Economics of Family

This is page i
Printer: Opaque this
Family Economics
Martin Browning Pierre-André Chiappori

Yoram Weiss
January 2011
ii
This is page iii
Contents
Introduction 3
I Models of Household Behavior 11

1 Facts 13
1.1 Marriage, divorce and remarriage . . . . . . . . . . . . . . . 13
1.1.1 Marital status . . . . . . . . . . . . . . . . . . . . . . 13
1.1.2 Marital histories . . . . . . . . . . . . . . . . . . . . 14
1.1.3 Flows . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.1.4 Transitions . . . . . . . . . . . . . . . . . . . . . . . 16
1.1.5 Households . . . . . . . . . . . . . . . . . . . . . . . 17
1.2 Marriage, work and wages . . . . . . . . . . . . . . . . . . . 18
1.2.1 Time use . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.2 Wages . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.2.3 The marriage premium . . . . . . . . . . . . . . . . 20
1.3 Who Marries Whom? . . . . . . . . . . . . . . . . . . . . . 21
1.4 Children . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.1 Fertility . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.2 Children under different household arrangements . . 25
1.5 Saving and life stages . . . . . . . . . . . . . . . . . . . . . 26
1.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.7 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . 31
2 The gains from marriage 81

2.1 Public goods . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.2 Specialization and Increasing Returns to Scale . . . . . . . . 89
2.3 Imperfect Credit Markets . . . . . . . . . . . . . . . . . . . 91
2.4 Risk sharing . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.5 Children . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.5.1 Technology and preferences . . . . . . . . . . . . . . 94
2.5.2 Intact families . . . . . . . . . . . . . . . . . . . . . 95
2.5.3 Divorce with no transfers . . . . . . . . . . . . . . . 96
2.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . 97
2.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3 Preferences and decision making 103

3.1 Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
iv
3.2 Household production . . . . . . . . . . . . . . . . . . . . . 107

3.2.1 The general framework. . . . . . . . . . . . . . . . . 107
3.2.2 Children . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.3 The unitary model . . . . . . . . . . . . . . . . . . . . . . . 111
3.4 Non-cooperative models . . . . . . . . . . . . . . . . . . . . 112
3.4.1 No public goods. . . . . . . . . . . . . . . . . . . . . 112
3.4.2 One public good. . . . . . . . . . . . . . . . . . . . . 113
3.4.3 Altruism and the Rotten Kid Theorem. . . . . . . . 116
3.4.4 Many public goods. . . . . . . . . . . . . . . . . . . 117
3.5 Cooperative models: the collective approach . . . . . . . . . 119
3.5.1 Decision processes . . . . . . . . . . . . . . . . . . . 120
3.5.2 Assuming efficiency . . . . . . . . . . . . . . . . . . 121
3.5.3 Distribution factors . . . . . . . . . . . . . . . . . . 123
3.5.4 Modeling efficiency . . . . . . . . . . . . . . . . . . . 124
3.5.5 Pareto weights and ‘power’ . . . . . . . . . . . . . . 128
3.5.6 Household utility . . . . . . . . . . . . . . . . . . . . 129
3.5.7 Caring . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3.5.8 Children . . . . . . . . . . . . . . . . . . . . . . . . . 132
3.5.9 The unitary model as a special case . . . . . . . . . 133
3.5.10 The Rotten Kid Theorem revisited . . . . . . . . . . 135
3.5.11 Bargaining models . . . . . . . . . . . . . . . . . . . 137
3.5.12 Other approaches . . . . . . . . . . . . . . . . . . . . 142
3.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4 The collective model: a formal analysis 157

4.1 Collective demand functions: a general characterization . . 157
4.1.1 The collective household utility function. . . . . . . 157
4.1.2 Structural and observable demand. . . . . . . . . . . 157
4.1.3 The Slutsky matrix for collective demands. . . . . . 159
4.1.4 Distribution factors . . . . . . . . . . . . . . . . . . 161
4.1.5 Larger households . . . . . . . . . . . . . . . . . . . 162
4.1.6 Children . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.2 Duality in the collective model . . . . . . . . . . . . . . . . 164
4.2.1 The collective expenditure function. . . . . . . . . . 164
4.2.2 Indirect utilities . . . . . . . . . . . . . . . . . . . . 166
4.2.3 Welfare . . . . . . . . . . . . . . . . . . . . . . . . . 167
4.3 The case of purely private consumptions . . . . . . . . . . . 169
4.3.1 The sharing rule. . . . . . . . . . . . . . . . . . . . . 169
4.3.2 Caring preferences . . . . . . . . . . . . . . . . . . . 171
4.3.3 Indirect utilities . . . . . . . . . . . . . . . . . . . . 172
4.4 Application: labor supply with private consumption . . . . 173
4.4.1 The general setting . . . . . . . . . . . . . . . . . . . 173
4.4.2 Egoistic preferences and private consumption . . . . 174
4.4.3 Collective labor supply . . . . . . . . . . . . . . . . . 175
4.5 Public goods . . . . . . . . . . . . . . . . . . . . . . . . . . 177
v
4.5.1 Lindahl prices . . . . . . . . . . . . . . . . . . . . . . 177

4.5.2 The conditional sharing rule. . . . . . . . . . . . . . 179
4.5.3 Application: labor supply, female empowerment and
expenditures on public good . . . . . . . . . . . . . . 180
4.6 Household production in the collective model . . . . . . . . 183
4.6.1 The basic model . . . . . . . . . . . . . . . . . . . . 185
4.6.2 Domestic production and time use . . . . . . . . . . 186
4.6.3 Empirical issues . . . . . . . . . . . . . . . . . . . . 193
4.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5 Empirical issues for the collective model 199

5.1 What are the objects of interest? . . . . . . . . . . . . . . . 199
5.2 Data without price variation. . . . . . . . . . . . . . . . . . 202
5.2.1 Necessary and sufficient conditions for a collective
model. . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.2.2 Identifiability. . . . . . . . . . . . . . . . . . . . . . . 204
5.2.3 Private consumption. . . . . . . . . . . . . . . . . . . 206
5.2.4 Assignability. . . . . . . . . . . . . . . . . . . . . . . 207
5.3 Observing price responses. . . . . . . . . . . . . . . . . . . . 209
5.3.1 Testing the collective model . . . . . . . . . . . . . . 209
5.3.2 Identifying the collective model . . . . . . . . . . . . 211
5.3.3 A simple example . . . . . . . . . . . . . . . . . . . 214
5.3.4 The revealed preference approach . . . . . . . . . . . 217
5.4 The case of labor supply . . . . . . . . . . . . . . . . . . . . 218
5.4.1 Egoistic preferences and private consumption . . . . 218
5.4.2 Extensions . . . . . . . . . . . . . . . . . . . . . . . 221
5.5 Empirical evidence. . . . . . . . . . . . . . . . . . . . . . . . 223
5.5.1 Evidence against the unitary model. . . . . . . . . . 223
5.5.2 Evidence on the collective model. . . . . . . . . . . . 227
5.5.3 Estimating the collective model . . . . . . . . . . . . 229
5.5.4 Concluding remarks. . . . . . . . . . . . . . . . . . . 236
5.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
6 Uncertainty and Dynamics in the Collective model 243

6.1 Is commitment possible? . . . . . . . . . . . . . . . . . . . . 244
6.2 Modeling commitment . . . . . . . . . . . . . . . . . . . . . 251
6.2.1 Full commitment . . . . . . . . . . . . . . . . . . . . 251
6.2.2 Constraints on commitment . . . . . . . . . . . . . . 252
6.2.3 Endogenous Pareto weights . . . . . . . . . . . . . . 253
6.3 Efficient risk sharing in a static context . . . . . . . . . . . 254
6.3.1 The collective model under uncertainty . . . . . . . 254
6.3.2 Efficient risk sharing in a one-commodity world . . . 258
6.3.3 Efficient risk sharing in a multi-commodity context:
an introduction . . . . . . . . . . . . . . . . . . . . . 261
6.3.4 Econometric issues . . . . . . . . . . . . . . . . . . . 263
vi
6.4 Intertemporal Behavior . . . . . . . . . . . . . . . . . . . . 266

6.4.1 The unitary approach: Euler equations at the house-
hold level . . . . . . . . . . . . . . . . . . . . . . . . 266
6.4.2 Collective Euler equations under ex ante efficiency . 268
6.4.3 The ex ante inefficiency case . . . . . . . . . . . . . 270
6.4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . 271
6.5 Divorce . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
6.5.1 The basic model . . . . . . . . . . . . . . . . . . . . 272
6.5.2 Divorce under transferable utility and the Becker-
Coase theorem . . . . . . . . . . . . . . . . . . . . . 276
6.5.3 Divorce and investment in children: a non transfer-
able utility example . . . . . . . . . . . . . . . . . . 281
6.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
II Equilibrium Models of the Marriage Market 291

7 Matching on the Marriage Market: Theory 293
7.1 Stable Matching without transfers: the Gale-Shapley Algo-
rithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
7.2 Stable Matching with transferable utilities: the Becker-Shapley-
Shubik model . . . . . . . . . . . . . . . . . . . . . . . . . . 298
7.2.1 The basic framework . . . . . . . . . . . . . . . . . . 298
7.2.2 Assortative mating . . . . . . . . . . . . . . . . . . . 304
7.2.3 Matching with a continuum of agents . . . . . . . . 308
7.2.4 Multidimensional matching . . . . . . . . . . . . . . 311
7.3 Matching with general utilities . . . . . . . . . . . . . . . . 312
7.4 Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
7.4.1 The basic framework . . . . . . . . . . . . . . . . . . 316
7.4.2 Search and Assortative Mating . . . . . . . . . . . . 321
7.5 Bargaining In Marriage (BIM) . . . . . . . . . . . . . . . . 324
7.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
8 Sharing the gains from marriage 333

8.1 Determination of shares with a finite number of agents . . . 334
8.1.1 The two men - two women case . . . . . . . . . . . . 334
8.1.2 Bounds on levels . . . . . . . . . . . . . . . . . . . . 337
8.1.3 Bounds on Differences . . . . . . . . . . . . . . . . . 339
8.2 The continuous case . . . . . . . . . . . . . . . . . . . . . . 341
8.2.1 Basic results . . . . . . . . . . . . . . . . . . . . . . 342
8.2.2 A tractable specification . . . . . . . . . . . . . . . . 345
8.2.3 Comparative Statics . . . . . . . . . . . . . . . . . . 348
8.2.4 Taxation . . . . . . . . . . . . . . . . . . . . . . . . . 353
8.2.5 An example . . . . . . . . . . . . . . . . . . . . . . . 354
vii
8.2.6 Matching on preferences: Roe vs. Wade and female

empowerment . . . . . . . . . . . . . . . . . . . . . . 361
8.3 Matching with general utilities . . . . . . . . . . . . . . . . 369
8.3.1 Recovering individual utilities: the general strategy . 370
8.3.2 A specific example . . . . . . . . . . . . . . . . . . . 371
8.4 Matching by Categories . . . . . . . . . . . . . . . . . . . . 377
8.5 Appendix: Extreme Value distributions . . . . . . . . . . . . 382
8.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
9 Investment in Schooling and the Marriage Market 391

9.1 Is pre-marital investment efficient? . . . . . . . . . . . . . . 392
9.2 The Basic Model . . . . . . . . . . . . . . . . . . . . . . . . 395
9.2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . 395
9.2.2 Assumptions . . . . . . . . . . . . . . . . . . . . . . 396
9.2.3 The Marriage Market . . . . . . . . . . . . . . . . . 397
9.2.4 Investment Decisions . . . . . . . . . . . . . . . . . . 400
9.2.5 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . 401
9.3 Gender Differences in the Incentive to Invest . . . . . . . . 405
9.3.1 The Household . . . . . . . . . . . . . . . . . . . . . 406
9.3.2 The Impact of the Wage Gap . . . . . . . . . . . . . 408
9.3.3 The Impact of Household Roles . . . . . . . . . . . . 410
9.3.4 Division of Labor and Career Choice . . . . . . . . . 411
9.3.5 Why Women May Acquire More Schooling than Men 413
9.4 A Numerical Example . . . . . . . . . . . . . . . . . . . . . 414
9.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
9.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
9.7 Appendix: Existence and Uniqueness of Equilibrium . . . . 424
10 An equilibrium model of marriage, fertility and divorce 435

10.1 A model of the Marriage Market . . . . . . . . . . . . . . . 435
10.1.1 Individual Choices . . . . . . . . . . . . . . . . . . . 436
10.1.2 Aggregation . . . . . . . . . . . . . . . . . . . . . . . 440
10.1.3 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . 440
10.2 An Example . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
10.3 Income uncertainty and ex-post heterogeneity . . . . . . . . 445
10.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
10.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
10.5.1 Properties of the expected utility, with and without
children . . . . . . . . . . . . . . . . . . . . . . . . . 449
10.5.2 Properties of the trigger for having children, θc . . . 450
10.5.3 Properties of the trigger for marriage, θm . . . . . . 450
10.5.4 Calculations for the example . . . . . . . . . . . . . 452
10.5.5 Low costs of raising children . . . . . . . . . . . . . 453
10.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454
viii
11 Marriage, Divorce, Children 459

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
11.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
11.2.1 The technology and preferences . . . . . . . . . . . . 462
11.2.2 The legal framework . . . . . . . . . . . . . . . . . . 464
11.2.3 The meeting’s technology . . . . . . . . . . . . . . . 464
11.3 Household structure and child care . . . . . . . . . . . . . . 465
11.3.1 Intact family . . . . . . . . . . . . . . . . . . . . . . 465
11.3.2 Separation, custody and voluntary transfers . . . . . 466
11.4 Transfers, the interim perspective . . . . . . . . . . . . . . . 471
11.4.1 Partial equilibrium . . . . . . . . . . . . . . . . . . . 474
11.5 Divorce . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
11.6 Fertility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
11.7 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
11.8 Further issues . . . . . . . . . . . . . . . . . . . . . . . . . . 484
11.8.1 The custody assignment . . . . . . . . . . . . . . . . 484
11.8.2 Mandated and contingent contracts . . . . . . . . . 485
11.8.3 Transfers, the ex-ante perspective . . . . . . . . . . . 490
11.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
11.10Appendix: Contingent contracts . . . . . . . . . . . . . . . . 492
11.10.1 The choice of contract . . . . . . . . . . . . . . . . . 492
11.11References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
This is page ix
List of Tables
1.1 Marital Status of Men and Women, over 20 Years Old, in
different Countries and Years . . . . . . . . . . . . . . . . . 31
1.2 Cohabitation in the US and Denmark by Age of the House-
hold Head . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.3 Household Arrangements, Denmark 2000 . . . . . . . . . . . 32
1.4 Household Arrangements, USA, 2000-2005 . . . . . . . . . . 33
1.5 Marital History by Age and Sex, US, 2001 (percents) . . . . 33
1.6 Marital History, US, of the 1931-36 and 1937-41 Birth Cohorts 34
1.7 Marital History of the NLS Panel . . . . . . . . . . . . . . . 34
1.8 Individuals Living Alone, various countries . . . . . . . . . . 35
1.9 Individuals Living Alone, Denmark . . . . . . . . . . . . . . 35
1.10 Hours of Work and Leisure per Day . . . . . . . . . . . . . 36
1.11 Hours per Day of Home Production, Childcare and Shopping 37
1.12 Labor Force Participation of Women and Men in Ten Countries 38
1.13 Marital Status at Age 35, by Gender and Education at Age 35 38
1.14 Years of Schooling at Marriage and at Age 35, by Gender
and Marital Status at Age 35 . . . . . . . . . . . . . . . . . 39
1.15 Completed Fertility for Two US Cohorts . . . . . . . . . . . 39
1.16 Living Arrangements of U.S. Children, Aged less than 18,
by Year . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.17 Child Support and Alimony Received by Mothers with Chil-
dren 0-18 (in 1982-84 dollars) by Mother’s Age and Time
Period . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.18 Consumption Through Life Stages . . . . . . . . . . . . . . 40
1.19 Descriptive Regression for Log Nondurable Consumption . . 41
2.1 Bounds for the relative cost of equivalent bundles . . . . . . 89

2.2 Work patterns and material welfare of family members . . . 97
5.1 Distribution factors . . . . . . . . . . . . . . . . . . . . . . . 224

5.2 Empirical collective studies . . . . . . . . . . . . . . . . . . 225
5.3 Budget shares for husbands and wives . . . . . . . . . . . . 232
x
This is page xi
List of Figures
1.1 Marriage Rates for Selected Countries. Source: Eurostat. . . 42
1.2 Divorce Rates for Selected Countries. Source: Eurostat. . . 43
1.3 Marriage and Divorce Rates, US 1940-2002. Source: National
Center of Health Statistics. . . . . . . . . . . . . . . . . . . 44
1.4 Marriage, Divorce and Remarriage Rates, US 1921-1989.
Source: National Center of Health Statistics. . . . . . . . . . 45
1.5 Entry into First Marriage, US, HRS Cohort. Source: Health
and Retirement Survey, 1992 Wave. . . . . . . . . . . . . . . 46
1.6 Entry into First Marriage, US, NLS Panel. Source: National
Longitudinal Survey, Youth,1979. . . . . . . . . . . . . . . . 47
1.7 Hazard of Divorce for two Birth Cohorts, US. Source: Na-
tional Longitudinal Survey, Health and Retirement Survey. 48
1.8 Hazard of Reamarriage, US. Source: National Longitudinal
Survey, Health and Retirement Survey. . . . . . . . . . . . . 49
1.9 Households by Type: Selected Years, US. Source: US Census. 50
1.10 Households by Size, Selected Years, US. Source: US Census. 51
1.11 Full-Time Workers, by Marital Status and Sex, US Birth
Cohort 1945-54. Source: Current Population Surveys. . . . . 52
1.12 Full-Time Workers, by Marital Status and Sex, US Birth
Cohort 1960—69. Source: Current Population Surveys. . . . 53
1.13 Full-Time Female Workers, by Marital Status and Child, US
Birth Cohort 1945-54. Source: Current Population Surveys. 54
1.14 Full-Time Female Workers, by Marital Status and Child, US
Birth Cohort 1960-69. Source: Current Population Surveys. 55
1.15 Work Patterns of Husbands and Wives (Aged 40-60). Source:
Current Population Surveys. . . . . . . . . . . . . . . . . . . 56
1.16 Work Patterns of Husbands and Wives (Aged 30-40). Source:
1.17 Weekly Wages of FT Workers, by Marital Status and Sex,
US Birth Cohort 1945-54. Source: Current Population Surveys. 58
1.18 Weekly Wages of FT Workers, by Marital Status and Sex,
US Birth Cohort 1960-69. Source: Current Population Surveys. 59
1.19 Log Wages Differnces between Married and Singles, by Sex,
US 1968—2005. Source: Current Population Surveys. . . . . 60
1.20 Women Working Full Time, by Marital Status, US 1968—2005. 61
1.21 Age Differences between Husbands and Wives, US 1968-
2005. Source: Current Population Surveys. . . . . . . . . . . 62
1.22 Age Pyramid , US 1950. Source: US Census. . . . . . . . . . 63
xii
1.23 Age Pyramid , US 2005. Source: US Census. . . . . . . . . . 64

1.24 Completed Education by Sex, Age 30-40, US 1968-2005.
Source: Current Population Surveys. . . . . . . . . . . . . . 65
1.25 Education of Spouses, by Husband’s Year of Birth, US. Source:
1.26 Spouse Education by own Education, Ages 30-40,US 1970-
79. Source: Current Population Surveys. . . . . . . . . . . . 67
1.27 Spouse Education by own Education, Ages 30-40, US 1996-
2005. Source: Current Population Surveys. . . . . . . . . . . 68
1.28 Husband-Wife Correlations of Schooling and Wages, Ages
25-40, US 1976-2005. Source: Current Population Surveys. . 69
1.29 Divorce Rates of Men+Women by Schooling, US 1973-2005
(% divorced age 30 − 35 year t) /(% marrried age 25 − 30
year t − 5). Source: Current Population Surveys. . . . . . . 70
1.30 Completed Fertility, US by Birth Cohort. Source: National
Center of Health Statistics. . . . . . . . . . . . . . . . . . . 71
1.31 Number of Children 0-18 of US Women, Aged 35-45 at Se-
lected Years. Source: Current Population Surveys. . . . . . 72
1.32 Birth Rates and Median Age at First Marriage, US 1900-
2000. Source: National Center of Health Statistics. . . . . . 73
1.33 Completed Fertility by Generation. Source: Eurostat. . . . . 74
1.34 Mean Age at First Marriage by Generation, Mean Age at
First Marriage by Generation. Source: Eurostat. . . . . . . 75
1.35 Age of Women at First Birth by Generation. Source: Eurostat. 76
1.36 Percent of US Mothers Receiving Child Support, by Marital
Status. Source: Panel Study of Income Dynamics. . . . . . . 77
1.37 Consumption and household size - more educated wives.
Source: UK Family Expenditure Surveys. . . . . . . . . . . 78
1.38 Consumption and household size - less educated wives. Source:
UK Family Expenditure Surveys. . . . . . . . . . . . . . . . 79
2.1 Preferences over two public goods . . . . . . . . . . . . . . . 84

2.2 Gains from public goods. . . . . . . . . . . . . . . . . . . . 85
3.1 Demand for public good . . . . . . . . . . . . . . . . . . . . 150

3.2 The demand for public goods with altruism. . . . . . . . . . 151
3.3 Household demands for public goods. . . . . . . . . . . . . . 152
3.4 The utility possibility frontier. . . . . . . . . . . . . . . . . 153
3.5 Linear and non linear generalized Samuelson index . . . . . 154
3.6 All utilities in the shaded area correspong to the same μ . . 154
3.7 All μ in the shaded cone correspond to the same ua . . . . 155
3.8 The effects of changes in prices . . . . . . . . . . . . . . . . 155
3.9 The RKT utility possibility frontier. . . . . . . . . . . . . . 156
4.1 Collective price responses . . . . . . . . . . . . . . . . . . . 196

xiii
4.2 A potentially compensating variation. . . . . . . . . . . . . 196

4.3 An actually compensating variation. . . . . . . . . . . . . . 197
5.1 A Shift in the Sharing Rule . . . . . . . . . . . . . . . . . . 221
6.1 Game tree for investment in education . . . . . . . . . . . . 250

6.2 Pareto frontiers in marriage and divorce, no public goods . 279
6.3 Divorce without transferable utility . . . . . . . . . . . . . . 281
7.1 Positive Assortative Mating . . . . . . . . . . . . . . . . . . 310

7.2 The slope condition . . . . . . . . . . . . . . . . . . . . . . . 315
8.1 Imputations when z12 + z21 > z22 + z11 . . . . . . . . . . . 335

8.2 Imputations when z12 + z21 < z22 + z11 . . . . . . . . . . . 336
8.3 A linear upward shift . . . . . . . . . . . . . . . . . . . . . . 347
8.4 Maximum husband’s utility as a function of the wife’s taste
- intermediate ESW . . . . . . . . . . . . . . . . . . . . . . 365
8.5 Maximum husband’s utility as a function of the wife’s taste
- small ESW . . . . . . . . . . . . . . . . . . . . . . . . . . 366
8.6 An Increase in the Risk of Pregnancy . . . . . . . . . . . . . 368
8.7 Pareto frontier . . . . . . . . . . . . . . . . . . . . . . . . . 373
8.8 Husband’s and Wife’s Utilities, Public Consumption and the
Husband’s Private Consumption . . . . . . . . . . . . . . . 375
8.9 Impact of a Rightwards Shift in Female Income Distribution
on Husband’s and Wife’s Utilities . . . . . . . . . . . . . . 376
8.10 Impact of a Rightwards Shift in Female Income Distribu-
tion on Public Consumption and the Husband’s Private Con-
sumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
8.11 Parameters of the predicted log wage distribution of men
and women and sex ratio, U.S. 1976-2005. Source: Current
Population Surveys. . . . . . . . . . . . . . . . . . . . . . . 385
8.12 Cumulative distributions of predicted hourly wages of men
and women . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
8.13 The surplus of married men and women in 1976, female to
male ratio = 1.045 . . . . . . . . . . . . . . . . . . . . . . . 387
8.14 The surplus of married men and women in 2005, female to
male ratio = 0.984 . . . . . . . . . . . . . . . . . . . . . . . 388
8.15 Wife’s relative share in the surplus, women are the majority
(r=1.1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
8.16 Wife’s relative share in the surplus, women are the majority
(r=0.9) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
9.1 Regions for Marriage and Investment . . . . . . . . . . . . 429

9.2 Equilibrium with Strictly Positive Assortative Matching . . 430
1
9.3 Mixed Equilibrium with More Educated Men than Educated

Women . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
9.4 The Impact of an Increase in the Wage of Educated Men
Combined with a Reduction in the Wage of Educated Women432
9.5 The Impact of an Increase in the Wife’s Work at Home . . 433
9.6 The Impact of a Decrease in the Wife’s Work at Home Com-
bined with an Increase in the Wage of Educated Women . . 434
10.1 Expected utility profiles. . . . . . . . . . . . . . . . . . . . . 455

