
i
U ( a)
dx2
(14.100)
so that at any level of consumption the gradient of a type B consumers indierence curve will be steeper than that of a type A. The equilibrium of a
competitive market in annuity supply is now determined and shown to be inecient. The possibility for a social security program to raise welfare is then
discussed.
An equilibrium in the supply of contracts is defined as follows.
Contract equilibrium
An equilibrium is a set of contracts such that:
(i) no contract in the equilibrium set makes a negative profit;
(ii) no contract outside the equilibrium set would make a positive profit if
oered.
The properties of the equilibrium are given in Theorem 14.2.
Theorem 66 (Eckstein, Eichenbaum and Peled)
(i) There can be no pooling equilibrium.
(ii) If an equilibrium exists it is a separating
equilibrium
in which consumers
n
o
1
of type A purchase a contract of the form a1 ,
and the contract purchased
A
n
o
B
by type B consumers is 1+ B , 1 .
B
B
(iii) For suciently small values of > 0, there is no equilibrium.
464
Figure 14.1: Non-Existence of Pooling Equilibrium

Proof. (i) If a pooling equilibrium
were to
exist it must satisfy the zeroprofit condition aA + aB = A aA + B aB for the pooling rate of return
and aA , aB are the purchases of annuities by As and Bs. Since pooling implies
aA = aB = a, = 1+
, 1 < < 1 . In Figure 14.1 a potential pooling
A +B
B
A
equilibrium contract is shown by P . All other contracts with return must also lie
on the line joining and . The relative slopes of the indierence curves then
implies that a contract such as that is preferred to P by group A consumers
and makespositive profit always exists regardless of the location of P on the
line , .
(ii) It is a condition of equilibrium that each contract oered must make nonnegative profits. The contract purchased by type B consumers cannot therefore
be more favourable than that which gives full insurance and consumption levels
B
xB
1 = x2 = 1+B nor can the contract oered to type As be preferred by the Bs
to the full insurance contract. Hence if a separating equilibrium exists it must
have full insurance for the Bs and the type As being supplied with the contract
defined as the solution to

a1
max U ( a1 ) + A U
,
(14.101)
A
{a1 }
subject to the restriction that type Bs prefer their full insurance contract
a1
[1 + B ] U
U ( a1 ) + B U
.
(14.102)
1 + B
A
This proves (ii).
(iii) The potential separating equilibrium of (ii) with type As contract at
SA and Bs at SB is shown in Figure 14.2. This equilibrium will not exists of
465
Figure 14.2: Separating Equilibrium

the line describing actuarially fair contracts is as shown in (a). In that case a
contract such as P 00 will be preferred by both types and make positive profits.
This will always be possible for small . The separating equilibrium does exist
in (b) since no pooling contract can make positive profit. This proves (iii).
Having now characterised the competitive equilibrium, it is possible to consider the eect of introducing social security. The distinction between contracts
that can be oered by private firms and social security is that a social security
program can cross-subsidise the provision of annuities to one group with the
proceeds from another. If a competitive firm attempted to do this, it would
be undercut by another firm that oered a slightly more attractive version of
the profit-making contract. To show that circumstances exist in which social
security can raise welfare, consider each consumer having to participate in a
mandatory program which involved the purchase of a units of annuity with an
actuarially fair return = 1+
based on the population survival probabilA +B
ities. Figure 14.3 shows how the social security program raises welfare. The
forced purchase of annuities with value a at the actuarially fair rate leaves at
most a to consume in period 1. Beginning at this point on the population fair rate line, the two groups can then purchase annuities along their own
fair-return lines. This modifies the budget constraints from their initial position
in (a) to the new position in (b) with a consequent change in the separating
equilibrium. The new separating equilibrium is determined by the intersection
of the indierence curve of type Bs through the full insurance point with the
fair-return line of type As. In (b) this intersection is shown above the initial
indierence curve of type As. Therefore in this case social security leads to a
Pareto improvement. However, study of the Figure will show that this need not
be the case.
The outcome of this analysis is that social security may not always be able to
466
Figure 14.3: Social Security

improve upon the equilibrium of the competitive market but, even when it can,
it will not lead to a Pareto ecient outcome. The government is constrained
by the same informational limitations as the competitive market and has only
the advantage of being able to cross-subsidise between contracts. This is not
sucient to achieve the Pareto ecient outcome of full insurance for both types.
It should also be noted that when the social security program does lead to a
Pareto improvement, the type As would prefer to exchange private annuities for
social security whereas the type Bs would prefer the opposite.
An alternative analysis of the eect of adverse selection is given in Abel
(1986a). That analysis is generalised by the inclusion of a bequest motive and
private savings. However, only a pooling equilibrium in the provision of private
annuities is considered. It is shown that the rate of return on private annuities
is below the population (fair) rate and falls when social security, which is fair,
is introduced. The eect of social security on the capital stock is ambiguous.
14.6.3
Moral hazard
Moral hazard can be introduced into the analysis of social security by assuming that consumers can take action which aect their longevity but cannot be
observed by firms providing annuities or by the government. This can be incorporated into the framework used above by assuming that the probability of
survival is dependent upon the level of consumption of some health-related activity so that = (e). If the consumption raises longevity then 0 (e) > 0 and
if it reduces it 0 (e) < 0.
The central reason why the competitive supply of annuities in such an economy is inecient is that consumers, when choosing their consumption level of
e, take account only of the direct cost of the activity. Since the price of a fair
14.7. CONCLUSIONS
467
annuity is based upon (e), the health-related activity also has an indirect cost
via the eect it has upon the price of annuities. A consumer maximising subject
to fixed prices does not take this into account. This externality generates an
inecient equilibrium as discussed in Chapter 10.
These ideas have been formalised by Davies and Kuhn (1992). They show
that the nature of the externality leads to an over-investment in the health
related activity in the sense that a reduction in investment accompanied by a
actuarially fair reduction in the price of annuities would raise welfare. Furthermore, since the government is constrained by the same informational deficiency
as private firms, provision of social security at fair returns cannot raise welfare
and will reduce it if mandatory purchases are above private purchases. Unlike
adverse selection where the government could pool risks when the firms could
not, the government has no advantage over firms in the case of moral hazard.
14.7
Conclusions
Many aspects of social security have been addressed in this chapter from the
design of an optimal program in the Diamond economy to the interaction of social security and asymmetric information. In terms of broad policy conclusions,
the most compelling must be that there are many circumstances in which the
introduction of a social security program can raise welfare and, of those considered above, only myopia involved any irrationality on the part of the agents in
the economy.
Without uncertainty, social security could be justified by dynamic ineciency in the economy even when the retirement decision was made endogenous.
More surprisingly, the conflicting eects of an increase in the rate of population
growth lead to there being several possible relations between it and the level of
welfare and, consequently, between the rate of population growth and the structure of social security. With aggregate uncertainty, social security could play a
limited redistributive role between fortunate and unfortunate generations. Individual uncertainty introduced asymmetric information and, as the government
need have no informational advantage over other agents, it was not always possible for social security to raise welfare. Finally, it should also be noted that
voting mechanisms were shown to be an inecient means of determining the
level of social security to be provided since there was no obvious connection
between the level they determined and the optimal level.
Chapter 15
Debt and Taxes

15.1
Introduction
In the policy analyses of previous chapters it was invariably assumed that the
government revenue requirement was pre-determined and the policy instruments
were chosen to maximise welfare given this fixed requirement. The satisfaction of
the revenue constraint implicitly prevented the government issuing any debt. In
a static setting there is some sense in this procedure since, by its very nature, the
static setting prevents a comprehensive analysis of the eects of the borrowing
and repayment process. Introducing time and considering the intertemporal
maximisation of welfare permits the endogenisation of government debt and
allows the determination of its optimal level to become part of the overall policy
formulation process. It also allows the interaction between debt and taxation
to be explored. The employment of both debt and taxation as instruments of
government policy will aect the capital market of the economy and, through
this, the dynamic evolution of the economy and the eventual steady state.
The content of this chapter will reflect these intertemporal issues. Section 2
is concerned with the eects of the maintenance of a constant stock of debt upon
the long-run equilibrium of the economy. Section 3 then analyses the relation
between lump-sum taxation and debt and characterises the optimal combination
of the two instruments. The focus of Section 4 is upon debt neutrality and the
circumstances in which this does, and does not, hold. Section 5 of the chapter
is more concerned with tax policy than debt and it studies the optimisation
of income and interest taxation in an economy with an heterogeneous population. In this case, both intragenerational and intergenerational redistribution
are important.
It should be noted that the policy optimisation problems studied in this
chapter are complicated by the infinite timespan of the economy. This leads to
the maximisation being subject to the infinite set of constraints which describe
the evolution of the economy. Two ways of characterising the solutions to such
problems are illustrated. The first is based upon the methodology of dynamic
469
470
CHAPTER 15. DEBT AND TAXES
programming; in particular the principle of optimality. The second approach

reduces the dimensionality of the problem by assuming that the chosen allocations achieve Pareto optimality between generations. This provides a shadow
interest rate with which future variables can be discounted.
15.2
The eect of debt
The eect of national debt has long been a contentious issue, with the debate
stretching back to at least Ricardo (1817). It is also a subject at the heart
of policy analysis since the issuing of debt is an important practical policy
tool. The conservative perspective upon debt is that it can only be harmful to
welfare and that governments should do all they can to minimise borrowing.
Alternatively, the philosophy of Keynsianism is that debt is simply another
tool of policy and should be employed whenever advantageous. To assess these
competing viewpoints, it is clear that a dynamic economy must be analysed as
its eects are felt through its issuing, servicing and redemption.
The existence of debt instruments as an alternative to investment in physical
capital as a means of saving for consumers alters the relationship between the
level of savings and available capital for the following period. In fact, capital
accumulation must be less for any given level of private saving. The taxes
required to service debt also have an eect upon the behaviour of consumers and
the equilibrium of the economy. Given these observations, the question remains
as to whether debt is always harmful to welfare. This section will determine
both the positive eects of debt on the long-run equilibrium of the economy
and its welfare eects. This is undertaken for both internal and external debt.
The analysis is set in the Diamond economy, with a fixed labour supply, and a
diagrammatic framework is employed that simplifies the derivation of results.
15.2.1
External and internal debt
The starting point of the analysis is to distinguish between internal and external
debt. This is necessary since they have distinctly dierent eects upon the
stationary equilibrium and the welfare level of consumers.
External debt is debt owed to agents located abroad, that is the holders of the
debt are not resident in the economy that issued the debt. On such debt there
is an interest charge that must be paid each period and this must be financed
by either further borrowing or by taxation. The important features of such debt
are that it is not in competition with physical capital as a savings instrument
for the consumers of the issuing economy but its servicing and repayment lead
to a flow of resources out of the economy.
In contrast, internal debt is held by residents of the economy in which it
is issued. With internal debt the government borrows o its own citizens by
providing bonds which compete with private capital. Private savings are divided
between the two investment instruments. When there is no uncertainty bonds
and capital will be perfect substitutes, so in equilibrium they must pay the same
15.2. THE EFFECT OF DEBT
471
rate of return. The cost of financing internal debt is again met by either further
borrowing or by taxation. In contrast to external debt, internal debt does not
lead to any resources being transferred away from the economy that issues the
debt.
15.2.2
Eects of debt
The analysis now considers the real eect upon the steady state equilibrium,
and the utility level of each consumer, of internal and external debt, starting
with external debt. The analytical technique employed, as in Diamond (1965),
is to characterise the steady state of the economy as the simultaneous solution
to a pair of equations for a fixed stock of debt. The consequence of a change in
the level of debt can then be found by determining its eect upon the graphs of
these equations.
External Debt
It has already been noted that the analysis is conducted for a fixed stock of
debt. Before it can be undertaken, it is first necessary to clarify the sense in
which the level of debt is fixed since alternative interpretations are possible.
If the stock of debt were fixed in absolute terms, it would eventually have no
eect upon the economy as population growth took place and the level of debt
per capita became asymptotically zero. Therefore, rather than fix the absolute
stock of debt, it is assumed that the level of debt per young consumer remains
constant and the eect of changes in this ratio are analysed.
Denote the quantity of external debt by in period t by Dt1 and the debt per
D1
young consumer ratio, Htt , by d1t . If the ratio is to remain constant, then the
growth of the absolute stock of debt is governed by
1
Dt+1
= [1 + n] Dt1 ,
(15.1)
where n is the growth rate of the population. The payment of interest upon the
debt is financed either by new borrowing or through taxation. With interest
rate rt , the payment of interest upon the debt in period t is
rt Dt1 ,
(15.2)
nDt1 .
(15.3)
and new borrowing is

The dierence between (15.2) and (15.3) given by
[rt n] Dt1 ,
(15.4)
represents that part of the interest payments that must be financed by taxation.
The tax employed to generate the required revenue is assumed to take the
form of a lump-sum tax on the young generation. Given (15.4), the level of this
472
tax for each consumer when young is

Tt = [rt n]
Dt1
= [rt n] d1t .
Ht
(15.5)
With this tax payment included, the budget constraint of a consumer born in
period t becomes
xt+1
t
= wt [rt n] d1t ,
(15.6)
xtt +
1 + rt+1
ct , is given by
so that full income, M
ct = wt [rt n] d1t .
M
(15.7)
From the maximisation of the utility function U = U xtt , xt+1

, subject to
t
(15.6), the savings function for a member of generation t is
ct , rt+1 = wt [rt n] d1t xtt M

ct , rt+1 .
st = st M
(15.8)
Capital market equilibrium requires that the capital-labour ratio is related
to the level of savings by
ct , rt+1 ,
[1 + n] kt+1 = st M
(15.9)
so that the interest rate satisfies
ct , rt+1
ct , rt+1
st M
wt [rt n] d1t xtt M
= f0
.
rt+1 = f 0
1+n
1+n
(15.10)
Given rt , (15.10) implicitly determines a relationship between wt and rt+1 . This
relation describes the locus of pairs of wage rate and interest
rate for which
capital demand is equal to supply and is denoted by rt+1 = wt ; rt , d1t and is
dependent upon the level of internal debt via the determination of the level of
savings.
Profit maximisation by firms occurs when the wage rate satisfies the equality
wt = f (kt ) kt f 0 (kt ) and the interest rate rt = f 0 (kt ), where kt is the capitallabour ratio. Together these imply the factor price frontier (see 14.4.1) wt =
(rt ) with 0 (rt ) = kt < 0 and 00 (rt ) = f100 > 0. It should be noted that
(rt ) is independent of the level of debt.
The long-run steady-state
equilibrium of the economy, with rt = r all t,

occurs when r = w; r, d1 and w = (r) are satisfied simultaneously, where
d1 is the constant stock of debt. If an equilibrium exists, the steady state will
be stable if the curve is steeper (in absolute value) than the curve. It
is assumed that the curve is downward sloping, that an equilibrium exists
(so the curves intersect) and that the equilibrium is stable and unique. These
assumptions give the configuration of Figure 15.1 which also shows part of the
473
Figure 15.1: Equilibrium and Stability
Figure 15.2: Eect of Increase in External Debt

path of the economy starting from an initial point r0 . It can be seen that, given
the assumptions placed on the gradients of the curves, the adjustment process
is convergent to the equilibrium at the intersection of the two curves.
To discover the eect of changes in the level of external debt it is only
necessary to determine how the change aects the curve. From (15.10) it is
evident that the eect will depend upon sgn.{rt n}. If rt = n, so that the
Golden rule is achieved, there will be no eect. From dierentiating (15.10),
holding rt constant, it can be found that
st
f 00 [n rt ] w
rt+1
t
=
.
t
d1
1 + n f 00 rst+1
(15.11)
The assumption that the curve is downward sloping restricts that denominator
of (15.11) to be positive. The numerator is positive if rt > n and negative for
rt < n. Hence, for given rt , the curve pivots around the point r = n as the
level of external debt increases. This is illustrated in Figure 15.2 below.
It is now straight forward to exhibit the eect of an increase in debt upon
474
Figure 15.3:
Figure 15.4:
the long-run equilibrium. In the steady state

!
!
1
1
,
r
,
r
s
w
[r
n]
d
s
(r)
[r
n]
d
r = f0
= f0
.
1+n
1+n
(15.12)
From dierentiating (15.12) the eect of an increase in external debt upon the
steady state interest rate is
s
f 00 [n r] w
r
=
.
s
s
d1
1 + n f 00 r
+ f 00 d1 w
(15.13)
The three possible outcomes, which depend upon the sign of n r, are shown
in Figure 15.3.
In both cases (a) and (c) where n r is non-zero, the increase in external
debt moves the steady state interest rate, determined by the intersection of the
curves, further from the Golden rule interest rate. If r < n, then the steady
state wage rate w rises and if r > n, then w falls. When the economy is already
475
Figure 15.5: Steady State Eect

at the Golden rule interest rate then the dierential increase in debt has no
eect.
The eect upon the welfare level of a consumer via changes in the steady
state levels of r and w is found by dierentiating the utility function, defined
following (15.7), taking account of the demand function and using the factor
price frontier to eliminate the wage rate. Doing this gives
k
dU
dr
1 dr
=
U
+
[r
n]
+
d
.
(15.14)
[r
n]
1
dd1
dd1 1 + r
dd1
The three terms in the bracket on the right-hand side of (15.14) are related to
the taxes required to finance the debt, the change in the tax burden due to
interest rate changes and the change in factor payments respectively.
Theorem 15.1 summarises the eects of external debt.
Theorem 67 (Diamond)
(i) If r > n, a dierential increase in external debt raises r, reduces w and
reduces utility;
(ii) If r < n, an dierential increase in external debt reduces r, raises w and
may increase utility;
(iii) If r = n, an dierential increase in external debt has no eect.
Proof. (i) When r > n the first and second bracketed terms of (15.14) are
dr
positive. As dd
1 has the sign of [r n], the third term is also positive. It is
then clear that utility will fall as the level of debt increases.
(ii)When r < n the first two terms of (15.14) are negative and the third
positive. Hence there are osetting eects which arise due to r falling, but w
rising.
(iii) Obvious from (15.13) and (15.14).
The reasoning behind the conclusions of this theorem are that when r > n,
the capital stock is too small and the levying of a tax to service the debt has
the eect of taking the economy further from the optimal level of capital. The
476
opposite applies when r < n and eciency is increased by the reduction in the
capital stock. This gain has to be oset against the direct loss in welfare due to
the imposition of the tax and the net eect may still be a reduction in welfare.
Internal Debt
In the case of internal debt, the issue of debt results in the government borrowing
from its own citizens. This leads to competition between government debt
and physical capital for consumer savings. Internal debt therefore aects the
economy through both the taxes required to service the debt and through its
eect on the capital market.
Denoting the ratio of internal debt to the number of young consumers in
period t by d2t , the savings function of a consumer of generation t is now given
by
st = st wt [rt n] d2t , rt+1 .

(15.15)
In (15.15), the term [rt n] d2t captures the tax payment required to service the
debt. On the supply side of the capital market, there is now both private capital
and government bonds. Equilibrium requires that the aggregate level of saving
must be equal to the total stock of capital and debt, or
2
St Ht st = Kt+1 + Dt+1
.
(15.16)
Expressing (15.16) in terms of quantities per member of generation t + 1 gives

the alternative equilibrium condition
st
= kt+1 + d2t .
1+n
(15.17)
The curve for the case of internal debt is found by substituting from (15.17)
into rt+1 = f 0 (kt+1 ) to give
rt+1 = f 0
st wt [rt n] d2t , rt+1

d2t .
1+n
(15.18)
It follows from (15.18) that for constant rt

rt+1
=
d2t
i
h
st
+
1
+
n
f 00 [rt n] w
t
t
1 + n f 00 rst+1
Assuming that consumption is a normal good, so 0 <

rt+1
d2t
st
wt
(15.19)
< 1, it follows that
> 0. An increase in internal debt, holding rt constant, therefore causes

the curve to move out to the right. The eect of this upon the steady state is
to lower the wage rate but to increase the interest rate, regardless of the initial
position. This is shown in Figure 15.4.
15.3. OPTIMISATION OF DEBT AND LUMP-SUM TAXES
477
Figure 15.6: Eect of an Increase in Internal Debt

The change in steady state utility brought about by the increase in debt can
be found by dierentiating the utility function to be
k + d2 dr
dU
= U1 [r n] 1 +
.
(15.20)
dd2
1 + r dd2
The eects of internal debt are summarised in the following theorem.
Theorem 68 (Diamond)
(i) If r > n, utility is decreased by a dierential increase in internal debt.
(ii) If r < n, utility is increased by extra debt.
(iii) If r = n there is no eect.
dr
Proof. As dd
2 > 0, the second bracketed term in (15.20) is positive. The
conclusions then follow immediately.
The explanation behind this theorem is that when r > n the increase in
debt raises the interest rate and moves the equilibrium further from the optimal
point of r = n. This must reduce welfare. When r < n the increase in debt
causes r to move closer to n. This eect is sucient to raise welfare despite the
competition for savings between government debt and physical capital.
15.3
Optimisation of debt and lump-sum taxes
In a static economy with identical consumers, the only role that lump-sum taxation can have is to collect revenue (and, of course, it is the most ecient method
of revenue collection). Other than this, there is no role for it to play in raising
eciency or welfare. The same cannot be said in an overlapping generations
context since the potential ineciency of the competitive equilibrium opens up
additional avenues by which lump-sum taxation can aect welfare. By varying
the timing of taxation over the life-cycle it is possible for lump-sum taxes to affect the savings decisions of consumers and, through this mechanism, the level of
the capital stock. This aects the intertemporal evolution of the economy. The
478
employment of lump-sum taxation also makes it possible for the government

to achieve intergenerational transfers of income. These observations makes the
study of the design of lump-sum taxes a worthwhile exercise.
The purpose of this section is to characterise an optimal lump-sum tax policy
for an economy with an homogeneous population of consumers. The homogeneity assumption removes the need to consider intragenerational transfers and
focuses attention upon the intergenerational consequences of taxation. It also
overcomes some of the objections to the use of lump-sum taxes expressed in
Chapter 2. Before undertaking this, it is first necessary to determine the relationship between lump-sum taxes and internal debt since there are important
interconnections between the two policy instruments.
15.3.1
Lump-sum taxation and debt
The provision of government debt and the use of lump-sum taxation are both
methods of aecting the life-cycle behaviour of consumers. When viewed as
such, it is not surprising that there is a significant degree of substitutability
between the two instruments. To investigate this substitutability formally, it
is assumed that each household in generation t, in a Diamond economy with
variable labour supply, is subject to a lump-sum tax of Ttt in the first period of
their life and one of Ttt+1 in the second period. Either, or both, of these may
be negative. With these taxes, the lifetime budget constraint of a consumer is
xtt +
xt+1
Ttt+1
t
= wt `t Ttt
.
1 + rt+1
1 + rt+1
(15.21)
Using R to denote the level of government revenue to be collected, the government budget constraint is
t
= R + [1 + tt ] Dt1 Dt ,
Ht Ttt + Ht1 Tt1
(15.22)
where Dt is the quantity of debt issued in period t. In per-capita terms the

government budget constraint becomes
Ttt +
t
Tt1
e + [1 + rt ] dt1 dt ,
=R
1+n
1+n
(15.23)
Dt
t1
R
e
with dt1 D
Ht1 , dt Ht and R Ht . Since the lump-sum tax in the
first period of life, Ttt , aects consumers savings decisions, the capital market
equilibrium condition becomes
where e
kt+1
Kt+1
Ht+1
[1 + n] e
kt+1 = wt `t Ttt xtt dt ,
(15.24)
is the level of capital per young consumer. It is important
to distinguish this from the capital labour ratio kt+1

are not equal when labour supply is variable.
Kt+1
Ht+1 `t+1
since the two
479
t1
t+1
t
A tax policy is a sequence of lump-sum taxes ..., Tt1
, Tt1
, Ttt , Ttt+1 , Tt+1
, ...
and a debt policy is a sequence of government debt issues {..., dt1 , dt , dt+1 , ...}.
Together with the factor price frontier, (15.21), (15.22) and (15.24) describe the
equilibrium of the economy for any given tax policy and debt policy.
The following theorem, originally due to Bierwag et al. (1969), relates tax
policies and debt policies.
t1
t+1
t
Theorem 69 (Bierwag, Grove, Khang) The combined tax policy ..., Tt1
, Tt1
, Ttt , Ttt+1 , Tt+1
, ...
and debt policy {...,ndt1 , dt , dt+1 , ...} lead to the same
o real equilibria as:
t1 bt
t bt+1 bt+1
b
b
(i) a tax policy ..., Tt1 , Tt1 , Tt , Tt , Tt+1 , ... and no debt, where Tbtt =
Ttt + dt and Tbtt+1 =nTtt+1 [1 + rt+1 ] dt ;
o
n
o
(ii) a tax policy ..., Tet1 , 0, Tet , 0, Tet+1 , ... and a debt policy ..., det1 , det , det+1 , ... ,
t1
Ttt+1
1+r
nt+1
t+1
Ttt+1
1+rt+1 ;
o
t+2
t
, 0, Tbtt+1 , 0, Tbt+1
, ... and
(iii) a tax policy ..., 0, Tbt1
where Tbtt+1 = [1 + rt+1 ] Ttt + Ttt+1 and dbt = dt + Ttt ;
where Tett = Ttt +
and det = dt
n
o
a debt policy ..., dbt1 , dbt , dbt+1 , ... ,
and
n
o
(iv) a tax policy {..., Tt1 , Tt , Tt+1 , ...} and a debt policy ..., dt1 , dt , dt+1 , ... ,
h
ih
i
Ttt+1
Tt
Ttt+1
t+1
.
Ttt + 1+r
and dt = dt 2+rtt+1 2+r
where Tt = 1+r
2+rt+1
t+1
t+1
Proof. (i) With the tax policy alone the equilibrium conditions become
xtt +
and
xt+1
Tbtt+1
t
= wt `t Tbtt
,
1 + rt+1
1 + rt+1
Tbt
e
Tbtt + t1 = R,
1+n
[1 + n] e
kt+1 = wt `t Tbtt xtt .
(15.25)
(15.26)
(15.27)
Contrasting (15.24) and (15.27) shows that the capital market will be unaected
by the change in policy if Tbtt = Ttt + dt . Substituting into (15.26) and equating
to (15.23) gives Tbtt+1 = Ttt+1 [1 + rt+1 ] dt . It can also be seen that these
changes leave the consumers budget constraint unaected. Parts (ii) to (iv) are
proved in an identical manner.
The relevance of Theorem 15.3 is that it shows that a restricted lump-sum
tax policy, either one employing uniform taxes (case iv) or one with taxes levied
in only one period of life (cases ii and iii), and an appropriate debt policy is
as eective as lump-sum taxes dierentiated across the lifecycle. It is arguable
that the uniform lump-sum tax policy requires less information than the dierentiated policy so that employing a uniform tax can economise on information
with no loss in potential welfare. The theorem also shows that in characterising
the optimal lump-sum tax policy, any one of the five combinations of tax and
debt policies will lead to the same real equilibrium.
480
15.3.2
Optimal lump-sum taxation and debt
Theorem 15.3 shows that in order to demonstrate the eect of an optimal lumpsum tax policy it is sucient to consider only one of the five alternative combinations of lump-sum taxes and debt. The representation chosen here will be
to adopt (i) and to set government debt equal to zero in all periods. Once the
results are derived, the translation can easily be made to one of the other forms
of policy. The technique used to characterise the choice of policy is the principle
of optimality of dynamic programming which has been employed in this context
by Diamond (1973), Atkinson and Sandmo (1980) and Park (1991).
The aim of the government is assumed to be the maximisation of the discounted sum of utility from period s, in which the maximisation takes place,
onwards. With discount rate , the social welfare function is that used in Chapter
13 Section 3.5.2 and is given by
X
Ttt+1
t V Ttt +
,
(15.28)
1 + rt+1
t=s
Ttt+1
where V Ttt + 1+r
is the indirect utility function of the generation born in
t+1
t. It is clear from the budget constraint in (15.21) that it is the discounted sum
of lump-sum taxes that determines welfare rather than the individual taxes.
The maximisation is constrained by the behaviour of the household and by
production possibilities. The maximisation of utility by each consumer subject to (15.21)
determines
commodity
demands and
labour supply
of the form
T t+1
T t+1
T t+1
t
t
t
xtt = xtt Ttt + 1+r
= xt+1
, xt+1
Ttt + 1+r
and `t = `t Ttt + 1+r
.
t
t
t+1
t+1
t+1
With a working population of Ht each choosing to undertake hours `t of employment, output with a capital stock of Kt is given by F (Kt , Ht , `t ). Assuming
that the production function is homogeneous of degree one,
output is given in

e
Kt
kt
e
terms of capital per young consumer, kt = Ht , as Ht `t f `t . Output is allocated between consumption, government purchases, R, and savings. With no
depreciation, the demand for output equals its supply when
F (Kt , Ht , `t ) + Kt = Ht xtt + Ht1 xtt1 + R + Ht st .
(15.29)
Expressing this equality in terms quantities of per worker (15.29) becomes

!
e
xt
kt
e + [1 + n] kt+1 kt .
= xtt + t1 + R
(15.30)
`t f
`t
1+n
The equality of savings in period t to capital in t + 1 requires that

[1 + n] e
kt+1 = wt `t Ttt xtt ,
which can be used in (15.30) to give the alternative representation of production

possibilities
" !
#
e
xt
k
t
e
e + Ttt = 0.
(15.31)
wt t1 R
kt + `t f
`t
1+n
481
Equation (15.30) can also be solved to express the capital stock of period t + 1
as a function of variables at t. Doing this gives
!
#
"
t
e
x
1
k
t
t1
t
e
e+e
kt+1 =
xt
(15.32)
`t f
R
kt .
1+n
`t
1+n
The optimal choice of lump-sum taxation is now characterised by the use

of the maximum principle of dynamic programming. The government enters
period s with the capital stock already determined by the savings decision of
period s 1. In addition, the lump-sum taxes aecting generation s 1 are
also pre-determined. The maximum welfare that can be achieved from period
s
Ts1
s1
s onwards is therefore a function of e
ks and Ts1
+ 1+r
. This maximum value
s
s
T
function is denoted e
ks , T s1 + s1 ; s or just (s) when the meaning is
s1
1+rs
clear. From the basic recursive relation of dynamic programming (see Stokey
and Lucas with Prescott (1989)), the maximum value at s must be equal to
the maximised value of period ss indirect utility, constrained by productive
feasibility, plus the maximum value attainable in s + 1 discounted. That is
(
"
" !
#
#
)
e
xss1
ks
s
e
e
(s) = max
V (Zs ) + s ks + `s f
ws
R + Ts + (s + 1) ,
`s
1+n
{Tss ,Tss+1 }
(15.33)
Tss+1
Tss+1
s
s
e
where Zs = Ts + 1+rs+1 and (s + 1) = ks+1 , Ts + 1+rs+1 ; s + 1 .
The necessary conditions for the maximisation in (15.34) with respect to Tss
and Tss+1 are
"
"
#
#
e
V
ws
(s + 1)
`s
0 ks
+ s
f ws f
`s
+ s +
Zs
Zs
`s
Zs
Zs
"
#
#
"
e
(s + 1)
xss
1
`s
ks
+
f f0
=0
(15.34)
1+n
Zs
`s
Zs
e
ks+1
and
"
"
#
#
e
(s + 1)
`s
1
1
ws
1
V
0 ks
+ s
+
f ws f
`s
Zs 1 + rs+1
Zs
`s
Zs 1 + rs+1
Zs
1 + rs+1
"
#
#
"
e
1
(s + 1)
xss
1
`s
0 ks
+
= 0.
(15.35)
f f
e
1+n
Zs
`s
Zs 1 + rs+1
ks+1
Comparison of (15.35) and (15.36) shows immediately that s = 0. The reason

for this is that the flexibility in the lump-sum taxes allows Tss to be varied
whilst keeping Zs constant. This renders the production constraint ineective.
Conditions (15.35) and (15.36) therefore both reduce to
"
#
#
"
e
(s + 1)
(s + 1)
xss
1
V
`s
0 ks
+
+
f f
= 0.
Zs
Zs
1+n
Zs
`s
Zs
e
ks+1
(15.36)
482
Dierentiating (15.34) with respect to the state variables e

ks and Zs1 gives
(s)
(s + 1)
=
[1 + f 0 ] ,
(15.37)
e
1+n
ks
e
ks+1
and
1
(s)
(s + 1) xss1
=
.
Zs1
1+n
e
ks+1 Zs1 1 + n
(15.38)
Equations (15.37) to (15.39) provide the characterisation of the optimal policy.