10.2 Expected utility profiles for example . . . . . . . . . . . . . 456
10.3 Equilibrium Divorce Rates . . . . . . . . . . . . . . . . . . . 457
11.1 Effect of Transfer on Mother’s Work, Child’s Consumption

and Child’s Utility . . . . . . . . . . . . . . . . . . . . . . . 503
11.2 The Pareto Utility Frontier . . . . . . . . . . . . . . . . . . 504
11.3 Optimal Transfers and Child’s Utility if the Mother Rem-
maries or Remains Single . . . . . . . . . . . . . . . . . . . 505
11.4 Expected Utilities of Father, Mother and Child at the Time
of Divorce . . . . . . . . . . . . . . . . . . . . . . . . . . . . 506
11.5 Potential Marriage Market Equilibria, with and without Chil-
dren . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
11.6 The Impact of Deviation to not having a Child on the Ex-
pected Life Time Utility when all other Couples have a Child 508
11.7 The Impact of Deviation to Having a Child on the Expected
Life Time Utility when all other Couples Have no Child . . 509
11.8 with Children under Different Payment Schemes and with-
out Children . . . . . . . . . . . . . . . . . . . . . . . . . . 510
11.9 The Impact of Deviation to a Fixed Payment when all other
Fathers give a Contingent Payment . . . . . . . . . . . . . . 511
11.10The Impact of Deviation to a Contingent Payment when all
other Fathers give no Contingent Payment . . . . . . . . . . 512
2
This is page 3
Introduction
The existence of a nuclear family is to a large extent dictated by nature.
According to Aristotle (Politics, Book1 part 2) "there must be a union of
those who cannot exist without each other; namely, of male and female,
that the race may continue (and this is a union which is formed, not of
deliberate purpose, but because, in common with other animals and with
plants, mankind have a natural desire to leave behind them an image of
themselves)". However, families are also economic units that share con-
sumption, coordinate work activities, accumulate wealth and invest in chil-
dren. Indeed, Aristotle adds that "The family is the association established
by nature for the supply of men’s everyday wants".
Economists’ interest in the family dates back to Cantillon (1730), Smith
(1776) and Malthus (1798). These authors investigated the connections
between economic circumstances and the size of the population. In partic-
ular, they discussed the subsistence wage and family size that can support
a stable work force over time, including the current workers and their de-
scendants that will replace them.1 The main economic decision discussed
in this context was the timing of marriage as a means to control fertility.2
Later writers, including Mill (1848) and Le Play (1855), have shifted atten-
tion to the impact of the family on the standard of living of its members,
via self production, insurance and redistribution of family resources. An
important issue in this context was the allocation of bequests among sib-
lings, which can affect marriage patterns, the incentives of children (and
parents) to work and save and the distribution of wealth in society.3
The role of the family has changed drastically in recent times. In mod-
ern societies, individuals can enter marriage and exit out of it almost at
1 "A man must always live by his work, and his wages must at least be sufficient to
maintain him. They must even upon most occasions be somewhat more; otherwise it
would be impossible for him to bring up a family, and the race of such workmen could
not last beyond the first generation." (Smith, 1776, p. 67).
2 "Universally, the practice of mankind on the subject of marriage has been much
superior to their theories; and however frequent may have been the declamations on the
duty of entering into this state, and the advantage of early unions to prevent vice, each
individual has practically found it necessary to consider of the means of supporting a
family, before he ventured to take so important a step." Malthus (1798, Book 4, Chapter
14).
3 Based on his empirical study of family budgets, Le Play (1872) argued that the
insurance role of the family, is better served by the British system of inheritance that
allows a flexible allocation of bequests than by the French system that imposes equal
division.
4
will, avoiding pregnancy is easy, child mortality is low, and both singles
and married partners can choose whether or not to have children. Due to
technological changes, the importance of the family as productive unit has
declined sharply and it has become much more common for both husband
and wife to work in the labor market. There is higher turnover and some
individuals transit through several marriages, being single during a larger
part of their life time. As marriages break and new marriages are formed,
the traditional division of labor between husbands and wives, especially in
taking care of the children, is put under pressure and transfers between ex-
spouses and custody arrangements are required to maintain the welfare of
children under variable family arrangements. Despite the higher turnover,
and the changing household roles, marriage patterns in modern societies
have some sustainable features, the most notable of which is the high cor-
relation in the schooling attainments of husband and wives.
A unified approach to the family which is applicable to modern societies
was first provided by Gary Becker (1973, 1974, 1991). This approach ties
within family allocations of time and goods to the aggregate patterns of
marriage and divorce. The important insight of Becker’s approach is that
when each man (woman) can choose among several alternative spouses,
competition over spouses matters. Then, the options of each particular
person willing to marry depend on whether individuals of the opposite sex
are willing to marry him\her. Therefore, an equilibrium concept must be
applied such that in existing marriages, no one wants to become single or
wants and can replace the current spouse. This broader perspective can
address the stability of alternative matching profiles in the society at large
and ultimately explain the assortative matching patterns and high marital
turnover that one observes in modern societies. Thus, assortative matching
by schooling in the society can be linked to the fact that within households,
the schooling of husband and wife complement each other.
Our book builds on Becker’s work and the subsequent literature in em-
pirical and theoretical family economics. There are two major strands to
the recent economics literature on the family: what happens inside existing
unions and who marries whom. Although the two strands of the literature
have obvious mutual implications and sometimes meet, they are largely
distinct (as can be seen from the largely disjoint set of contributors to the
two strands). A principal aim of the book is to move forward the merging
of the two strands (as well as providing a state of the art discussion of the
two strands). Accordingly, we divide the book into two parts. The first part
covers the decision making within families; the second part of the book ex-
amines the aggregate patterns in the marriage market and how the actions
of different couples are interrelated.
Given the current, active state of the field, several different modeling
strategies exist. Concerning the behavior of families, we explicitly recog-
nize that spouses in marriage care about each other and their common
children and yet may have conflicting interests. This situation allows for
5
two distinct solution concepts; one is a non-cooperative self-enforcing out-

come, the other is cooperative solution that is efficient and requires binding
commitments, enforced by formal or informal agreements (Lundberg and
Pollak, 1993). We discuss both alternatives, but the main message is that
one can test which approach better fits the available data on consumption
and work behavior of married couples. We show that even if the partners
cooperate and act efficiently, the observed behavior, in terms of the con-
sumption and work choices will generally differ from that of a single decision
maker. The differences arises from the recognition that changes in prices or
incomes that influence the family budget constraint can also influence the
relative "power" of the partners. For instance, transfers of income between
husband and wife (or parents to children) which do not affect total fam-
ily income have no impact on family behavior according to he traditional
"unitary" model of the family but have systematic testable effects under
the "collective" model that we propose.
There are also two approaches to model the competition over spouses
and the division of gains from marriage. One strategy emphasizes frictions
such that one can meet only with few and random members of the oppo-
site sex, before entering marriage (Mortensen, 1988). The other approach
ignores frictions, assuming that it is relatively easy to meet many partners
in a short period of time (Gale and Shapley, 1968, Shapley and Shubik,
1972, Roth and Sotomayor, 1990). In each of these cases, one can further
distinguish between a ‘no transfers’ case in which partners must accept
the characteristics of their spouse, good or bad, and the ‘transfers’ case in
which spouses can effectively compensate (reward) for deficient (attractive)
attributes. We discuss all these cases, showing their different implications
for marriage patterns and for the division of the gains from marriage.
The economic approach to the family can be contrasted with that of bi-
ologists and sociologists. These two fields use different methods which may
yield different testable predictions. For instance, in discussing sex roles
economists often rely on the principle of comparative advantage. Thus, a
mother will spend more time with child than her husband if the ratio of her
productivity at home relative to her market wage exceeds that of father.
The partners can then divide the gains in total output resulting from spe-
cialization. In biology, unequal division of labor is ascribed to the ability of
women to have only few children from different men, while a man can have
many children from different women. Thus, the mother is usually willing
to invest more resources in the child than the father who can potentially
free ride on her desire to invest in the children. Hence, men will compete
for women who will select the most trustworthy man she can (based on
some signals) but the end outcome is that men will spend less effort on
each offspring. (Trivers, 1972). A significant difference between these two
accounts is that the comparative advantage argument rests on transfers of
resources across spouses, that is, exchange that makes both parties better
off (Bergstrom, 1996 and Cox, 2007). As another example, sociologists often
6
motivate assortative matching by inherent preferences to marry someone

similar in terms of predefined attributes such as social status or ethnicity.
In such a case, assortative mating is mainly constrained by groups size, and
minorities are more likely to marry outside the group (Lewis and Oppen-
heimer, 2000). Economists obtain the a similar outcome, but the groups are
formed in equilibrium as a consequence of optimal individual search and
investment decisions (Burdett and Coles, 1999, Chiappori et al., 2009). As
these examples illustrate, economists bring to bear a large degree of free
choice to individual agents subject to resource constraints and some ag-
gregate consistency (an equilibrium) that makes all the individual choices
mutually feasible and sustainable.
The economic approach to the family shares many features of the em-
ployment relationship that is widely discussed by economists. In both cases
the issues of matching and the division of the surplus arise, as well invest-
ment and effort spent in search. However, there are important differences
that originate from the non economic aspect of the marriage relationship.
First, some initial blind trust in the form of love is required to undertake
commitments between the two partners. Based on such commitments, the
partners can coordinate work and investment decision that increase their
gains from marriage and stabilize their marriage, ex post. Second, the pres-
ence of children, who are "public goods" for the parents, strongly influences
entry into marriage and separation decisions. The partners cannot simply
part and go their separate ways because they still care and are legally re-
sponsible for their children. These two differences make the analysis of the
family radically different from the analysis of the employment relationship.
This book is intended for economists. It should be accessible to any
graduate student in economics and has sufficient material for one or two
semesters of lectures on family economics. Although somewhat technical,
we verbalize and illustrate the main ideas so that the book can also be
useful for scholars from other fields who wish to understand the economic
approach, without necessarily agreeing with it. However, the book will be
useful mainly for those interested in modern societies with high marital
turnover. Important problems that face traditional societies are not cov-
ered in this book. We do not discuss intergenerational transfers and dy-
nastic households. Nor do we discuss the important issues related to the
demographic transition from a high to low population growth. For a recent
account see Razin and Sadka (1995), Laitner (1997) and Hotz et al (1997) in
the Handbook of Population Economics, edited by Stark and Rosenzweig.
The first chapter of the book presents some basic facts about the marriage
market and the family. The chapter is intended to motivate the analysis
that follows in the rest of the book by showing how marriage and fertility
interact with economic variables such as work, wages and investment in
schooling. We display data showing that married men work more and have
higher wages than single men, while the opposite patterns hold for women.
We also document the patterns of assortative matching and show how they
7
were affected by the rising investments in schooling and the higher labor
force participation of women. The subsequent chapters are then divided
into two parts; the first part (chapters 2-6) provides a micro level analysis
of family behavior and the second part (chapters 7-11) provides a macro
level analysis of marriage patterns and their welfare implications.
Chapter 2, addresses the question ‘why marry’ and we discuss several
broad sources of potential material gains from marriage, such as sharing
consumption and coordination of work and investment decisions. Chapter
3 provides a basic theoretical framework for the analysis of family behav-
ior. The framework is intentionally broad, including features such as altru-
ism, public and private goods and interaction of several family members
(including children) who may act independently or cooperatively. We com-
pare the traditional "unitary" model that treats the family as if it is a
single decision maker to alternative models that allow family members to
have different views on the decisions that are to be made. We present both
non-cooperative and cooperative variants of these non-unitary models. In
particular, we discuss the "collective approach" which assumes efficiency
and a stable rule for allocating family resources and provides a tractable
way for predicting family behavior and its response to varying economic
conditions (see Chiappori, 1982 and Browning and Chiappori, 1998). Chap-
ter 4 discusses in detail the collective model and its testable implications. A
particular emphasis is given to testing efficiency, an assumption embedded
in all cooperative models of the family. We also discuss the normative im-
plications of the collective assumption which replaces conventional analyzes
of household welfare with an analysis of individual welfare. Chapter 5 dis-
cusses how to empirically recover individual preferences within the house-
hold and the associated decision rules implicit in the collective model. This
chapter also summarizes the main empirical findings. It is shown that the
unitary model is often rejected but efficiency is not rejected. Importantly,
the rule for sharing the marital gains can be identified (up to a constant)
and it is found to respond systematically to marriage market conditions
such as sex ratios and divorce laws.
Chapter 6 extends the static framework and considers family choice over
time and under conditions of uncertainty. We address the new strategic is-
sues that arise in a dynamic setup and the important role of commitments.
Partners anticipate upon marriage that a negative future shock in match
quality may cause separation, which will reduce their benefits from col-
lective goods, including children. Based on this anticipation, they choose
how much to invest in children and how much to consume each period.
To attain efficient investment and consumption outcomes, commitments
made at the time of marriage are usually required. For instance, a binding
contact, enforceable by law, can be signed at the time of marriage which
determines the proportion of family assets that each partner would receive
upon divorce.
Chapter 7 provides an extensive and integrated analysis of matching
8
models. The main question here is ‘who marries whom’. To address this,
we discuss models with and without frictions. Usually, there is less sorting
when there are frictions or when utility is transferable within couples, but
the reasons differ. With frictions, individuals are willing to compromise
rather than wait for a more suitable match. With transferable utility, a
less attractive spouse can bid for a more attractive spouse by giving up
part of his\her share in the gains from marriage. Chapter 8 discusses in
detail how the shares in the marital gains are determined jointly with the
equilibrium matches when frictions are assumed away. The main insight is
that the individual traits of two married partners, such as their schooling
or income, are insufficient to determine the division. Rather, due to com-
petitive forces and the endogeneity of the equilibrium matching, it is the
distribution of traits in the population at large that determines the out-
come. Chapter 9 uses the same friction-less approach to address premarital
investments, such as schooling, whereby individuals can accumulate assets
that will influence their prospects of marriage and their share in gains from
marriage. We emphasize the contrast between inherited traits such as eth-
nicity and acquired traits, such as schooling. Both kinds of traits influence
marriage patterns but acquired traits are also affected by these patterns.
In this case, a rational expectations analysis is required to deal with the
two way feedbacks that arise. We apply such equilibrium analysis to discuss
the interesting reversal in the education attainments of men and women,
whereby women who in the past invested less than men in schooling now
invest more than men do.
Chapters 10 and 11 introduce search frictions to address turnover in the
marriage market, allowing for divorce and remarriage. We examine the wel-
fare implications of turnover for men, women and their children. We also
discuss the role of different laws governing divorce, custody and child sup-
port. These chapters provide a less alarming perspective on divorce than is
adopted by many observers. We recognize that the emotional components
of a match are subject to unanticipated shocks and that divorce and re-
marriage allow the replacement of a bad match by a better one. Moreover,
in a search environment, couples that received negative shocks can more
easily find a new partner when many couples, rather than few choose to
divorce.
Several graduate students at the University of Tel-Aviv assisted us: Linor
Kinkzade and Avi Tillman assisted with processing data for the presenta-
tion of facts in Chapter 1; Uri Tal and Ellana Melnik-Shoef assisted with
programming the numerical examples; Ellana Melnik-Shoef also went over
all the chapters, checking proofs, references and the clarity of exposition;
Evan Finesilver assisted with the creation of the index. Bernard Salanié
provided very useful comments on several chapters. We are very grateful
to all of them.
Finally, several colleagues, including Richard Blundell, Jens Bonke, François
Bourguignon, Olivier Donni, Ivar Ekeland, Bernard Fortin, Mette Gørtz,
9
Murat Iyigun, Guy Lacroix, Valérie Lechene, Arthur Lewbel, Thierry Magnac,
Costas Meghir, Sonia Oreffice and Bernard Salanié, have also contributed
to the development of this book as co-authors in related papers. We thank
them for their valuable contributions.
Becker, Gary, ‘A Theory of Marriage: Part 1’, Journal of Political

Economy, 81 (1973), 813-846.
Becker, Gary, ‘A Theory of Marriage: Part 2’, Journal of Political
Economy, 82 (1974), s11-s26.
Becker, Gary, Treatise on the Family, (Cambridge Mass: Harvard
University Press, 1991).
Bergstrom, Theodore C., ‘Economics in a Family Way’, Journal of Eco-
nomic Literature, 34 (1996), 1903-1934.
Browning, M., and Chiappori, P. A., ‘Efficient Intra-Household Alloca-
tions: A General Characterization and Empirical Tests’, Econometrica,
66 (1998), 1241-1278.
Burdett, Kenneth and Melvyn G. Coles, ‘Long-Term Partnership Forma-
tion: Marriage and Employment’, The Economic Journal, 109 (1999),
307-334.
Cantillon, Richard (1730) Essai sur la Nature du Commerce in
Général (Essay on the Nature of Trade in General), edited and
translated by Henry Higgs, London: Frank Cass and Co.
Chiappori, Pierre-Andre, ‘Collective Labor Supply and Welfare’, Jour-
nal of Political Economy, 100 (1992), 437-67.
Cox, Donald, ‘Biological Basics and the Economics of the Family’, Jour-
nal of Economic Perspectives, 21 (2007), 91-108.
Gale, David and Lloyd Shapley, ‘College Admissions and the Stability of
Marriage’, American Mathematical Monthly, 69 (1962), 9-15.
Hotz, V. Joseph, Klerman, Jacob Alex, and Robert J. Willis, ‘The Eco-
nomics of Fertility in Developed Countries’, in Rosenzweig, Mark R. and
Stark, Oded (eds.), Handbook of Population and Family Economics,
(Elsevier, Vol 1, 1997).
Smith, Adam (1776), The Wealth of Nations, (edited by Edwin Can-
nan, New York: The Modern Library, 1937).
Laitner, John, ‘Intergenerational and Interhousehold Economic Links’, in
Rosenzweig, Mark R. and Stark, Oded (eds.), Handbook of Population
and Family Economics, (Elsevier, Vol 1, 1997).
Le Play, Pierre Guillaume Frédéric (1855), Les Ouvriers Europeens,
reprinted in Catherine Bodard, editor and translator, Frederic Le Play
on Family, Work, and Social Change (Chiago Il: University of Chicago
Press, 1982).
Le Play, Pierre Guillaume Frédéric (1872), La Reforme Sociale, edited
and translated by Catherine Bodard, Frederic Le Play on Family,
Work, and Social Change (Chicago Il: University of Chicago Press,
1982).
10
Lewis, Susan K. and Valerie K. Oppenheimer, ‘Educational Assorta-

tive Mating across Marriage Markets: Non-Hispanic Whites in the United
States’, Demography, 37 (2000), 29-40.
Lundberg, Shelly and Robert A. Pollak, ‘Separate Spheres Bargaining
and the Marriage Market’, Journal of Political Economy, 101 (1993),
988-1010.
Malthus, Thomas Robert (1798), An Essay on the Principle of Pop-
ulation, (Edited by Donald Winch, Cambridge: Cambridge Texts in the
History of Political Thought, 1992).
Mill, John Stuart (1848), Principles of Political Economy with
some of their Applications to Social Philosophy, London: Longmans,
Green and Co., ed. William J. Ashley, 1909, Seventh edition.
Mortensen, Dale T., ‘Matching: Finding a Partner for Life or Otherwise’,
American Journal of Sociology, 94 (1998, supplement), s215-s240.
Roth, Alvin E. and Marilda A. Oliveira Sotomayor, Two Sided Match-
ing: A Study in Game-Theoretic Modeling and Analysis, (New
York: Cambridge University Press, 1990).
Shapley, Lloyd and Martin Shubik, ‘The Assignment Game 1: The Core’,
International Journal of Game Theory, 1 (1972), 111-130.
Razin, Assaf and Sadka, Efraim, Population economics, (MIT Press,
1995).
Trivers, R. L., ‘Parental investment and sexual selection’, In Bernard
Grant Campbell (Ed.), Sexual selection and the descent of man,
(Aldine, 1972), 136-179.
Part I
Models of Household
Behavior
11
12
This is page 13
1
Facts
The purpose of this chapter is to present some basic and general facts
about the marriage and the family. The chapter is intended to motivate
the analysis that follows in the rest of the book by showing how presum-
ably non-economic activities, such as marriage and fertility, interact with
economic considerations such as work, wages and schooling.
1.1 Marriage, divorce and remarriage

1.1.1 Marital status
Marriage is a "natural state". Table 1.1 shows the proportions (the ‘stocks’)
in each marital state for three different years for six high income countries.
These figures show that most of the adult population (aged 20 or older) is
married at any given time (about 55 − 70 percent). However the proportion
of the adult population that is married has declined in most countries in the
last fifty years. This trend is accompanied by a larger proportion of never
married and a higher proportion of divorced individuals, with little change
in the proportion of widowed (because of the offsetting effects of reduced
marriage and mortality). In all countries and at all times, the proportion
of never married women is significantly lower than the proportion of never
married men; this is partially attributable to the fact that men marry
later. Explaining these cross-country regularities and trends is a major
challenge for demographers and economists. Factors that may be related
to the changes in marital status that we see in this table are: changes in the
age structure; delays in marriage; the relative attractiveness of alternative
household arrangements; higher turnover and longer life spans.
There are some notable differences among the countries, reflecting differ-
ent social norms and legal regimes. As an obvious example, the low stock of
divorcees in Italy reflects the fact that divorce was illegal until 1970. As an-
other example, the increasing proportion of ‘never married’ in Denmark can
be attributed to cohabitation, which has become common in Scandinavia.
Strictly, cohabitation should be seen as an alternative state and an extra
column should be included in Table 1.1, but it has been relatively unim-
portant in most countries until recently. To give some idea of the level
and changes in cohabitation we present numbers from the US and Den-
mark in Table 1.2; this gives the proportion of couples (classified by the
head’s age) who live together who are not formally married. As can be seen,
14 1. Facts
there appear to be age, period and cohort effects. That is, cohabitation is
more common amongst the young; at any given age cohabitation is more
common amongst younger cohorts and cohabitation rates are higher now
than twenty years ago. Dramatically, in Denmark, 80 percent of those aged
20−24 who live together choose not to marry. Comparing the two countries
we note that the rate of convergence between US rates and Danish rates,
if any, is very slow.
The propensity to cohabit rather than marry is associated with having
children. In Tables 1.3 and 1.4 we show the proportion of households with
children, conditioned on whether the household head is single, married or
cohabiting for Denmark and the US, respectively. We see that, in each age
group, married couples have more children than cohabiting couples who in
turn have more children than singles. Moreover, the proportion of cohab-
iting couples declines sharply with age. We can thus think of cohabitation
as a "partial marriage" involving less investment in children and a lower
commitment to a long term relationship.
1.1.2 Marital histories

Modern societies are characterized by marriage and divorce at will. Thus,
although marriage appears to be the preferred state, one need not be mar-
ried to the same person and, in fact, there is substantial turnover, especially
among the young. Table 1.5 provides data on marriage histories by age in
2001. Among those who were 50 − 59 years old in 2001, only 6 percent were
never married, but about 31 percent of men and 26 percent of women had
been married more than two times and about 40 percent of each gender di-
vorced at least once. Widowhood at old ages is much more common among
women and they are also more likely to be divorced when old.
A more refined picture of the marital histories is provided in Table 1.6
that records the marital history of the 1931-36 and 1937-41 US birth cohorts
at different years as they age, separately for men and women. It is seen that
the proportions of men and women in their first marriage tend to first rise
and then decline, while the proportions in the second or third marriages and
the proportion divorced rise. Women are more likely to be in a first marriage
when young but less likely to be in a first marriage when old. In addition
women are substantially more likely to be divorced when old, suggesting
that women find it more difficult to maintain their first marriage and to
remarry. For both cohorts we find an increase in the proportion divorced
during the period 1970-1980, suggesting that the "divorce revolution" in
the US that occurred in this period affected all couples and not only the
newly married. However, in the more recent cohort individuals are more
likely to be divorced at any given age.
Table 1.7 provides information form the NLS Panel that includes indi-
viduals who were aged 11 − 21 years old in 1979 and then followed up until
2000. By age 35, most men (81 percent) and women (89 percent) were
1. Facts 15
married at least once. However, the divorce rate has been substantial too
and 35 percent of the women (26 percent of the men) had divorced at least
once. By age 35, most men and women reported that they had finished
their schooling but 21 percent of women and 16 percent of men have done
so after marriage. About 16 percent of the women had a child prior to
marriage.
1.1.3 Flows
The numbers presented so far refer to stocks but we are also interested in
flows into and out of marriage. Figures 1.1 and 1.2 describe the crude mar-
riage and divorce rates for a selection of high income countries. In contrast
to Table 1.1 that provides information on the stocks in different marital
states, these graphs describe the flows into the married and divorced states
in a given year as proportions of the adult population. The picture is quite
clear; starting in 1960 marriage rates have declined and divorce rates have
risen in all the displayed countries. Divorce rates started to rise sharply in
the late 1960’s with a weak tendency for convergence around 3 − 4 percent
per year for some countries; but about 6 percent in the US and about 1
percent in Spain and Italy. The fact that divorced rates went up in many
countries at about the same time suggest a common trigger, such as the
anti-pregnancy pill (see Michael, 1988). Given that about 60 percent of the
population is married in these late years, the implied probability that an
average marriage will break up is roughly 2 percent per year (4 percent in
the US).
Figures 1.3 and 1.4 provide a longer perspective of the marriage and
divorce rates in the US (see also Stevenson and Wolfers (2007)). Figure 1.3
shows marriage and divorce rates per thousand. As can be seen, following a
short episode of increase in the marriage rate after World War 2 (reflecting
delayed marriages and divorces during the war) the marriage rate declined
slightly from 1950 to 2000 with some ups and downs in between. In contrast,
there is an abrupt change in the divorce rate starting at about 1965 with a
doubling in the rate form 1965 to 1975. Although the crude marriage and
divorce rates are informative, much more useful are hazard rates (that is,
the proportion per the relevant groups at risk). Figure 1.4 shows hazards
of marriage, divorce and re-marriage form 1922 to 1988 in the US.1 This
figure too shows the abrupt change in divorce rates after 1965. At about
the same time the remarriage rate increased relative to the marriage rate,
indicating a higher marital turnover. The presence of many divorcees raises
the incentive of any given couple to divorce, because it would be easier to
remarry following separation (see Chiappori and Weiss (2006)).
1 Unfortunately it is not possible to extend the series beyond 1988.