The structure of the optimal policy is easily seen if the move is now made
= (s+1)
=
so that (15.38)
to the steady state. In a steady state (s)
e
ks
e
ks+1
e
k
reduces to
(15.39)
1 + n = [1 + f 0 ] ,
which is the modified Golden rule. Unrestricted lump-sum taxes can therefore
take the economy to the modified Golden rule equilibrium, even with variable
labour supply, since they cause no distortion on any of the markets. This result
should be contrasted to the results on social security in Chapter 14 Section 4.1
where the pension aected the retirement decision. This eect is absent in the
present framework.
In the steady state, (15.39) reduces to
x2
1
=
,
(15.40)
Z
1 + n e
k Z 1 + n
where xi is the steady state level of consumption in the ith period of life. From
the budget constraint (15.21) dierentiation with respect to Z shows that in the
steady state
`
x1
1 x2
w
=1+
.
(15.41)
Z
Z
1 + r Z
Substituting from (15.41) and (15.42) then yields
.
=
Z
1 + n e
k
(15.42)
Since V
Z is the private marginal utility of income, (15.40) and (15.43) show that
the optimal lump-sum tax policy will achieve the modified Golden rule and will
equate the private marginal utility of income to the discounted return of further
investment in capital.
15.3.3
Summary
This section has detailed the connections between lump-sum tax policy and
debt policy and has shown the equivalence of unrestricted lump-sum tax policies to combinations of debt and restricted lump-sum taxes. Combined with
the analysis of optimal lump-sum taxes, these results show how the modified
15.4. DEBT NEUTRALITY
483
Golden rule equilibrium can be sustained by a number of combinations of policy

instruments.
In assessing these results, it should be borne in mind that the economy under
consideration was restricted by the assumption of an homogeneous population.
This limits the scope for the worthwhile dierentiation of lump-sum taxes, since
consumers are only distinguished by their date of birth, and therefore makes
feasible a policy that achieves the optimal outcome. If there were further degrees of dierentiation of the consumers, for example by endowment of labour,
then lump-sum taxes could be dierentiated across the lifecycle and over the
endowment. A debt policy and a uniform lump-sum tax policy would then not
be equivalent to the optimal fully-dierentiated set of lump-sum taxes. There
would then be less justification for expecting the optimum to be achievable since
this would also require intragenerational redistribution which could not be fully
undertaken with uniform lump-sum taxes. However, as shown by Park (1991),
the modified Golden rule will be satisfied by the correct choice of uniform lumpsum taxes and debt policy even with a heterogeneous population but other
taxes will be required to supplement them in order to achieve intragenerational
distributional aims. Since, in practice, populations are not homogeneous, this
observation justifies the study of policies that are not equivalent to the set of
optimal lump-sum taxes.
15.4
Debt neutrality
The concept of debt neutrality, or Ricardian equivalence, can be traced back to

Ricardo (1817). A brief summary of some aspects of debt neutrality relevant
for the analysis of social security has already been given in Chapter 14. In
this section the emphasis will be upon proving the basic neutrality result and
exploring its limitations.
The essence of debt neutrality is that the real economy is unaected by
whether the government chooses to raise revenue by using taxation or through
the issue of debt. Although the payment of taxes clearly reduces wealth, the
issue of debt would seem to leave the wealth of consumers intact since it simply
displaces private capital from the portfolios of consumers. However, this overlooks the fact that the issue of debt implies future tax liabilities for its servicing
and redemption. Discounted back to the present, these future liabilities reduce
the present value of wealth by precisely the same amount as the tax payment.
Since wealth is identical in both cases, so must the equilibrium outcome. This
simple argument is the foundation of the theory of debt neutrality.
The next subsection will formally demonstrate the neutrality result for two
extreme cases. The first considers debt which is redeemed after a single period and the second involves consumers with infinite lives. The second case is
intended to motivate the mechanism by which intergenerational altruism generates the neutrality result in the case of finite lives. The main neutrality result
is then presented and some extensions and limitations are discussed.
484
15.4.1
Two special cases
Debt neutrality will now be demonstrated for two special cases in which the
mechanism behind the result is particularly transparent. These follow the simplifying approach of Barro (1974) in which the government gives an equal number of bonds to each of the identical consumers in the economy. If after the
receipt of these bonds, the consumers do not perceive their wealth to have increased, then debt neutrality is said to apply.
It is assumed that the gift of debt is a one-o event and that taxation must be
used to service the debt and to meet its repayment. When future tax liabilities
are taken into account, the present discounted value of wealth becomes
1 = 0 + D DT L,
(15.43)
where 0 is the consumers wealth before receiving the bonds, 1 the level of
wealth after receiving the bonds, D the value of bonds received and DT L the
discounted tax liabilities arising due to the issue of the debt. Two dierent assumptions on the structure of repayment of the debt are now shown to generate
the same neutrality conclusion.
Under the first set of assumptions the bond is paid o with interest the year
after issue and the same set of consumers is alive in both years. In this case,
with interest rate r, the interest and principle paid the year after issue is given
by [1 + r] D. Discounting this back to the present shows that DT L has value
DT L =
[1 + r] D
= D.
1+r
(15.44)
Substituting from (15.45) into (15.44) shows immediately that the issue of the
debt has no net eect upon wealth. This is a consequence of the recipients of the
bonds realising that they face the tax liability for its repayment. The receipt of
the bond is therefore not treated as an increase in wealth.
The second set of assumptions involve a bond that is never redeemed but on
which interest is paid in perpetuity and consumers who have infinite lifespans.
In this case the discounted tax liabilities must be equal to the discounted value
of the stream of interest payments. This gives
Z
DT L =
rDert dt,
(15.45)
0
where rD is the tax payment per period. Thus DT L = D and the bond again
has no eect upon net wealth.
The special feature of these two cases that makes the neutrality result so
transparent is that the consumers who receive the bonds also pay the future
taxes used to finance them. In an overlapping generations economy, it would
appear at first sight that such a result could only apply in very particular circumstances since the set of living consumers changes each period. The important insight of Barro (1974) was to show how intergenerational altruism could
support the neutrality result in general circumstances.
485
Figure 15.7: Receipts and Payments for One-Period Debt
15.4.2
Intergenerational altruism
The discussion of the examples above emphasised the importance of the recipient
of debt financing its servicing and repayment via taxation. In an overlapping
generations economy this will in general not occur. For example, Figure 15.5
indicates the pattern of receipts and payments if a one-period bond was given
to the young of generation t and redeemed through taxes upon generation t + 1.
If the generations were not linked in any way, the gift of debt in this economy
would raise the wealth of generation t, reduce that of generation t + 1 and would
have real eects upon the equilibrium.
The linkages between generations that are needed to support the neutrality
result in an overlapping generations economy can be introduced by assuming
that each consumer gives birth to a number of descendants. If the consumer
cares about the welfare of their descendants, so that intergenerational altruism
is present, then it is likely that they will choose to leave a bequest. The eect of
the debt transfer can then be neutralised by changes in the size of the bequest
and, as will be shown below, since the initial size of bequest was optimal it will
also be optimal to neutralise the eect of the debt issue. Conversely, it may be
the case that a consumer cares about their forebear and reflects this by giving
a gift to their forebear. The neutrality result can then be attained by changes
in the size of the gift osetting the debt issue.
To formalise these arguments, consider first the case of concern for descendants. Assume that each consumer has 1 + n descendants and that the utility
of a typical member of generation t is assumed to be dependent upon the utility
they derive from consumption and the return from the well-being of their descendants. Adopting an additively separable functional form for simplicity, the
utility function can then be written as
U (t + 1)
U (t) = U xtt , xt+1
+
,
t
1+
(15.46)
where U (t + 1) is the welfare level of each descendent and is the discount

rate. The budget constraints of the consumer, incorporating the taxes required
to finance the debt and the bequests given and received, are
xtt + st = wt Tt ,
(15.47)
+ [1 + n] bt = [1 + rt+1 ] [st + bt1 ] .

xt+1
t
(15.48)
and
486
These budget constraints embody the assumptions that each consumer in generation t has 1 + n descendants and leaves a bequest of bt . They receive a bequest
of bt1 in the second period of their lives and, since this was actually bequethed
in the previous period, it receives interest at the market rate. Lump-sum taxes,
Tt , are paid in the first period of life.
The necessary conditions for the optimal choice of consumption and bequest
can be derived from (15.47) - (15.49) as
U
xtt
U
xt+1
t
and
= 1 + rt+1 ,
(15.49)
1
U (t + 1)
U
[1
+
n]
,
1+
bt
xt+1
t
(15.50)
with bt = 0 if (15.51) is a strict inequality. Noting that an increase in the bequest

is simply an increase in second period lump-sum income for the descendants
which, after the addition of interest, must be valued at the marginal utility
of consumption, it follows that U b(t+1)

= [1 + rt+2 ] xU
t+2 so (15.51) can be
t
t+1
written as
U
xtt
U
xt+2
t+1
1 + rt+2
.
[1 + ] [1 + n]
(15.51)
Letting Dt be the value of bonds issued in period t the budget constraint of

the government in period t is
[1 + rt ] Dt1 = Dt + Ht Tt .
By defining dt =
Dt
Ht ,
(15.52)
(15.53) can be alternatively written

[1 + rt ] dt1 = [1 + n] dt + [1 + n] Tt .
(15.53)
With bequests, equilibrium in period t satisfies the usual conditions ((13.39)

and (13.40)) for the determination of the interest rate and wage rate, and the
capital market equilibrium condition
st + bt1 = [1 + n] kt1 + dt .
(15.54)
To characterise the steady state, the level of debt and the lump-sum tax
rate are held constant at d and T . Substituting from the steady state version of (15.53) and (15.54) into (15.48) and (15.49), the steady state levels of
consumption satisfy

1 + f0
1
0
x = f kf [1 + n] k + b
d ,
(15.55)
1+n
and

1 + f0
x2 = [1 + n] [1 + f 0 ] k [1 + n] b
d .
1+n
(15.56)
487
The steady state conditions for consumer choice (15.50) and (15.52) reduce to
U
x1
U
x2
= 1 + f 0,
(15.57)
and
[1 + n] [1 + ] 1 + f 0 .
(15.58)
From (15.59) the bequest motive will only be operative in the steady state if
[1 + n] [1 + ] = 1 + f 0 . Since > 0, this implies that the equilibrium interest
rate must be above the growth rate of population so the capital-labour ratio
must be below the Golden rule level for bequests to be positive.
It is now possible to prove the first neutrality result.
Theorem 70 (Barro) If the bequest motive is operative, so that bequests are
positive both before and after any change in the level of bonds, such a change has
no eect upon either the short run equilibrium or the steady state equilibrium.
Proof. To see that there is no eect upon the steady state equilibrium, it is
sucient to observe that the consumption levels hare determined
by conditions
i
0
(15.56) dependent upon the values of k and b 1+f
d.
It
is
then
clear that
1+n
any change in d can be oset by a change in b, leaving k and the steady state
consumption levels unchanged. Such a change also leaves (15.58) and (15.59)
satisfied.
To show that a change in the level of debt is neutral in the short run, consider
the government wishing to increase d to d + de starting at time t. It can achieve
e
this by giving each member of generation t 1 a quantity
h
i d of bonds, which
n
they will sell to generation t, and by selling bonds 1+n de to each member of
generation t. To
h finance
i the interest payments on the bonds held by generation
rt n e
t 1 a tax of 1+n d is levied on each member of generation t. With these
changes, the budget constraint of a member of generation t 1 in period t is
e
xtt1 = [1 + rt ] [st1 + bt2 ] [1 + n] bt1 + [1 + rt ] d,
(15.59)
and the constraints facing generation t are
xtt = wt Tt st
and
rt n e
d,
1+n
= [1 + rt+1 ] [st + bt1 ] [1 + n] bt .

xt+1
t
(15.60)
(15.61)
The capital market equilibrium condition becomes

e
st + bt1 = [1 + n] kt + dt + d.
(15.62)
To show that the change in the level of debt can

byha changes
h be neutralised
i
i
1+rt e
t+1
in the level of bequests, let bt1 be increased by 1+n d and bt by 1+r
de
1+n
488
h
i
t n
e Substituting these changes in (15.60) to (15.63)
and let st fall by r1+n
d.
shows that the change in debt is cancelled from the equilibrium conditions and
that the levels of consumption remain unchanged. Neutrality therefore applies.
To prove an identical result for the case of concern about forebears, assume
that the utility of a consumer in t takes the form
U (t 1)
U (t) = U xtt , xt+1
+
,
t
1+
(15.63)
where U (t 1) is the welfare level of the consumers forebearer. Denoting the

gift received by a forebear by gt , the budget constraints are
1
(15.64)
gt = wt Tt ,
xtt + st +
1+n
and
xt+1
= [1 + rt+1 ] st + gt+1 ,
t
(15.65)
where these follow from the observation that each forebearer has [1 + n] descendants. The necessary conditions for the optimal choice of consumption and gift
can be calculated to be
U
xtt
U
xt+1
t
and
= 1 + rt+1 ,
1 + n U (t 1)
U
[1 + rt+1 ]
,
1+
gt
xt+1
t
(15.66)
(15.67)
with gt = 0 if (15.68) is a strict inequality. Exploiting the fact that an increase

in gift is equivalent to an increase in lump-sum income, (15.68) can be written
as
U
U (t 1)
=
.
(15.68)
gt
xtt1
The steady state levels of consumption satisfy
1
1
0
[g + [1 + f 0 ] d] ,
x = f kf [1 + n] k
1+n
(15.69)
and
x2 = [1 + n] [1 + f 0 ] k + [g + [1 + f 0 ] d]
(15.70)
and the steady state version of (15.68) is

1 + f0
1+n
.
1+
(15.71)
Gifts can be positive only when (15.72) is an equality which requires the economy
to be overcapitalised relative to the Golden rule.
Using these conditions, the second neutrality theorem follows.
489
Theorem 71 (Carmichael) If the gift motive is operative, so that gifts are positive, both before and after any change in the level of bonds, such a change has
no eect upon either the short run equilibrium or the steady state equilibrium.
Proof. The proof is essentially identical to that of Theorem 15.4. In the
steady state it is the value of g + [1 + f 0 ] d that is relevant so that under the
conditions of the theorem any change in d can be oset by a change in g. In the
short run a change in the level of gifts and savings can also be found that will
oset the change in level of debt.
After the publication of Barros (1974) demonstration of the neutrality result for economies with bequests, the comments of Feldstein (1976), Buchanan
(1976) and Barros (1976) reply were directed to the discussion of whether the
neutrality proposition was applicable when the rate of interest was less than
the rate of growth. As shown in Theorem 15.5, which is based on Carmichael
(1982), neutrality does apply in this latter case if the gift motive, rather than the
bequest motive, is operative. This latter result, and the related work of Buiter
(1980), has itself been criticised by Burbidge (1983) for the form of utility function employed. The contention of Burbidge is that consistency of treatment
between descendants and forebears places additional restrictions upon the rates
of discounting that can be employed. However, as noted in the reply of Buiter
and Carmichael (1984), the utility function is constructed as the representation of an underlying preference ordering and if the preference ordering does
not restrict discount rates nor should the utility function. Viewed in this ways,
the discount factors in (15.47) and (15.64) implicitly adjust for the number of
descendants and forebears respectively.
Given the importance of an operative bequest motive, it is worth clarifying
the conditions under which bequests will be positive. Drazen (1978) provided
a partial analysis of this issue but the first substantive result was given in Weil
(1987). To demonstrate this result, assume that in the economy with concern
for descendants there is no debt and that the utility function is also additively
separable in consumption with the form

U (t 1)
U (t) = U xtt + U xt+1
+
t
1+
(15.72)
The steady state equilibrium in then the solution to
x1 = f kf 0 [1 + n] k + b = w (k) [1 + n] k + b,
(15.73)
x2 = [1 + n] [1 + f 0 ] k [1 + n] b = [1 + n] [1 + r (k)] k [1 + n] b,
(15.74)
and
[1 + n] [1 + ] = 1 + f 0 = 1 + r (k) 1 + r .
(15.75)
The equality in (15.77) captures the assumption that the bequest motive is
operative.
If the optimal level of bequests is zero, (15.76) will be satisfied when evaluated at b = 0. Substituting (15.74) and (15.75) into (15.76) then shows that
490
the optimal value of the bequest will satisfy

sgn. {b} = sgn. {U 0 (w (k) [1 + n] k) [1 + r ] U 0 ([1 + r ] [1 + n] k)} .
(15.76)
Now denote the savings function of the equivalent economy without bequest
motive (i.e. the solution to U 0 (w s) - [1 + r] U 0 ([1 + r] s) = 0) by s =
s (w, r). Using this definition, (15.78) is equivalent to
sgn. {b} = sgn. {[1 + n] k s (w, r)} .
(15.77)
Assuming that the identity [1 + n] k s (w, r) = 0 has a unique solution k,

which would be the steady state capital labour ratio in the economy without
bequest motive, the following theorem can be proved.

Theorem 72 (Weil) If s (w, r) [1 + n] k > 0 k 0, k and s (w, r)
[1 + n] k < 0 k > k, then the bequest motive
is operative in the steady state if
1
0
and only if 1+
> 1+n
,
where
r
=
f
k
is
the
steady state interest rate of the
1+r
corresponding economy without bequest motive.
Proof. From (15.79), the bequest will only be positive if [1 + n] k s (w, r)
> 0. The implies, from the assumptions of the theorem, that the steady state
capital stock, k , of an economy with an operative bequest motive satisfies
1
1+n
k > k. From (15.77), the economy with bequest motive satisfies 1+
= 1+r
and, since k > k, r < r. Hence the bequest motive can only be operative
1
when 1+
> 1+n
1+r .
The assumptions employed in the proof of this theorem regarding the relation
of s (w, r)[1 + n] k to k are equivalent to the assumption that the unique steady
state of the economy without bequest motive is stable. From the conclusion of
the theorem, it can be seen that the bequest motive will be operative provided
the discount factor applied to descendants utility is suciently great. That is,
there must be a sucient degree of intergenerational concern for bequests to be
positive.
This section has shown how bequest and gift motives can extend the twoperiod lived consumers of the overlapping generations economy into infinitely
lived dynasties connected by intergenerational altruism. Once this has been
done, it is not surprising that neutrality results which apply for infinitely lived
consumers then apply in the overlapping generations framework. It is important
to stress that these results rely on changes in the level of bequests or gifts being
able to oset changes in the pattern of debt and taxes. If neither of these
motives are operative or the level of transfer is small, this may not be possible.
15.4.3
Generalisations
One aspect of the neutrality results described above that has received attention
is the assumption of certainty. Blanchard (1985) has considered the possibility
that the lifespan of each consumer may be of random length. The introduction
of such uncertainty has the implication that those who benefit from any debt
491
issue will have a probability of less than one of being alive to face any future
tax payments. In the analysis of Blanchard each consumer has a constant probability of their life continuing (which gives rise to the description of such an
economy as embodying perpetual youth) and the population remains constant
as new consumers are born at each point in time to replace those who have died.
To prevent unintended bequests, a life insurance market oering insurance at
fair terms is assumed to be operational. When there is no bequest motive, neutrality does not hold in such an economy. This analysis is extended by Buiter
(1988) to show that a necessary and sucient condition for debt neutrality to
apply in such an economy is that the sum of the probability of death and the
growth rate of population is identically zero. This is clearly violated in the
Blanchard analysis except when the consumers have a zero probability of death
and are therefore infinitely lived.
Although the focus to this point has been upon debt neutrality, an economy
with an operative gift or bequest motive is capable of neutralising a broad range
of fiscal policies. Since Bernheim and Bagwell (1988) show that almost any policy can be neutralised, the fact that this seems refuted by practice suggests
that the economy in which such propositions can be derived is not a successful
representation of reality. To attempt to achieve debt neutrality without the
neutrality of all policy, Abel and Bernheim (1991) note that changes in the level
of debt are often neutralised by exploiting only a few links in the intergenerational chain whereas the neutralisation of other policies requires the exploitation
of many links. This observation motivates the introduction of friction into the
intergenerational altruism via the derivation of utility from the act of giving,
imperfect knowledge of the later generations preferences and social norms that
govern bequests. The first two allow approximate neutrality in the short run
but drive the marginal propensity to consume to zero. The third avoids this
conclusion but does lead to the conclusion that an exogenous increase in wealth
of any one consumer can never be a Pareto improvement. Due to these unpalatable conclusions, the introduction of friction in this way cannot be claimed to
be a successful mechanism for retaining debt neutrality whilst eliminating more
general neutralities.
Finally, it should be noted again that debt neutrality fails if the gift or bequest motives are not operative. In addition, it can also fail when the change in
policy leads to redistribution between consumers with diering marginal propensities to consume. This did not occur in the theorems above since all consumers
were identical and the policy aected each equally. It can also occur when lifetimes are uncertain and insurance markets are imperfect (Abel (1986)) and when
the tax instruments are not lump-sum. As yet, neither empirical evidence (see
Bernheim (1987) and Sweeney (1988)) nor experimental evidence (Cadsby and
Frank (1991)) has confirmed or refuted whether neutrality applies in practice.
492
15.5
Income and interest taxation
In a dynamic economy an income tax leads to the double taxation of savings:

once when the income is received and then again when the interest income from
savings is received. This fact has often been used to support the contention,
for example in Meade (1975), that an expenditure tax which avoids the double
taxation would be preferable to an income tax. This view overlooks the fact that
it is necessary to establish that the double taxation leads to an inferior outcome.
Simply counting the number of instances of distortion is not a sucient analysis.
More relevant for the present discussion is the potential dynamic ineciency of
an overlapping generations economy. If the economy is on a path with excessive
capital accumulation then there may be a gain to introducing a distortion into
the savings decision that reduces the level of capital. Since interest and income
taxes have dierent eects upon the steady state equilibrium of the economy,
the design of the tax system should take advantage of this by combining the
two instruments. These issues are now addressed for nonlinear taxes on income
and interest in an economy with a heterogeneous population.
The optimal combination of income and interest taxation has been approached using two dierent methods. Atkinson and Sandmo (1980) and Park
(1990) employ the dynamic programming method described in Section 3. The
results of Atkinson and Sandmo are restricted to an economy in which consumers
only dier with respect to their date of birth and are otherwise homogeneous.
Such a setting does not explore the full potential of the taxes for eecting intragenerational transfers. Although Park considers an heterogeneous population,
few concrete results are derived. The alternative approach, developed in Ordover and Phelps (1975, 1979), Ordover (1976), and Phelps and Riley (1978), is
based on the observation that what links generations t and t + 1 are the capital
stock provided by the members of generation t and their claim to consumption
in the second-period of life. The maximisation of social welfare can then be
reduced to a single-period problem by assuming that the welfare of generation
t is maximised subject to the constraints that the capital stock provided for
generation t + 1 does not fall below some chosen level and that second-period
consumption is not above some fixed level. This alternative approach is now
described.
The details of the analyses dier in the objective functions of the government
and the manner in which the constraining levels of capital stock and secondperiod consumption are chosen. Ordover and Phelps (1975), Ordover (1976),
and Phelps and Riley (1978) all employ a Rawlsian maxi-min objective for
the government whereas Ordover and Phelps (1979) consider a more general
Bergson-Samuelson social welfare function. In Ordover and Phelps (1975) the
levels of capital stock and second-period consumption were arbitrarily chosen
and it was required that they were held constant over time. Although simple,
this procedure cannot, in general, maximise any social welfare function satisfying
the Pareto criterion (2.49) since opportunities to raise welfare are being missed.
Phelps and Riley (1978) showed how the requirement that the capital stock
and level of second-period consumption be Pareto optimal in the allocations
15.5. INCOME AND INTEREST TAXATION
493
between generations could be given a simple representation and incorporated

into the optimisation and this made possible the general analysis of Ordover
and Phelps (1979) which is now described.
The heterogeneous population is introduced by combining the Mirrlees economy of Chapter 5 with the Diamond overlapping generations economy. Each
consumer lives for two periods, consuming in both but working only in the
first period. Within a generation, consumers are dierentiated by their level
of ability. Denoting the ability level by s, 0 s S, S finite, the cumulative distribution of s is given by (s) with the normalisation (S) = 1. This
normalisation restricts the generations to be of equal size. The corresponding
density function is (s). Writing the consumption levels and labour supply (in
hours) of a consumer of ability s born in t as xtt (s) , xt+1
(s) and `t (s), the
t
consumption levels and eective labour supply per member of generation t are
Z S
1
xt (s) =
xtt (s) (s) ds,
(15.78)
0
x2t (s) =
and
zt (s) =
xt+1
(s) (s) ds,
t
(15.79)
s`t (s) (s) ds.
(15.80)
The preferences of each consumer are given by the utility function U = U xtt , xt+1
, `t
t
and this is maximised subject to the budget constraint
xtt +
xt+1
t
= wt s`t T` (wt s`t ) Ts xt+1
.
t
1 + rt+1
(15.81)
In (15.83),
T` (wt s`t ) is the income tax function, with wt the wage rate, and
Ts xt+1
the tax on second-period consumption. The latter tax is essentially
t
equivalent to a tax on savings. To simplify the analysis by eliminating the
need for discounting in the government budget, it is assumed that the tax on
second-period consumption is pre-paid in the first period of life.
Given consumer choices xtt (s) , xt+1
(s) and `t (s), total dierentiation of the
t
utility function with respect to s gives
z0
z0
dU
xt+10
+ U`t t U`t 2t .
= Uxtt xt0
t + Uxt+1
t
t
ds
s
s
(15.82)
Applying
the argumentof (5.13) - (5.15) that s0 = s must minimise U (s0 )
U xtt (s) , xt+1
, it follows that
(s) , zts(s)
0
t
t+10
Uxtt xt0
+ U`t
t + Uxt+1 xt
t
and hence that
zt0
= 0,
s
z0
dU
= U`t 2t 0.
ds
s
(15.83)
(15.84)
494
Equation (15.86) is the first-order condition for self-selection that must constrain
the choice of tax functions. The final part of the description
of the consumer is
to solve the identity U (s) = U xtt (s) , xt+1

(s)
,
`
(s)
to
write
t
t
xtt (s) = xtt U (s) , xt+1

(s) , `t (s) .
t
(15.85)
The production
function
for the economy is given by the constant returns
e
kt is the level of capital stock per member
to scale function F kt , zt where e
of the young generation. The identities wt = Fz e

kt , zt and rt = Fk e
kt , zt
determine the wage rate and interest rate via optimal choice of factors by firms.
The capital stock left to generation t + 1 is given by
e
kt+1 = F e
kt , zt + e
(15.86)
kt x1t x2t1 .
The optimisation problem facing the government in period t is to maximise

the welfare level of generation t determined by the social welfare function
t
W =
subject to the constraints
U xtt (s) , xt+1

(s) , `t (s) (s) ds,
t
x2t
(15.87)
xt+1
(s) (s) ds,
t
(15.88)
xtt (s) (s) ds x2t1 e

kt+1 ,
(15.89)
F e
kt , zt + e
kt
and the dierential equation (15.86). In (15.90) is the maximum permissible

mean level of second period consumption to which the members of generation t
are entitled and in (15.92) e
kt+1 is the minimum level of capital stock they must
endow to the next generation. For this maximisation, xt+1
(s) and `t (s) are
t
the control variables, U (s) is the state variable and xtt (s) is determined from
(15.87). This formulation leads to the Hamiltonian
H (s) = U (s) (s) + t x2t xt+1

(s) (s)
t
i
U` `t (s)
t
t
2
t
e
e
e
+ F kt , zt + kt xt (s) xt1 kt+1 (s) + (s)
,
s
(15.90)
where t and t are the Lagrange multipliers on the constraints and t (s) is
the co-state variable.
The next step is to exploit the requirement that x2t and e
kt+1 are Pareto
optimal to simplify (15.92). It follows from the interpretation of Lagrange mult
t
tipliers that W
= t > 0 and W
= t < 0 where W t is the maximum
x2
e
t
kt+1
15.5. INCOME AND INTEREST TAXATION
495
value function for the optimisation. From these

W t
t
e
kt+1
x2t
t =
> 0.
t =
2 |W
W
t
xt
Similarly,
(15.91)
e
kt+1
W t
x2t1
1
e
kt
,
t =
|
=
W
W t
x2t1
1 + Fk e
kt , zt
e
kt
(15.92)
which, when stepped one period forward, gives

W t+1
e
kt+1
1
x2t
.
|W t+1 = W t+1
=
2
xt
e
1
+
F
,
z
k
e
k
t+1 t+1
kt+1
(15.93)
The allocation is Pareto ecient when the gradients given in (15.93) and (15.95)
are equal so that the indierence curve of the social welfare function, defined
kt+1 , are tangential for periods t and t + 1. Therefore, at a point
over x2t and e
on the locus of Pareto optima,
t
1
.
=
t
1 + Fk e
kt+1 , zt+1
(15.94)
The restriction in (15.96) can be substituted into the Hamiltonian to give the
simplified form
U` `t (s)
H (s) = U (s) (s) + t (s)
s
t+1
2
x
x
(s)
t
t
+F e
kt xtt (s) x2t1 e
+t
kt+1 (s) .
kt , zt + e
et+1 , zt+1
1 + Fk k
(15.95)
x2
=F e
kt+1 , zt+1 and denote t = 1+rt e
kt+1 . From the
Next define rt+1
t+1
second-period budget constraint of consumers it can be seen that t is the
market value in period t of one-period bonds sold to consumers of generation t
and which will be redeemed in t + 1. The Hamiltonian can then be written as
U` `t (s)
t
H (s) = U (s) (s) + (s)
s
t+1
xt (s)
t
t
2
e
e
e
+ F kt , zt + kt xt (s) xt1 kt+1 (s) . (15.96)
+ t
1 + rt+1
496
Optimising with respect to xt+1

(s) provides the necessary condition
t
xtt (s)
`t
1
xtt (s)
t
t

(s) + (s)
+ U`xt+1 = 0,
t+1
U`xtt t+1
t
1 + rt+1
s
xt (s)
xt (s)
(15.97)
for the determination of the tax schedule Ts ().
From condition (15.99), two theorems can be proved.
Theorem 73 (Ordover and Phelps) The marginal rate of interest income tax,
Ts0 (), is zero at s = S.
Proof. The transversality condition for the optimisation is that t (S) = 0.
At s = S this reduces (15.99) to
1
xtt (S)
t
(S) = 0.
(15.98)
1 + rt+1
xt+1
(S)
t
As t > 0 and (S) > 0, (16.73) shows that there is no distortion in choice.
Theorem 74 (Ordover and Phelps) When the utility function is separable between consumption and labour, interest income is tax exempt.

, `t .
Proof. The separable utility function is written as U xtt , xt+1
t
Then
U`xtt
xtt (s)
+ U`xt+1
t
xt+1
(s)
t
xtt (s)
+ U` xtt +1
xt+1
(s)
t
= U` xt+1 + U` xtt +1 = 0,
= U` xtt
xt (s)
(15.99)
xt +1
t
where the second equality follows from using xt+1
= t t . Equation
(s)
xt
t
h
i
xtt (s)
(s)
=
0
for
all
s
which
proves the
(15.101) implies that t 1+r1 xt+1
(s)
t+1
t
theorem.
The argument behind Theorem 15.7 is essentially equivalent to that of Theorem 5.6: given the upper-bound upon ability, there is no gain from having
the choices of that consumer distorted. Theorem 15.8 is dierent in nature and
shows that the tax on interest income is redundant when the utility function is
weakly separable and an optimal income tax is employed. Despite the dierent
eects of the two taxes, there is nothing to be gained by having an interest tax
in addition to a nonlinear income tax when preferences are separable. Without
separability, both taxes will be employed.
15.6
Conclusions
The analysis of tax policy in dynamic economies involves consideration of both

intragenerational and intertemporal allocations. Compared with the static analysis, there is also a broader range of instruments available since the use of debt
15.6. CONCLUSIONS
497
becomes a meaningful option. This chapter has attempted to capture these

issues and to illustrate alternative methods of analysis.
The eects of debt were demonstrated in an economy without intergenerational altruism. Although an increase in external debt can rarely lead to an
increase in welfare, in an overcapitalised economy an increase in internal debt
will. The policy relevance of this finding has to be considered in the light of the
Ricardian equivalence proposition that changes in debt have no real eect upon
the economy. A proof of this proposition was given but it was also argued that
it required a particular form of intergenerational altruism and that there were
numerous circumstances in which equivalence does not apply.
In dynamically ecient economies with a homogeneous population, the introduction of an income tax or an interest tax simply adds a distortion and
reduces welfare. With a heterogeneous population intragenerational distribution also becomes relevant. Despite this, an interest tax is redundant when the
utility function is separable and the optimal nonlinear income tax is employed.
Chapter 16
References
Aaron, H. (1966) The social insurance paradox, Canadian Journal of Economics and Political Science, 32, 371 - 374.
Aaron, H. (1982) Economic Eects of Social Security, Washington: The
Brookings Institute.
Abel, B. (1986a) Capital accumulation and uncertain lifetimes with adverse
aelection, Econometrica, 54, 1079 - 1097.
Abel, A.B. (1986b) The failure of Ricardian equivalence under progressive
wealth taxation, Journal of Public Economics, 30, 117 - 128.
Abel, A.B. and B.D. Bernheim (1991) Fiscal policy with impure intergenerational altruism, Econometrica, 59, 1687 - 1711.
Abrams, B.A. and M.A. Schmitz (1984) The crowding out eect of government transfers on private charitable contributions: cross sectional evidence,
National Tax Journal, 37, 563 - 568.
Ahmad, E. and N.H. Stern (1984) The theory of reform and Indian indirect
taxes, Journal of Public Economics, 25, 259 - 298.
Allen, F. (1982) Optimal linear income taxation with general equilibrium
eects on wages, Journal of Public Economics, 17, 135 - 143.
Allingham, M.G. and A. Sandmo (1972) Income tax evasion: a theoretical
analysis, Journal of Public Economics, 1, 323 - 338.
Andersen, P. (1977) Tax evasion and labour supply, Scandinavian Journal
of Economics, 79, 375 - 383.
Anderson, R. and J.G. Ballentine (1976) The incidence and excess burden
of a profits tax under imperfect competition, Public Finance, 31, 159 - 176.
Andreoni, J. (1988a) Privately provided public goods in a large economy:
the limits of altruism, Journal of Public Economics, 35, 57 73.
Andreoni, J. (1988b) Why free ride?, Journal of Public Economics, 37, 291
- 304.
Andreoni, J. (1989) Giving with impure altruism: applications to charity
and Ricardian equivalence, Journal of Political Economy, 97, 1447 - 1458.
Andreoni, J. (1990) Impure altruism and donations to public goods: a
theory of warm-glow giving, Economic Journal, 100, 464 -477.
499
500
CHAPTER 16. REFERENCES
Aoki, M. (1971) Marshallian external economies and optimal tax-subsidy