16 1. Facts
1.1.4 Transitions
The most direct information on marital turnover within cohorts is given
by the transition rates across marital states. To show these we use two
different data sources. The first is the HRS which provides us with marital
histories for a cohort born between 1931 and 1941 that reported (retro-
spectively) its marital status history in 2000. The second data source is the
NLS(Y) which provides information on marital status up to age 40 for a
younger cohort born between 1958 and 1965. Figures 1.5 and 1.6 present
the annual transition rate from never married to first marriage of men and
women for the two cohorts respectively. For both cohorts the entry rate
into first marriage first rises and then declines as most individuals who
wish or can marry have already married. The short phase of rising rates of
entry indicates a delay associated with premarital investments and learn-
ing about one’s potential spouse. However, women enter first marriage at
a higher rate than men, suggesting that their gain from early marriage is
higher.
Figure 1.7 presents the rate of dissolution of the first marriage by the
duration of marriage for the same two cohorts. For each cohort we break
up those who are married into those who married before the median age
for that cohort (‘early marriages’) and those who marry latter than the
median age (‘late marriages’). These figures illustrate two important facts.
First, the hazard of divorce is first rising and then declines and, second, the
divorce hazard at any marriage duration is generally lower for later mar-
riages. These two features are the consequence of the interplay between
sorting and acquisition of information of match quality. The hazard of di-
vorce is initially rising with duration of marriage because partners learn
about each other. As new information arrives, some marriages break. How-
ever, with the passage of time, the weak matches are eliminated and the
remaining marriages are increasingly stable. Similarly, the higher stability
of late marriages can be ascribed to longer premarital search and courtship,
which eliminates some of the potentially weak matches (see Becker, Lan-
des and Michael, 1977 and Weiss and Willis, 1993, 1997). Although these
features are common to the two cohorts, there is a very large difference in
divorce rates between the two cohorts. For any duration of the first mar-
riage, the younger cohort reports a divorce rate that is about twice as high.
This reflects the general rise in the divorce rates during the period 1965-
1975. All of the divorces of the younger cohort, born in 1958-65, happened
after the divorce revolution, while most of the divorces of the cohort born
between 1931 and 1941 happened before 1975.
Similarly to the first marriage rate, the remarriage rate of divorced indi-
viduals first rises with the time since divorce (indicating experimentation)
and then declines sharply, because those divorcees who remained unmarried
for a long period are less suitable or less willing to remarry (see Figure 1.8).
The remarriage rate is much higher among the younger cohort, correspond-
1. Facts 17
ing to their higher divorce rate. Thus, later cohorts are characterized by
higher turnover which is reflected in both higher divorce rates and higher
remarriage rates. The remarriage rates of men and women are similar at
the early part of the 1958-1965 birth cohort. For earlier cohorts that are
observed later in life, men remarry at substantial higher rates than women,
especially at high ages. This reflects the fact that the ratio of eligible men
to eligible women decreases because women marry earlier and live longer,
so more of them are either divorced or widowed at late age. The remar-
riage options of men are further enhanced by the fact that the wage gap
between female and male earning capacity is increasing with age because,
on average, males had accumulated more work experience.
Comparing Figure 1.8 with Figures 1.5 and 1.6, we see that for both co-
horts, the remarriage rates of those who remarry quickly exceed the rates
of entry into first marriage. This suggests that some individuals are en-
dowed with marital attributes that make them attractive in any marriage,
whether it be the first or second.
1.1.5 Households
Marriage usually involves at least two people living together in the same
household. This allows the sharing of housing and other consumption goods.
The benefits from such sharing opportunities depend on the household size.
Clearly, living in the same household does not require marriage and more
than one family (or an extended family) can live in the same household.
In Tables 1.8 and 1.9 we present some statistics on the prevalence of one
person households. Table 1.8 shows that the proportion living alone ranges
from 5 percent in Iberia to over 20 percent in Scandinavia. Given that there
are significant material gains from living in many person households, the
high level of people living alone in some countries represents a considerable
potential loss of material well-being. Table 1.9 (for Denmark) shows that
the latter high proportions are not simply a result of older people or younger
people living alone, although the rates are higher for these groups. For
example, the proportion of 40 year old living alone in Denmark is higher
than the overall proportion for France.
Figures 1.9 and 1.10 give statistics on living arrangements over time for
the US. We see that the proportion of households that are ‘married with
children’ has declined from 40 percent in 1970 to 24 percent in 2000 and
the proportion of ‘married without children’ has hardly changed. There
have been sharp increases in the proportion of single person households,
from 17.1 percent to 25.5 percent and in ‘other’ households (whether ‘fam-
ily’ or ‘non-family’), from 12.3 percent to 21.7 percent. Figure 1.10 shows
the corresponding changes in household size. As can be seen, the propor-
tion of large households (5+ members) has halved and the proportion of
single person households has increased by about one half. Taken together,
these figures suggest a substantial reduction in the gains from sharing con-
18 1. Facts
sumption goods within households. One possible reason is that technolog-

ical advances in home appliances allow singles to obtain household goods
more cheaply, even without sharing with others (see Greenwood and Guner,
2005).
1.2 Marriage, work and wages

1.2.1 Time use
Marital status is strongly correlated with the allocation of work time and
the market wages that individuals receive. Thus, compared with singles,
married men work more in the market and have higher wages, while mar-
ried women work less in the market, receiving lower market wages. This
pattern may result from two different effects. First, the division of labor be-
tween the married partners, whereby, on average, wives takes a larger part
of the household chores. Second, selection into marriage, whereby those
willing and able to marry are high wage males with prospective strong
market attachment and low wage females with prospective weak market
attachment.
Time budget data allows a closer look at the relationship between marital
status and the allocation of time. Such data is presented in Tables 1.10 (paid
work and leisure) and 1.11 (for housework and some of its components)
for four countries, where for each country we provide information for two
time periods. A number of robust (if unsurprising) regularities can be seen.
Most importantly, in all countries and for all marital states, men work
more than women in the market and women do more housework than men.
Over time, married women increase their market work (see Table 1.10) and
reduce their non-market work (see Table 1.11), while married men increase
non market work and reduce market work (although Canada provides some
exceptions). However, this trend toward equalization is quite slow and by
2000, the gender gap in household roles remains large. When the children
are less than 5 years old, women work in the market less than half the
time that men do (2.8 versus 6.4 hours per day in the US) and about
twice as much at home (2.7 versus 1.2 hours per day on child care and
2.6 versus. 1.4 hours per day on home production in the US).2 Although
technological advance has substantially reduced the time that women spend
on household chores such as cooking and cleaning (from 3.7 hours a day to
2.6 hours a day in the US) the amount of time spent with children by both
fathers and mothers has risen. Time spent on shopping has not changed
much over time and women continue to spend about twice as much time on
2 Note that child care is underreported since it is a residual category in time use
diaries. Typically respondents record some other activity they are doing even when they
are also looking after their children.
1. Facts 19
shopping than men, irrespective of their marital status. Note that gender
differences in the allocation of time, whereby men work in the market and
less at home, are also present among unmarried men and women, perhaps
reflecting the higher average market wages of men. However, the difference
in the allocation of time of married men and women are more pronounced,
indicating an added role for the division of labor within couples. Another
salient feature of these statistics is that although single men enjoy more
leisure than single women, hours of leisure are about the same for married
men and women, suggesting some coordination of leisure activities (see
Aguiar and Hurst, 2006, and Burda et al, 2006). These averages, however,
mask quite large differences across households; in some households we see
one partner having twice as much leisure as the other (see Browning and
Gørtz, 2006).
Similar patterns are observed in aggregate data. Table 1.12 presents sta-
tistics for ten countries on labor force participation. These statistics show
clearly that historically men have participated more than women but this
gap is narrowing as the participation of women rises (except in Japan) and
the participation of men declines. In Figures 1.11 and 1.12, we report a
more detailed examination of labor force participation for the US. These
figures give the proportion of full time workers by age and marital status
for two birth cohorts, 1945-54 and 1960-69.3 We see a very clear pattern. At
any age, married men are more likely to be fully employed than single men
and married women are less likely to be fully employed than single women.
Married men are substantially more likely to be fully employed than mar-
ried women, suggesting a division of labor between married partners. This
gap in labor market attachment initially rises with age (and time) and then
declines within cohorts; it also declines across cohorts at given ages (com-
pare Figure 1.12 to 1.11). These patterns can be related to the impact of
children on the division of labor. When couples have young children, mar-
ried women are more likely to reduce their labor force participation and,
therefore, the participation gap between men and women is larger. Figures
1.13 and 1.14 compare the work patterns of married and divorced women
and also show a strong impact of having children. Divorced women with
children 0 − 18 work more than married women with children 0 − 18, sug-
gesting that, due to the absence of partner and limited transfers, division of
labor between parents is not feasible and divorced women with children are,
therefore, "pushed" into the labor market. The higher participation rate of
young married women in the younger cohort relative to the older cohort
is associated with lower fertility, a delay in having children and a higher
participation rate for mothers of young children in the younger cohort.
The gap in labor market attachment between married men and women
3 In each subsample, we count the number of fully employed individuals and divide
by the number of all individuals, including those who do not work.

20 1. Facts
may not capture the full extent of the division of labor within couples,
because no control is made for the behavior of the spouse. In Figures 1.15
and 1.16 we display the work patterns of individuals who are married to
each other for two age groups, women aged 40−60 and 30−40 respectively.
As seen, the most common situation before 2000 was that the husband
works full time and the wife works part time or does not work in the
market at all. The differences between the age groups in the earlier years
probably reflect the presence of children in the household. However, with
time, the proportion of such couples has declined and the proportion of
couples in which both partners work full time has risen sharply, reflecting
the increase in the participation of married women into the labor force. On
the other hand, the proportion of couples in which she is full-time and he
is not remains small.
1.2.2 Wages
The gender differences in the employment of married individuals are closely
related to the gender differences in market wages, because a wage gap may
lead to different household choices for the husband and wife, based on
comparative advantage. But, in parallel, differences in past and expected
participation can cause different rates of investment in human capital that
result in lower wages for married women compared with married men (see
Mincer and Polacheck, 1974, and Weiss and Gronau, 1981).
Figures 1.17 and 1.18 display the development over time of weekly wages
(in logs) of US full time workers by marital status for two birth cohorts,
1945-54 and 1960-69. The graphs show that married men have consistently
the highest wage among men while never married women have the high-
est wage among women. In recent cohorts, divorced women are the lowest
paid group, while in earlier cohorts the married women had the lowest
pay. Within each cohort, these differences in log wages by marital status
increase with age (and time), reflecting the cumulative effects of marital
status on the acquisition of labor market experience. In contrast, the differ-
ences in wages by marital status decline with time as we move towards the
more recent birth cohort, holding age constant. This reflects the stronger
attachment of married women to the labor market noted above. As married
women participate more, their wage becomes more similar to that of men
and marital status becomes less important as a determinant of the wages.
1.2.3 The marriage premium

The proportional wage gap between married and single individuals is often
(and somewhat misleadingly) referred to as the "marriage premium" which
is positive for men and negative for women. In Figure 1.19, we compare
married men to divorced and never married men and married women to
divorced and never married women. We make these wage comparisons for
1. Facts 21
individuals who are 30 to 39 old, using three year averages. We see that
the marriage premium of both men and women has risen over time but
the rise is sharper for women. The rise of the marriage premia is consistent
with the notion that when fewer individuals marry, the quality of partners
that do marry relative to those who do not rises. The sharper increase
in the marriage premium for women in Figure 1.19 is a reflection of the
rising participation of married women (see Figure 1.20) which is associated
with higher wages and schooling (see Goldin, 2006). Because we report wage
patterns only for women who work full time, an increase in the participation
of married women can increase the marriage premium if the added workers
are of relatively high ability (see Mulligan and Rubinstein, 2008).
1.3 Who Marries Whom?

Marriages are not formed randomly. Rather individuals sort themselves into
marriage based on the attributes of both partners, because interactions in
individual attributes generate mutual gains from marriage. For instance,
an educated man may benefit more from marrying an educated woman
than a less educated man, who may even resent having a wife who is more
educated than him. Similarly, a marriage in which both partners are similar
in age may create higher gains than a marriage with a large discrepancy
in ages. Consequently, ‘suitable marriages’ are more likely to form and
less likely to dissolve. This means that the observed attributes of married
individuals may be quite different from the attributes of men and women in
general. Additionally, assortative mating arises in which men and women
with similar characteristics, such as age, race and education, marry each
other.
Figure 1.21 records the distribution of age differences among married
couples in the US. In most marriages, the husband is older than the wife
but this proportion of such couples had declined from about 70 percent
during 1968-78 to about 60 percent in 2000-2005. Among couples in which
the wife is of the same age or older than her husband, the sharpest increase
is in the proportion of couples in which the wife is older by 3 or more years
than her husband, which has risen from about 5 percent in 1968-78 to
about 13 percent in 2000-2005. Together, these trends suggest a moderate
but steady reduction of age difference over time. This reduction in age
differences is partially influenced by the changes in the age distributions
of men and women (see Figures 1.22 and 1.23). Over time, the sex ratio
of women to men has increased, especially at older ages because women
live longer. This excess supply of older women raises, to some extent, the
likelihood that men who are 30 − 40 will marry older women, although as
we have seen, an increasing proportion of the older women remain single.
Couples often sort based on schooling; see Lewis and Oppenheimer (2000).
22 1. Facts
This process is driven not only by the mutual gain from marriage, but also
by the availability of partners with different levels of schooling in the pop-
ulation and the chance of meeting them in school or the work place (see
Oppenheimer, 2000). The US (and other countries) has experienced a dra-
matic increase in the stock of educated women relative to educated men
(see Figure 1.24). This change in relative supplies had a marked effect on
the patterns of assortative mating by schooling (see Figure 1.25). While the
proportion of couples in which the husband and wife have the same school-
ing has remained stable at about 50 percent, the past pattern whereby
in 30 percent of the couples the husband is more educated has been re-
placed by the opposite pattern whereby in 30 percent of the couples the
wife has a higher degree. Figures 1.26 and 1.27 show the distribution of the
spouse’s education for husbands and wives with different level of schooling,
by cohort of birth. At lower levels of schooling (up to high school gradu-
ates), each gender mainly marries with individuals of the opposite sex with
similar education. This was not the case for higher levels of education for
earlier cohorts but becomes more common with time as the distributions
of education among women and men become more similar. In particular,
we see a large increase in the marriages in which husband and wife have
some college education. Because the number of women with some college
education has risen sharply relative to men, we see that husbands with
some college have replaced wives with high school by wives with some col-
lege, while wives with some college replaced men with college and higher
degree by men with some college. However, at higher levels of schooling,
BA and more, where women are still relatively scarce we see that men of
high education marry down while women with college education marry up.
We should note that between the two periods, the proportion of couples in
which both spouses are highly educated has risen while the proportion in
which both are less educated declined. In this regard, the rise in education
of men and women combined with assortative matching in schooling has
contributed to the trend of rising inequality between households.
In contrast to other attributes, such as country of origin or race, schooling
is an acquired attribute and investment in schooling is partially motivated
by the prospect of marriage as well as enhanced market power (see Goldin,
Katz, and Kuziemko, 2006 and Chiappori, Iyigun and Weiss, 2006). In Ta-
bles 1.13 and 1.14, we present some evidence on the interaction between
marital status and investment in schooling from the NLS panel. As seen in
Table 1.13, more educated men and women are more likely to be married
and less likely to be separated or divorced at age 35 (after they have com-
pleted most of their schooling). The proportion of unmarried women at age
35 rises with schooling which is not the case for men. Table 1.14 presents
mean cumulated schooling for men and women at marriage and at age 35.
This Table shows, unsurprisingly, that most of the schooling acquired up
to age 35 is taken prior to the first marriage. Those who married and never
divorced acquired about 4 months of additional schooling during marriage
1. Facts 23
out of 13.8 years, while those who married and divorced acquired about 6
months for men and 10 months for women after their first marriage, which
is a relatively large effect given that these are means in which most women
have no extra schooling after marriage.
Having considered schooling, it is natural to consider wages. Figure 1.28
provides a comparison of husband-wife correlations in wages and school-
ing (measured here in years). We examine the correlation in wages in two
ways; wages (in logs) and wage residuals (in logs) netting out observable
differences in schooling and age.4 Thus the correlations in residuals repre-
sent correlations in unobservable factors that affect the wages of the two
spouses.5 The correlation by school years is relatively stable over time, at
about 0.65. The correlations in wage residuals are also stable at a low level
of about 0.1. However the correlations in wages rise from 0.2 to about 0.4.6
The difference between the correlations for schooling and wages is strik-
ing. Some of the difference may be due to spurious factors such as higher
measurement error for wages, the use of wages at the ‘wrong’ point in
the life-cycle, the imputation of wages for non-participants etc.. However
there may also be systematic reasons for the difference. For instance, the
stronger sorting by education may be due to similar educations facilitating
joint consumption and reducing conflicts on the choice of public good. In
contrast, specialization within the household generates a negative corre-
lation between the spouses’ wages. The rise in the correlation for wages
can then be attributed to a reduction in specialization within households
associated with the rise in female labor force participation.
One reason for couples to sort based on schooling is that the schooling
levels of the two spouses complement each other in generating marital sur-
plus. Weiss and Willis (1977) found supporting evidence for this hypothesis
showing that, among couples with the same schooling, divorce declines with
schooling. We should then also expect that, as the proportion of couples
in which both partners are highly educated rises, education will have a
stronger impact in reducing the probability that a given man or woman
will divorce. Figure 1.29 shows that this is indeed the case.
4 Wages were imputed for men and women who did not work at all or worked less
than 10 hours a week.

5 The wage correlation will be higher than for the residuals since the latter removes
the correlation due to age and schooling.

6 We have also calculated the correlations between the percentiles of husbands and
wives in the respective (log) distributions of men and women each year. The correlation
in wage percentiles is slightly higher than the correlation in wages but the trend over
time is very similar. The correlation in residual percentiles is the same as the correlation
in residuals.
24 1. Facts
1.4 Children
Children are the most important ‘products’ of the family. The decision
about how many children to have, when to have them and how to care for
them interacts importantly with a whole host of other decisions including
schooling, marriage, divorce and re-marriage.
1.4.1 Fertility
As we saw, for marriage and divorce there is considerable heterogeneity
across countries and time and this is even more true for fertility. Figure
1.30 presents the time path for completed fertility for cohorts of US women
born between 1903 and 1956.7 The most important feature of this figure is
that there are significant variations across cohorts in the mean number of
children per woman. Thus, women born early in the century had about 2.2
children, those born in the mid-1930’s (the mothers of the ‘baby-boom’)
had over three children and those born in the fifties had close to two. Table
1.15 shows the change in the distribution of children born for women born
in the mid-1930’s and in the late 1950’s. As can be seen the change in the
mean is partly a result of fewer women born in the 1930’s being childless
and partly a result of these women having larger families, conditional on
having a child at all. Particularly striking is that the modal family size
for the older cohort is 4+ but only 2 for the younger cohort. Figure 1.31
shows data on the number of children less than 18 of US women (married
or single), aged 35−45, at different periods of time.8 As seen, the reduction
in fertility and marriage rates during the second half of the 20’th century
is associated with a decrease in the proportion of women with more than 3
children and an increase in the proportion of women with no children, while
the proportion of women with 1 or 2 children remained unchanged at about
half. By 2000-2005, the proportion of women with children is still high (67
percent) indicating that the natural desire to have children remains strong.
Figure 1.32 shows that the birth rate fluctuates dramatically over time.
We see a large increase from the mid-1930’s to the early 1960’s and then
7 Completed fertility is defined as the mean number of children born to women of

a given generation at the end of their childbearing years. This is calculated by adding
the fertility rates by age of the mother observed for successive years, when the cohort
has reached the age in question (in general, only ages between 15 and 49 years are
considered). In practice, the fertility rates for older women can be estimated using the
rates observed for previous generations, without waiting for the cohort to reach the end
of the reproductive period.
8 Table 1.15 and Figure 20 provide different but complementary information. The table
shows completed fertility whereas the figure shows the number of children less than 18
living with the mother. Therefore, the proportion of women who have no children living
with them in the figure is larger than the proportion of women who never had children
in the table.
1. Facts 25
a sharp decrease. This is consistent with Figure 1.30 which shows a peak
in fertility for mothers born in the mid-1930’s; this is the baby boom gen-
eration. The median age at first marriage has also increased at the same
period suggesting fewer "forced marriages" (see Michael, 1988 and Goldin
and Katz, 2002).
Figure 1.33 presents evidence on completed fertility for a cross-section
of six western European countries for women born between 1931 and 1967.
In common with the USA, all of theses countries display a falling pattern
from the mid-1930’s, although the US has a much higher value in the early
years (3.1 as compared to 2.65 for the highest European values). Thus all
these countries indicate a ‘baby-bust’ even though the trends show signif-
icant differences across countries. For example, Italy has the lowest values
throughout this period with a steady decline from 2.3 to 1.5 children per
women. In contrast, the Netherlands starts off with a high value of 2.6 and
falls quickly by about 0.7 children in 1946 and then falls much more slowly
over the next twenty years by about 0.2 children. Most dramatic is the case
of Spain which has the highest value in the early 1940’s (at 2.6 children per
woman) and one of the lowest 25 years latest (at 1.6).
The timing of children is also of interest. In Figures 1.34 and 1.35 we
show the timing of first marriage and first birth for the same countries
as in Figure 1.33. There is a clear relationship between reduced fertility
and the delay in marriage. On the average, age of first child is only two
years after year of marriage (28 and 26, respectively for the latest cohort
born in 1963). In these figures, marriage does not include cohabitation. In
most countries the latter is low for women born before 1960 but for some
countries there is considerable cohabitation. For example, the dramatic rise
in the Danish age at first marriage largely reflects the fact that marriage
before the birth of a child is increasingly rare amongst younger cohorts.
1.4.2 Children under different household arrangements

One consequence of the increasing marital turnover is the sharp rise in the
number of children who live in single parent and step parent households
(see Table 1.16). In the US, 2005, 68 percent of children less than 18 years
old lived with two parents (including step parents), 23 percent lived only
with their mother and 5 percent lived only with their father whilst the rest
lived in households with neither parent present. The impact of living with
single parents on the children depends on the amount of transfers between
unmarried parents. Generally, such transfers are small with a substantial
proportion of eligible mothers receiving no transfer at all. Only about half
of eligible women receive any child support and when a transfer is received
it is about 20 percent of the mother’s income (see Table 1.17 and Figure
1.36). The consequence is that divorced mothers have less than half of the
family income of married mothers and, therefore, children living with single
mothers are often in poverty. The impact of marital turnover on children is
26 1. Facts
a major policy concern and much research has been directed to the analysis
and measurement of this effect (see Weiss and Willis, 1985, Chiappori and
Weiss, 2006, Piketty, 2003, Gruber, 2004, and Bjorklund and Sundstrom,
2006).
1.5 Saving and life stages

Progression through life-stages has a major impact on consumption, saving
and wealth. In the savings literature the traditional picture of the life-cycle
is very circumscribed. Agents are born, they receive education, they work
and then retire and finally they die. Within such an environment the natural
emphasis is on financing schooling decisions, smoothing consumption in the
presence of income fluctuations and saving for retirement and bequests.
When we take account of leaving home, marrying, having children, divorce
and remarriage, a much more nuanced pattern emerges.
The empirical evidence suggests that savings rates vary substantially
across family types. The evidence for the US presented in Avery and Ken-
nickel (1991), Bosworth et al (1991) and Lupton and Smith (2006) sug-
gests that couples without children have the highest savings rate and lone
parents have the lowest rate. Avery and Kennickel (1991) show that mar-
ried couples have the highest wealth and the highest savings rate whereas
divorced people dissave from substantial wealth holdings. Bosworth et al
(1991) investigate more closely the variations with children and show that
households with children present save less than those without. The latter
group is largely split between younger couples, many of whom will have
children later and those who have children who have left home. Lupton
and Smith (2000) use three waves of the PSID and concentrate on changes
in savings rates consequent on transitions between marital states. Finally,
Zagorsky (2005) presents evidence based on the NLSY79 that suggest that
the wealth of divorcees is much lower than the wealth of continuously mar-
ried individuals and those who never married. Overall, the main finding is
that transitions into being married raises savings rates9 and transitions out
of being married lowers them. Although all these studies present a consis-
tent picture, much still remains to be found out about saving and marital
status.
Measuring (or even defining) wealth and/or savings in surveys is fraught
with difficulties. Using consumption we can illustrate some of the patterns
associated with children more clearly. To do this we shall break the evo-
lution of married life into four life stages: being a couple before having
children; having young children in the household; having only older chil-
9 Blow, Browning and Ejrnæs (2009) find similar results for the transition into mar-
riage using UK expenditure data.

1. Facts 27
dren in the household and living together after the children have left; see
Apps and Rees (2009), chapter 5 for a similar analysis using Australian
data. Unfortunately in cross-sections we do not observe whether younger
households that do not currently have children will have them in the future.
On the other side, for older households with no children present, we do not
observe whether they have had children. Instead we take the earliest life
stage to be being a couple with no children and the wife aged less than 41
and the fourth life stage to be having no children with the wife aged over
40.10 Table 1.18 presents some facts on income, nondurable expenditures
and budget shares for some goods. The data are drawn from the Canadian
Family Expenditure Surveys (FAMEX) for 1986, 1990 and 1992. 11 We se-
lect out households in which the husband reports less than 35 hours of
full-time work in the year to take account of long spells of unemployment
and retirement. There is no selection on the wife’s labor force participation.
The top panel of Table 1.18 gives details of income and nondurable expen-
diture. Through the four life stages, expenditure is highest when there are
older children present and drops significantly when they leave home. This is
partially reflected in the evolution of income but changes in income are not
the sole driving force, as can be seen from the expenditure/income ratio.
To show this more clearly, Table 1.19 presents the results from regressing
log nondurable expenditure on log income and dummies for the last three
life-stages. As can be seen, even when we control for income the life-stage
has a large and highly significant impact on nondurable expenditures. The
bottom panel of Table 11.18 shows how patterns of demand, conditional on
total expenditure, evolve through life stages. In the earliest period budget
shares for restaurants and alcohol and tobacco are high. These fall on the
arrival of the first child and budget shares for food at home rise. As chil-
dren age, more is spent (relatively) on clothing. Interestingly, although the
post-children life stage patterns show some reversion to the pre-children
patterns the two are not the same, even though net income is similar.
The impact of children on consumption emerges even more clearly if we
follow quasi-panels through time. To do this we use UK Family Expendi-
ture Surveys from 1968 to 1995.12 We consider only married or cohabiting
couples. To construct quasi-panel data we first construct cohorts accord-
ing to the wife’s age and her level of education (‘minimum’ or ‘more than
minimum’). We then take cell means for each cohort and year. That is, we
1 0 Browning and Ejrnæs (2009) present a quasi-panel analysis on UK data that takes
into account that some younger ‘childless’ households will never have children and some
older ‘childless’ households have never had them.
1 1 We use the FAMEX since it is the only large expenditure survey that collects infor-
mation on annual expenditures. Most budget surveys employ a two week diary which
induces problems with infrequency.
1 2 We use the UK data since it gives a very long time series of cross-sections with
consistent coding throughout the period.

28 1. Facts
have means for, say, high educated households aged 37 in 1981 and those
aged 38 in 1982. This allows us to follow quasi-individuals through time.
We consider cohort/year means of log nondurable consumption and equiv-
alent household size. To construct the latter we first assign each member a
consumption weight according to their age; we take values of 0.1, 0.15, 0.25,
0.35 and 0.65 for children aged 0−2, 3−4, 5−10, 11−16 and 17−18 respec-
tively. Each adult is given a weight of unity. We then sum these weights for
each household and raise this to the power 0.7 to capture scale effects.13 In
Figures 1.37 and 1.38 we show the smoothed paths of cohort means of log
nondurable consumption and equivalent household size against the wife’s
age. As can be seen, the patterns of consumption and family size coincide
very closely. The variation over the life-cycle is substantial and much larger
than variation induced by fluctuations in income or employment.
1.6 References
[1] Aguiar, Mark and Erik Hurst, ‘Measuring Trends in Leisure: The Allo-
cation of Time over Five Decades’, NBER Working Paper No. W12082,
(2006).
[2] Apps, Patricia and Ray Rees, Public Economics and the House-
hold, (Cambridge, Cambridge University Press, 2009).
[3] Avery, Robert B. and Arthur B. Kennickel, ‘Household Saving in the
US’, Review of Income and Wealth, 37 (1991), 409-432.
[4] Bailey, Martha, ‘More Power to the Pill: The Impact of Contraceptive
Freedom on Women’s Life Cycle Labor Supply’, Quarterly Journal
of Economics, 121 (2006), 289-320.
[5] Becker, Gary, Landes, Elisabeth M. and Robert T. Michael,
‘Economic-Analysis of Marital Instability’, Journal of Political
Economy, 85 (1977), 1141-1187.
[6] Bjorklund, Anders and Marianne Sundstrom, ‘Parental Separation and
Children’s Educational Attainment: A Siblings Analysis on Swedish
Register Data’, Economica, 73 (2006), 605-624.
[7] Bosworth, Barry, Burtless, Gary and John Sabelhaus, ‘The Decline in
Saving: Evidence from Household Surveys’, in Brainard, William C.,
(eds.) and Perry, George L., (eds.), Brookings Papers on Economic
Activity: Macroeconomics, (Brookings Institution Press, 1991).
[8] Browning, Martin and Mette Ejrnæs, ‘Consumption and children’,
The Review of Economics and Statistics, 91 (2009), 93-111.
1 3 This scheme follows the suggestion in Browning and Ejrnæs (2009). Adopting dif-
ferent (plausible) weights or scale factors gives similar results.

1. Facts 29
[9] Browning, Martin and Mette Gørtz ,‘Spending Time and Money
within the Household’, University of Oxford, Working Paper No. 288,
(2006).
[10] Burda, Michael C., Hamermesh, Daniel S. and Philippe Weil, ‘The
Distribution of Total Work in the EU and US’, IZA Discussion Paper
No. 2270, (2006).
[11] Chiappori, Pierre-Andre, Iyigun, Murat, and Yoram Weiss, ‘Invest-
ment in Schooling and the Marriage Market’, American Economic
Review, 99 (2009), 1689-1713.
[12] Chiappori, Pierre-Andre and Yoram Weiss, ‘Divorce, Remarriage, and
Welfare: A General Equilibrium Approach”, Journal of the Euro-
pean Economic Association, 4 (2006), 415-426.
[13] Goldin, Claudia, ‘The Quiet Revolution that Transformed Women’s
Employment, Education, and Family’, American Economic Re-
view, 90 (2006), 1-21.
[14] Goldin, Claudia and Lawrence F. Katz, ‘The Power of the Pill: Oral
Contraceptives and Women’s Career and Marriage Decisions’, Jour-
nal of Political Economy, 110 (2002), 730-770.
[15] Goldin, Claudia, Katz, Lawrence F., and Ilyana Kuziemko, ‘The
Homecoming of American College Women: The Reversal of the Col-
lege Gender Gap’, Journal of Economic Perspectives, 20 (2006),
133-156.
[16] Greenwood, J. and N. Guner, ‘Marriage and divorce since World War
II: Analyzing the Role of Technological Progress on the Formation of
Household’, University of Rochester, research report 8, (2005).
[17] Gruber, Jonathan, ‘Is Making Divorce Easier Bad for Children? The
Long-Run Implications of Unilateral Divorce’, Journal of Labor
Economics, 22 (2004), 799-833.
[18] Lewis, Susan K. and Valerie K. Oppenheimer, ‘Educational Assor-
tative Mating across Marriage Markets: Non-Hispanic Whites in the
United States’, Demography, 37 (2000), 29-40.
[19] Lupton, Joseph P. and James P. Smith, ‘Marriage, Assets and Savings’,
in Grossbard-Shechtman, Shoshana A., (eds.), Marriage and the
Economy, (Cambridge University Press, 2003).
[20] Mazzocco, Maurizio and Shintaro Yamaguchi, ‘Labor Supply, Wealth
Dynamics and Marriage Decisions’, University of California Los An-
geles, California Center for Population, Research Working Paper No.
CCPR-065-06, (2006).
30 1. Facts
[21] Michael, Robert, ‘Why did the U.S. Divorce Rate Double within a
Decade?’, Research in Population Economics, 6 (1988), 367-399.
[22] Mincer, Jacob and Solomon Polachek, ‘Family Investments in Human
Capital: Earnings of Women’, Journal of Political Economy, 82
(1974), S76-S108.
[23] Mulligan, Casey B. and Yona Rubinstein, ‘Selection, Investment, and
Women Relative Wages since 1975’, Quarterly Journal of Eco-
nomics, 123 (2008), 1061-1110.
[24] Piketty, Thomas, ‘The Impact of Divorce on School Performance: Ev-
idence from France, 1968-2002’, Centre of Economic Policy Research,
Discussion Paper No. 4146, (2003).
[25] Stevenson, Betsey and Justin Wolfers, ‘Marriage and Divorce: Changes
and their Driving Forces’, Journal of Economic Perspectives, 21
(2007), 27-52.
[26] Weiss, Yoram and Reuben Gronau, ‘Expected Interruptions in Labor
Force Participation and Sex-Related Differences in Earnings Growth’,
Review of Economic Studies, 48 (1981), 607-619.
[27] Weiss, Yoram and Robert J. Willis, ‘Children as Collective Goods and
Divorce Settlements”, Journal of Labor Economics, 3 (1985), 268-
292.
[28] Weiss, Yoram and Robert J. Willis, ‘Transfers among Divorced Cou-
ples: Evidence and Interpretation’, Journal of Labor Economics,
11 (1993), 629-679.
[29] Weiss, Yoram and Robert J. Willis, ‘Match Quality, New Information,
and Marital Dissolution’, Journal of Labor Economics, 15 (1997),
S293-S329.
[30] Zagorsky, Jay L., ‘Marriage and Divorce’s Impact on Wealth’, Journal
of Sociology, 41 (2005), 406-424.
1. Facts 31
1.7 Tables and Figures
Year Never married Married Divorced Widowed

Women Men Women Men Women Men Women Men
Canada
1951 17.6 24.0 71.4 71.5 0.4 0.3 10.6 4.3
1981 15.3 21.5 69.6 73.5 3.5 2.5 11.3 2.5
2003 19.1 26.0 63.5 66.1 7.1 5.4 10.3 2.5
Denmark
1950 18.2 21.8 67.5 70.5 3.9 2.7 10.7 5.1
1980 16.8 24.5 62.1 65.5 7.2 5.8 14.0 4.2
2002 24.8 33.5 51.8 54.2 12.8 3.7 10.5 8.6
France
1954 16.7 21.6 62.6 71.7 2.2 1.5 18.6 5.3
1980 16.5 23.0 63.4 70.4 4.3 3.1 15.8 3.5
2001 24.8 31.9 53.2 59.0 7.8 6.2 14.2 3.0
Italy
1950 24.1 28.5 61.3 66.5 0.3 0.3 14.3 4.8
1980 16.7 22.2 66.3 73.1 1.5 1.3 15.6 3.5
2001 22.0 29.8 60.7 65.8 1.7 1.3 15.7 3.1
United Kingdom
1950 19.8 20.7 65.5 73.7 0.7 0.5 13.9 5.1
1980 14.0 19.8 66.2 72.9 4.2 3.3 15.6 4.0
2002 22.9 30.1 54.0 58.1 9.8 7.9 13.2 3.9
USA
1950 11.8 17.0 72.3 76.1 2.7 2.2 13.3 4.7
1980 14.1 19.7 64.1 71.4 7.9 6.0 13.9 2.9
2002 18.5 25.0 56.3 61.5 14.3 10.8 10.9 2.7
Source: Census of different countries
TABLE 1.1. Marital Status of Men and Women, over 20 Years Old, in different
Countries and Years
32 1. Facts
Age group USA Denmark

1980 1990 2000 1980 1990 2000
20 − 24 11.5 25.6 36.3 59.2 76.1 80.8
25 − 29 7.0 12.4 20.2 25.4 41.7 53.1
30 − 34 3.8 7.2 11.7 10.5 20.8 20.4
35 − 39 2.0 5.1 7.0 6.2 9.7 11.5
40 − 44 1.6 3.4 5.3 4.4 6.6 9.0
45 − 49 1.3 2.4 5.1 3.9 5.9 9.0
50 − 54 1.2 2.2 4.1 3.9 5.6 7.2
55 − 59 1.2 1.6 5.0 4.1 5.4 8.2
60 − 64 1.4 1.8 3.1 4.4 5.3 6.1
Source: US Census and Statistics Denmark
TABLE 1.2. Cohabitation in the US and Denmark by Age of the Household Head
Single head Married couples Cohabiting couples

Age % of % of single % of % of married % of % of cohab.
all HH HH with child. all HH with child. all HH with child.
20 − 24 69.1 2.1 5.3 51.7 25.6 14.2
25 − 29 42.0 8.3 22.4 72.8 35.7 33.4
30 − 34 28.9 23.4 46.2 88.0 25.0 61.5
35 − 39 25.7 37.4 56.1 91.7 18.3 73.6
40 − 44 25.4 37.4 61.1 82.7 13.4 66.3
45 − 49 25.4 20.0 64.5 51.4 10.2 43.0
50 − 54 23.6 6.9 69.0 18.7 7.5 20.0
55 − 59 24.3 1.8 69.8 5.5 6.0 7.9
60 − 64 26.4 0.3 68.7 1.5 4.9 2.5
Source: Statistics Denmark.
TABLE 1.3. Household Arrangements, Denmark 2000
1. Facts 33
Single head Married couples Cohabiting couples

Age % of % of single % of % of married % of % of cohab
all HH HH with child. all HH with child. all HH with child.
20 − 24 57.4 15.4 27.9 58.3 14.7 36.5
25 − 29 42.1 21.5 47.7 65.4 10.2 39.1
30 − 34 34.4 29.5 59.1 77.9 6.5 47.1
35 − 39 33.6 36.2 61.7 83.9 4.7 47.1
40 − 44 33.9 33.3 62.3 77.2 3.8 37.6
45 − 49 34.8 22.6 62.4 56.3 2.8 20.8
50 − 54 35.5 9.5 62.0 28.6 2.5 12.8
55 − 59 36.1 3.4 62.0 11.0 1.9 5.2
60 − 64 38.3 1.4 60.2 4.4 1.5 1.6
Source: Current Population Surveys
TABLE 1.4. Household Arrangements, USA, 2000-2005
Sex,Age NumberofMarriages Divorced Widowed

0 1 2 3+ 1+ Now Ever Now Ever
Women
30 − 34 21.7 67.3 10.0 1.0 78.3 9.3 18.6 0.4 0.6
35 − 39 15.6 66.8 15.7 1.8 84.4 13.7 28.1 0.6 1.1
40 − 49 10.5 65.1 19.8 3.3 89.5 16.8 35.4 2.4 3.5
50 − 59 6.4 65.2 22.1 4.1 93.6 17.9 38.9 7.1 9.5
60 − 69 4.1 72.9 17.4 3.1 95.9 12.6 28.4 19.7 23.3
Men
30 − 34 29.5 60.8 8.7 1.1 70.5 7.0 15.4 − 0.3
35 − 39 21.5 66.2 10.9 1.4 78.5 12.5 22.9 0.2 0.5
40 − 49 14.2 65.1 17.1 3.6 85.8 12.5 29.5 12.5 1.3
50 − 59 6.3 62.6 23.2 8.0 93.7 16.9 40.8 1.8 2.9
60 − 69 4.3 67.5 21.3 6.8 95.7 9.7 30.9 4.5 7.6
Source: U.S. Census Bureau, Survey of Income and Program
Participation (SIPP), 2001 Panel, Wave 2 Topical Module
TABLE 1.5. Marital History by Age and Sex, US, 2001 (percents)
34 1. Facts
Birth cohort 1931-36

1960 1965 1970 1975 1980 1985 1990
Married, men 60.84 74.25 74.56 70.25 65.06 61.39 57.45
first time women 69.45 69.00 65.28 58.92 52.51 45.36 38.75
Married, second men 2.08 5.57 8.26 12.16 15.15 18.88 20.21
time or more women 4.95 8.79 11.28 12.95 13.87 13.84 11.5
Divorced, men 3.05 3.73 5.02 8.67 8.31 7.76 8.97
first time women 4.45 6.37 7.86 13.25 13.50 15.64 17.24
Never married men 26.90 14.64 10.74 9.01 8.30 7.90 7.52
women 20.10 13.30 10.64 9.29 6.78 8.41 8.03
Birth cohort 1937-41
Married, men 19.59 60.57 71.33 67.38 62.49 58.52 55.09
first time women 47.13 66.95 66.05 58.52 50.61 44.05 36.72
Married, second men 0.47 2.60 7.47 12.27 16.92 19.24 21.49
time or more Women 2.01 6.31 10.17 12.89 14.96 16.33 16.82
Divorced, men 1.04 4.20 5.40 6.90 9.86 10.52 9.78
first time women 2.65 5.59 8.88 11.03 16.68 18.48 21.40
Never married men 63.90 26.43 13.57 9.88 8.32 7.66 7.24
women 47.70 20.22 12.54 10.24 9.23 8.81 8.59
Source: Health and Retirement Survey, 1992
TABLE 1.6. Marital History, US, of the 1931-36 and 1937-41 Birth Cohorts
Marital and educational Males Females

Status N=2085 N=2202
Not married no child at age 35 0.15 0.08

Married before age 36 0.81 0.89
Had child before age 36 0.67 0.78
Divorced before age 36 0.26 0.35
Finished school before age 36 0.90 0.90
Had child before first marriage 0.10 0.16
Married before finishing school 0.16 0.21
Source: National Longitudinal Survey, Youth, 1979
TABLE 1.7. Marital History of the NLS Panel
1. Facts 35
%
Belgium 9
Denmark 21.9
Germany 17
Greece 9
Spain 5
France 13
Ireland 8
Italy 10
Netherlands 14
Austria 14
Portugal 5
Finland (2000) 23
UK 13
USA 13.7
Source: Census of different countries
TABLE 1.8. Individuals Living Alone, various countries
Age group Percentage living alone

18-30 19.8
31-40 13.2
41-50 13.4
51-60 17.7
61-70 25.9
71+ 51.9
Source: Statistics Denmark
TABLE 1.9. Individuals Living Alone, Denmark
36 1. Facts
USA Canada UK Norway

Year of survey 1975 2003 1971 1998 1975 2000 1971 2000
Paid work
Single men 5.55 5.39 5.31 5.30 6.48 4.80 6.01 5.18
Single women 4.39 4.71 4.84 4.27 4.11 3.49 4.17 3.59
Married men, no children 6.13 6.32 6.37 6.39 6.64 5.91 6.20 5.89
Married women, no child 3.42 4.51 3.38 4.39 3.38 3.99 3.12 4.60
Married men, child 5-17 7.17 6.40 6.16 6.80 6.70 6.17 6.06 5.78
Married women, child 5-17 2.71 3.68 1.97 4.08 2.46 3.54 1.86 4.28
Married men, child < 5 6.98 6.39 6.13 6.21 6.59 6.16 6.58 5.73
Married women, child < 5 1.55 2.81 1.11 2.64 0.82 2.45 0.91 2.58
Leisure
Single men 6.94 6.82 7.20 7.29 6.49 7.22 5.91 7.04
Single women 6.23 6.04 5.86 6.43 6.05 6.44 5.13 6.86
Source: Multinational Time Use Study
TABLE 1.10. Hours of Work and Leisure per Day
1. Facts 37
USA Canada UK Norway

Year of survey 1975 2003 1971 1998 1975 2000 1971 2000
Home production
Single men 1.05 1.27 1.19 1.14 0.61 1.28 1.12 1.19
Single women 2.06 1.72 1.84 2.03 2.47 2.34 2.74 1.77
Child care
Single men 0.03 0.08 0.03 0.04 0.02 0.03 0.02 0.04
Single women 0.36 0.48 0.15 0.43 0.23 0.47 0.19 0.33
Married men, no children ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗
Married women, no child ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗
Shopping
Single men 0.24 0.35 0.31 0.41 0.24 0.33 0.21 0.28
Single women 0.49 0.49 0.23 0.57 0.53 0.55 0.29 0.46
Source: Multinational Time Use Study
TABLE 1.11. Hours per Day of Home Production, Childcare and Shopping
38 1. Facts
Country Male Participation Rates Female Participation Rates

1965 1975 1985 1995 2005 1965 1975 1985 1995 2005
US 80.7 77.9 76.3 75.0 73.3 39.3 46.3 54.5 58.9 59.3
Canada 79.9 78.4 77.4 72.7 72.7 33.8 44.4 54.9 57.3 61.4
Australia 85.1 82.2 76.7 74.6 73.0 34.8 44.5 47.1 74.7 58.1
Japan 81.1 81.2 77.9 77.5 73.1 48.8 44.8 47.6 49.3 47.7
France 79.2 74.4 68.4 63.4 63.3∗ 38.2 41.7 46.4 48.2 51.1∗
Germany 80.9 73.4 70.1 68.1 63.9∗ 40.0 39.3 41.1 47.1 49.6∗
Italy 77.5 70.6 65.3 61.6 61.1∗ 27.8 26.8 30.7 34.4 38.2∗
Nether. NA 80.0 73.8 69.8 72.7 NA 29.5 37.9 48.1 57.8
Sweden 82.2 77.0 72.5 68.9 67.8∗ 46.6 55.2 61.5 59.5 59.7∗
UK 85.4 81.2 76.1 72.0 70.5 41.7 46.6 50.7 53.5 56.2
Source: Comparative Civilian Labor Force Statistics, 10 Countries, 1960-2005,
US Department of Labor, 2006.
Note: ∗ Observation from 2004.
TABLE 1.12. Labor Force Participation of Women and Men in Ten Countries
Marital Status School years

Women Men
< 12 12 13 − 15 16+ < 12 12 13 − 15 16+
Unmarried 9.7 9.2 10.6 14.3 19.7 18.8 18.9 18.6
Married 28.3 50.8 50.2 61.9 34.9 50.8 52.7 64.0
Separated 31.4 19.1 22.0 11.4 26.1 15.2 16.9 7.8
Divorced 30.6 20.9 17.2 12.7 19.4 17.2 11.6 9.6
Source: National Longitudinal Study, Youth, 1979.
TABLE 1.13. Marital Status at Age 35, by Gender and Education at Age 35
.
1. Facts 39
Marital Status at age 35

Women Men
Yearsofeducation Years of education
% at age35 at marr. % at age 35 at marr.
Never married 8.5 13.9 − 15.7 13.6 −
Married,never divorced 54.0 13.9 13.5 56.0 13.8 13.5
Married 37.6 12.7 11.9 28.3 12.3 11.9
with prior separation
Source: National Longitudinal Study, Youth, 1979.
TABLE 1.14. Years of Schooling at Marriage and at Age 35, by Gender and
Marital Status at Age 35
Number of Children
0 1 2 3 4+
Born 1932-1936 10.2 9.6 21.7 22.7 35.8
Born 1956-1960 19.0 16.4 35.0 19.1 10.5
Source: US Census
TABLE 1.15. Completed Fertility for Two US Cohorts
Children with Children Children

Year two parents with mother with father
1950 93 5.9 1.3
1960 91 7.4 1.2
1970 87 10.7 1.5
1980 81 16.8 2.7
1990 76 19.5 4.4
2000 72 21.9 6.3
2005 68 23.4 4.7
Source: US Census (percentage)
TABLE 1.16. Living Arrangements of U.S. Children, Aged less than 18, by Year
40 1. Facts
Mother’s Age 20-30 31-40 41-60

1979-92 Div. Mar. Div. Mar. Div. Mar.
Prop. with CS>0 0.452 0.062 0.509 0.068 0.428 0.037

CS, if CS>0 1905 1320 2947 1797 3660 1859
Mother’s Inc. if CS>0 10728 7190 15230 11444 17138 12858
Mother’s Inc. 8834 5218 12952 7618 13398 7725
Family Inc. if CS>0 11210 25868 16085 36945 20499 44399
Family Inc. 9918 23867 14045 33247 17894 38095
Observations 8071 74900 14410 107108 7536 57936
1993-2004 Div. Mar. Div. Mar. Div. Mar.
Prop. with CS>0 0.463 0.049 0.502 0.054 0.454 0.034

CS, if CS>0 1920 1664 2959 2368 4023 2775
Mother’s Inc. if CS>0 11351 9195 16873 13854 21958 17564
Mother’s Inc. 9699 7086 14544 10989 17779 13550
Family Inc. if CS>0 11731 27313 17644 39240 23929 49227
Family Inc. 10825 26298 15720 39599 21675 49148
Observations 4171 40686 12312 88472 10427 63332
Source: Current Population Surveys.
TABLE 1.17. Child Support and Alimony Received by Mothers with Children
0-18 (in 1982-84 dollars) by Mother’s Age and Time Period
No children Children, at Children No children

wife ≤ 40 least one ≤ 6 all aged > 6 wife > 40
Sample size 1, 255 2, 367 1, 965 1, 217
Net income 50, 060 48, 425 52, 889 50, 045
Nondur expend 23, 484 25, 768 27, 947 21, 560
Ex/inc ratio 0.50 0.56 0.56 0.47
Selected budget shares (%)
Food at home 16.1 22.1 23.1 21.7
Restaurants 10.9 6.6 7.0 8.1
Clothing 13.7 11.6 13.6 10.9
Alc and tob 9.2 6.1 6.0 7.7
Recreation 13.6 11.3 12.9 11.1
Source: Canadian Family Expenditure Surveys.
All monetary values in 1992 Canadian dollars
TABLE 1.18. Consumption Through Life Stages
1. Facts 41
Variable Constant Log income Stage 2 Stage 3 Stage 4

Coefficient 3.38 0.62 0.12 0.14 −0.08
t-value 78 13 16 −8
R2 = 0.51
Source: Canadian Family Expenditure Surveys.
TABLE 1.19. Descriptive Regression for Log Nondurable Consumption
42 1. Facts
18
16
14
12
rate per 1000, 20+
10
6
4
2
0
19
19
19
19
19
19
60
67
74
81
88
95
D enm ark S pain Franc e Italy Netherlands UK US A
FIGURE 1.1. Marriage Rates for Selected Countries. Source: Eurostat.

1. Facts 43
6
rate per 1000, 20+
0
19
19
19
19
19
19
60
67
74
81
88
95
D enm ark S pain Franc e Italy Netherlands UK US A
FIGURE 1.2. Divorce Rates for Selected Countries. Source: Eurostat.

44 1. Facts
18 6
16
5
Marriage rate per 1000 population
14
12 4
10
3
8
6 2
4
1
2
0 0
1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000
marriage rate divorce rate
FIGURE 1.3. Marriage and Divorce Rates, US 1940-2002. Source: National Cen-
ter of Health Statistics.
1. Facts 45
200 Fir s t Ma r r ia g e p e r 1 ,0 0 0
s in g le w o me n , 1 5 - 4 4 y e a r s
180 o ld .
Div o r c e p e r 1 ,0 0 0 ma r r ie d
160 w o me n , 1 5 - 4 4 y e a r s o ld .
Re ma r r ia g e p e r 1 ,0 0 0
140 w id o w e d a n d d iv o r c e d
w o me n , 1 5 - 4 4 y e a r s o ld .
120
100
80
60
40
20
0
19 22 19 28 19 34 194 0 194 6 1952 1958 1964 1970 1 976 19 82 19 8
FIGURE 1.4. Marriage, Divorce and Remarriage Rates, US 1921-1989. Source:

National Center of Health Statistics.
46 1. Facts
0.30
Men
percent entering first marriage
0.25
Women
0.20
0.15
0.10
0.05
0.00
14 16 18 20 22 24 26 28 30 32 34 36 38 40
age
FIGURE 1.5. Entry into First Marriage, US, HRS Cohort. Source: Health and
Retirement Survey, 1992 Wave.
1. Facts 47
0.30
Men
percent entering first marriage
0.25
Women
0.20
0.15
0.10
0.05
0.00
14 16 18 20 22 24 26 28 30 32 34 36 38 40
age
FIGURE 1.6. Entry into First Marriage, US, NLS Panel. Source: National Lon-
gitudinal Survey, Youth,1979.
48 1. Facts
0.08
E arly m arriage, N LS c ohort
Late m arriage, N LS c ohort
E arly m arriage, H R S c ohort

percent entering divorce
0.06
Late m arriage, H R S c ohort
0.04
0.02
0.00
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
d u ra tio n o f firs t m a rria g e
FIGURE 1.7. Hazard of Divorce for two Birth Cohorts, US. Source: National
Longitudinal Survey, Health and Retirement Survey.
1. Facts 49
0.35
Men, NLS
0.30 W om en, NLS

per cent entering second marriage
Men, HR S
0.25 W om en, HR S
0.20
0.15
0.10
0.05
0.00
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
duration of first div orce
FIGURE 1.8. Hazard of Reamarriage, US. Source: National Longitudinal Survey,

Health and Retirement Survey.
50 1. Facts
100%
90% 24.1
26.3
30.9
80% 40.3
70%
60% 29.8 28.7
29.9
50%
30.3
40% 16.0
14.8
30% 12.9
10.6 9.7 10.7
20% 8.6
5.6
14.9 14.8
10% 14.0
11.5
3.6 4.6 5.7
0% 1.7
1970 1980 1990 2000
Other non family housholds Women living alone Men living alone
Other family households Married couples without own children Married couples with own children
FIGURE 1.9. Households by Type: Selected Years, US. Source: US Census.

1. Facts 51
100%
12.8 10.4 10.4
90% 20.9
80% 15.5 14.6
15.7
70% 15.8
17.3 16.4
17.5
60%
17.3
50%
40% 32.3 33.1

31.4
30% 28.9
20%
22.7 24.6 25.5
10% 17.1
0%
1970 1980 1990 2000
1 person 2 people 3 people 4 people 5 people or more
FIGURE 1.10. Households by Size, Selected Years, US. Source: US Census.