structure, Econometrica, 39, 35 - 53.
Arrow, K.J. (1950) A diculty in the concept of social welfare, Journal
of Political Economy, 58, 328 - 346.
Arrow, K.J. (1951a) An extension of the basic theorems of welfare economics, in J. Neyman (ed.), Proceedings of the Second Berkeley Symposium
on Mathematical Statistics and Probability, Berkeley: University of California
Press.
Arrow, K.J. (1951b) Social Choice and Individual Values, New York: Wiley.
Arrow, K.J. (1963) The role of securities in the optimal allocation of riskbearing, Review of Economic Studies, 31, 91 - 96.
Arrow, K.J. (1965) Aspects of the Theory of Risk-Bearing, Helsinki: Yrjo
Jahnssonin Saatio.
Arrow, K.J. (1969) The organisation of economic activity: issues pertinent to the choice of market versus non-market allocation in The Analysis and
Evaluation of Public Expenditures: The PPB System, Washington D.C.: Joint
Economic Committee of the Congress of the United States.
Arrow, K.J. and G. Debreu (1954) Existence of equilibrium for a competitive economy, Econometrica, 22, 265 - 290.
Arrow, K.J. and F. Hahn (1971) General Competitive Analysis, San Francisco: Holden Day.
Arrow, K.J. and R.C. Lind (1970) Uncertainty and the evaluation of public
investment decisions, American Economic Review, 60, 364 - 378.
Ashenfelter, O. and J. Heckman (1973) Estimating labour supply functions
in G.C. Cain and N.W. Watts (eds.), Income Maintenance and Labour Supply,
Chicago: Rand McNally.
dAspremont, C. and L.-A. Grard-Varet (1979) Incentives and incomplete
information, Journal of Public Economics, 11, 25 - 45.
Atkinson, A.B (1972) Maxi-min and optimal income taxation, (unpublished).
Atkinson, A.B. (1970) On the measurement of inequality, Journal of Economic Theory, 2, 244 - 263.
Atkinson, A.B. (1987) On the measurement of poverty, Econometrica, 55,
749 - 764.
Atkinson, A.B. and A. Sandmo (1980) Welfare implications of the taxation
of savings, Economic Journal, 90, 529 - 549.
Atkinson, A.B. and N.H. Stern (1974) Pigou, taxation and public goods
Review of Economic Studies, 41, 119 - 128.
Atkinson, A.B. and J. E. Stiglitz (1972) The structure of indirect taxation
and economic eciency, Journal of Public Economics, 1, 97 - 119.
Atkinson, A.B. and J.E. Stiglitz (1976) The design of tax structure: direct
versus indirect taxation, Journal of Public Economics, 6, 55 - 75.
Atkinson, A.B. and J.E. Stiglitz (1980) Lectures on Public Economics, London: McGraw-Hill.
Auerbach, A.J. (1979) Wealth maximisation and the cost of capital, Quarterly Journal of Economics, 93, 433 - 446.
501
Auerbach, A.J. (1983) Taxation, corporate financial policy and the cost of
capital, Journal of Economic Literature, 21, 905 - 940.
Auerbach, A.J. and M.A. King (1983) Taxation, portfolio choice, and debtequity ratios: a general equilibrium model, Quarterly Journal of Economics,
97, 587 - 609.
Aumann, R.J. (1964) Markets with a continuum of traders, Econometrica,
34, 1 - 17.
Averch, H. and L.L. Johnson (1962) Behaviour of the firm under regulatory
constraint, American Economic Review, 52, 1053 - 1069.
Balasko and Shell (1980) The overlapping-generations model, I: the case of
pure exchange without money, Journal of Economic Theory, 23, 281 - 306.
Balasko and Shell (1981) The overlapping-generations model, II: the case
of pure exchange with money, Journal of Economic Theory, 24, 112 - 142.
Balasko, Y., D. Cass and K. Shell (1980) Existence of competitive equilibrium in a general overlapping-generations model, Journal of Economic Theory,
23, 307 - 322.
Balcer, Y. (1980) Taxation of externalities: direct versus indirect, Journal
of Public Economics, 13, 121 - 129.
Baldry, J.C. (1986) Tax evasion is not a gamble, Economics Letters, 22,
333 - 335.
Ball, M.J. (1973) Recent empirical work on the determinants of relative
house prices, Urban Studies, 10, 213 - 233.
Ballantine, J.G. and I. Eris (1975) On the general equilibrium analysis of
tax incidence, Journal of Political Economy, 83, 633 - 644.
Baron, D. (1989) The design of regulatory mechanisms and institutions,
in R. Schmalansee and R. Willig (eds), Handbook of Industrial Organisation,
Amsterdam: North Holland.
Barone, E. (1935) The ministry of production in the collectivist state, in
F.A. von Hayek (ed.), Collectivist Economic Planning, London: Routledge.
Barro, R. J. (1974) Are government bonds net wealth?, Journal of Political
Economy, 82, 1095 - 1117.
Barro, R.J. (1976) Reply to Feldstein and Buchanan, Journal of Political
Economy, 84, 343 - 349.
Baumol, W.J., E.E. Bailey and R.D. Willig (1977) Weak invisible hand
theorems on the sustainability of multiproduct natural monopoly, American
Economic Review, 67, 350 - 365.
Baumol, W.J. and D.F. Bradford (1970) Optimal departures from marginal
cost pricing, American Economic Review, 60, 265 - 283.
Baumol, W.J. and J. Benhabib (1989) Chaos: significance, mechanism and
economic applications, Journal of Economic Perspectives, 3, 77 - 105.
Baumol, W.J. and W.E. Oates (1988) The Theory of Environmental Policy,
Cambridge: Cambridge University Press.
Becker, W., H.-J. Bchner and S. Sleeking (1987) The impact of public
transfer expenditures on tax evasion, Journal of Public Economics, 34, 243 252.
502
Beckerman, W. (1979) The impact of income maintenance payments on

poverty in Britain 1975, Economic Journal, 89, 261 - 279.
Berge, C. (1963) Topological Spaces, London: Oliver and Boyd.
Bergson, A. (1938) A reformulation of certain aspects of welfare economics,
Quarterly Journal of Economics, 68, 233 - 252.
Bergson, A. (1973) On monopoly welfare losses, American Economic Review, 63, 853 - 870.
Bergstrom, T.C. (1976) Regulation of externalities, Journal of Public Economics, 5, 131 - 138.
Bergstrom, T.C. and R.C. Cornes (1983) Independence of allocative eciency from distribution in the theory of public goods, Econometrica, 51, 1753
- 1765.
Bergstrom, T.C., L. Blume and H. Varian (1986) On the private provision
of public goods, Journal of Public Economics, 29, 25 - 49.
Bergstrom, T.C., L. Blume and H. Varian (1992) Uniqueness of Nash equilibrium in private provision of public goods, Journal of Public Economics, 49,
391 - 392.
Bernheim, B.D. (1987) Ricardian equivalence: an evaluation of theory and
evidence, NBER Macro Annual, 263 - 304.
Bernheim, B.D. and K. Bagwell (1988) Is everything neutral?, Journal of
Political Economy, 96, 308 - 338.
Besley, T. and I. Jewitt (1990) Optimal uniform taxation and the structure
of consumer preferences, in G.D. Myles (ed.), Measurement and Modelling in
Economics, Amsterdam: North-Holland.
Beveridge, W. (1942) Social Insurance and Allied Services, Cmd. 6404, London: HMSO.
Bhatia, K.B. (1981) Intermediate goods and the incidence of the corporation income tax, Journal of Public Economics, 16, 93 - 112.
Bhatia, K.B. (1982a) Intermediate goods and the theory of tax incidence,
Public Finance, 37 , 318 - 338.
Bhatia K.B. (1982b) Value-added tax and the theory of tax incidence,
Journal of Public Economics, 19, 203 - 223.
Bhatia, K.B. (1986) Taxes, intermediate goods, and relative prices: the
case of variable coecients, Journal of Public Economics, 31, 197 - 213.
Bierwag, G. O., M. A. Grove and C. Khang (1969) National debt in a
neo-classical growth model: comment, American Economic Review, 59, 205 210.
Bigelow, J.P. (1993) Inducing eciency: externalities, missing markets, and
the Coase theorem, International Economic Review, 34, 335 - 346.
Blackorby, C. and D. Donaldson (1978) Measures of relative inequality and
their meaning in terms of social welfare, Journal of Economic Theory, 18, 59
- 80.
Blanchard, O.J. (1985) Debt, deficits and finite horizons, Journal of Political Economy, 93, 223 - 247.
Blundell, R. (1992) Labour supply and taxation: a survey, Fiscal Studies,
13, 15 - 40.
503
Blundell, R. and A. Lewbel (1991) The information content of equivalence
scales, Journal of Econometrics, 50, 49 - 68.
Boadway, R.W., M. Marchand and P. Pestieau (1991) Optimal linear income taxation in models with occupational choice, Journal of Public Economics, 46, 133 - 162.
Boadway, R.W. and D.E. Wildasin, (1989) A median voter model of social
security, International Economic Review, 30, 307 - 328.
Bohm, P. (1971) An approach to the problem of estimating the demand for
public goods, Swedish Journal of Economics, 73, 55 - 66.
Bohm, P. (1972) Estimating demand for public goods: an experiment,
European Economic Review, 3, 55 - 66.
Bohm, P. (1984) Revealing demand for an actual public good, Journal of
Public Economics, 24, 135 - 151.
Boldrin, M. and M. Woodford (1990) Equilibrium models displaying endogeneous fluctuations and chaos, Journal of Monetary Economics, 25, 189 222.
Bonnisseau, J.-M. (1991) Existence of Lindahl equilibria in economies with
nonconvex production sets, Journal of Economic Theory, 54, 409 - 416.
Bordignon, M. (1990) Was Kant right? Voluntary provision of public goods
under the principle of unconditional commitment, Economic Notes, 342 - 372.
Bordignon, M. (1993) A fairness approach to income tax evasion, Journal
Bs, D. (1986) Public Enterprise Economics, Amsterdam: North-Holland.
Bourguignon, F. (1979) Decomposble income inequality measures, Econometrica, 47, 901 - 920.
Bradford, D.F. (1981) The incidence and allocation eects of a tax on
corporate distributions, Journal of Public Economics, 15, 1 - 22.
Brandts, J. and C.A.M. de Bartolome (1992) Population uncertainty, social
insurance, and actuarial bias, Journal of Public Economics, 47, 361 - 380.
Break, G.F. (1957) Income taxes and the incentive to work: an empirical
study, American Economic Review, 47, 529 - 549.
Brito, D.L. and W.H. Oakland (1977) Some properties of the optimal income tax, International Economic Review, 18, 407 - 423.
Brookshire, D.S., M.A. Thayer, W. Schulze and R. DArge (1982) Valuing public goods: a comparison of survey and hedonic approaches, American
Brown, C.V. and E. Levin (1974) The eects of income taxation on overtime, Economic Journal, 84, 833 - 848.
Brown, S.J. and D.S. Sibley (1986) The Theory of Public Utility Pricing,
Cambridge: Cambridge University Press.
Browning, E.K. (1975) Why the social insurance budget is too large in a
democratic society, Economic Inquiry, 13, 373 - 388.
Buchanan, J. M. (1976) Barro on the Ricardian equivalence theorem, Journal of Political Economy, 84, 337 - 342.
Buchanan, J. and M. Kafoglis (1963) A note on public goods supply,
American Economic Review, 53, 403 - 414.
504
Buhmann, B., L. Rainwater, G. Schmaus and T.M. Smeeding (1989) Equivalence scales, well-being, inequality, and poverty: sensitivity estimates across
ten countries using the Luxembourg income study (LIS) database, Review of
Income and Wealth, 34, 115 - 142.
Buiter, W.H. (1980) Crowding out of private capital formation by government borrowing in the presence of intergenerational gifts and bequests, Greek
Buiter, W.H. (1988) Death, birth, productivity growth and debt neutrality, Economic Journal, 89, 279 - 293
Buiter, W.H. and J. Carmichael (1984) Government debt: comment,
Burbidge, J.B. (1983a) Social security and savings plans in overlappinggenerations models, Journal of Public Economics, 21, 79 - 92.
Burbidge J.B. (1983b) Government debt in an overlapping-generations model
with bequests and gifts, American Economic Review, 73, 222 - 227.
Burtless, G. and J.A. Hausman (1978) The eect of taxation on labour supply: evaluating the Gary negative income tax experiment, Journal of Political
Economy, 86, 1103 - 1130.
Bush, W.C. and L.S. Mayer (1974) Some implications of anarchy for the
distribution of property, Journal of Economic Theory, 8, 401 - 412.
Cadsby, C.B. and M. Frank (1991) Experimental tests of Ricardian equivalence, Economic Inquiry, 29, 645 - 664.
Cagan, P. (1965) The Eects of Pension Plans on Aggregate Savings, New
York: National Bureau of Economic Research.
Callan, T. and B. Nolan (1991) Concepts of poverty and the poverty line,
Journal of Economic Surveys, 5, 243 - 261.
Carmichael, J. (1982) On Barros theorem of debt neutrality: the irrelevance of net wealth, American Economic Review, 72, 202 - 213.
Cass, D. and J.E. Stiglitz (1972) Risk aversion and wealth eects on portfolios with many assets, Review of Economic Studies, 39, 331 - 354.
Chamberlin, J. (1974) Provision of collective goods as a function of group
size, American Political Science Review, 68, 707 - 716.
Chamberlin, J. (1976) A diagrammatic exposition of the logic of collective
action, Public Choice, 26, 59 - 74.
Chen, P. (1990) Prices vs. quantities and delegating price authority to a
monopolist, Review of Economic Studies, 57, 521 - 529.
Chipman, J. (1970) External economies of scale and competitive equilibrium, Quarterly Journal of Economics, 84, 347 - 385.
Christiansen, V. (1980) Two comments on tax evasion, Journal of Public
Economics, 13, 389 - 393.
Christiansen, V. (1981) Evaluation of public projects under optimal taxation, Review of Economic Studies, 48, 447 - 457.
Christiansen, V. (1984) Which commodity taxes should supplement the
income tax?, Journal of Public Economics, 24, 195 - 220.
Christiansen, V. (1988) Choice of occupation, tax incidence and piecemeal
tax revision, Scandinavian Journal of Economics, 90, 141 - 159.
505
Christiansen, V. (1993) A normative analysis of capital income taxes in the
presence of aggregate risk, The Geneva Papers on Risk and Insurance Theory,
18, 55 - 76..
Christiansen, V. and E.S. Jansen (1978) Implicit social preferences in the
Norwegian system of indirect taxation, Journal of Public Economics, 10, 217
- 245..
Clarke, E.H. (1971) Multipart pricing of public goods, Public Choice, 11,
17 - 33.
Clotfelter, C.T. (1983) Tax evasion and tax rates: an analysis of individual
returns, Review of Economics and Statistics, 65, 363 - 373.
Coase, R.H. (1960) The problem of social cost, Journal of Law and Economics, 3, 1 - 44.
Collet, P. and J. Eckmann (1980) Iterated Maps on the Interval as Dynamical
Systems, Basel: Birkhauser.
Corlett, W.J. and D.C. Hague (1953) Complementarity and the excess burden of taxation, Review of Economic Studies, 21, 21 - 30.
Cornes, R. and T. Sandler (1984a) Easy riders, joint production and public
goods, Economic Journal, 94, 580 - 598.
Cornes, R. and T. Sandler (1984b) The theory of public goods: non-Nash
behaviour, Journal of Public Economics, 23, 367 - 379.
Cornes, R. and T. Sandler (1985) The simple analytics of pure public good
provision, Economica, 52, 103 - 116.
Cornwall, R.R. (1977) The concept of general equilibrium in a market economy with imperfectly competitive producers, Metroeconomica, 29, 55 - 72.
Coulter, F.A.E., F.A. Cowell and S.P. Jenkins (1992) Dierences in needs
and assessment of income distributions, Bulletin of Economic Research, 44, 77
- 124.
Cournot, A.A. (1838) Recherches sur les Principes Mathmatiques de la
Thorie des Richesses, Paris: Librairie des Sciences Politiques et Sociales.
Cowell, F.A. (1977) Measuring Inequality, Oxford: Phillip Allan.
Cowell, F.A. (1981) Taxation and labour supply with risky activities, Economica, 48, 365 - 379.
Cowell, F.A. (1985a) The economic analysis of tax evasion, Bulletin of
Economic Research, 37, 163 - 193.
Cowell, F.A. (1985b) Tax evasion with labour income, Journal of Public
Economics, 26, 19 - 34.
Cowell, F.A. and J.P.F. Gordon (1988) Unwillingness to pay, Journal of
Cowell, F.A. and K. Kuga (1971) Additivity and the entropy concept: an
axiomatic approach to inequality measurement, Journal of Economic Theory,
25, 131 - 143.
Cowling, K.G. and D.C. Mueller (1978) The social costs of monopoly power,
Economic Journal, 88, 727 - 748.
Cowling, K.G. and M. Waterson (1976) Price-cost margins and market
structure, Economica, 43, 267 - 274.
Cramer, J.S. (1969) Empirical Econometrics, Amsterdam: North-Holland.
506
Craig, B. and R.G. Batina, (1991), The eects of social security in a life
cycle family labor supply simulation model, Journal of Public Economics, 46,
199 - 226.
Crane, S.E. and F. Nourzad (1986) Inflation and tax evasion: an empirical
analysis, Review of Economics and Statistics, 68, 217 - 223.
Crawford, V.P. and D.M. Lilien (1981) Social security and the retirement
decision, Quarterly Journal of Economics, 95, 505 - 529.
Cremer, H. and F. Gahvari (1993) Tax evasion and optimal commodity
taxation, Journal of Public Economics, 50, 261 - 275.
Cremer, H., M. Marchand and P. Pestieau (1990) Evading, auditing and
taxing, Journal of Public Economics, 43, 67 - 92.
Cripps, M.W. and G.D. Myles (1989) Price normalisations in general equilibrium models of imperfect competition, (unpublished).
Cross, R. and G.K. Shaw (1982) On the economics of tax aversion, Public
Finance, 37, 36 - 47.
Dales, J.H. (1968) Pollution, Property and Prices, Toronto: University of
Toronto Press.
Dalton, H. (1920) The measurement of the inequality of incomes, Economic Journal, 30, 348 - 361.
Danziger, S., R. Haverman and R. Plotnick (1981) How income transfer
programs aect work, savings, and the income distribution: a critical review,
Journal of Economic Literature, 19, 975 - 1028.
Dasgupta, D. and J.-I. Itaya (1992) Comparative statics for the private
provision of public goods in a conjectural variations model with heterogeneous
agents, Public Finance, 47, 17 - 31.
Dasgupta, P., P. Hammond and E. Maskin (1979) The implementation of
social choice rules: some general results on incentive compatibility, Review of
Economic Studies, 46, 185 - 216.
Dasgupta, P., P. Hammond and E. Maskin (1980) On imperfect information
and optimal pollution control, Review of Economic Studies, 47, 857 - 860.
Dasgupta, P., A. Sen and D. Starret (1973) Notes on the measurement of
inequality, Journal of Economic Theory, 6, 180 - 187.
Dasgupta, P. and J.E. Stiglitz (1972) On optimal taxation and public production, Review of Economic Studies, 39, 87 - 103.
Davidson, C. and L.W. Martin (l985) General equilibrium tax incidence
under imperfect competition: a quantity-setting supergame snalysis, Journal
of Political Economy, 93, 1212 - 1223.
Davies, J.B. and P. Kuhn (1992) Social security, longevity, and moral hazard, Journal of Public Economics, 49, 91 - 106.
Davis, O. and A. Whinston (1962) Externalities, welfare and the theory of
games, Journal of Political Economy, 70, 241 - 262.
Deardor, A.V. (1976) The optimum growth rate for population: comment, International Economic Review, 17, 510 - 514.
Deaton, A. S. (1981) Optimal taxes and the structure of preferences,
Econometrica, 49, 1245 - 1260.
507
Deaton, A.S. (1977) Equity, eciency, and the structure of indirect taxation, Journal of Public Economics, 8, 299 - 312.
Deaton, A.S. (1979) The distance function in consumer behaviour with applications to index numbers and optimal taxation, Review of Economic Studies,
46, 391 - 405.
Deaton, A.S. and J. Muellbauer (1986) On measuring child costs: with
applications to poor countries, Journal of Political Economy, 94, 720 - 744.
Deaton, A.S., J. Ruiz-Castillo and D. Thomas (1989) The influence of
household composition on household expenditure: theory and Spanish evidence,
Journal of Political Economy, 97, 179 - 200.
Deaton, A.S. and N.H. Stern (1986) Optimally uniform commodity taxes,
taste dierence and lump-sum grants, Economics Letters, 20, 263-266.
Debreu, G. (1951) The coecient of resource utilisation, Econometrica,
19, 273-292.
Debreu, G. (1954a) Representation of a preference oredering by a numerical function, in R.M. Thrall, C.H. Coombs and R.L. Davis (eds.) Decision
Processes, New York: Wiley.
Debreu, G. (1954b) Valuation equilibrium and Pareto optimum, Proceedings of the National Academy of Sciences of the U.S.A., 40, 588 - 592.
Debreu, G. (1959) Theory of Value, New York: Wiley.
Debreu, G. (1970) Economies with a finite set of equilibria, Econometrica,
38, 387 - 392.
Debreu, G. (1980) Existence of competitive equilibrium in K.J. Arrow
and M.D. Intriligator (eds.), Handbook of Mathematical Economics, Amsterdam:
North-Holland.
Debreu, G. and H. Scarf (1963) A limit theorem on the core of an economy,
International Economic Review, 4, 235 - 246.
Decoster, A. and E. Schokkaert (1989) Equity and eciency of a reform of
Belgian indirect taxes, Recherches Economiques de Louvain, 55, 155 - 176.
Delipalla, S. and M. Keen (1992) The comparison between ad valorem and
specific taxation under imperfect competition, Journal of Public Economics,
49, 351 - 367.
Desai, M. and A. Shah (1988) An econometric approach to the measurment
of poverty, Oxford Economic Papers, 40, 505 - 522.
Diamond, P.A. (1965) National debt in a neo-classical growth model,
Diamond, P.A. (1967) The role of the stock market in a general equilibrium
model with technological uncertainty, American Economic Review, 57, 759 776.
Diamond, P.A. (1973a) Consumption externalities and imperfect corrective
pricing, Bell Journal of Economics, 4, 526 - 538.
Diamond, P.A. (1973b) Taxation and public production in a growth setting, in J.A. Mirrlees and N.H. Stern (eds.), Models of Economic Growth,
London: Macmillan.
Diamond, P.A. (1975) A many-person Ramsey tax rule, Journal of Public
Economics, 4, 227 - 244.
508
Diamond, P.A. (1977) A framework for social security analysis, Journal