52 1. Facts
0.8
0.6
0.4
0.2
married men divorced men
nevermarried men married women
divorced women nevermarried women
0
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
FIGURE 1.11. Full-Time Workers, by Marital Status and Sex, US Birth Cohort
1945-54. Source: Current Population Surveys.
1. Facts 53
0.8
0.6
0.4
0.2
nevermarried men married women
divorced women nevermarried women
0
23
24
25
26
27
28
29
30
31
32
33
34
35
36
FIGURE 1.12. Full-Time Workers, by Marital Status and Sex, US Birth Cohort
1960—69. Source: Current Population Surveys.
54 1. Facts
0.8
0.6
0.4
0.2
0
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
Married child0-18 Divorced child0-18
Married no child Divorced no child
FIGURE 1.13. Full-Time Female Workers, by Marital Status and Child, US Birth
Cohort 1945-54. Source: Current Population Surveys.
1. Facts 55
0.8
0.6
0.4
0.2
0
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
Married child0-18 Divorced child0-18
Married no child Divorced no child
FIGURE 1.14. Full-Time Female Workers, by Marital Status and Child, US Birth
Cohort 1960-69. Source: Current Population Surveys.
56 1. Facts
0.8
0.6
0.4
0.2
0
68
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
Husband FT Wife FT Husband FT Wife not FT Husband not FT Wife FT
FIGURE 1.15. Work Patterns of Husbands and Wives (Aged 40-60). Source:
Current Population Surveys.
1. Facts 57
0.8
0.6
0.4
0.2
0
68
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
Husband FT Wife FT Husband FT Wife not FT Husband not FT Wife FT
FIGURE 1.16. Work Patterns of Husbands and Wives (Aged 30-40). Source:
58 1. Facts
6.8
6.6
6.4
6.2
6
nevermarried men married wome
divorced women nevermarried wom e
5.8
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
FIGURE 1.17. Weekly Wages of FT Workers, by Marital Status and Sex, US
Birth Cohort 1945-54. Source: Current Population Surveys.
1. Facts 59
6.8
6.6
6.4
6.2
6
married divorced
nevermarried married
divorced nevermarried
5.8
23
24
25
26
27
28
29
30
31
32
33
34
35
36
FIGURE 1.18. Weekly Wages of FT Workers, by Marital Status and Sex, US
Birth Cohort 1960-69. Source: Current Population Surveys.
60 1. Facts
0.4
0.2
0
0
4
97
97
97
97
98
98
98
99
99
99
00
00
_1
_1
_1
_1
_1
_1
_1
_1
_1
_1
_2
_2
68
72
73
76
79
82
85
88
91
94
99
02
19
19
19
19
19
19
19
19
19
19
19
20
-0.2
-0.4
married versus divorced, men
married versus never married, men
married versus divorced, w omen
married versus never married, w omen
FIGURE 1.19. Log Wages Differnces between Married and Singles, by Sex, US
1968—2005. Source: Current Population Surveys.
1. Facts 61
0.8
0.6
0.4
0.2
0
68
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
married divorced nevermarried
FIGURE 1.20. Women Working Full Time, by Marital Status, US 1968—2005.

62 1. Facts
0.6
0.4
0.2
0
8
70
72
74
76
78
84
86
88
90
92
94
00
02
04
6
9
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
diff=0 diff1=1-3 diff=4+ diff=-2-1 diff=-3+
FIGURE 1.21. Age Differences between Husbands and Wives, US 1968-2005.

1. Facts 63
FIGURE 1.22. Age Pyramid , US 1950. Source: US Census.

64 1. Facts
FIGURE 1.23. Age Pyramid , US 2005. Source: US Census.

1. Facts 65
0.4
0.3
0.2
0.1
0
68
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
Some college_male College_male MA and PHD_male
Some college_female College_female MA and PHD_female
FIGURE 1.24. Completed Education by Sex, Age 30-40, US 1968-2005. Source:

66 1. Facts
0.6
0.4
0.2
0
30
33
36
39
42
45
48
51
54
57
60
63
66
69
19
19
19
19
19
19
19
19
19
19
19
19
19
19
Hus band=Wife Hus band>Wife Wife>Hus band
FIGURE 1.25. Education of Spouses, by Husband’s Year of Birth, US. Source:

1. Facts 67
100% 0.4
1.2 0.2
1.0 3.9 0.5
2.7 1.7
3.6 4.5
6.7 9.4
8.7 10.0 16.0 19.6
35.3
80% 18.3 34.2
32.3 39.2 28.3 spouse MA
25.9 Degree +
65.0
38.3 spouse Bach
60% Degree
68.6
spouse Some
29.3 College
51.4 30.3 40.9
40% spouse HS
52.5 Graduate
59.4 23.6
56.1 spouse less t
19.5 High School
20%
22.9 13.6 30.4
8.6 17.6
17.7 18.3
8.2 8.3 5.8
4.9 1.9 1.7
0% 1.1 0.8
w ife husband w ife husband w ife husband w ife husband w ife husband
less than High school High school graduate Some college Bachelor´s degree Master´s degree +
FIGURE 1.26. Spouse Education by own Education, Ages 30-40,US 1970-79.

68 1. Facts
100% 0.3
1.4 0.6
2.2 1.5 2.0
5.3 4.4
7.0 8.1 14.0
11.4 10.2 17.7
18.4 20.1
37.4
80% 27.6
21.4 43.2
26.1
spouse MA
32.0 Degree +
spouse Bach
60% 47.2 51.4
Degree
42.6
47.8 spouse Some
College
57.6 43.2 spouse HS

40% 36.1
Graduate
53.9
63.7 spouse less
53.9 High School
20.8 23.3
20% 29.4
25.3 12.8
14.8
13.3 10.8
11.5 7.3
7.8 4.3
4.3 2.5 1.0 0.7
0% 0.6 0.3
wife hus band wife hus band wife hus band wife hus band wife hus band
les s than High s chool High s chool graduate Som e college Bachelor´s degree Mas ter´s degree +
FIGURE 1.27. Spouse Education by own Education, Ages 30-40, US 1996-2005.

1. Facts 69
0.8
0.6
0.4
0.2
0
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
Log wages Log residuals School years
FIGURE 1.28. Husband-Wife Correlations of Schooling and Wages, Ages 25-40,

US 1976-2005. Source: Current Population Surveys.
70 1. Facts
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
73
75
77
79
81
83
85
87
89
91
93
95
97
99
01
03
05
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
High School Drop Out High School College MA & PHD
FIGURE 1.29. Divorce Rates of Men+Women by Schooling, US 1973-2005 (%

divorced age 30 − 35 year t) /(% marrried age 25 − 30 year t − 5). Source: Current
Population Surveys.
1. Facts 71
100 3.5
95 3
proportion
90 2.5
85 2
80 1.5
75 1
1900 1905 1910 1915 1920 1925 1930 1935 1940 1945 1950 1955 1960
P roportion ever m arried wom en 40+

P roportion ever having a birth (white c ohort only)
Com pleted fertility rate
FIGURE 1.30. Completed Fertility, US by Birth Cohort. Source: National Center

of Health Statistics.
72 1. Facts
100% 3.9 4.1

6.1
10.5
10.4 11.6
12.5
14.0
80%
27.5 28.4
29.0
26.8
60%
24.9 23.2
40% 23.0
25.4
20%
33.3 32.8
25.6 27.0
0%
1970_1975 1980_1985 1990_1995 2000_2005
child0 child1 child2 child3 child4+
FIGURE 1.31. Number of Children 0-18 of US Women, Aged 35-45 at Selected

Years. Source: Current Population Surveys.
1. Facts 73
4 26
25
3.5
24
3
Birth rate
23
2.5
22
2
21
1.5 20
1920 1930 1940 1950 1960 1970 1980 1990 2000
birth rate median age at first marriage (women)
FIGURE 1.32. Birth Rates and Median Age at First Marriage, US 1900-2000.
Source: National Center of Health Statistics.
74 1. Facts
2.7
Average number of children per woman
2.5
2.3
2.1
1.9
1.7
1.5
30
33
36
39
42
45
48
51
54
57
60
63
19
19
19
19
19
19
19
19
19
19
19
19
Denmark Spain France Italy United Kingdom Netherlands
FIGURE 1.33. Completed Fertility by Generation. Source: Eurostat.

1. Facts 75
29
28
27
26
Age
25
24
23
22
30
33
36
39
42
45
48
51
54
57
60
63
19
19
19
19
19
19
19
19
19
19
19
19
Denm ark S pain F ranc e Italy United K ingdom Netherlands
FIGURE 1.34. Mean Age at First Marriage by Generation, Mean Age at First
Marriage by Generation. Source: Eurostat.
76 1. Facts
28
27
26
Age
25
24
23
30
33
36
39
42
45
48
51
54
57
60
63
19
19
19
19
19
19
19
19
19
19
19
19
Denmark Spain France Italy Netherlands UK
FIGURE 1.35. Age of Women at First Birth by Generation. Source: Eurostat.

1. Facts 77
0.8
0.6
0.4
0.2
0
85
86
87
88
89
90
91
92
93
94
95
96
97
99
01
03
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
20
divorced married never married
Note: PSID , Mother's age is 20-60. Mother's eligiblle children 0-18 do not include children born in new marriage.
FIGURE 1.36. Percent of US Mothers Receiving Child Support, by Marital Sta-

tus. Source: Panel Study of Income Dynamics.
78 1. Facts
FIGURE 1.37. Consumption and household size - more educated wives. Source:
UK Family Expenditure Surveys.
1. Facts 79
2.8
.1
Equivalent household size

2.6
0
Log expenditures
-.1
2.4
-.2
2.2
-.3
2
20 30 40 50 60
Age of wife
Log expenditures Household size
FIGURE 1.38. Consumption and household size - less educated wives. Source:
UK Family Expenditure Surveys.
80 1. Facts
This is page 81
2
The gains from marriage
From an economic point of view, marriage is a partnership for the pur-
pose of joint production and joint consumption. However, consumption
and production are broadly defined to include goods and services such as
companionship and children. Indeed, the production and rearing of children
is the most commonly recognized role of the family. But there are other
important gains from marriage, both economic and emotional.1 Although
the economic gains may not be the most important motivation for living
together with someone (‘marrying’), we focus on them here and examine
five broad sources of potential material gain from marriage, that is, why
"two are better than one":2
1. The sharing of public (non rival) goods. For instance, both partners
can equally enjoy their children, share the same information and use
the same home.
2. The division of labor to exploit comparative advantage and increasing
returns to scale. For instance, one partner works at home and the
other works in the market.
3. Extending credit and coordination of investment activities. For ex-

ample, one partner works when the other is in school.
4. Risk pooling. For example, one partner works when the other is sick
or unemployed.
5. Coordinating child care, which is a public good for the parents.
We emphasize that the gains discussed here are only potential - if they
are realized to their full extent and who benefits from them is the subject
matter of much of the rest of this book. We shall cast our discussion in
terms of two agents who choose to live together but many of the points
apply generally to a many person household. We also note that the gains
1 In this book we shall often make a distinction between the material gains and the
non-material gains and assume that the latter do not impinge upon valuations of the
latter. This is done mainly for tractability. Generally, the two sets of factors need not
be additive and the economic gains could interact with the "quality of match".
2 According to of Ecclesiastes (4: 9-10) ; "Two are better than one, because they have
a good reward for their toil. For if they fall, one will lift up the other; but woe to one
who is alone and falls and does not have another to help. Again, if two lie together, they
keep warm; but how can one keep warm alone?"
82 2. The gains from marriage
for one person may be different depending on the potential partner. In later
sections of the book we shall expand and elaborate on many of the issues
presented in this chapter.
2.1 Public goods

We begin with the most obvious potential gain, the publicness of some con-
sumption. Some of the consumption goods of a family are public (non-rival)
and both partners can consume them equally; expenditures on housing,
children and heating are clear examples.3 The sharing of housing clearly
requires that the partners live in the same household. However, parents
may enjoy their children (not necessarily equally) even if the parents live
in different households. In this respect, children continue to be a public
good for the parents even if the marriage dissolves. In practice, most goods
display some publicness and some privateness. For example, housing has a
strong public element in that both partners share the location and many
of the amenities of the house. Nonetheless there is some private element if,
for example, one or both of the partners requires a room of their own or if
there is some congestion.4
To illustrate some of the issues, we begin with a simple situation in which
we have two agents, a and b, and two goods.5 One of the goods is a purely
public good, Q, and the other is a single purely private good, q. We denote
the incomes of these persons y a and y b , respectively, and normalize the
prices of the two goods to unity. To focus on the issues associated with
sharing, we shall also assume that the two agents do not care for each
other and each has a private utility function that is used to order their own
levels of private and public goods; in the next chapter we return to this
issue. Let q s denote the consumption of the private good by person s and
let the felicity (private utility) functions be given by us (Q, q s ) for s = a, b.
If the two agents live apart then each individual s solves
maxs us (Q, q s )
Q,q
subject to Q + q s = y s (2.1)
³ ´
Let the optimal choices be Q̂s , q̂ s respectively. If the agents live together,
they can pool their income and their joint budget constraint is
Q + qa + qb = ya + yb . (2.2)
3 ‘Public’ refers to the point of view of the two partners only. Such goods are sometimes
known as collective goods or local public goods.

4 As famously noted by Virginia Wolfe in "A room of one’s own".
5 In all that follows we assume that a is female and b is male.
2. The gains from marriage 83
If the preferences of both partners are increasing in the level of the public
good then the two will always be potentially better off by living together
in the sense that we can find feasible allocations that Pareto dominate the
separate living case. Suppose, for example, that Q̂a ≥ Q̂b ; then the couple
can set:
Q = Q̂a , q b = q̂ b and q a = q̂ a + Q̂b (2.3)
Such an allocation is feasible given the joint income and it maintains or im-
proves the welfare of both b and a. This demonstration can be generalized
to any number of private and public goods. A couple can always replicate
the private consumption of the two partners as singles, purchase the max-
imal amount of each public good that the partners bought as singles and
still have some income left over.
This result relies on the assumption that both partners have positive
marginal utility from Q. Although a standard assumption, one can think
of realistic situations in which preferences are not monotone in the public
good; for example, for heating, too much may be as bad as too little and the
partners may differ in what is the optimal level of heating. Then, there may
be no gains from marriage at all, despite the reduced costs resulting from
sharing. An obvious example is one in which the public good is beneficial
for one partner and a nuisance to the other. Then publicness can be a curse
rather than a blessing, because it may be impossible to avoid the jointness
in consumption. Clearly, potential partners with such opposing preferences
would not marry. In general, some concordance of preferences is required
to generate gains from marriage (Lich-Tyler, 2003). Positive gains from
marriage require that the preferred sets for each partner, relative to the
situation when single, have a non-empty intersection on the budget line if
they live together. This is illustrated in Figure 2.1 for two people who have
the same income. In the left panel the two partners have preferences such
that, if there are no other gains, they will not choose to live together. In
the right panel they can find feasible allocations if they live together which
give both more than if they live apart.
In the example of the last paragraph, we do not have any private goods;
if we do have a private good then there may be possibilities for compen-
sation to achieve positive gains from marriage. To see the nature of the
requirements, suppose we have two public goods (Q1 , Q2 ) and one private
good. The program is:
max ua (Q1 , Q2 , q a ) (2.4)

a b a b
subject to Q1 + Q2 + q + q ≤ y + y
¡ ¢ ³ ´
and ub Q1 , Q2 , q b ≥ ub Q̂b1 , Q̂b2 , q̂ b .
We need to ³ show the´solution of this program exceeds the utility of a as

single, u Q̂a1 , Q̂a2 , q̂ a . Because the minimum cost required to obtain the
a
Q2 Q2
ua ua
y=ya +yb
Core
ub ub
ya =yb
Q1 Q1
FIGURE 2.1. Preferences over two public goods
level of welfare that b had as single is y b , it is possible to give a private con-

sumption level of at most y a without hurting b. Thus, a sufficient condition
for positive gains from marriage is
³ ´ ³ ´
ua Q̂b1 , Q̂b2 , y a > ua Q̂a1 , Q̂a2 , q̂ a . (2.5)
That is, it is possible to ‘bribe’ a to conform to b’s preferences for public

goods by giving her additional private consumption. By a similar logic
³ ´ ³ ´
ub Q̂a1 , Q̂a2 , y b > ub Q̂b1 , Q̂b2 , q̂ b (2.6)
is also a sufficient condition. Which of these two conditions is relevant

depends on the initial wealth of the parties. If b is wealthier and public
goods are normal goods then he would consume more public goods when
single, and it would be easier to satisfy condition (2.5) and attract a into
the marriage.
We return now to the simple case with one public good and one private
good and monotone preferences and illustrate some further issues ¡ associ-
¢
ated with sharing. Specifically, suppose that ua (Q, q a ) = q a Q, ub Q, q b =
s s
q b Q. If the two live separately then we have Q̂s = q̂ s = y2 and us = ( y2 )2
for s = a, b. If they live together, they have household income of y a + y b .
a b
The efficient program is to set Q̂ = y +y 2 and then divide the remaining
ya +yb
household income so that q a + q b = 2 . This gives a utility possibility
frontier of:
∙ ¸
aya + yb 2 ya + yb 2
u =( ) − ūb where ūb ∈ 0, ( ) . (2.7)
2 2
Figure 2.2 illustrates the case when y a = 1 and y b = 3. The Pareto frontier
in this case is given by ua + ub = 4. Not all points on this frontier will
be realized, because each partner has some reservation utility to enter the
marriage (if the gains from sharing public goods are the only gain). Alone,
partner a obtains ua = 14 and partner b obtains ub = 94 . Clearly, these
individual utility levels are well within the frontier and any choice of ūb
between 94 and 15 4 will give both partners more than they would receive if
they lived separately.
uA
4
Utility possibility frontier
Core
0.25
2.25 4 uB
FIGURE 2.2. Gains from public goods.
This example has two related special features that are due to the assumed
preferences. First, the level of the public good is independent of the distri-
bution of the private good but this will not generally be the case. Second,
the utility possibility frontier is linear (with a slope of −1) but generally
it will be nonlinear (see Bergstrom, Blume and Varian, 1986).6 Despite
6 It is possible for the public good to be independent of the division of income also
when the Pareto frontier is concave. This is the case, for instance, when ui = ln Q+β ln ci .
y a +yb β(ya +y b )
Then Q = 1+β
and, for 0 < ca < 1+β
, the slope of the utility frontier is
this simplicity, this example brings out a number of important ideas. First,
there are potentially large gains from the publicness of goods, which arises
from the complementarity between the incomes that the partners bring
into marriage. Second, although the distribution of the gains may not be
uniquely determined, there may exist a unique efficient level of the public
good, which depends only on the joint income of the partners. Thus the
partners may agree on the level of the public good and restrict any dis-
agreement to the allocation of private goods. Third, if there are cultural
or legal constraints that limit inequality within the family then the high
income person may not want to marry. For example, equal sharing in this
example gives b a utility level of 2, which is lower than his utility level if
single. Thus the gains from publicness are outweighed by the requirement
to share with the partner. Finally, even if the final allocation is not Pareto
efficient it may still pay to live together (if the allocation gives utility levels
inside the UPF but above the singles levels).
That there are potential gains from the publicness of some consumption
is uncontroversial. We would like to quantify how large these gains are.
To do this we use the concept of ‘equivalent income’ which is the amount
of income needed by two singles to achieve the same outcome as when
they live together. There are two outcomes of interest: buying the same
bundle and achieving the same utility levels (see Browning, Chiappori and
Lewbel (2003)). For the former, we compute the cost of buying the bundle
that the couple buys and the cost of the same bundle for each of partners if
single. The ratio of what the two partners would spend if single to what the
couple pays is the ‘relative cost of an equivalent bundle’. For our example
this bundle is Q = 2 and q a and q b are such that q a + q a = 2. Whatever
the distribution of the private good, the same bundle of goods would cost
6 units since each has to be given a level of public good equal to 2. The
relative cost of an equivalent bundle is thus 1.5 so that the couple, if single,
would need 50% more income to buy the bundle they consume as a couple.
Although the calculation of the relative cost of an equivalent bundle gives
the two agents the same bundle and hence the same utility as when living
together, the cost of achieving the same utility level may be lower since
agents may choose to substitute away from the bundle they had when
married. In our example, the utilities when together are ua = 2q a and
ub = 2 (2 − q a ). If a is single then she spends half her money on the public
good and half on the private good. Hence she needs an income y a that
solves:
µ a¶µ a¶
a a y y p
2q = u = ⇒ y a = 8q a (2.8)
2 2
p
Similarly, b needs an income of y b = 8 (2 − q a ) so that the relative cost
dua β(y a +yb )

dub
=1− (1+β)ca
.
of equivalent utilities is:

√ a p
ya + yb 8q + 8 (2 − q a )
= (2.9)
3+1 4
√
For example, if q a = 0.5 then y a = 2 and y b = 12 ' 3.46 so that the cost
of achieving the same utilities when single as when together is 5.46 and the
relative cost of equivalent utilities is 1.375.
To use the ‘relative cost of an equivalent bundle’ with household ex-
penditure data, we need to identify which goods are public and which are
private and also to estimate budget shares for these goods for couples. To
compute the ‘relative cost of equivalent utilities’ we need more information.
Specifically, we need to know both the distribution of the private good in
the couple household and preferences when single. This is a significantly
higher informational level.
Rather than distinguishing goods into being entirely private or public,
one can use a parameter η j that indicates how ‘public’ is each particular
good. Thus, if the quantity of good j bought in the market is qj , then
together the two partners can obtain qja + qjb = ηj qj units of consumption
where ηj is between 1 and 2. We refer to η j as the degree of jointness of good
j. If ηj = 2 then good j is purely public and 2qj is available for consumption
which is necessarily the same for the two agents: qja = qjb = qj . If η j = 1
then good j is purely private and any allocation qja + qjb = qj is feasible.
Generally, the share that each¡ one¢receives of this total ¡ must¢ satisfy the
restrictions that qj ≥ qja ≥ η j − 1 qj and qj ≥ qjb ≥ ηj − 1 qj to allow
for the non-exclusion of each person from the public element of the good.
As η j rises and the good becomes more public, the utility frontier shifts up
and, at the same time, the set of possible divisions narrows. In the demand
literature this is known as Barten scaling (formally the Barten scale for
¡ ¢−1
good j equals the inverse of the degree of jointness ηj ); see, for example,
Deaton and Muellbauer (1980), chapter 8. In the next chapter we shall
discuss household production in more detail; for now it suffices to note that
Barten scaling defines a simple household production technology in which
n market goods are transformed into n household commodities in a linear
and non-joint way. ThePcost of giving each partner the consumption they
have when together is nj=1 ηj qj and an index of the degree of publicness
is Pn
j=1 η j qj Pn
η= = j=1 η j ω j (2.10)
x
where qj is the couple’s demand for good j, ω j is the budget share for
good j in the married household (recall that all prices are normalized to
unity) and x is total expenditure. This index will vary from household to
household even if all households have the same technology (the same η j ’s)
since different couples spend in different ways. It gives an upper bound on
the cost of providing the same level of utility when the partners are single
as when they were together because, as discussed above, the actual cost
may be lower since the singles may optimize and choose different bundles
than when together. In the example given above we have η 1 = 2, η 2 = 1
and ω 1 = ω 2 = 0.5 so that the relative cost of an equivalent bundle is 1.5,
as derived above.
Although we can conceptually formulate precise measures of the gains
from the jointness of goods, in practice we have very little idea of how im-
portant these gains are. As an informal application of the Barten approach,
we consider the expenditure patterns of US households in taken from the
Consumer Expenditure Survey (2003) and assign a degree of jointness to
each of the composite commodities such as food, housing, clothing, etc..
Table 2.1 gives details for a nine commodity grouping.7 For each commod-
ity we assign a minimum and maximum for the jointness of the good (η j )
and then we compute the minimum and maximum values of the jointness
of total expenditure (‘consumption’). We do this for three different income
groups (gross household incomes of $10,000-$20,000, $30,000-$40,000 and
$50,000-$70,000, respectively) to allow that demand patterns differ between
rich and poor. Of course, the bounds for jointness are somewhat arbitrary
but they capture the idea that food, for example, is mostly private and
housing is largely public. The implied scales for rich and poor do not vary
much; this reflects the fact that public goods are a mix of necessities (hous-
ing) and luxuries (durables, transport and cars). The relative costs are
bounded between singles needing one third and two thirds as much as cou-
ples to buy the equivalent bundles.
The bounds in Table 2.1 are rather wide. To pin down the values more
precisely we need to make additional (and strong) assumptions and use the
data more carefully. Lazear and Michael (1980) use a single cross-section
family expenditure survey and estimate that two single individuals can al-
most double their purchasing power by forming a union. However, their
identification rests on very strong identifying assumptions. Browning, Chi-
appori and Lewbel (2003) use Canadian nondurable expenditure data on
cross-sections of single people and two person households and employ a
Barten scheme of the variety outlined above. This exploits the variation
in relative prices that arises from changes over time and variations across
provinces. The estimates are only for nondurables and services and exclude
housing and durables. They estimate that a couple who share private ex-
penditures equally when married require 41 percent more total expenditure
to replicate the bundles when single; that is, the relative cost of an equiva-
lent bundle, η, is 1.41. This is at the low end of the bounds given in Table
2.1, perhaps because housing and durables are not included.
7 Housing includes the costs of housing plus utilities and house operations. Durables
are white goods, furniture and small durables. Electronic goods are included under
entertainment. Transportation includes all transportation costs except for the purchase
of cars. We exclude health and education expenditures.
Low Medium High

income income income
Net hhold income - - $12, 761 $33, 381 $56, 360
Degree of
jointness
Min Max Budget shares (×100)
Food 1 1.2 19.7 17.9 16.3
Alcohol and tobacco 1 1.2 3.1 2.5 2.8
Housing 1.5 1.9 38.6 34.2 33.7
Durables 1.5 1.9 3.8 4.8 4.8
Clothing 1 1.2 5.6 4.9 4.6
Transportation 1.3 1.7 11.9 13.8 14.2
Car purchases 1.5 1.9 10.3 12.4 14.1
Entertainment 1.3 1.8 5.2 7.8 7.5
Personal care 1 1.5 1.8 1.8 1.7
Relative cost of an equivalent bundle
Minimum 131.5 132.2 133.0
Maximum 166.5 168.1 169.2
TABLE 2.1. Bounds for the relative cost of equivalent bundles
2.2 Specialization and Increasing Returns to Scale

The idea that agents can gain by specializing in different tasks is one of the
most venerable and useful in economics. Becker, in particular, has empha-
sized this when considering the gains from marriage (see Becker, 1991). To
illustrate its application within the family we consider a very simple house-
hold production model. Suppose that we have two people a and b who can
spend their time in market work or home production of a single non-market
good denoted by z. For a single person the household production function
is:
z = xt (2.11)
where t denotes time spent on production and x denotes purchased goods.
This production function displays increasing returns to scale in the sense
that doubling the inputs of home production time and market purchases
raises output by a factor of more than two (see Crossley and Lu (2005) for
evidence on the returns to scale for food preparation). Expenditure on the
market good is given by x = w (1 − t), where ws is the market wage for
person s.
We assume that agents only derive utility from the amount of z con-
sumed. This assumption implies that any agent is indifferent between time
spent on household production and time spent in market work. We as-
sume that other uses of time (leisure and personal care) are held fixed and
normalize the total amount of work time to unity. Given this, an agent liv-
ing alone will choose to maximize the output of the home produced good
subject to 0 ≤ t ≤ 1 and person s, when single, sets:
1 ws
ts = , zs = (2.12)
2 4
If the couple lives together, we assume that the household production
function is given by: ¡ ¢
z = x ta + tb (2.13)
so that a and b are perfect substitutes in home production. Observe that
total output is determined by the aggregate time spent at home by both
partners and the total amount of goods purchased by the family in the
market. The household budget constraint is
x = wa (1 − ta ) + wb (1 − tb ) (2.14)
Thus the agents living together can produce aggregate output:

¡ ¢¡ ¡ ¢¢
z = ta + tb wa (1 − ta ) + wb 1 − tb (2.15)
We assume that z is a private good which can be divided between the two
partners and that the partners agree to maximize the total output available
to both of them. If they set the time allocation to the optimal levels for
a b
singles their total output will be w +w 2 , which is larger than the aggre-
a b
gate output if they live separately, w +w 4 . This outcome, which is due to
increasing returns, is similar to the gains from jointness discussed in the
previous section. However, the couple acting together can improve even on
this higher output if their wages differ. To see this, suppose that wa > wb
and set ta = 0 and tb = 1; thus the higher wage person specializes in mar-
ket work and the lower wage person specializes in home production. This
gives a total output of the home produced good of wa which is, of course,
a b
greater than the output with no specialization w +w 2 . It can be shown that
this choice maximizes aggregate output. Comparing the results for a single
person household and a couple, we see that there is always a positive gain
¡ ¢ a b
from marriage of max wa , wb − w +w 4 . The gain due to specialization ac-
¡ ¢ a b
cording to comparative advantage is given by max wa , wb − w +w 2 which
is zero if and only if the wages are the same.
This example illustrates the potential gains from specialization but the
specific implications depend on a number of special features of this model.
First, the two partners are assumed equally productive at home production.
This can be trivially extended to allow for different fixed productivities in
which case specialization will depend on the ratios of productivity in the
market (that is, the wage) to productivity at home of the two partners.
Second, the technology is linear in the time inputs. If, instead, we allowed
for some concavity and complementarity between partners time use, spe-
cialization need not occur and interior solutions would arise. Yet we would
still expect the high wage spouse to work more in the market when wages
differ.
As emphasized by Becker (1991, chapter 2), comparative advantage can
be developed via differential investments or learning by doing. Within mar-
riage or in the market each party can use their own human capital to a larger
extent, yielding convexity and dynamic increasing return. In particular, if
one partner may specialize in home production while the other specializes
in market work then both of them acquire skills relevant to their specific
activity. Thus, a small innate difference can be magnified, and strengthen
the incentives to specialize (see Chicilinsky 2005, Pollak 2007).
There is ample evidence for a division of labor within the household
(see Chapter 1). Married men work longer hours in the market and have
substantially higher wages than unmarried men. Married women have lower
wages and work more at home than unmarried women; see Gronau 1986,
Korenman and Neumark, 1992 and Daniel, 1992.
2.3 Imperfect Credit Markets

Consider two potential partners denoted by a and b. Each person lives for
two periods which we denote by 1 and 2. Utility in period t is derived from
consumption and the per period utility is
u (ct ) = ln ct . (2.16)
For simplicity, we assume that the discount factor is unity and the real rate
is zero. Each person has an initial wage of 1 that he/she can augment by
spending the first period in school, obtaining a second period wage of w.
If there is a perfect capital market, a person can smooth his consumption
through borrowing and will set c1 = c2 = c . Thus, with investment in
schooling, one can obtain c = w2 each period, while without investment
consumption each period will be 1. Investment is profitable if the increase
in wage is sufficient to compensate for the earnings forgone in the first
period, that is if the second period wage w exceeds 2. However, if borrowing
is impossible there is no investment in schooling since consumption in the
first period would be zero.
Now assume that a and b marry each other. Under a perfect capital
market, marriage will not influence their investment choices. However, if
there is an imperfect capital market, marriage allows a couple to partially
overcome the no borrowing constraint. This is accomplished by extending
credit within the family, whereby one partner (b, say) works in the market
while the other goes to school. To evaluate the potential gains from mar-
riage, consider an efficient program that maximizes the utility of partner a
given that partner b receives the lifetime utility he would have in the single
state, without schooling. With our choice of units, life time utility in the
absence of investment is 0. We thus solve
max {ln ca1 + ln ca2 } (2.17)

ln cb1 + ln cb2 ≥ 0
ca1 + cb1 = 1
ca2 + cb2 = 1+w
A necessary condition for efficiency is that consumption in each period

is distributed between the partners so as to equalize the ratios of their
marginal utilities from consumption in the two periods
u0 (ca1 ) u0 (cb1 )
= . (2.18)
u0 (ca2 ) u0 (cb2 )
With a logarithmic utility function, this implies that the consumption of

both partners must grow at the same rate, 1+w. Using the requirement that
1
the lifetime utility of partner b remains zero, we obtain that cb1 = (1+w)− 2
1
and cb2 = (1+w) 2 . Because the consumption of a grows at the same rate, her
lifetime utility will be positive if and only if the first period consumption,
1
ca1 = 1 − (1 + w)− 2 , exceeds that of b. A brief calculation will confirm that
this is true whenever w > 3.
We conclude that the potential for coordination of investment activities
through credit can motivate marriage when credit markets are not oper-
ative. Notice that marriage does not completely eliminate the borrowing
constraint, because only one person will invest in schooling and he/she will
do so only at higher rates of return from schooling than in the case of per-
fect capital market. An important aspect of this example is that individuals
who are ex-ante identical may voluntarily agree to pursue different careers,
allowing both partners to share in the gains from this efficient program.
Obviously, specialization in investment activities can also be motivated by
differences in innate abilities. Typically, the family will choose to invest in
the person with the higher return from human capital investment. In either
case, commitments are crucial for the implementation of such a program,
see Dufwenberg (2002). A woman will be hesitant to support her husband
through medical school if she expect him to break the marriage (and marry
a young nurse) when he finishes.
Evidence of implicit credit arrangements within marriage is sometimes
revealed at the time of divorce, when the wife claims a share of her ex-
husband’s earnings on the grounds that she supported him in school; see
Borenstein and Courant (1989). However, recent empirical work casts doubt
on the importance of liquidity constraints for schooling choices see Carneiro
and Heckman (2003). This important issue is still a matter of controversy;
see Acemoglu and Pischke (2001).
2.4 Risk sharing

Individuals who face idiosyncratic income risk have an obvious incentive
to provide mutual insurance. This can be done within the family. Here we
present a simple example. Consider two risk averse partners with random
incomes, y s , s = a, b. Acting alone, if there are no possibilities for saving or
borrowing, each partner will have an expected utility given by E(us (y s ))
respectively. Acting together, they can trade consumption in different states
of nature. To see the potential gains from trade, consider the maximization:
max E(ua (ca ))
subject to E(ub (y a + y b − ca )) ≥ E(ub (y b )).
Clearly, setting in each state ca = y a and cb = y b , is a feasible solution
which will replicate the allocations in the single state. However, the optimal
risk sharing rule is
u0 (ca ) = λu0 (cb ) (2.19)
where λ is a positive constant. That is, the slope of the utility frontier,
a
given by − u0(c )
u0(cb )
is equalized across all states, where a state is defined by
the realized sum of the individual incomes, y a + y b , that is, total family
income. Otherwise, both partners can be made better off by transferring
resources to a person in a state where his marginal utility of consumption is
relatively high, taking resources away from him in another state where his
marginal utility is relatively low. Following this optimal rule, both partners
can be made strictly better off, provided that their incomes are not perfectly
correlated (or that risk aversions differ).
A strong testable implication of efficient risk sharing is that the con-
sumption of each family member varies only with family income. That is,
holding family income constant, the idiosyncratic shocks to individual in-
comes will induce transfers between the partners, but consumption levels
will remain the same.
Depending upon the particular risk, the potential gains from mutual in-
surance can be quite large. For instance, Kotlikoff and Spivak (1981) who
consider the risk of uncertain life, in the absence of an annuity market,
estimate that the gains that a single person can expect upon marriage
are equivalent to 10 to 20 percent of his wealth. In a different applica-
tion, Rosenzweig and Stark (1989) show that marriages in rural India are
arranged between partners who are sufficiently distant to significantly re-
duce the correlation in rainfall, thereby generating gains from insurance.
Hess (2004) finds that couples with a higher correlation in incomes are more
likely to divorce, suggesting that effects of mutual insurance on the gains
from marriage are higher when the partners’ incomes are less correlated.
Shore (2007) finds that the correlation in spouses’ earnings respond to the
business cycle; it is higher for couples whose marriage spans longer periods
of high economic activity.
2.5 Children
2.5.1 Technology and preferences
One of the principal gains from marriage is the production and rearing of
children. Although the biological and emotional gains may dominate here,
we can also consider the economic aspects. In particular, we wish to discuss
the gains to the child that arise from living with their natural parents in an
intact family. Consider two partners, a and b, who choose to have a child (or
some other fixed number of children) denoted by k. We allow that the two
partners have alternative uses for their time; in this case they can spend
time in child care, ta and tb , respectively or in market work at the wages
wa and wb . In this example we shall assume that there is a single private
good with market purchases of q of this good being allocated between the
three family members in amounts ca , cb , ck . The utility of children depends
additively on their consumption of goods and the time spent with each of
the parents:
uk = ck + αta + βtb , (2.20)
where the parameters α and β represent the efficiency of parents a and b,

respectively, in childcare. This is, of course, a very special assumption and
implies that consumption can fully compensate the child for the absence
of parents and that the two parents’ childcare time are perfect substitutes.
Usually we assume that α and β are positive (perhaps an arguable assump-
tion for teenagers). The utility of each parent is assumed to be multiplica-
tive in their own consumption and the child’s utility level:
us = cs uk for s = a, b. (2.21)
Thus, children are assumed to be a public good to their natural parents

and both care about their welfare.
We consider here situations in which parents differ in their earning capac-
ity and efficiency in child care. The linearity of the parents’ utility func-
tions in their own consumption implies that the parents would agree on
an efficient program that maximizes the joint "pie" that is available for
distribution between them.8 That is, the parents would agree to:
© a¡ ¡ ¢ ¢ ª
max w (1 − ta ) + wb 1 − tb − ck (ck + αta + βtb ) (2.22)
ta ,tb ,ck
subject to 0 ≤ ts ≤ 1, for s = a, b
8 Thus, the amount of time spent on the child is determined by efficiency considera-
tions, independently of the distribution of the consumption good. The two stage decision
process, whereby production and distribution are separable, is an important consequence
of transferable utility that will be discussed later in the book.
2.5.2 Intact families

We have three regimes, depending on the parameter values. We always
assume:
w b > wa , α > β
implying that the high wage spouse, b, has a comparative advantage in

market work and the low wage person a has comparative advantage in
home production:
wb wa
> (2.23)
β α
If both wages are high relative to efficiency at home production, (if wa > α
and, consequently, wb > β) then both parents will work full-time in the
market and use only market goods for caring for the child. Conversely, if
both wages are low relative to efficiency at home production (if wb < β
and wa < α) then parents will use only time to care for the child. An
intermediate case is the one in which the high wage partner, b, has ab-
solute advantage in market work and the low wage person a has absolute
advantage working at home ,
α > wa , β < wb
For this intermediate case b will spend all his time in market work and a
will spend all her time looking after the child. This intermediate case has
two distinct sub-cases that differ in the expenditures on the child. For case
1 we have:
wb > α. (2.24)
In this case, the intact family spends part of its income on child goods,
b
ck > 0. Specifically, ta = 1 and tb = 0 and ck = w 2−α . The utility of
b
the child is then uk = w 2+α and the utility possibility frontier facing the
parents is given by
(wb + α)2
ua + ub = . (2.25)
4
In case 2 we have the converse:
wb < α, (2.26)
which gives ck = 0. In this case, the utility of the child is uk = α and the
UPF facing the parents is then given by
ua + ub = wb α. (2.27)
2.5.3 Divorce with no transfers

What happens if the partners split and one of the partners receives custody,
without any transfers? It is quite likely that if the marriage breaks up and
the parents live in separate households, the utility of the non custodial
parent from the child is reduced. Nevertheless, it is only natural that the
non-custodial parent continues to care about the child and for simplicity
we shall continue to assume that the utility of both parents is given by
(2.21). We shall further assume that only the custodial parent can spend
time with the child. If custody is assigned to parent b he will work fulltime
in the market ( tb = 0 since wb > β) and will set ck = 0.5wb = uk . If
custody is assigned to parent a she will work part time to finance her own
consumption, setting t = 0.5, but will spend no money on child goods
(since α > wa ). In this case, the child’s utility is uk = 0.5α. If we now
choose the custodial parent to maximize the welfare of the child, we obtain
a very simple rule for the assignment of custody. In the absence of post
divorce transfers, the high wage parent b should obtain custody if and only
if his\her wage, wb , exceeds the efficiency of the low wage spouse a at home,
α.
Table 2.2 compares the utility of the child when the parents are mar-
ried and separated, when custody is assigned optimally for the two cases
discussed above. We also show the utilities of each parent when they are
separated and the sum of their utilities when they are married. Examining
the entries in the table, it is seen that the child is always worse off when the
parents split, because the custodial parent spends less time with the child
or less goods on the child. We also have that at least one of the parents is
worse off materially when the parents live apart, because their post divorce
payoffs are below the utility possibility frontier in an intact family. That
is:
wa wb (wb + α)2
( )2 + ( )2 < . (2.28)
2 2 4
Such results are quite typical and can be traced to the inefficient allocation
of time following divorce. For example, for case 2 the custodial parent is
pushed into the labor market, despite her comparative advantage in child
care. The custodial parent who chooses how much time to spend with the
child does not (or cannot) take into account the interests of the other
parent, which is the source of the inefficiency. Following separation, the
non- custodial parent can be better off than the custodial parent, because
they can free ride on the custodial parent who takes care of the child. This
is the case if the low wage parent a is the custodial parent and also holds
if the high wage parent b is the custodian and 2wa > wb . Thus, although
the child is better off under the custody of the parent who is more efficient
in caring for it, this parent may be better off if the other parent had the
custody. The most natural way to deal with this "hot potato" problem, as
well as with the low welfare of the child, is to force the non custodian parent
Case 1: wb > α, b is the custodian

Married Separated
Family Work at Work at
member home Utility home Utility
b 2 a b
(w +α) w w
a 1 4 0 2
b
b 0 0 ( w2 )2
wb +α wb
k − 2 − 2
Case 2, wb < α, a is the custodian
Family Work at Work at
member home Utility home Utility
1 αwa
a 1 wb α 2 4
αwb
b 0 0 2
α
k − α 2
Note: when married the utility of a and b is shared.
TABLE 2.2. Work patterns and material welfare of family members
to pay child support. Post divorce transfers will be discussed in detail in a

subsequent chapter, but it should be noted at the outset that, in practice,
custodial mothers often receive no transfer from the ex-husband and when
they do the transfer is often quite small.
There is ample evidence that children with single or step parents are
worse off than children in intact families (see Argys et al, 1998, Hethering-
ton and Stanley-Hagan, 1999), suggesting that the break up of marriage
can be quite costly. However, Piketty (2004) and Bjorklund and Sundstrom
(2006) show that much of the differences in child attainments precede the
divorce, so that the reduction in the child’s welfare is caused by a bad
quality of the match (for example, fights between the parents) rather than
the divorce itself. In either case, the risk of separation may reduce the
incentives to produce children and to specialize in home production.
2.6 Concluding remarks

None of the gains that we have discussed in this chapter actually require
the traditional family institution. If all goods and work activities are mar-
ketable, there is no need to form marriages to enjoy increasing returns or to
pool risks. In fact, the role of the family varies depending on market con-
ditions and vice versa. For instance, with good medical or unemployment
insurance one does not need to rely on his spouse. Similarly, sex and even
children can be obtained commercially. Nevertheless, household production
persists because it economizes on search, transaction costs and monitoring.
However, to fully exploit these advantages requires a durable relationship.
This shifts attention to the question which types of partnerships are likely
to last.
Gains from human partnerships need not be confined to a couple of the
opposite sex. One also observes ”extended families” of varying structures
which coordinate the activities of their members and provide self insurance.
The prevalence of male-female partnerships has to do with sexual attrac-
tion which triggers some initial amount of blind trust. (The Bible is quite
right in puzzling over why ”shall a man leave his father and mother and
cleave unto his wife”9 .) Equally important is a strong preference for own
(self produced) children. These emotional and biological considerations are
sufficient to bring into the family domain some activities that could be
purchased in the market. Then, the accumulation of specific ”marital cap-
ital” in the form of children, shared experience and personal information
increases the costs of separation and creates incentives for a lasting re-
lationship. In this sense, there is an accumulative effect where economic
considerations and investments reinforce the natural attachment. Other
glues, derived from cultural and social norms also support lasting rela-
tionships. But in each case customs interact with economic considerations.
The weaker is the market, the more useful is the extended family and social
norms (commands) are added to the natural glue.
Keeping these considerations in mind, we can now address the ques-
tion which activities will be carried out within the family. One argument
is that the family simply fills in gaps in the market system, arising from
thin markets, or other market failures, see Locay (1990). Another line of
argument (see Pollak (1985)) is that the family has some intrinsic advan-
tages in monitoring (due to proximity) and in enforcement (due to access
to non-monetary punishments and rewards). A related but somewhat dif-
ferent argument is that family members have already paid the (sunk) costs
required to acquire information about each other, see Ben-Porath (1980).
Thus, credit for human capital investments may be supplied internally ei-
ther because of a lack of lending institutions or because a spouse recognizes
the capacity of her partner to learn and is able to monitor the utilization
of his human capital better than outsiders. Similarly, annuity insurance is
provided internally, either because of lack of annuity markets or because
married partners have a more precise information on their spouse’s state of
health than the market at large. It is clear that these three considerations
interact with each other and cannot be easily separated. The main insight
is that the gains from marriage depend on the state of the market and must
be determined in a general equilibrium context.
9 Genesis 2: 24.
2.7 References
[1] Acemoglu, Daron and Jorn-Steffen Pischke, ‘Changes in the Wage
Structure, Family Income, and Children’s Education’, European Eco-
nomic Review, 45 (2001), 890-904.
[2] Argys, Laura M., H. Elizabeth Peters, Jeanne Brooks-Gunn and Judith
R. Smith Argys, ‘The Impact of Child Support on Cognitive Outcomes
of Young Children’, Demography, 35 (1998), 159-173.
[3] Becker, Gary, Treatise on the family, (Cambridge Mass: Harvard

[4] Ben-Porath, Yoram, ‘The F-Connection: Families, Friends, and Firms

and the Organization of Exchange’, Population and Development
Review, 6 (1980), 1-30.
[5] Bergstrom, Theodore, Lawrence Blume and Hal Varian, ‘On the Private
Provision of Public-Goods’, Journal of Public Economics, 29 (1986),
25-49.
[6] Bjorklund, Anders and Marianne Sundstrom, ‘Parental Separation and

Children’s Educational Attainment: A Siblings Analysis on Swedish Reg-
ister Data’, Economica, 73 (2006), 605-624.
[7] Borenstein, Severin and Paul N. Courant, ‘How to Carve a Medical

Degree: Human Capital Assets in Divorce Settlements’, The American
Economic Review, 79 (1989), 992-1009.
[8] Browning, Martin, Pierre-André Chiappori and Arthur Lewbel, ‘Esti-

mating Consumption Economies of Scale, Adult Equivalence Scales, and
Household Bargaining Power’, CAM Working Paper No. 2003-12, (2006).
[9] Bureau of Labor Statistics, (2004), http://www.bls.gov/cex/2003.
[10] Carneiro, Pedro and James J. Heckman, ‘The evidence on credit con-
straints in post-secondary schooling’, Economic Journal, 112 (2002),
705-734.
[11] Chichilnisky, Graciela, ‘The Gender Gap’, Review of Development

Economics, 12 (2008), 828-844.
[12] Crossley, Thomas and Yuqian Lu, ‘Exploring the returns to scale in
food preparation (baking penny buns at home)’, IFS Working Paper No.
W05/03, (2005).
[13] Daniel, Kermit, ‘Does Marriage Make Men More Productive?’, Uni-
versity of Chicago, Population Research Center 92-2, (1992).
[14] Deaton, Angus and John Muellbauer, Economics and Consumer

Behavior, (Cambridge: Cambridge University Press, 1980).
[15] Dufwenberg, Martin, ‘Marital investments, time consistency and emo-

tions’, Journal of Economic Behavior & Organization, 48 (2002),
57-69.
[16] Gronau, Reuben, ‘The Theory of Home Production: The Past Ten
Years’, Journal of Labor Economics, 15 (1997), 197-205.
[17] Hetherington, E. Mavis and Margaret M. Stanley-Hagan, ‘Stepfami-

lies’, in Lamb, Michael E. (eds.), Parenting and Child Development
in "Nontraditional" Families, (Mahwah, New Jersey: Lawrence Erl-
baum Associates, 1999).
[18] Hess, Gregory D., ‘Marriage and Consumption Insurance: What’s Love
Got to Do with It?’, Journal of Political Economy, 112 (2004), 290-
318.
[19] Korenman, Sanders and David Neumark, ‘Marriage, Motherhood, and

Wages’, The Journal of Human Resources, 27 (1992), 233-255.
[20] Kotlikoff, Laurence J. and Avia Spivak, ‘The Family as an Incomplete

Annuities Market’, The Journal of Political Economy, 89 (1981),
372-391.
[21] Lazear, Edward P. and Robert T. Michael, Allocation of income

within the household, (Chicago: University of Chicago Press, 1988).
[22] Lich-Tyler, Stephen, ‘Preference-based Assortative Matching’, unpub-

lished manuscript, University of Michigan at Ann Arbor, (2003).
[23] Locay, Luis, ‘Economic Development and the Division of Production

between Households and Markets’, The Journal of Political Econ-
omy, 98 (1990), 965-982.
[24] Piketty, Thomas, ‘The Impact of divorce on school performance: evi-

dence from France, 1968-2002’, CEPR Discussion Paper Series No. 4146,
(2003).
[25] Pollak, Robert A., ‘Allocating Time: Individuals’ Technologies and

Household Technology’, Washington University in St. Louis, (2007).
[26] Pollak, Robert A., ‘A Transaction Cost Approach to Families and

Households’, Journal of Economic Literature, 23 (1985), 581-608.
[27] Rosenzweig, Mark R. and Oded Stark, ‘Consumption Smoothing, Mi-

gration, and Marriage: Evidence from Rural India’, The Journal of
Political Economy, 97 (1989), 905-926.
[28] Shore, Stephen H.,’For Better, for Worse: Intra-Household Risk-

Sharing over the Business cycle’, Review of economics and statistics,
forthcoming.
This is page 103
3
Preferences and decision
making
3.1 Preferences
In the last chapter we informally reviewed the gains from marriage in some
generality. The existence of potential gains from marriage is not sufficient to
motivate marriage and to sustain it. Prospective mates need to form some
notion as to whether families realize the potential gains and how they are
divided. In this chapter we consider these issues in a very specific context.
The context is a two person (woman a and man b) household1 in which the
only (static) decision is how much to spend on various market goods that
are available at fixed prices, given fixed total household expenditure on
all goods. Although very special this context allows us to discuss formally
many of the issues that will be used in other contexts in later chapters.
Some commodities are private and some public. Private goods are con-
sumed non-jointly by each partner and public goods, such as heating, are
consumed jointly and non-exclusively by the two partners. In other words,
private goods are characterized by an exclusion restriction property: the
fact that I consume a particular apple de facto excludes anyone else from
consuming the same apple. With public goods, on the contrary, no such
restriction exists: that I enjoy seeing a beautiful painting on my wall does
not preclude my spouse from enjoying it just as much (or even disliking it).
Several remarks can be made at that point. First, several commodities are
sometimes used publicly and sometimes privately; for instance, I can drive
my car alone to go to work, or the whole family may take a ride together.
Second, the privateness or publicness of a good is quite independent of the
type of control existing on that good and who exerts it; typically, parents
have control over the (private) consumption of their young children. Finally,
and more crucially, there exist subtle interactions between the (‘technical’)
nature of a good and how it enters the member’s utilities. The private
consumptions of member a certainly enter a’s utility; but it also may enter
b’s - we then call it an externality. Conversely, some commodities, although
public by nature, may in fact be exclusively consumed by one member; for
instance, although both spouses may in principle watch television together
without exclusion, one of them may simply dislike TV and never use it.
Throughout most of the book, we assume, to keep things simple, that any
1 Children will be introduced at a later point.