Diamond, P. A., L.J. Helms and J. A. Mirrlees (1980) Optimal taxation in
a stochastic economy, Journal of Public Economics, 14, 1 - 29.
Diamond, P.A. and J. A. Mirrlees (1971) Optimal taxation and public production 1: Production eciency and 2: Tax rules, American Economic Review,
61, 8 - 27 and 261 - 278.
Diamond, P.A. and J. A. Mirrlees (1973) Aggregate production with consumption externalities, Quarterly Journal of Economics, 87, 1 - 24.
Diamond, P.A. and J. A. Mirrlees (1986) Payroll-tax financed social insurance with variable retirement, Scandinavian Journal of Economics, 88, 25 50.
Diamond, P.A. and J.E. Stiglitz (1974) Increases in risk and in risk aversion, Journal of Economic Theory, 8, 337 - 360.
Dickson, V.A. and W. Yu (1989) Welfare losses in Canadian manufacturing under alternaitve oligopoly regimes, International Journal of Industrial
Organisation, 7, 257 - 267.
Dierckx, I, C. Matutes and D. Neven (1988) Indirect taxation and Cournot
equilibrium, International Journal of Industrial Organisation, 6, 385 - 399.
Dierker, H. and B. Grodal (1986) Non-existence of Cournot-Walras equilibrium in a general equilibrium model with two oligopolists, in W. Hildenbrand
and A. Mas-Collel (eds.), Contributions to Mathematical Economics, Amsterdam: North-Holland.
Dixit, A.K. (1970) On the optimum structure of commodity taxes, American Economic Review, 60, 295 - 301.
Dixit, A.K. (1979) Price changes and optimum taxation in a many-consumer
economy, Journal of Public Economics, 11, 143 - 157.
Donaldson, D. and J.A. Weymark (1980) A single-parameter generalisation
of the Gini indices of inequality, Journal of Economic Theory, 22, 67 - 86.
Drazen, A. (1978) Government debt, human capital, and bequests in a
life-cycle model, Journal of Political Economy, 86, 505 - 516.
Dreze, J.H. (1964) Some postwar contributions of French economists to theory and public policy with special emphasis on problems of resource allocation,
Dreze, J.H. (1970-1) Market allocation under uncertainty, European Economic Review, 2, 133 - 165.
Dreze, J.H. (1974) Investment under private ownership: optimality, equilibrium and stability, in J.H. Dreze (ed.) Allocation Under Uncertainty: Equilibrium and Optimality, New York: Macmillan.
Due, D. and H. Sonnenschein (1989) Arrow and general equilibrium theory, Journal of Economic Literature, 27, 565 - 598.
Eaton, J. and H.S. Rosen (1980) Labor supply, uncertainty, and ecient
taxation, Journal of Public Economics, 14, 365 374.
Ebert, U. (1992) A reexamination of the optimal nonlinear income tax,
509
Ebrahimi, A. and C. J. Heady (1988) Tax design and household consumption, Economic Journal (Supp.), 98, 83 - 96.
Ebrill, L.P. and S.M. Slutsky (1982) Time, congestion and public goods,
Journal of Public Economics, 17, 307 334.
Ebrill, L.P. and S.M. Slutsky (1990) Production eciency and optimal pricing in intermediate-good related industries, International Journal of Industrial
Organisation, 8, 417 - 442.
Eckstein, Z., M. Eichenbaum and D. Peled (1985) Uncertain lifetimes and
the welfare enhancing properties of annuity markets and social security, Journal of Public Economics, 26, 303 - 326.
Edgeworth, F.Y. (1925) The pure theory of monopoly in Papers Relating
to Political Economy, New York: Burt Franklin.
Eeckhoudt, L. and P. Hansen (1982) Uncertainty and the partial loss oset
provision, Economics Letters, 9, 31 - 35.
Eichhorn, W. (1988) On a class of inequality measures, Social Choice and
Welfare, 5, 171 - 177.
Enders, W. and H.E. Lapan (1992) Social security taxation and intergenerational risk sharing, International Economic Review, 23, 647 - 658.
Engel, E. (1895) Die lebenskosten Belgischer arbeiter-familien frher und
jetzt, International Statistical Institute Bulletin, 9, 1- 74.
Feehan, J.P. (1989) Pareto-eciency with three varieties of public input,
Public Finance, 44, 237 - 248.
Feige, E.L. (1979) How big is the irregular economy?, Challenge, 5 - 13.
Feige, E.L. (1981) The UKs unobserved economy, Journal of Economic
Aairs, 1, 205 - 212.
Feige, E.L. and R.T. McGee (1983) Swedens Laer curve: taxation and
the unobserved economy, Scandinavian Journal of Economics, 85, 499 - 519.
Feldstein, M.S. (1973) On the optimal progressivity of the income tax,
Feldstein, M.S. (1974) Social security, induced retirement and aggregate
capital accumulation, Journal of Political Economy, 82, 905 - 926.
Feldstein, M.S. (1976a) On the theory of tax reform, Journal of Public
Economics, 6, 77 - 104.
Feldstein, M.S. (1976b) Perceived wealth in bonds and social security: a
comment, Journal of Political Economy, 84, 331 - 336.
Feldstein, M.S. (1985) The optimal level of social security benefits, Quarterly Journal of Economics, 100, 303 - 320.
Feldstein, M.S. (1987) Should social security benefits be means tested?,
Feldstein, M.S. (1990) Imperfect annuity markets, unintended bequests,
and the optimal age strcture of social security benefits, Journal of Public Economics, 41, 31 - 43.
Fields, G.S. and J.C.H. Fei (1978) On inequality comparisons, Econometrica, 46, 303 - 316.
Fisher, I. (1930) The Theory of Interest, New York: Macmillan.
510
Flemming, J.S. (1977) Optimal payroll taxes and social security funding,
Fogelman, F., M. Quinzii and R. Guesnerie (1978) Dynamic processes for
tax reform theory, Journal of Economic Theory, 17, 200 - 226.
Foley, D.K. (1970) Lindahls solution and the core of an economy with
public goods, Econometrica, 38, 66 - 72.
Foster, J.E. (1983) An axiomatic characterisation of the Theil measure of
income inequality, Journal of Economic Theory, 31, 105 - 121.
Foster, J.E., J. Greer and E. Thorbecke (1984) A class of decomposable
poverty measures, Econometrica, 52, 761 - 767.
Friedland, N., S. Maital and A. Rutenberg (1978) A simulation study of
income taxation, Journal of Public Economics, 10, 107 - 116.
Friedman, J.W. (1977) Oligopoly and the Theory of Games, Amsterdam:
North-Holland.
Gabszewicz, J. J. and J.-P. Vial (1972) Oligopoly a la Cournot in general
equilibrium analysis, Journal of Economic Theory, 4, 381 - 400.
Gale, D. (1955) The law of supply and demand, Mathematica Scandinavica, 3, 155 - 169.
Gale, D. (1973) Pure exchange equilibrium of dynamic economic models,
Journal of Economic Theory, 6, 12 - 36.
Gary-Bobo, R.J. (1989) Cournot-Walras and locally consistent equilibria,
Geanakoplos, J.D. and H.M. Polemarchakis (1984) Intertemporally separable, overlapping-generations economies, Journal of Economic Theory, 34, 207
- 215.
Geeroms, H. and H. Wilmots (1985) An empirical model of tax evasion and
tax avoidance, Public Finance, 40, 190 - 209.
Gigliotti, G.A. (1984) Total utility, overlapping generations and social security, Economics Letters, 15, 169 - 173.
Gini, C. (1912) Variabilita Mutabilita, Bologna.
Glazer, A. (1989) The social discount rate under majority voting, Public
Finance, 44, 384 - 393.
Gordon, J.P.F. (1989) Individual morality and reputation costs as deterrents to tax evasion, European Economic Review, 33, 797 - 805.
Gorman, W.M. (1959b) Are social indierence curves convex?, Quarterly
Journal of Economics, 73, 485 - 496.
Gorman, W.M. (1978) Tricks with utility functions, Part 1, LSE Econometrics Programme, Discussion Paper No. B1.
Gorman, W.M. and G.D. Myles (1987) Characteristics, in The New Palgrave Dictionary of Economics, London: Macmillan.
Grandmont, J.-M. (1985) On endogeneous competitive business cycles,
Green, J. (1988) Demographics, market failure and social securiy in S.M.
Wachter (ed.), Social Security and Private Pensions, Lexington: Lexington
Books.
511
Green, J. and J.-J. Laont (1977) Characterization of satisfactory mechanisms for the revelation of preferences for public goods, Econometrica, 45, 427
- 438.
Green, J. and J.-J. Laont (1979) Incentives in Public Decision Making,
Amsterdam: North-Holland.
Green, J. and E. Sheshinski (1976) Direct versus indirect remedies for externalities, Journal of Political Economy, 84, 797 - 808.
Griliches, Z. (1971) Hedonic price indexes for automobiles: an econometric analysis of quality chage, in Z. Griliches (ed.), Price Indexes and Quality
Change: Studies in New Methods of Measurement, Cambridge, MA: Harvard
University Press.
Grinols, E.L. (1985) Public investment and social risk-sharing, European
Grossman, S.J. and O.D. Hart (1979) A theory of competitive equilibrium
in stock market economies, Econometrica, 47, 293 - 330.
Groves, T. (1973) Incentives in teams, Econometrica, 41, 617 - 631.
Groves, T. and J. Ledyard (1977) Optimal allocation of public goods: a
solution to the free rider problem, Econometrica, 45, 783 - 809.
Groves, T. and M. Loeb (1975) Incentives and public inputs, Journal of
Guesnerie, R. (1977) On the direction of tax reform, Journal of Public
Economics, 7, 179 - 202.
Guesnerie, R. and J.-J. Laont (1978) Taxing price makers, Journal of
Economic Theory, 19, 423 - 455.
Gutmann, P.M. (1977) The subterranean economy, Financial Analysts
Journal, 26 - 28.
Hahn, F.H. (1973) On optimum taxation, Journal of Economic Theory,
6, 96 - 106.
Hahn, F.H. (1982) Money and Inflation, Oxford: Basil Blackwell.
Hansson, I. and C. Stuart (1989) Social security as trade among living
generations, American Economic Review, 79, 1182 - 1195.
Harberger, A.C. (1954) Monopoly and resource allocation, American Economic Review, 45, 77 - 87.
Harberger, A.C. (1962) The incidence of the corporation income tax, Journal of Political Economy, 70, 215 - 240.
Harris, R.G. and J.G. MacKinnon (1979) Computing optimal tax equilibria, Journal of Public Economics, 11, 197 - 212.
Hart, O.D. (1975a) On the optimality of equilibrium when the market structure is incomplete, Journal of Economic Theory, 11, 418-443.
Hart, O.D. (1975b) Some negative results on the existence of comparative
statics results in portfolio theory, Review of Economic Studies, 42, 615 - 621.
Hart, O.D. (1982) A model of imperfect competition with Keynsian features, Quarterly Journal of Economics, 97, 109 - 138.
Hart, O.D. (1985) Imperfect competition and general equilibrium: an overview
of recent work in K.J. Arrow and S. Honkapohja (eds.), Frontiers of Economics,
Oxford: Basil Blackwell.
512
Hatta, T. (1986) Welfare eects of changing commodity tax rates toward

uniformity, Journal of Public Economics, 29, 99 - 112.
Hausman, (1985) Taxation and labour supply in A.J. Auerbach and M.
Feldstein (eds.), Handbook of Public Economics, Amsterdam: North-Holland.
Heady, C. and P. Mitra (1980) The computation of optimum linear taxation. Review of Economic Studies, 47, 567 - 585.
Heller, W.P. and D.A. Starrett (1976) On the nature of externalities, in
S.A.Y. Lin (ed.), Theory and Measurement of Economic Externalities, New
York: Academic Press.
Hellwig, M.F. (1981) Bankruptcy, limited liability, and the ModiglianiMiller theorem, American Economic Review, 89, 155 - 170.
Hellwig, M.F. (1986) The optimal linear income tax revisited, Journal of
Helpman, E. and E. Sadka (1978) The optimal income tax, Journal of
Hicks, J.R. (1939) Value and Capital, Oxford: Clarendon Press.
Hildenbrand, W. and A. Kirman (1988) Equilibrium Analysis, Amsterdam:
North-Holland.
Hirshleifer, J. (1964) Ecient allocation of capital in an uncertain world,
American Economic Review, Papers and Proceedings, 54, 77 - 85.
Hirshleifer, J. (1966) Investment decisions under uncertainty: applications
of the state-preference approach, Quarterly Journal of Economics, 80, 252 277.
Holmstrom, B. (1979) Groves scheme on restricted domains, Econometrica, 47, 1137 - 1144.
Houthakker, H.S. (1960) Additive preferences, Econometrica, 28, 244 257.
Hu, S.C. (1978) On the dynamic behaviour of the consumer and the optimal
provision of social security, Review of Economic Studies, 45, 437 - 445.
Hu, S.C. (1979) Social security, the supply of labor, and capital accumulation, American Economic Review, 69, 274 - 283.
Hu, S.C. (1982) Social security, majority-voting equilibrium and dynamic
eciency, International Economic Review, 23, 269 - 287.
Hurwicz, L. (1979a) Outcome functions yielding Walrasian and Lindahl
allocations at Nash equilibrium points, Review of Economic Studies, 46, 217 225.
Hurwicz, L. (1979b) On allocations attainable through Nash equilibria,
Intriligator, M.D. (1971) Mathematical Optimization and Economic Theory,
Englewood Clis: Prentice Hall.
Ireland, N.J. (1977) Ideal prices vs. prices vs. quantities, Review of Economic Studies, 44, 183 - 186.
Ireland, N.J. (1978) Roys identity and monopoly welfare loss, in D.A.
Currie and W. Peters (eds.), Contempory Economic Analysis, London: Croom
Helm.
513
Isaac, R.M., K.F. McCue and C.R. Plott (1985) Public goods in an experimental environment, Journal of Public Economics, 26, 51 - 74.
Isaac, R.M. and J.M. Walker (1988) Group size eects in public goods provision: the voluntary contribution mechanism, Quarterly Journal of Economics,
102, 179 - 199.
Isaac, R.M., J.M. Walker and S.H. Thomas (1984) Divergent evidence
on free riding: an experimental examination of possible explanations, Public Choice, 43, 113 - 149.
Isachsen, A.J. and S. Strom (1980) The hidden economy: the labour market
and tax evasion, Scandinavian Journal of Economics, 82, 304 - 311.
Jenkins, S.P. (1991) The measurement of income inequality, in L. Osberg
(ed.), Economic Inequality and Poverty: International Perspectives, New York:
M.E. Sharpe.
Johansen, L. (1963) Some notes on the Lindahl theory of determination of
public expenditures, International Economic Review, 4, 346 358.
Johansen, L. (1977) The theory of public goods: misplaced emphasis?,
Jones, R.W. (1965) The structure of simple general equilibrium models,
Jones, R.W. (1971) Distortions in factor markets and the general equilibrium model of production, Journal of Political Economy, 79, 437-459.
Kaizuka, K. (1965) Public goods and decentralisation of production, Review of Economics and Statistics, 47, 118 - 120.
Kanbur, S.M.R. (1980) Risk taking and taxation, Journal of Public Economics, 15, 163 - 184.
Kanbur, S.M.R. and G.D. Myles (1992) Policy choice and political constraints, European Journal of Political Economy, 8, 1 - 29.
Kanbur, S.M.R. and M. Tuomala (1994) Inherent inequality and the optimal graduation of marginal tax rates, Scandinavian Journal of Economics, 96,
275 - 282.
Karni, E. and I. Zilcha (1986) Welfare and comparative statics implications
of fair social security, Journal of Public Economics, 30, 341 - 357.
Katona, G. (1964) Private Pensions and Individual Savings, Ann Arbor:
Survey Research Centre, Institute for Social Research, University of Michigan.
Kay, J.A. (1983) A general equilibrium approach to the measurement of
monopoly welfare loss, International Journal of Industrial Organisation, 1,
317 - 331.
Kay, J.A. and M. Keen (1983) How should commodities be taxed?, European Economic Review, 23, 339 - 358.
Kay, J.A. and M.A. King (1990) The British Tax System, Oxford: Oxford
University Press.
Kehoe, T.J. and D.K. Levine (1985) Comparative statics and perfect foresight in infinite horizon economies, Econometrica, 53, 433 - 453.
Kemp, M.C. and Y.-K. Ng (1976) On the existence of social welfare functions, social orderings and social decision functions, Economica, 43, 59 - 66.
514
Kihlstrom, R.E. and J.-J. Laont (1983) Taxation and risk taking in general
equilibrium models with free entry, Journal of Public Economics, 21, 159 - 181.
Kim, O. and M. Walker (1984) The free rider problem: experimental evidence, Public Choice, 43, 3 - 24.
King, M.A. (1975) Taxation, corporate financial policy, and the cost of
capital, Journal of Public Economics, 4, 271 - 279.
King, M.A. (1977) Public Policy and the Corporation, London: Chapman
and Hall.
Kolm, S.-C. (1969) The optimal production of social justice, in J. Margolis
and H. Guitton (eds.), Public Economics, London: Macmillan.
Kolm, S.-C. (1973) A note on optimum tax evasion, Journal of Public
Economics, 2, 265 - 270.
Kolm, S.-C. (1976) Unequal inequalities I and II, Journal of Economic
Theory, 12, 416 - 442 and 13, 82 - 111.
Konishi, H. (1990) Final and intermediate goods taxes in an oligopolistic
economy with free entry, Journal of Public Economics, 42, 371 - 386.
Koopmans, T.C. (1957) Three Essays on the State of Economic Science,
New York: McGraw-Hill.
Koskella, E. (1983) On the shape of tax schedule, the probability of detection, and the penalty schemes as deterrents to tax evasion, Public Finance, 38,
70 - 80.
Kotliko, L.J. (1979) Social security and equilibrium capital intensity,
Quarterly Journal of Economics, 93, 233 - 253.
Kotliko, L.J. (1989) On the contribution of economics to the evaluation
and formation of social insurance policy, American Economic Review, Papers
and Proceedings, 77, 184 - 190.
Kotliko, L.J., A. Spivak and L.H. Summers (1982) The adequacy of savings, American Economic Review, 72, 1056 - 1069.
Laont, J.-J. (1977) More on prices vs. quantities, Review of Economic
Studies, 44, 177 - 82.
Laont, J.-J. (1987) Incentives and the allocation of public goods, in A.J.
Auerbach and M. Feldstein (eds.), Handbook of Public Economics, Amsterdam:
North-Holland.
Laont, J.-J. and E. Maskin (1982) The theory of incentives: an overview,
in W. Hildenbrand (ed.), Advances in Economic Theory, Cambridge: Cambridge
University Press.
Lambert, P. (1989) The Distribution and Redistribution of Income: A Mathematical Analysis, Oxford: Basil Blackwell.
Lancaster, K.J. (1966) A new approach to consumer theory, Journal of
Political Economy, 74, 132 - 157.
Landskroner, Y., J. Paroush and I. Swary (1990) Tax evasion and portfolio
decisions, Public Finance, 45, 409 - 422.
Lange, O. (1942) The foundations of welfare economics, Econometrica,
10, 215-228.
Layard, P.R.G. and A.A. Walters (1978) Microeconomic Theory, London:
McGraw-Hill.
515
Lee, M.L. and S.-W. Chao (1988) Eects of social security on personal
saving, Economic Letters, 28, 365 - 368.
Lerner, A. (1970) On optimal taxes with an untaxable sector, American
Lewbel, A. (1989) Household equivalence scales and welfare comparisons,
Lewbel, A. (1991) Cost of characteristics indices and household equivalence
scales, European Economic Review, 35, 1277 - 1293.
Lewis, G.W. and D.T. Ulph (1988) Poverty, inequality and welfare, Economic Journal, 98, 117 - 131.
Lindahl, E. (1919) Positive Losung, die Gerechtigkeit der Besteuerung, Lund,
reprinted as Just Taxation - A Positive Solution in R.A. Musgrave and A.T.
Peacock (eds.), Classics in the Theory of Public Finance, London: Macmillan.
Lipsey, R.G. and K. Lancaster (1956) The general theory of second best,
Little, I.M.D. (1950) A Critique of Welfare Economics, Oxford: Clarendon
Press.
Lockwood, B. (1995) Multi-firm regulation without lump-sum taxes, Journal of Public Economics, (forthcoming).
Lopez-Garcia, M.-A. (1991) Population growth and pay-as-you-go social
security in an overlapping generations model, Public Finance, 46, 93 - 110.
Lorenz, M.C. (1905) Methods of measuring the concentration of wealth,
Publications of the American Statistical Association, 9, 209 - 219.
Lydall, H.F. (1968) The Structure of Earnings, Oxford: Clarendon Press.
Mailath, G. and A. Postlewaite (1990) Asymmetric-information bargaining
problems with many agents, Review of Economic Studies, 57, 351 -368.
Mangasarian, O.L. (1969) Nonlinear Programming, New York: McGraw-Hill.
Marrelli, M. (1984) On indirect tax evasion, Journal of Public Economics,
25, 181 - 196.
Marrelli, M. and R. Martina (1988) Tax evasion and strategic behaviour of
the firms, Journal of Public Economics, 37, 55 - 69.
Marwell, G. and R.E. Ames (1981) Economists free ride, does anyone else?,
Mas-Collel, A. (1980a) The Theory of General Economic Equilibrium, Cambridge: Cambridge University Press.
Mas-Colell, A. (1980b) Eciency and decentralization in the pure theory
of public goods, Quarterly Journal of Economics, 84, 625 - 641.
Mas-Colell, A. and J. Silvestre (1989) Cost share equilibria: a Lindahlian
approach, Journal of Economic Theory, 47, 239 - 256.
McClements, L.D. (1977) Equivalence scales for children, Journal of Public
Economics, 8, 191 - 210.
McGuire, M. (1974) Group homogeneity and aggregate provision of a pure
public good under Cournot behaviour, Public Choice, 18, 107 - 126.
McKenzie, L. (1955) Competitive equilibrium with dependent consumer
preferences, in National Bureau of Standards and Department of the Air Force,
The Second Symposium on Linear Programming, Washington, D.C.
516
McLure, C.E. (1972) A diagrammatic exposition of the Harberger model

with one immobile factor, Journal of Political Economy, 80, 56 - 82.
McLure, C.E. (1975) General equilibrium incidence analysis: the Harberger
model after ten years, Journal of Public Economics, 80, 56 - 82.
Meade, J.E. (1952) External economies and diseconomies in a competitive
situation, Economic Journal, 62, 54 - 67.
Meade, J.E. (1975) The Intelligent Radicals Guide to Economic Policy, London: Allen and Unwin.
Meyer, R.A. (1971) Externalities as commodities, Amercian Economic
Review, 61, 736 - 740.
de Meza, D. (1982) Generalised oligopoly derived demand with an application to tax induced entry, Bulletin of Economic Research, 34, 1 - 16.
Mieszkowski, P.M. (1967) On the theory of tax incidence, American Economic Review, 75, 250 - 262.
Mieszkowski, P.M.(1969) Tax incidence theory: the eects of taxes on the
distribution of income, Journal of Economic Literature, 7, 1103 - 1124.
Milleron, J.-C. (1972) Theory of value with public goods: a survey article,
Mirrlees, J.A. (1971) An exploration in the theory of optimum income taxation, Review of Economic Studies, 38, 175 - 208.
Mirrlees, J.A. (1972) On producer taxation, Review of Economic Studies,
39, 105 - 111.
Mirrlees, J.A. (1975) Optimal commodity taxation in a two-class economy,
Mirrlees, J.A. (1976) Optimal tax theory: a synthesis, Journal of Public
Economics, 6, 327 - 358.
Mirrlees, J.A. (1986) The theory of optimal taxation in K.J. Arrow and
M.D. Intrilligator (eds.), Handbook of Mathematical Economics, Amsterdam:
North-Holland.
Modigliani, F. and M.H. Miller (1958) The cost of capital, corporation
finance and the theory of investment, American Economic Review, 48, 261 297.
Montgomery, W.D. (1972) Markets in licenses and ecient pollution control
programs, Journal of Economic Theory, 5, 395 - 418.
Moore, J. and R. Repullo (1988) Subgame perfect implementation, Econometrica, 56, 1191 - 1220.
Mork, K.A. (1975) Income tax evasion: some empirical evidence, Public
Finance, 30, 70 - 76.
Mossin, J. (1968) Taxation and risk-taking: an expected utility approach,
Economica, 35, 74 - 82.
Muellbauer, J. (1974) Household composition, Engel curves and welfare
comparisons between households, European Economic Review, 5, 103 - 122.
Muellbauer, J. (1977) Testing the Barten model of household composition
eects and the cost of children, Economic Journal, 87, 460 - 487.
Muench, T.J. (1972) The core and the Lindahl equilibrium of an economy
with public goods: an example, Journal of Economic Theory, 4, 241 - 255.
517
Munk, K.J. (1978) Optimal taxation and pure profit, Scandinavian Journal of Economics, 80, 1 - 19.
Munk, K.J. (1980) Optimal taxation with some non-taxable commodities,
Murakami, Y. and T. Negishi (1964) A formulation of external economy,
Murty, M.N. and R. Ray (1987) Sensitivity of optimal commodity taxes to
relaxing leisure/goods separability and to the wage rate, Economics Letters,
24, 273 - 277.
Muzondo, T.R. (1978) Mixed and pure public goods, user charges and
welfare, Public Finance, 33, 314 - 330.
Myerson, R.B. and M.A. Satterthwaite (1983) Ecient mechanisms for
bilateral trading, Journal of Economic Theory, 29, 265 - 281.
Myles, G.D. (1987) Tax design in the presence of imperfect competition:
an example, Journal of Public Economics, 34, 367 - 378.
Myles, G.D. (1989a) Ramsey tax rules for economies with imperfect competition., Journal of Public Economics, 38, 95 - 115.
Myles, G.D. (1989b) Imperfect competition and the taxation of intermediate goods, Public Finance, 44, 62 - 74.
Myles, G.D. (1994a) Welfare loss with intermediate goods, in W. Eichhorn
(ed.), Models and Measurement of Welfare and Inequality, Berlin: SpringerVerlag.
Myles, G.D. (1994b) Ramsey pricing in imperfectly competitive economies,
(unpublished).
Myles, G.D. (1995) Imperfect competition and industry-specific input taxes,
Public Finance Quarterly, (forthcoming).
Myles, G.D. and R.A. Naylor (1995a) A model of tax evasion with group
conformity and social customs, European Journal of Political Economy, (forthcoming).
Myles, G.D. and R.A. Naylor (1995b) Optimal auditing and social customs, (unpublished).
Nash, J.F. (1950) The bargaining problem, Econometrica, 18, 155 - 162.
Negishi, T. (1961) Monopolistic competition and general equilibrium, Review of Economic Studies, 28, 196 - 201.
Newbery, D.M. (1970) A theorem on the measurement of inequality, Journal of Economic Theory, 2, 264 - 266.
Newbery, D.M. (1986) On the desirability of input taxes, Economics Letters, 20, 267 - 270.
Ng, Y.-K. (1985) Welfare Economics, Basingstoke: MacMillan.
Nicholson, J. (1976) Appraisal of dierent methods of estimating equivalence scales and their results, Review of Income and Wealth, 22, 1 - 11.
Oakland, W.H. (1972) Congestion, public goods and welfare, Journal of
Oakland, W.H. (1987) Theory of public goods in A.J. Auerbach and M.
Feldstein (eds.), Handbook of Public Economics, Amsterdam: North-Holland.
518
Ordover, J.A. (1976) Distributive justice and optimal taxation of wages and
interest in a growing economy, Journal of Public Economics, 5, 139 - 160.
Ordover, J.A. and E.S. Phelps (1975) Linear taxation of wealth and wages
for intragenerational lifetime justice: some steady-state cases, American Economic Review, 65, 660 - 673.
Ordover, J.A. and E.S. Phelps (1979) The concept of optimal taxation in
the overlapping-generations model of capital and wealth, Journal of Public
Economics, 12, 1 - 26.
Osana, H, (1972) Externalities and the basic theorems of welfare economics, Journal of Economic Theory, 4, 401 - 414.
Osana, H. (1973) On the boundedness of an economy with externalities,
Otani, Y. and J. Sicilian (1977) Externalities and problems of nonconvexity
and overhead costs in welfare economics, Journal of Economic Theory, 14, 239
- 251.
Pareto, V. (1909) Manuel dconomie Politique, Paris: Girard and Briere.
Parish, R.M. (1972) Economic aspects of pollution control, Australian
Economic Papers, 11, 32 - 43.
Park, N.-H. (1991) Steady-state solutions of optimal tax mixes in an overlappinggenerations model, Journal of Public Economics, 46, 227 - 246.
Parks, R.P. (1991) Pareto irrelevant externalities, Journal of Economic
Theory, 54, 165 - 179.
Peacock, A. and G.K. Shaw (1982) Tax evasion and tax revenue loss,
Peck, R.M. (1989) Taxation, risk and returns to scale, Journal of Public
Economics, 40, 319 - 330.
Pencavel, J.H. (1979) A note on income tax evasion, labor supply, and
nonlinear tax schedules, Journal of Public Economics, 12, 115 - 124.
Pezzey, J. (1992) The symmetry between controlling pollution by price and
controlling it by quantity, Canadian Journal of Economics, 25, 983 - 991.
Pfingsten, A. (1986) Distributionally-neutral tax changes for dierent inequality concepts, Journal of Public Economics, 30, 385 - 393.
Phelps, E.S. and J.G. Riley (1978) Rawlsian growth: dynamic programming of capital and wealth for intergeneration maxmin justice, Review of
Economic Studies, 45, 103 - 120.
Pigou, A.C. (1920) The Economics of Welfare, New York:Macmillan.
Pissarides, C.A. and G. Weber (1989) An expenditure-based estimate of
Britains black economy, Journal of Public Economics, 39, 17 - 32.
Pollak, R.A. and T.J. Wales (1979) Welfare comparisons and equivalence
scales, American Economic Review, Papers and Proceedings, 69, 216 - 221.
Prais, S.J. (1953) The estimation of equivalent-adult scales from family
budgets. Economic Journal, 63, 791 - 810.
Prais, S.J. and H.S. Houthakker (1955) The Analysis of Family Budgets,
Department of Applied Economics Monograph 4, Cambridge: Cambridge University Press.
519
Pratt, J.W. (1964) Risk aversion in the small and in the large, Econometrica, 32, 122 - 136.
Pyatt, G. (1985) An axiomatic approach to the Gini coecient and the
measurement of welfare, Advances in Econometrics, 3, 87 - 109.
Rader, T. (1972) The Theory of General Equilibrium, New York: Academic
Press.
Radner , R. (1982) Equilibrium under uncertainty, in K.J. Arrow and
M.D. Intrilligator (eds.), Handbook of Mathematical Economics, Amsterdam:
North-Holland.
Ramsey, F.P. (1927) A Contribution to the theory of taxation, Economic
Journal, 37, 47-61.
Rao, V.V. (1981) Measurement of deprivation and poverty based on the
proportion spent on food, World Development, 9, 337 - 353.
Rapanos, V.T. (1991) The incidence of corporate income tax under variable
returns to scale: comment, Public Finance, 44, 335 - 336.
Ray, R. (1986a) Sensitivity of optimal commodity tax rates to alternative
demand functional forms, Journal of Public Economics, 31, 253 - 268.
Ray, R. (1986b) Redistribution through commodity taxes: the non-linear
Engel curve case, Public Finance, 41, 277 - 284.
Reddaway, W.B. (1970) Eects of the Selective Employment Tax, London:
HMSO.
Regan, D.H. (1972) The problem of social cost revisited, Journal of Law
and Economics, 15, 427 - 437.
Reinganum, J. and L. Wilde (1985) Income tax compliance in a principalagent framework, Journal of Public Economics, 26, 1 - 18.
Reinganum, J. and L. Wilde (1986) Equilibrium verification and reporting
policies in a model of tax compliance, International Economics Review, 27,
739 - 760.
Revesz, J.T. (1986) On some advantages of progressive indirect taxation,
Rey, M. (1965) Estimating tax evasions: the example of the Italian General
Sales Tax, Public Finance, 20,366 - 392.
Ricardo, D. (1817) The Principles of Political Economy and Taxation, London: M. Dent and Sons.
Rob, R. (1982) Asymptotic eciency of the demand revealing mechanism,
Rob, R. (1989) Pollution claim settlements under private information,
Journal of Economic Theory, 47, 307 -333.
Robbins, L. (1932) An Essay on the Nature and Significance of Economic
Science, London: George Allen and Unwin.
Roberts, D.J. (1973) Existence of Lindahl equilibrium with a measure space
of consumers, Journal of Economic Theory, 6, 355 - 381.
Roberts, D.J. (1974) The Lindahl solution for economies with public goods,
Roberts, J. and H. Sonnenschein (1977) On the foundations of the theory
of monopolistic competition, Econometrica, 45, 101 - 113.
520
Roberts, K.W.S. (1980a) Possibility theorems with interpersonally comparable welfare levels, Review of Economic Studies, 47, 409 - 420.
Roberts, K.W.S. (1980b) Interpersonal comparability and social choice theory, Review of Economic Studies, 47, 421 - 439.
Roberts, K.W.S. (1980c) Social choice theory: the single and multi-profile
approaches, Review of Economic Studies, 47, 441 - 450.
Roberts, K.W.S. (1980d) The limit points of monopolistic competition,
Roberts, M.J. and M. Spence (1976) Euent charges and licences under
uncertainty, Journal of Public Economics, 5, 193 - 208.
Romer, T. (1976) On the progressivity of the utilitarian income tax, Public
Finance, 31, 329 - 340.
Rosen, S. (1974) Hedonic prices and implicit markets: product dierentiation in pure competition, Journal of Political Economy, 82, 34 - 55.
Rothbarth, E. (1943) Note on a method of determining equivalent income
for households of dierent composition, in C. Madge (ed.), War-Time Pattern
of Savings and Expenditure, Cambridge: Cambridge University Press.
Rothschild, M. and J.E. Stiglitz (1973) Some further results on the measurment of inequality, Journal of Economic Theory, 6, 188 - 204.
Rothschild, M. and J.E. Stiglitz (1976) Equilibrium in competitive insurance markets: an essay in the economics of incomplete information, Quarterly
Rowntree, B.S. (1901) Poverty: A Study of Town Life, London: Macmillan.
Rowntree, B.S. (1941) Poverty and Progress, London: Longman.
Rowntree, B.S. and G.R. Lavers (1951) Poverty and the Welfare State, London: Longman.
Rubinstein, A. (1982) Perfect equilibrium in a bargaining model, Econometrica, 50, 97 - 109.
Rubinstein, A. (1985) A bargaining model with incomplete information
about time preferences, Econometrica, 53, 1151 - 1172.
Sadka, E. (1976) On income distribution incentive eects and optimal income taxation, Review of Economic Studies, 43, 261 - 268.
Samuelson, P.A. (1954) The pure theory of public expenditure, Review of
Economics and Statistics, 36, 387 - 389.
Samuelson, P.A. (1955) Diagrammatic exposition of a pure theory of public
expenditure, Review of Economics and Statistics, 37, 350 - 356.
Samuelson, P.A. (1958) An exact consumption-loan model of interest with
or without the social contrivance of money, Journal of Political Economy, 66,
467 - 482.
Samuelson, P.A. (1964) Discussion, American Economic Review, Papers
and Proceedings, 59, 93 - 96.
Samuelson, P.A. (1975a) Optimum social security in a life-cycle growth
model, International Economic Review, 16, 539 - 544.
Samuelson, P.A. (1975b) The Optimum Growth Rate for Population, International Economic Review, 16, 531 - 538.
521
Samuelson, P.A. (1976) The optimum growth rate for population: agreement and evaluations, International Economic Review, 17, 516 - 525.
Samuelson, P.A. (1977) Rearming the existence of reasonable BergsonSamuelson social welfare functions, Economica, 44, 81 - 88.
Samuelson, P.A. (1986) Theory of optimal taxation, Journal of Public
Economics, 30, 137-143.
Sandmo, A. (1972) Discount rates for public investment under uncertainty,
Sandmo, A. (1973) Public goods and the technology of consumption, Review of Economic Studies, 49, 517 - 528.
Sandmo, A. (1974) A note on the structure of optimal taxation, American
Sandmo, A. (1975) Optimal taxation in the presence of externalities,
Swedish Journal of Economics, 77, 86 - 98.
Sandmo, A. (1977) Portfolio theory, asset demand and taxation: comparative statics with many assets, Review of Economic Studies, 44, 369 - 379.
Sandmo, A. (1981) Income tax evasion, labour supply, and the equityeciency tradeo, Journal of Public Economics, 16, 265 - 288.
Sargent, T.J. (1979) Macroeconomic Theory, New York: Academic Press.
Sawhill, I.V. (1988) Poverty in the US: why is it so persistent?, Journal
of Economic Literature, 26, 1073 - 1119.
Scitovsky, T. (1951) The state of welfare economics, American Economic
Review, 41, 303 - 315.
Scotchmer, S. (1987) Audit classes and tax enforcement policy, American
Seade, J. (1977) On the shape of optimal tax schedules, Journal of Public
Economics, 7, 203 - 236.
Seade, J. (1980) The stability of Cournot revisited, Journal of Economic
Theory, 23, 749 - 752.
Seade, J. (1982) On the sign of the optimum marginal income tax, Review
of Economic Studies, 49, 637 - 643.
Seade, J. (1985) Profitable cost increases, Warwick Economic Research
Paper, No. 260.
Sen, A.K. (1973) On Economic Inequality, Oxford: Oxford University Press.
Sen, A.K. (1976) Poverty: an ordinal approach to measurement, Econometrica, 44, 219 - 231.
Sen, A.K. (1977) On weights and measures, Econometrica, 45, 1539 1572.
Sen, A.K. (1979) Issues in the measurement of poverty, Scandinavian
Sen, A.K. (1986) Social choice theory in K.J. Arrow and M.D. Intriligator
(eds.) Handbook of Mathematical Economics, Amsterdam: North-Holland.
Seneca, J.J. and M.K. Taussig (1971) Family equivalence scales and personal income tax exemptions for children, Review of Economics and Statistics,
53, 253 - 262.
522
Shell, K. (1971) Notes on the economics of infinity, Journal of Political