104 3. Preferences and decision making
particular commodity is either purely public or purely private, although

many of our results would extend to more general settings.
We introduce some notation that will be used throughout the chapter.
There are N public goods, and the market purchase of public good j is
denoted Qj ; the N -vector of public goods is given by Q. Similarly, private
goods are denoted qi with the n-vector q. Each private good bought is
divided between the two partners so that a receives qia of good i and b
receives qib = qi − qia . Hence the vectors a and b receive are¡qa and q¢b
respectively, with qa +qb = q. An allocation is a N +2n-vector Q, qa , qb .
The associated market prices are given by the N -vector P and the n-vector
p for public and private goods respectively.
We assume that each married person has her or his own preferences ¡ over ¢
a
the allocation of¡ family resources.
¢ Denote a’s utility function by U Q, qa , qb
and b’s by U b Q, qa , qb . This general formulation allows that a is con-
cerned directly with b’s consumption and also that b’s consumption of pri-
vate goods impacts on a’s preferences between her own private goods and
the public goods. Any kind of externality is allowed. The presence of the
partner’s private consumption does not mean necessarily that members are
altruistic to each other; for instance, it could simply represent the partner’s
smoking that bothers the other member by reducing their utility. Through-
out the book, we assume, unless stated otherwise, preference orderings are
continuous and convex and can be represented by utility functions U s (.),
s = a, b, that are continuously differentiable and strictly concave.
In the subsequent chapters in the first half of this book we shall be
discussing the resolution of conflicts that arise between partners if U a (.)
and U b (.) represent different preferences. It is important to acknowledge,
however, that if marriage is sometimes a battleground, it is can also be a
playground. In the context of the family, love or affection might be oper-
ating and conflicts are thereby considerably attenuated. We return to this
below.
Although quite reasonable, the form just described is too general to be
used in most contexts - if only because it is difficult to incorporate such
preferences into a model in which agents live alone for some part of their
life-cycle. Consequently the literature generally takes a special case which
is known as caring.2 For this ¡ we ¢first assume agents a and b have felicity
functions ua (Q, qa ) and ub Q, qb respectively. The most general form has
¡ ¢ ¡ ¡ ¢¢
U a Q, qa , qb = W a ua (Q, qa ) , ub Q, qb
¡ ¢ ¡ ¡ ¢¢
U b Q, qa , qb = W b ua (Q, qa ) , ub Q, qb , (3.1)
where both W a (., .) and W b (., .) are monotone increasing functions. The
2 Sometimes the term altruistic is used for preferences taking this form. Pollak (2006)
has suggested the term deferential since each person defers to the judgment of the other
regarding their consumption.
3. Preferences and decision making 105
weak separability of these ‘social’ preferences represents an important moral

principle; a is indifferent between bundles qb , Q that b consumes whenever b
is indifferent (and similarly for b). In this sense caring is distinguished from
paternalism. Caring rules out direct externalities because a’s evaluation
of her private consumption qa does not depend directly on the private
goods that b consumes (and vice versa). A more commonly used form is
the restricted version:
¡ ¢ ¡ ¢
U a Q, qa , qb = ua (Q, qa ) + δ a ub Q, qb ,
¡ ¢ ¡ ¢
U b Q, qa , qb = ub Q, qb + δ b ua (Q, qa ) . (3.2)
Generally we take the weights δ a and δ b to be non-negative parameters

such that each person cares for the other but not as much as they care for
themselves. For this formulation, δ a = δ b = 0 corresponds to egotistic pref-
erences and δ s > 0 represents altruism. If δ a δ b = 1 then the two partners
have the same ordinal preferences.
Some authors use a slightly different representation of altruism, namely
¡ ¢ ¡ ¢ a ¡ ¢
U a Q, qa , qb = ua Q, qb + δ̃ U b Q, qa , qb
¡ ¢ ¡ ¢ b ¡ ¢
U b Q, qa , qb = ub Q, qb + δ̃ U a Q, qa , qb (3.3)
The logic here is that a should cares about b’s ultimate utility U b , which
includes also b’s altruistic feelings towards a. We can then think of (3.2) as
a reduced form obtained by the substitution:
¡ ¢ ¡ ¢ a ¡ ¢ b ¡ ¢
U a Q, qa , qb = ua Q, qb + δ̃ [ub Q, qb + δ̃ U a Q, qa , qb ] (3.4)
a b
If δ̃ δ̃ 6= 1 we have:
a
¡ ¢ 1 δ̃ ¡ ¢
U Q, qa , qb =
a
a b
a a
u (Q, q ) + a b
ub Q, qb (3.5)
1 − δ̃ δ̃ 1 − δ̃ δ̃
a b
Such a reduction yields logical results only if δ̃ δ̃ < 1. Clearly, too much
a b
caring (δ̃ δ̃ > 1) can lead to paradoxical results in which a puts negative
weights on both felicity functions. See Bergstrom (1989) and Bernheim and
Stark (1988) for further discussion and examples of how excessive altruism
can lead to unpalatable outcomes.
In some contexts we wish to impose stronger restrictions on preferences.
For example, we shall often consider only one private good. This can be
justified if prices are fixed by an appeal to the composite commodity the-
orem. In that case we can consider the unique private good to be ‘money’.
A second, particular case that we shall consider in many contexts relies on
the assumption of transferable utility (TU). This holds if we have egotistic
preferences and each felicity function can be (possibly after an increasing
transform and a renaming of the private goods) put into a form that is
similar to the Gorman polar form:
ua (Q, qa ) = f a (q2a , ..., qna , Q) + G (Q) q1a
¡ ¢ ¡ ¢
ub Q, qb = f b q2b , ..., qnb , Q + G (Q) q1b (3.6)
where G (Q) > 0 for all Q. Note that the G (.) function is identical for both
members, whereas the f (.) functions can be individual-specific. In words,
the transferable utility assumption implies that, for some well chosen car-
dinalization of individual preferences, the marginal utility of an additional
dollar spent on private consumption of commodity 1 is always the same
for both members. Hence utility can be ‘transferred’ between them (us-
ing commodity 1 transfers) at a fixed rate of exchange. Repeatedly in this
book, we shall develop examples in which the transferability assumption
drastically simplifies the problem to hand.
We shall often need to compare the utility of a given individual in two
different marital situations, for instance when married versus when single
(or divorced). Various assumptions can be made here. One extreme hy-
pothesis states that marriage may change preferences in an arbitrary way.
Then there is simply no relation between pre-marital and post-marital util-
ity functions - not a very useful property for our purpose. Conversely, we
may assume that preferences over commodities are not changed by mar-
riage. This by no means implies that the satisfaction derived from any
consumption is the same inside and outside marriage, but simply that the
ranking of the various consumption bundles is not affected by the individ-
ual’s marital status. With egotistic preferences, this will hold if the utility
in marriage, us , is related to the pre-marital preferences represented by the
utility function, ūs (Q, qs ) by:
us (Q, qs ) = F s (ūs (Q, qs )) (3.7)
s
where the mapping F (.) is strictly increasing. A particularly convenient
special case that we shall employ when we consider explicitly the full gains
from marriage is:
us (Q, qs ) = F (ūs (Q, qs ) + θs ) (3.8)
s
Here, θ represents non-monetary, marriage-specific aspects of s’s idiosyn-
cratic desire to be married. With caring preferences, the same obtains if
we normalize the contribution of the spouse’s utility to be uniformly zero
when the agent is single. This assumption has important consequences on
the empirical estimation of the models. If condition (3.8) is satisfied, then
the preferences of married individuals amongst private and public goods are
the same when married or single. These preferences can then be recovered
from data on singles’ behavior.
Finally, an intermediate assumption states that single and married indi-
viduals have the same basic preferences, but marriage involves a change in
the consumption technology, a concept we define in the next subsection.
3.2 Household production

3.2.1 The general framework.
Household activities are not limited to private or public consumptions.
They are also the source of important production activities that should
not be disregarded. In low income countries, a large fraction of GDP con-
sists of agricultural commodities produced at the household (or the village)
level. Even in high income economies, a significant fraction of individual
available time is spent on household production. This entails immediate
tasks (cleaning, cooking, etc.) but also long term investments in health,
education and others. In a sense, even such ‘commodities’ as love, affec-
tion or mutual care are ‘produced’ (and consumed) at the household level.
In Becker’s (1965) model, the only commodities that are ultimately con-
sumed by individuals are those produced at the household level. Becker
views goods that are purchased in the market as inputs in a production
system that transforms these purchased goods into final commodities that
are actually consumed (and enter individual utilities). These home pro-
duced goods can be either public or private for the two partners, denoted
by Cj and cj respectively. The production of commodities also requires time
inputs that are provided by the household members in addition to market
purchased
¡ goods.
¢ The technology is described by a production possibility
set Ω q, ta , tb that gives the possible vector of outputs (c, C) that can be
produced given a vector of market purchases q and the total time spent in
household production by each of the two partners, ta and tb . This allows
that time spent on any activity may produce many goods.
Household production function

A special case is when the feasible set can be described by household pro-
duction functions that specify the amount of each commodity that can be
produced given the amount of market goods and time assigned to that
commodity. We denote the vector of market goods assigned to commodity
j by qj and the time inputs of a and b devoted to commodity j by taj and
tbj , respectively. Thus:
¡ ¢
cj = f j qj , taj , tbj (3.9)
The associated constraints are:
P j
q = q
j
P
tsj = ts , s = a, b (3.10)
j
Each person has preferences¡ defined over household

¢ produced goods and
the vectors of time use, U s C, ca , cb , ta , tb for s = a, b, where ts is the
vector of time inputs for j. This framework allows time activities to have
two distinct roles. For example, a father who spends time with his child
contributes to the development of the child (through f j (.)) and may also
enjoy spending time with the child (captured by the presence of tbj in U b (.)).
Of course, either of these effects could be negative (although not both).
A standard problem with this approach is that the production function,
despite its conceptual interest, cannot be estimated independently of the
utility function unless the home produced commodities are independently
observable; see Pollak and Wachter (1975) and Gronau (2006). Observabil-
ity of outputs may be acceptable for agricultural production, or even for
children’s health or education; it is less likely for, say, cleaning, and almost
impossible for personal caring.
If only inputs are observed and not outputs we may be able to recover
information about the technology if we make auxiliary assumptions such
as constant returns to scale and assumptions on preferences. To illustrate
this, consider two partners who consume one single public good C and one
private good c such that a consumes ca and b consumes cb with preferences
given by us (C, cs ) , s = a, b. Assume that the private good is purchased in
the market and that the public commodity is produced using only the time
inputs of the two partners. That is,
¡ ¢
C = f ta , tb . (3.11)
Assuming that both partners participate in the labor market at wages wa
and wb respectively, it can then be shown that for any efficient allocation
the partners will minimize the cost of producing the public commodity in
terms of the forgone private commodity, yielding
¡ ¢
f1 ta , tb wa
= (3.12)
f2 (ta , tb ) wb
in any interior solution. If we assume constant returns to scale, we can
write: ¡ ¢
C = f ta , tb = tb φ (r) (3.13)
ta
for some function φ (r) where r = tb
. The condition (3.12) then reduces to:
0
φ (r) wa
= (3.14)
φ (r) − rφ0 (r) wb
The testable implication of this equality is that r only depends on the wage
ratio ω; this can be tested on a data set that reports wages and time spent
on household production. Defining,
φ0 (r)
h(r) = (3.15)
φ (r) − rφ0 (r)
this equation can be re-written as:
φ0 (r) 1
= 1
φ (r) r + h(r)
Integrating, we have:
ÃZ !
r
ds
φ (r) = K exp 1
0 s + h(s)
where K is an unknown constant. In other words, when outputs are not

observable, knowledge of the input supply (as a function of relative wages)
allows us to determine the household production function up to a multi-
plicative scale factor.
It is important to note that the assumptions of constant returns to scale
and no joint production ( in the sense that ta and tb do not appear directly
in the utility function) are critical for this particular identification result;
see Pollak and Wachter (1975) and Gronau (2006) for further discussion of
the role of these assumptions. A further issue that was not challenged in
this literature is whether or not the partners are cooperating. The example
above shows that in some cases it is sufficient to assume efficiency; other
assumptions may also guarantee identification.
Marital technology and indifference scales

Let us briefly come back to the previous discussion on the changes in pref-
erences that may result from marriage. The two extreme assumptions de-
scribed were either that there are no such changes (in the sense that an
individual’s preference relationship over consumption bundles was indepen-
dent of the person’s marital status) or that they were arbitrary (that is,
there is no relationship between pre- and post- marital utilities). The first
assumption is often too restrictive, whereas the second is too general to be
useful.
An intermediate approach, proposed by Browning, Chiappori and Lew-
bel (2003), relies on the notion of production technology. The idea is that
marriage leaves ordinal preferences over commodities unchanged, but al-
lows a different (and more productive) technology to be used. Formally,
they apply the simple Barten household production technology in which
n market goods are transformed into n household commodities in a linear
and non-joint way; see the discussion in chapter 2, section 2.1. This setting
allows us to separate the identification of preferences (which can be done
on a subsample of singles) and that of the production function (for which
household level data are needed). Not surprisingly, being able to observe
identical individuals operating under different technologies (that is, as sin-
gle or married) considerably facilitates identification. Browning et al show
that the model can be estimated from the observation of demand functions
for individuals and couples.
A crucial outcome of this approach is the computation of each member’s
indifference scale, defined as the minimum fraction of the household’s in-
come that this member would need to buy (at market prices) a bundle
of privately consumed goods that put her on the same indifference curve
over goods that she attained as a member of the household. Note that this
amount is different (and lower) than what would be needed to purchase,
as a single, the same bundle the member was consuming when married.
Indeed, an obvious effect of the household technology is that the prices
implicitly used within the household may differ from market prices; see
chapter 2, section 2. It follows that even for a given level of expenditures,
the consumption profile of a couple typically differs from that of single
individuals.
3.2.2 Children
Modeling children is a complex issue, and one in which even basic method-
ological choices may be far from innocuous in terms of normative implica-
tions. A general approach relies on two basic ideas. One is that, in general,
parents care about their children. This could take the form of parent s
caring directly about the goods that the child consumes:
¡ ¢
U s = U s Q, qa , qb , qk , ta , tb (3.16)
where ts are the time inputs of the parents and qk denotes the vector of
private consumption by the child. A widely used special case posits the
existence of a child utility function:
³ ´
uk = uk ta , tb , Q, qk (3.17)
Then the preferences of adult s can be defined recursively by:

¡ ¢ ³ ´
U s Q, qa , qb , ta , tb + κs uk Q, qk , ta , tb (3.18)
where κs is the weight that parent s gives to the children.3

Of course, this approach can be used with any number of children. De-
pending on the problem to hand, one may either introduce one sub-utility
per child or only distinguish between broader ‘classes’ (for example, boys
versus girls, younger children versus older ones, etc.). Timing introduces
additional issues, since parents care not only about their children but also
their grandchildren. Barro and Becker (1988) have introduced the concept
of dynastic utilities, whereby parents actually consider the sum of utility
levels of all of their descendents, weighted according to P some ‘distance’
∞ t
factor κs < 1. Then adult s’s utility takes the form us + t=1 (κs ) uk(t) ,
with the convention that uk(1) denotes the utility of s’s children, uk(2) of
his grandchildren, and so on. This model, which relies on restricting (3.16)
to (3.18), may have strong policy implications. For instance, government
subsidies given to children can be completely offset by lower contribution
3A

more general formulation would have utilities of the form us Q, qa , qb , uk .
of their parents without any effect on the final outcome. This type of neu-
trality is often termed Ricardian Equivalence.
It is important to note that in this context, children matter for the house-
hold’s choices, but only through the utility their parents derive from their
well-being. This is a strong assumption, that can be relaxed in two direc-
tions. First, one may, alternatively, consider the child as another decision
maker within the household. In this case a couple with one child would
be considered as a three person household. Whether a child should be
considered as a decision maker or not is a very difficult question, which
may depend on a host of factors (age, education, occupation, income, etc.);
moreover, its empirical translation introduces subtle differences that are
discussed below.
Secondly, throughout this book we stick to a standard practice in eco-
nomics, and we assume that preferences are given, that is, exogenous and
stable. This assumption may be acceptable for adults, but maybe less so
for children; after all, many parents invest time and resources into influ-
encing (or ‘shaping’) their children’s preferences regarding work, risk, or
‘values’ in some general sense. Indeed, a growing literature analyzes the
formation of preferences from an economic viewpoint, as a particular ‘pro-
duction’ process. These contributions are outside the scope of this book;
the interested reader is referred to Becker (1998).
3.3 The unitary model

We now consider how the partners in the household make decisions on how
to spend their time and money. To bring out the main ideas we consider
a context in which there are no time allocation decisions, income is given
and there is no household production. We take the incomes of a and b to
be given at levels Y a and Y b respectively and we assume that there is no
other income into the household. We further assume that household total
expenditure, x, is set equal to household income Y = Y a +Y b , so that there
is no borrowing or lending. The household budget constraint for allocations
is given by:
¡ ¢
P0 Q + p0 qa + qb = x (3.19)
In general the agents will differ on how to spend household income. There
are three broad classes of decision processes: the unitary assumption, non-
cooperative processes and cooperative processes.
The most widely used assumption is that
¡ choices ¢ are made according to
a ‘unitary’ household utility function Ũ Q, qa , qb . In subsection 3.5.6 we
shall investigate when such an assumption is justified, but for now we sim-
ply consider the consequences. One natural assumption, due to Samuelson
(1956), is to impose on the household utility function that it respects the
individual preferences in the sense that:

¡ ¢ ¡ ¡ ¢ ¡ ¢¢
Ũ Q, qa , qb = W U a Q, qa , qb , U b Q, qa , qb (3.20)
where W is a utility weighting function which is strictly increasing in the
individual utilities. The important feature of this weighting function is that
it is fixed and does not vary with prices or income. Given a unitary utility
function we define a household utility function over market goods by:
U (Q, q) = max
a
Ũ (Q, qa , q − qa ) (3.21)
q
Given this household utility function we can derive market demands in the
usual way; namely, it solves the program
max U (Q, q) subject to P0 Q + p0 q ≤ x
(Q,q)
We assume enough on preferences (continuous differentiability, strict con-

cavity), so that this leads to demands for market goods:
Q = Ξ (P, p, x) (3.22)
q = ξ (P, p, x) (3.23)
The unitary assumption has two important sets of implications. First, mar-
ket demand functions satisfy the usual Slutsky conditions; adding-up, ho-
mogeneity, symmetry and negativity of the Slutsky matrix; see, for ex-
ample, Mas-Colell, Whinston and Green (1995), chapter 3. Second, the
demands only depend on prices and total household income, and are inde-
pendent of the distribution of income; that is, they display income pooling.
As we shall see below the latter has been the focus of much testing in the
empirical literature.
3.4 Non-cooperative models

3.4.1 No public goods.
If we are not willing to assume a unitary utility function then we must spec-
ify a decision process. As always there are very many possibilities here but
we shall only explore a small subset of these. We begin with non-cooperative
procedures.4 If household behavior is modeled non-cooperatively, then no
binding agreements between members are assumed and the optimal deci-
sions need not be Pareto efficient. However, in some cases efficiency ob-
tains automatically as an outcome of independent decision making. Take
4 Several authors take the Nash position that any cooperative game should be pre-
ceded by a non-cooperative game. Some of the authors cited in this section only develop
a noncooperative interaction for this purpose.
the simple situation in which preferences are egotistic, and all commodi-
ties are privately consumed. The noncooperative solution boils down to the
following two programs:
max {ua (qa ) subject to p0 qa = Y a }

qa
© ¡ ¢ ª
max ub qb subject to p0 qb = Y b (3.24)
qb
In words, the noncooperative solution simply implies in that case that each
agent chooses his/her level of consumption independently of the other; that
is, they live side by side, but without any economic interaction.5 Then the
consumption of individual s is simply this individual’s demands at prices
p and income Y s . Denote
³ the demand functions
´ for s by ξ s (p, Y s ). Note
¡ ¢
that the allocation ξa (p, Y a ) , ξb p, Y b is Pareto efficient: clearly, the
utility of, say, a can only be increased by an income transfer from b, which
would reduce b’s welfare. Generally the associated household demands
¡ ¢ ¡ ¢
ξ p, Y a , Y b = ξa (p, Y a ) + ξb p, Y b (3.25)
will not satisfy income pooling or the Slutsky conditions. The special case
in which income pooling and the Slutsky conditions will hold is if the classic
aggregation conditions hold. That is, if the two agents have linear Engel
curves with each partner having the same slope for any good:
ξ ai = φai (p) + ϕi (p) Y a

ξ bi = φbi (p) + ϕi (p) Y b (3.26)
so that the household demand for good i is given by:

¡ ¢ ¡ ¢
ξ i p, Y a , Y b = φai (p) + φbi (p) + ϕi (p) Y a + Y b
= φai (p) + φbi (p) + ϕi (p) Y (3.27)
In this very special case income pooling holds in the sense that the house-
hold demands do not depend on the distribution of income. The distribution
of the goods within the household will, however, depend on the distribution
of income.
3.4.2 One public good.

Whenever a direct interaction between members is introduced - either be-
cause of public consumption, or because one member’s consumption has
an external effect on the other member’s well being - inefficiencies are very
5 Of course, this does not preclude the existence of non-economic interactions - love,
sex, conversation or others.

likely to appear. To bring out the essential features of the analysis, let
us assume that there is only one public good and one private good and
that each person has egotistic preferences; see Chen and Wooley (2001)
and Browning, Chiappori and Lechene (2010). Given that we have a public
good and individual incomes a natural, noncooperative process to consider
is a voluntary contributions game in which each person contributes to the
purchase of the public good and then uses any money remaining to buy the
private good for themselves. That is, the two agents have the problems:
© a¡ a b a
¢ a a a
ª
max u Q + Q , q subject to P Q + pq = Y
Qa ,q a
© ¡ ¢ ª
max ub Qa + Qb , q b subject to P Qb + pq b = Y b (3.28)
Qb ,q b
where Qs denotes agent s’s contribution to the public good. Assuming that
both goods are normal, this interaction has exactly one Nash equilibrium,
which can take one of two forms. In the first form, both agents contribute
to the public good. Since this is an interior solution for both we have:
uaQ ³ a
´ P
Q̂, q̂ =
uaq p
ubQ ³ ´ P
b
Q̂, q̂ b = (3.29)
uq p
If we sum the budget constraints we have:

¡ ¢
P Q̂ + p q̂ a + q̂ b = Y a + Y b (3.30)
³ ´
Thus we have three equations in three unknowns Q̂, q̂ a , q̂ b with a solu-
tion:
¡ ¢
q̂ a = ξ a P, p, Y a + Y b
¡ ¢
q̂ b = ξ b P, p, Y a + Y b
¡ ¢
Q̂ = Ξ P, p, Y a + Y b (3.31)
We conclude that the household’s market demand for both the public good,
Q̂, and the private good, q̂ = q̂ a +q̂ b depends only on total household income
Y a + Y b and not on how it is distributed. In other words, we have income
pooling even though we have a non-unitary model. This is an example of
the remarkable neutrality result due to Warr (1983) (see also Bergstrom,
Blume and Varian (1986) and Bernheim (1986)). This shows that while
income pooling is a necessary condition for the unitary model, it is not
sufficient.
It is important to note that income pooling here is a local property and
holds only as long as both partners contribute to the public good. The other
case we have to consider is the one in which only one person contributes.
If this is person a, the first order condition in (3.29) holds for her. Person
b spends all of his income on the private good, so that:
µ ¶
ubQ Yb P
Q̂, ≤ (3.32)
ubq p p
with a strict inequality if the agent is not on the margin of contributing

to the public good. In this case a redistribution of income from a to b will
generally change the market demand since b will increase his demand for
the private good and a generally will not change her demands to exactly
offset these revisions. Thus we have market demands:
Yb Yb
q̂ = ξ a (P, p, Y a ) +
= q̂ a +
p p
a
¡ a b
¢
Q̂ = Q̂ = Ξ P, p, Y , Y (3.33)
In both cases, the non-cooperative procedure leads to an inefficient out-

come (except for the cases in which one or the other has all of the income);
this is the standard under-provision for the voluntary contributions public
goods game. To see that for the case of an interior solution, add the two
first order conditions (3.29), yielding
uaQ ³ a
´ ub ³
Q
´ P
a
Q̂, q̂ + b Q̂, q̂ b = 2 ,
uq uq p
while Samuelson’s (1954) condition for an efficient allocation of public

goods requires that
uaQ ³ a
´ ub ³
Q
´ P
a
Q̂, q̂ + b Q̂, q̂ b = . (3.34)
uq uq p
That is, the sum of the willingness to pay for the public good of the two
partners, should equal to the opportunity cost of the public good in terms
of the private good. In this regard, there is an under provision of the public
good. 6
We now present an example to illustrate some of the points made here.
Normalize prices to unity, P = p = 1, and take preferences represented by
ua = q a Qα and ub = q b Q. The parameter α governs how much a likes the
public good; if α > 1 then she values it more than b if they have the same
private consumption. We set Y a = ρ and Y b = (1 − ρ) so that household
6 Results on dynamic contributions games suggest that inefficiencies cam be elimi-
nated if players contribute sequentially and cannot reduce previous contributions; see,
for example, Matthews (2006).
income is unity and ρ is a’s share of household income. It is straight forward

to show that the decisions of the agents are given by:
µ µ ¶ ¶
a 1 αρ
Q̂ = min max 0, ρ − ,
1 + 2α 1+α
µ µ ¶ ¶
b α (1 − ρ)
Q̂ = min max 0, (1 − ρ) − ,
1 + 2α 2
(with the demands for private goods being given by the budget constraints).
It is easiest to see the implications of this if we graph Q̂ = Q̂a + Q̂b against
a’s share of income, ρ. In figure 3.1 we do this for two values of α, 0.8 and
1.2. There are a number of notable features to figure 3.1. First, if b has all
the income (ρ = 0) then the level of public goods provision corresponds
to his preferred level; here a value of one half. If we now redistribute some
income from b to a we see that the level of the public good falls whether or
not a has a higher valuation for the public good (α ≶ 1). This is because
a uses all of her income for her own private good and b reduces spending
on both the public good and his private good. As we continue shifting
income from b to a the level of the public good falls until at some point a
starts to contribute. The level at which a starts to contribute is lower the
higher is the value of her liking for the public good (compare the curves for
α = 0.8 and α = 1.2). Once both partners are contributing to the public
good, further small transfers from b to a leave all allocations unchanged as
a increases her contributions and b reduces his in an exactly offsetting way
(this is the local income pooling result). At some point b’s income falls to
the point at which he stops contributing. This level of income is lower the
higher is the level of the provision of the public good. After this, transfers
of income from b to a cause a to increase her contribution to the public
good until she has all of the income (ρ = 1).
To illustrate that the level of provision of the public god is inefficiently
low for any value of ρ ∈ (0, 1), consider the case α = 1 and ρ = 0.5. The
equilibrium choices are Q̂ = 1/3 and q a = q b = 1/3. This gives each a utility
level of 1/9. If we instead impose that each contributes 0.25 to the public
good and spends 0.25 on their own private good then each has a utility
level of 1/8, which is a Pareto improvement on the equilibrium outcome.
3.4.3 Altruism and the Rotten Kid Theorem.