Economy, 79, 1002 - 1011.
Sheshinski, E. (1972a) Relation between a social welfare function and the
Gini index of inequality, Journal of Economic Theory, 4, 98 - 100.
Sheshinski, E. (1972b) The optimal linear income tax, Review of Economic
Studies, 39, 297 - 302.
Sheshinski, E. (1978) A model of social security and retirement decisions,
Sheshinski, E. and Y. Weiss (1981) Uncertainty and optimal social security
systems, Quarterly Journal of Economics, 96, 189 - 206.
Shorrocks, A.F. (1980) The class of additively decomposable inequality
measures, Econometrica, 48, 613 - 625.
Shorrocks, A.F. (1983) Ranking income distributions, Economica, 50, 3 17.
Shorrocks, A.F. (1984) Inequality decompositions by population subgroups,
Shoven, J.B. and J. Whalley (1972) A general equilibrium calculation of
the eects of dierential taxation of income from capital in the U.S., Journal
Simons, H.C. (1938) Personal Income Taxation, Chicago: University of
Chicago Press.
Smith, A. (1776) An Enquiry into the Nature and Causes of the Wealth of
Nations, London: Metheun.
Smith, A. (1981) The informal economy, Lloyds Bank Review, 45 - 61.
Smith, A. (1982), Intergenerational transfers as social insurance, Journal
Smith, A. (1983) Tax reform and temporary ineciency, Journal of Public
Economics, 20, 265 - 270.
Spence, M. (1977) Nonlinear prices and welfare, Journal of Public Economics, 8, 1 - 18.
Spicer, M.W. and S.B. Lundstedt (1976) Understanding tax evasion, Public Finance, 31, 295 - 305.
Srinivasan, P.V. (1989) Redistributive impact of optimal commodity taxes,
Economics Letters, 30, 385 - 388.
Srinivasan, T.N. (1973) Tax evasion: a model, Journal of Public Economics, 2, 339 - 346.
Starret, D.A. (1972) Fundamental nonconvexities in the theory of externalitites, Journal of Economic Theory, 4, 180 - 199.
Stern, N.H. (1976) On the specification of models of optimum income taxation., Journal of Public Economics, 6, 123 - 162.
Stern, N.H. (1987) The eects of taxation, price control and government
contracts in oligopoly and monopolistic competition, Journal of Public Economics, 32, 133 - 158.
Stiglitz, J.E. (1969a) The eects of income,wealth, and capital gains taxation on risk-taking, Quarterly Journal of Economics, 83, 262 - 283.
523
Stiglitz, J.E. (1969b) A re-examination of the Modigliani-Miller Theorem,
Stiglitz, J. E. (1973), Taxation, corporate financial policy, and the cost of
capital, Journal of Public Economics, 2, 1 - 34.
Stiglitz, J.E. (1974) On the irrelevance of corporate financial policy, American Economic Review, 64, 851 - 866.
Stiglitz, J.E. (1976) The corporation tax, Journal of Public Economics, 5,
303 - 311.
Stokey, N.L. and R.E. Lucas with E.C. Prescott (1989) Recursive Methods
in Economic Dynamics, Cambridge: Harvard University Press.
Stone, R. (1954) Linear expenditure systems and demand analysis: an application to the pattern of British demand, Economic Journal, 64, 511 - 527.
Sugden, R. (1984) Reciprocity: the supply of public goods through voluntary contributions, Economic Journal, 94, 772 787.
Sugden, R. (1985) Consistent conjectures and voluntary contributions to
public goods: why the conventional theory does not work, Journal of Public
Economics, 27, 117 - 124.
Svensson, L.G. and J.W. Weibull (1986) An upper bound on optimal income
taxes, Journal of Public Economics, 30, 165 - 181.
Sweeney, R.J. (1988) Wealth Eects and Monetary Theory, New York: Basil
Blackwell.
Taggart, R.A. (1980) Taxes and corporate capital structure in an incomplete market, Journal of Finance, 35, 645 - 659.
Takayama, A. (1985) Mathematical Economics, Cambridge: Cambridge University Press.
Takayama, N. (1979) Poverty, income inequality, and their measures: Professor Sens axiomatic approach reconsidered, Econometrica, 47, 747 - 758.
Theil, H. (1967) Economics and Information Theory, Amsterdam: NorthHolland.
Thompson, L.H., (1983) The social security reform debate, Journal of
Economic Literature, 21, 1425 - 1467
Townsend, P. (1979) Poverty in the United Kingdom, London: Penguin.
Tuomala, M. (1985) Simplified formulae for optimal linear income taxation, Scandinavian Journal of Economics, 87, 668 - 672.
Tuomala, M. (1990) Optimal Income Tax and Redistribution, Oxford: Clarendon Press.
Varian, H.R. (1980) Redistributive taxation as social insurance, Journal
Varian, H.R. (1984) Price discrimination and social welfare, American
Varian, H.R. (1992) Microeconomic Theory, New York: Norton.
Veall, M.R. (1986) Public pensions as optimal social contracts, Journal of
Venables, A.J. (1986) Production subsidies, import taris and imperfectly
competitive trade, in: Greenaway, D. and P.K.M. Tharakan (eds.), Imperfect
Competition and International Trade, Sussex: Wheatsheaf.
524
Vickrey, W. (1964) Discussion, American Economic Review, Papers and

Proceedings, 59, 88 - 92.
Virmani, A. (1989) Indirect tax evasion and production eciency, Journal
of Public Economics, 39, 223-237.
von Neumann, J. and O. Morgernstern (1953) Theory of Games and Economic Behaviour, Third Edition, Princeton: Princeton University press.
Walras, L. (1874) Elements dEconomie Politique Pure, Lausanne: L. Corbaz. English translation by W. Jae (1954), Elements of pure economics, London: Allen and Unwin.
Warr, P.G. (1983) The private provision of a pure public good is independent of the distribution of income, Economics Letters, 13, 207 - 211.
Weber, S. and H. Wiesmeth (1991) The equivalence of core and cost share
equilibria in an economy with public goods, Journal of Economic Theory, 54,
180 - 199.
Weil, P. (1987) Love thy children; reflections on the Barro debt neutrality
theorem, Journal of Monetary Economics, 19, 378 - 391.
Weitzman, M.L. (1974) Prices vs. quantities, Review of Economic Studies,
41, 477 - 491.
Weymark, J.A. (1979a) Generalised Gini inequality indices, Discussion
Paper No. 79-12, Department of Economics, University of British Columbia.
Weymark, J.A. (1979b) A reconciliation of recent results in optimal taxation theory, Journal of Public Economics, 12, 171 - 189.
Weymark, J.A. (1981) Undominated directions of tax reform, Journal of
Wilson, J.D. (1991a) Optimal public good provision in the Ramsey tax
model: a generalization, Economics Letters, 35, 57 - 61.
Wilson, J.D. (1991b) Optimal public good provision with limited lump-sum
taxation, American Economic Review, 81, 153 - 166.
Yamada, M. (1990) An analysis of optimal taxation with tax evasion,
Yitzhaki, S. (1974) Income tax evasion: a theoretical analysis, Journal of
Yohe, G.W. (1978) Towards a general comparison of price controls and
quantity controls under uncertainty, Review of Economic Studies, 45, 229 238.
Young, D. (1982) Voluntary purchase of public goods, Public Choice, 38,
73 - 86.

MR Regression

Uploaded by

Copyright:

Available Formats

MR Regression

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MR Regression

Uploaded by

Copyright:

Available Formats

Public Economics

2 General Equilibrium and Welfare Economics

Analysis in the competitive model

8.3.3 Taxation and finance . . . . . . . . . . . . .

Relaxing the Assumptions

11.8 Other forms of regulation . . . . . . . . . . . . . . . . . . . . . . 367

Introducing Real Time

13 Overlapping Generations Economies

15 Debt and Taxes

In the broadest interpretation, public economics is the study of economic policy,

of evaluating, in terms of welfare, the outcome of policy. To achieve the first

command economy implies no claim that such perfect control is possible, or

ering whether intervention would be beneficial is compelling. For example, if

Eciency versus equity

Following the discussion of the methodology that will be employed, it is clear

1.7. PUBLIC SECTOR INCOME AND EXPENDITURE

Public sector income and expenditure

Figure 1.1: General government total outlay as % of nominal GDP

Figure 1.2: Japanese Government Income 1991

1.7. PUBLIC SECTOR INCOME AND EXPENDITURE

Goods and services

Figure 1.3: Japanese Government Expenditure 1991

Figure 1.4: UK Government Income

Goods and services

Figure 1.5: UK Government Expenditure

1.7. PUBLIC SECTOR INCOME AND EXPENDITURE

Figure 1.6: Comparison of Tax Yields

(a) Federal Government

(b) State and Local

Figure 1.8: US State and Federal Expenditure

h marginal utility of income

General Equilibrium and

20CHAPTER 2. GENERAL EQUILIBRIUM AND WELFARE ECONOMICS

The Arrow-Debreu economy

This section introduces the constituent parts of the Arrow-Debreu economy,

2.2. THE ARROW-DEBREU ECONOMY

The institutional framework

The concept of a commodity is central to the economy and the flexibility in

22CHAPTER 2. GENERAL EQUILIBRIUM AND WELFARE ECONOMICS

2.2. THE ARROW-DEBREU ECONOMY

is strictly convex for any x

U h xh1 , ..., xhn ,

where each hj 0. Hence if firm j makes a profit of amount j , household h

24CHAPTER 2. GENERAL EQUILIBRIUM AND WELFARE ECONOMICS

xh xh1 , ..., xhn ,

to maximize their utility function U h () subject to the budget constraint

As individual demands may be negative, it is clearly possible for the aggregate

2.2. THE ARROW-DEBREU ECONOMY

Figure 2.1: A Typical Production Set

26CHAPTER 2. GENERAL EQUILIBRIUM AND WELFARE ECONOMICS

The representation of profit as an inner product is mathematically convenient

y j y1j , ..., ynj ,

As noted in the discussion of household demand, it is the level of aggregate

2.2. THE ARROW-DEBREU ECONOMY

Equation (2.17) expresses aggregate demand as a function of the price vector

28CHAPTER 2. GENERAL EQUILIBRIUM AND WELFARE ECONOMICS

summing over all households gives

Using the definition of aggregate supply in (2.15), (2.22) can be expressed as

2.2. THE ARROW-DEBREU ECONOMY

Equation (2.24) is the standard form of Walras law. In words, Walras