An important extension to this analysis is to move beyond the egoistic as-
sumption and to allow for altruism. In this case, if one person has most (or
all) of the income and cares for the other then they may make a transfer
of private goods to the poorer partner as well as being the sole contributor
to the public good. This adds flat segments at the extremes of figure 3.1,
as shown in figure 3.2. In this figure the demand for the public good if a’s
income share is between ρ1 and ρ4 is of the same form as for the egoistic
case of the last subsection. If, however, we have an extreme distribution of

income then the figure changes. For example, if b has most of the income
(ρ < ρ1 ) and cares for a then he will transfer some private good to her and
will be the only contributor to the public good (since ρ2 is the distribution
at which a starts to contribute). In this region we have three important im-
plications. First, there is local income pooling and small re-distributions
³ of
´
income within the household would not change the allocations Q̂, q̂ a , q̂ b .
Second, the outcome is efficient since b is an effective dictator; any other
feasible allocation will make b worse off. Third, the household demands
for private goods (q̂ a (P, p, Y ) + q̂ b (P, p, Y )) and public goods (Q̂ (P, p, Y ))
will satisfy the Slutsky conditions. Note, however, that the range of this
unitary-like behavior and efficiency will depend on the degree of altruism;
as drawn, b cares more for a than a cares for b (the interval [0, ρ1 ] is wider
than the interval [ρ4 , 1]).
In chapter 8 of his revised Treatise of 1991 Becker refers to the unitary
style implications (efficiency, income pooling and the Slutsky conditions) as
the Rotten Kid Theorem (RKT); see also Becker (1974). If one person has
enough income relative to the other and cares for them then they internalize
all decisions and the household behaves as though it is one. A corollary is
that each household member will be motivated to maximize total household
income. For example, if we have ρ < ρ1 and a can take some action that
lowers her income but increases b’s income by more, she will choose to do
it, safe in the knowledge that b will increase the transfer to her sufficiently
to make her better off. This presentation makes it clear that the scope of
the RKT (in this version) is limited; it only applies locally and requires
an extreme distribution of household income and altruism. In subsection
3.5.10 below we present a more general version of the RKT that is closer in
spirit to Becker’s original formulation in Becker (1974). This version widens
the scope at the cost of imposing restrictions on preferences.
3.4.4 Many public goods.

When we turn to the more realistic case with more than one public good,
the important features we saw above persist but some new ones emerge.
The main points can be seen in a model with no altruism, N public goods,
a single private good and prices normalized to unity. The voluntary contri-
butions model has:
© a¡ a ¢ ª
max
a a
u Q + Qb , q a subject to e0 Qa + q a = Y a
Q ,q
© ¡ ¢ ª
max ub Qa + Qb , q b subject to e0 Qb + q b = Y b (3.35)
Qb ,q b
³ ´
where e is an N -vector of ones. Let Q̂s1 , ...Q̂sN for s = a, b be a Nash
equilibrium.7 We say that person s contributes to good j if Q̂sj > 0. Let
ma (respectively, mb ) be the number of goods to which a (respectively, b)
contributes. Browning at al (2010) show that if all public goods are bought
(Q̂sj > 0 for at least one s) then either ma + mb = N or ma + mb = N + 1
(generally). This striking result shows that there is at most one public good
to which both partners contribute.8
To see the result informally, suppose that both partners simultaneously
contribute to two public goods, i and j. Then both set the marginal rates
of substitution between the two goods to unity (the relative prices) and
hence equalize the mrs’s:
uai ub
a = bi (3.36)
uj uj
Without restrictions on preferences and incomes, this is unlikely to hold.
Moreover, if it does hold, if we make an infinitesimal change in Y a or Y b
the property (3.36) will generally not hold.
If there is some overlap in contributions (ma + mb = N + 1) then we
have the local income pooling result, just as in the one public good case
when both contribute. The result that each partner has a set of public
goods which are his or her ‘domain’ suggests a gender division of allocation
within the household. Note, however, that the goods that each takes as
their domain is determined endogenously by preferences and the division
of income within the household. As we move from b having all the income
to a having all the income (holding total income constant) the number of
goods that she contributes to will generally rise and the number of goods
to which he contributes will generally fall.
We illustrate with an example with egoistic preferences from Browning
et al (2010) for the case of two public goods, G and H. Let the two partners
have preferences represented by the pair of Cobb-Douglas utility functions
5 8
ua (q a , G, H) = ln q a +
ln G + H
3 9
b b b 15 1
u (q , G, H) = ln q + ln G + ln H
32 2
The relative weights on the two public goods are 45 15
24 and 16 for a and
b respectively; that is, a likes good G relative to good H, more than b.
Figure 3.3 shows the purchases of public goods against a’s share of income.
When a has a low share of income (region I on the x-axis) she does not
7 We assume enough to ensure the existence of at least one Nash equilibrium. We do
not impose uniqueness.

8 This result is generic in the sense that it is possible to find ‘knife-edge’ configurations
of preferences and incomes for which the two partners contribute to more than one
common public good.
contribute to either public good (ma = 0 and mb = 2). In this region,

increases in a’s income share lead her to spend more on the private good
and lead b to spend less on both public goods. If her income is increased to
region II then she starts contributing to one of the public goods (good G
in this case) and he continues contributing to both (ma = 1 and mb = 2);
this is a region of income pooling. As we continue to take income from b
and give it to a we move to region III where she contributes to one public
good and he contributes to the other (ma = 1 and mb = 1). This is again
a region in which the distribution of income matters (locally). Regions IV
and V are analogous to regions II and I, with a and b exchanged. One
feature to note about this model is the point at which flat segments begin
is the same for the two goods (and household expenditures on the private
good).
Lundberg and Pollak (1993) introduce a model that is similar to the
many public goods version above, except that they restrict contributions
to exogenously given sets of public goods for each partner, which they
term separate spheres.9 They have two public goods and assume that each
partner has a public good to which they alone can contribute; this is their
‘sphere’ of responsibility or expertise. These spheres are determined by
social norms; this is the principal difference from the model developed in
this subsection in which the ‘sphere’ of influence depends on preferences and
the distribution of income within the household and is hence idiosyncratic
to each household.
3.5 Cooperative models: the collective approach

The main problem with non-cooperative procedures is that they typically
lead to inefficient outcomes. In a household context this is a somewhat
unpalatable conclusion. If each partner knows the preferences of the other
and can observe their consumption behavior (the assumption of symmetric
information) and the two interact on a regular basis then we would expect
that they would find ways to exploit any possibilities for Pareto improve-
ments. This does not preclude the existence of power issues; as we shall see,
the notion of ‘power’ plays a crucial role (and has a very natural interpreta-
tion) in cooperative models. The cooperative approach does recognize that
the allocation of resources within the household may (and generally) will
depend on the members’ respective ‘weights’; it simply posits that however
resources are allocated, none are left on the table.
There are various ways of modeling cooperative behavior. In what fol-
lows, we mainly explore the implications of the only restriction common to
all cooperative models, namely that household decisions are Pareto efficient,
9 In section 5 below we give a fuller account of the separate spheres model.

in the usual sense that no other feasible choice would have been preferred
by all household members. This approach was originally suggested by Chi-
appori (1988, 1992) and Apps and Rees (1988). Following Chiappori, we
refer to such models as collective and refer to households that always have
Pareto efficient outcomes as collective households. More specific represen-
tations, based on bargaining theory, are briefly discussed at the end at this
section. In the remainder of this chapter we briefly introduce the collective
model. Chapters 4 and 5 expand on this discussion.
The collective approach relies on two fundamental assumptions. First,
there exists a decision process in the household and it is stable. Second,
this process leads to Pareto efficient outcomes. We discuss these aspects
successively.
3.5.1 Decision processes

A fundamental assumption in unitary demand theory is that individual
preferences are stable, in the sense of not changing capriciously from mo-
ment to moment. This is not a logical requirement; in principle, the world
could be such that people are intrinsically inconsistent, and a person’s pref-
erences today are unconnected with those of yesterday. Clearly, in such a
world, very little could be say about individual behavior: a minimum level
of stability is necessary if we wish to make predictions based on our models.
The same requirement applies to any model aimed at describing the
behavior of a group. The notion of stability, in that case, must be given a
broader interpretation: it relates not only to preferences, but also to the
decision process. Again, the world could be such that a given household,
faced with the same environment in different time periods, adopts each time
a different decision process leading to different outcomes. And again, in such
a world not much could be predicted about household behavior. We rule
out these situations by assuming the existence of a stable decision process.
Formally, we define the fundamentals of the model as the preferences of the
members and the domestic technologies they can use. A decision process
is a mapping that associates, to given fundamentals and given vectors of
prices, incomes and factors that affect preferences and the decision process,
a probability distribution over the set of consumption bundles. Our first
basic assumption is thus the following:
Axiom 3.1 (Stability) Each household is characterized by a unique de-

cision process.
In words: there is a stable relationship between the fundamentals of the

model, the economic environment and the chosen outcomes. Note that, in
full generality, this relationship needs not be deterministic. It may be the
case, for instance, that in some circumstances the process will lead to ex-
plicit randomization.10 What the stability axiom requires in that case is

that the randomization be stable, in the sense that, keeping the funda-
mentals fixed, a household faced with the same economic environment will
always randomize in exactly the same way (that is, using the same proba-
bility). Nevertheless, in what follows we essentially consider deterministic
decision processes.
An important remark, however, is that while household members can
observe all the factors influencing the decision process, the econometrician
analyzing their behavior may not have such luck. If some determinants of
the process are not observed, we are in a classical situation of unobserved
heterogeneity. Then the model may (and typically will), for empirical pur-
poses, involve probability distributions (of unobserved heterogeneity), even
when the decision process in each household is purely deterministic. The
corresponding randomness should not be considered as intrinsic; it simply
reflects the econometrician’s inability to observe all the relevant variables.
Clearly, the stability axiom is not specific to the collective approach; any
model of group behavior must assume some kind of stability to be able
to make predictions. Most of the time, the stability is implicit in the for-
mulation of the model. For instance, in the unitary framework, a unique
utility is maximized under a budget constraint; the outcome is the solution
to the maximization problem. Similarly, in noncooperative models based
on Nash equilibria, the decision process selects, for given fundamentals and
a given environment, the fixed point(s) of the best response mapping. In
the collective approach, one way to justify the stability axiom could be to
assume that the household uses a well specified bargaining protocol, which
can be cooperative (Nash, Kalai-Smorodinsky) or noncooperative (for ex-
ample, Rubinstein’s ‘shrinking cake’ model). In all cases, the concept under
consideration exactly pins down the particular outcome of the negotiations
as a function of prices, incomes and factors which influence the bargain-
ing game (for example, the status quo points). But, of course, assuming
bargaining is by no means necessary for stability.
3.5.2 Assuming efficiency

The second key assumption of the collective approach is that the outcomes
of the decision process are always efficient, in the (usual) sense that no
alternative decision would have been preferred by all members. The effi-
ciency assumption is standard in many economic contexts, and has often
been applied to household behavior. For instance, the axiomatic concepts
of bargaining used in cooperative game theory typically assume efficiency,
1 0 For instance, a basic conclusion of second best theory is that in the presence of non
convexities, randomization may be needed to achieve Pareto efficiency. See Chiappori

(2009) for an application to collective labor supply.
and noncooperative models of bargaining under symmetric information typ-

ically generate Pareto efficient outcomes. Among the alternative approaches
that have been proposed in the literature, many, from simple dictatorship
(possibly by a ‘benevolent patriarch’, as in Becker, 1974) to the existence of
some household welfare function (as in Samuelson, 1956) assume or imply
Pareto efficiency. In the same line, the ‘equilibrium’ approaches of Becker
(1991) and Grossbard-Schechtman (1993), in which household members
trade at existing market prices, typically generate efficient outcomes.
Natural as it seems for economists, the efficiency assumption neverthe-
less needs careful justification. Within a static context, this assumption
amounts to the requirement that married partners will find a way to take
advantage of opportunities that make both of them better off. Because
of proximity and durability of the relation, both partners are aware of
the preferences and actions of each other. They can act cooperatively by
reaching some binding agreement. Enforcement of such agreements can be
achieved through mutual care and trust, by social norms and by formal
legal contracts. Alternatively, the agreement can be supported by repeated
interactions, including the possibility of punishment. A large literature in
game theory, based on several ‘folk theorems’, suggests that in such situa-
tions, efficiency should prevail.11
There are, however, two situations (at least) in which the efficiency as-
sumption may fail to apply. One is when existing social norms impose pat-
terns of behavior that may conflict with efficiency. One example for appar-
ent inefficiency is when, due to the traditional norms, the wife is expected
to stay at home and the husband to work in the market. Although such a
division of labor may have been efficient in the past, it certainly conflicts
with efficiency in modern societies in which women are often more educated
than their husband. Another example is the finding by Udry (1996) that
households in Burkina-Faso fail to allocate inputs efficiently among various
crops because of the socially imposed division of labor between genders,
which implies that some crops can only be grown by a particular gender.12
Secondly, some decisions are taken only once (or a few times), which
implies that the repeated game argument does not apply; see Lundberg and
Pollak (2003). Deciding whether to engage in a long training program or to
move to another city, are not frequent decisions. The kind of ‘equilibrium
punishments’ that are needed to implement efficient outcomes in repeated
games may then be unavailable. The main theoretical insight here is that for
1 1 Note, however, that folks theorems essentially apply to infinitely repeated interac-
tions.
1 2 A program of research in economics tries to explain existing institutions (including
social norms) as an efficient response to a particular context; the argument being that
competition will tend to promote the ‘best performing’ institutions. However, when
technology changes deviations from efficiency can arise, because social norms may change
slowly.
medium or long-term decisions, efficiency requires commitment; conversely,

any limitation of the members’ ability to commit may result in inefficient
outcomes. As we know, commitment within a household is only partial;
for instance, it is impossible to commit not to divorce, although marriage
contracts can be used to make divorce costly for one or both of the members.
For that reason, we will treat the issue of efficiency in a different manner
depending upon whether we deal with a dynamic or static context. In most
of the first half of this book the setting is static and efficiency is assumed.
However, in chapters 6, 7, 8 and 12, which discuss saving and investment,
we allow departures from efficiency.
In the remainder of this Chapter, we investigate the properties of mod-
els based on the collective approach. Before entering the technical analysis,
however, one point should be stressed. The great advantage of the collective
model is that we do not have to specify the (stable) mechanism that house-
holds use, but only assume that such a mechanism exists. In other words,
the collective strategy does not make assumptions about decision processes
(beyond efficiency); it rather lets the data tell what these processes are.
This naturally leads to two crucial questions. One is whether the efficiency
assumption is testable; that is, whether empirically falsifiable predictions
can be derived from it. The second question relates to identification: when
(and under which assumptions) is it possible to recover the decision process
from observable behavior? Answering these questions is a strong motivation
for what follows.
3.5.3 Distribution factors

The generality of the collective approach comes at a cost. While the effi-
ciency assumption restricts the set of possible allocations, we are still left
with ‘many’ of them - a continuum, in fact. If we want to make more specific
predictions on household behavior, additional information - more precisely,
additional sources of variations - may be useful. Among the various factors
that can influence household behavior, many have a ‘direct’ impact on ei-
ther preferences or the budget constraint. A change in prices, wages and
non labor income are likely to affect demands and labor supplies, simply
because they modify the set of options available. A more subtle influence
goes, indirectly, through the decision process. A change in the economic en-
vironment may not affect either the preferences or the budget opportunities
but still have an impact on the decision process. This idea is incorporated
into the collective model by introducing distribution factors. Any variable
that has an impact on the decision process but affects neither preferences
nor budget constraints is termed a distribution factor. In theory, a large
number of variables fit this description. Factors influencing divorce, either
directly (for example, the legislation governing divorce settlements and al-
imony payments) or indirectly (for example, the probability of remarriage,
which itself depends on the number of available potential mates - what
Becker calls the ‘marriage market’) should matter, at least insofar as the
threat (or the risk) of divorce may play a role in the decision process. Indi-
viduals’ income or wealth can also be used as distribution factors. Suppose,
for example, that earned and unearned income is given for any individual
and let Y s denote the total income of person s. Then total household in-
come, given by Y = Y a + Y b , is all that matters for the budget constraint.
For any given level of Y , the individual contribution of a to total income,
measured for instance by the ratio Y a /Y , can only influence the outcome
through its impact on the decision process; it is thus a distribution factor.
In the collective framework, changes in distribution factors typically lead
to variations in outcomes while the set of efficient allocations remains un-
changed; as such, it provides very useful information on the decision process
actually at stake in the household. For that reason, it is in general crucial
to explicitly take then into account in the formal model. In what follows,
therefore, z denotes a vector of distribution factors.
3.5.4 Modeling efficiency

The basic framework
The characterization of efficient allocations follows the standard approach.
The basic definition is that an allocation is Pareto efficient if making one
person better off makes the other worse off:
¡ ¢
Definition 3.1 An allocation Q, qa , qb is Pareto efficient if any other
allocation (q̄a , q̄b , Q̄) that is feasible:
0¡ ¢ ¡ ¢
P0 Q̄ + p q̄a + q̄b ≤ P0 Q + p0 qa + qb
¡ ¢ ¡ ¢ ¡ ¢
and¡ is such that
¢ ua Q̄, q̄a , q̄b > ua Q, qa , qb , it must be such that ub Q̄, q̄a , q̄b <
ub Q, qa , qb (and conversely).
In practice the basic definition is not very tractable and we often use one
of two alternative characterizations. A first characterization is:
Definition 3.2 For any given vector (P, ¡ p,x, z)¢of prices, total expenditure
and distribution factors, an allocation Q̄, q̄a , q̄b is Pareto efficient¡ if there¢
exists a feasible ūa , which may depend on (P, p,x, z), such that Q̄, q̄a , q̄b
solves the problem:
¡ ¢
max ub Q, qa , qb (3.37)
Q,qa ,qb
¡ ¢
subject to P0 Q + p0 qa + qb ≤ x (3.38)
a
¡ a b
¢ a
and u Q, q , q ≥ ū (P, p,x, z) (3.39)
Thus the Pareto efficient allocation can be derived from maximizing the
utility of one partner holding the utility of the other at a given level: among
all allocations providing a with exactly ūa , the efficient one(s) generate the
maximum possible utility for b. It goes without saying that this approach
- just like most microeconomics - should not be taken literally. No one
believes that agents actually write and solve a program such as (3.37). Our
claim is simply that when a decision process, whatever its exact nature,
always lead to efficient outcomes, then for any choice of prices, income and
distribution factors there exists a ūa such that the household behaves as if
it was solving program (3.37).
The solution to (3.37), when it exists (that is, if ūa is feasible), depends on
prices, total expenditure and on the arbitrary level ūa ; it can be denoted
as ub = Υ (P, p, x, ūa ). The set of all pairs (ūa , Υ (P, p, x, ūa )) when ūa
varies over a domain of feasible allocations for a is the set of all efficient
allocations; it is known as the Pareto frontier or utility possibility frontier,
UPF. Under the assumption that the utility functions are strictly concave
it is straight forward to show that the function Υ (.) is strictly concave in
ūa (see below). This allows us to write Program (3.37) in a different but
equivalent way. Let μ denote the Lagrange multiplier for constraint (3.39);
note that μ is always nonnegative (and is a function of (P, p,x, z)). Then
the program is equivalent to:
¡ ¢ ¡ ¢
max μua Q, qa , qb + ub Q, qa , qb (3.40)
Q,qa ,qb
under the constraint (3.38). The coefficient μ is known as the Pareto weight
for a. That is, a Pareto efficient outcome always maximizes a weighted sum
of the two individual utilities. A slightly modified form that keeps the
formal symmetry of the problem is sometimes used:
¡ ¢ ¡ ¢
max μ̃ua Q, qa , qb + (1 − μ̃) ub Q, qa , qb , (3.41)
Q,qa ,qb
where μ̃ ∈ [0, 1]. The Pareto weight plays a critical role in all that follows.
Finally, an equivalent formulation directly generalizes Samuelson’s
¡ house-¢
hold welfare. Specifically, take any smooth function W ua , ub , P, p,x, z
that is strictly increasing in its first two arguments, and consider the pro-
gram:
¡ ¡ ¢ ¡ ¢ ¢
max W ua Q, qa , qb , ub Q, qa , qb , P, p,x, z (3.42)
Q,qa ,qb
under the constraint (3.38). Clearly, a solution to (3.42) is Pareto efficient;

for otherwise some alternative allocation would increase both ua and ub ,
one of them (at least) strictly, but that would strictly increase W , a con-
tradiction. Conversely, any allocation that is Pareto efficient maximizes a
weighted sum of the form (3.40), which is a particular (linear) case of a
W function. This establishes equivalence: an allocation is Pareto efficient if
and only if there exists some W such that it maximizes (3.42) under budget
constraint.
This setting generalizes Samuelson because the welfare index W depends

on individual utilities, but also directly on prices, total expenditure and dis-
tribution factors. In other words, the household maximizes an index that is
price (and income) dependent, which distinguishes this setting from a uni-
tary representation. The surprising property is that under tsrict concavity
one can assume without loss of generality that the index W is indeed linear
(as in (3.40)). We shall come back to this issue below.
Ordinal versus cardinal representation of preferences

It is important to understand what, in the previous discussion, requires a
cardinal representation of preferences, and what can be defined using only
a standard, ordinal representation. The set of Pareto efficient allocations is
an ordinal concept; it is not modified when us is replaced with F (us ) for a
strictly increasing mapping F (.). Under smoothness conditions, the set is
one-dimensional, and therefore can be described by one parameter. How-
ever, the parametrization entails cardinality issues. For instance, a natural
parametrization is through the weight μ. But μ depends on the particu-
lar cardinal representation that has been adopted for ua and ub : if us is
replaced with F (us ), then the parameter μ corresponding to a particular
efficient allocation has to be modified accordingly. Moreover, the convex-
ity properties of the Pareto set are also of a cardinal nature. Assuming
smooth preferences, for any given price-income vector, one can find cardi-
nal representations of preferences such that the Pareto frontier is convex,
linear or concave. In most of what follows, we adopt the convention of al-
ways using a strictly concave representation of utilities. In this case, the
Pareto set ¡is strictly¢ convex.
¡ 0a Indeed,
¢ for a given price-income vector,
¡ take¢
a b 0b a
two points ū
¡ 0 0a 0b ¢ , ū and u , u on the Pareto frontier, and let Q, q , qb
and Q , q , q be the corresponding consumption vectors. The vector
¡ 00 00a 00b ¢ 1 ¡ ¢ 1 ¡ 0 0a 0b ¢
Q ,q ,q = Q, qa , qb + Q ,q ,q
2 2
satisfies the budget constraint, and by strict concavity,
¡ ¢ 1 ¡ ¢ 1 ¡ ¢ 1 1
us Q00 , q00a , q00b > us Q, qa , qb + us Q0 , q0a , q0b = ūs + u0s
2 2 2 2
¡ ¢ ¡ ¢
for s = a, b. We conclude that the point 12 ūa , ūb + 12 u0a , u0b belongs to
the interior of the Pareto set.
Graphically, on Figure 3.4, the Pareto set is indeed strictly convex. We
see that any point on the UPF can be defined either by its coordinate on the
horizontal axis, here ua , as in program (3.37), or by the negative of the slope
of the Pareto frontier at that point, here μ as in program (3.40). Given that
the UPF is strictly concave there is an increasing correspondence between
ūa and μ: a larger ūa (or μ) corresponds to an allocation that is more
favorable to a (hence less favorable for a). Note that the correspondence
between ūa and μ is one-to-one; that is, for any ūa , there exists exactly
one μ that picks up the efficient point providing a with exactly ūa , and
conversely for any μ there is only one allocation that maximizes (3.40)
under budget constraint, therefore only one corresponding utility level ūa .
We can also understand from ¡ Figure 3.5 why ¢ the maximization of gen-
eralized Samuelson index W ua , ub , P, p,x, z is equivalent to that of a
linear combination μua + ub . The maximization of a non linear index W
will select a point where the Pareto frontier is tangent to some indifference
curve of W . If −μ denotes the slope of the corresponding tangent, maximiz-
ing μua + ub leads to exactly the same point. Replacing W with its linear
equivalent can be done at any point, provided that μ varies adequately;
technically, this simply requires that:
∂W/∂ua
μ=
∂W/∂ub
The main drawback of the generalized index version is that a contin-

uum of different welfare indices lead to the same choices. Indeed, for any
function F strictly increasing in its first argument, the indices W and
W̄ = F (W, P, p,x, z) are empirically indistinguishable. The linear version,
from this perspective, has an obvious advantage in terms of parsimony; in
addition, it has a natural interpretation in terms of distribution of powers
(see below).
Finally, we may briefly discuss two particular cases. One obtains when
the cardinal representations of utilities are concave but not strictly concave.
In that case, the UPF may include ‘flat’ (that is, linear) segments (Figure
3.6). Then Program (3.37) is still equivalent to Program (3.40), but the
relationship between ūa and μ is no longer one-to-one. It is still the case
that for any ūa , exactly one μ picks up the efficient point providing a with
ūa . But the converse property does not hold; that is, to some values of μ are
associated a continuum of utility levels ūa ; graphically, this occurs when −μ
is exactly the slope of a flat portion of the UPF.13 This case is particularly
relevant for two types of situations, namely transferable utility on the one
hand (then the cardinalization is usually chosen so that the entire UPF is
a straight line) and explicit randomization on the other hand.
The second particular case relates to local non differentiability of utility
functions (Figure 3.7). Then the UPF may exhibit a kink, and the one-to-
one relationship breaks down for the opposite reason - namely, many values
of μ are associated with the same ūa .
1 3 However, a strictly quasi concave generalized welfare index would still pick up exactly
one point.
Stability and the Pareto weight

In what follows, we concentrate on deterministic decision processes. Then
the stability axiom has a very simple implication - namely that in pro-
gram (3.37), the coefficient ūa is a well-defined function of prices, income
and possibly distribution factors, denoted ūa (P, p, x, z). It follows that,
for given fundamentals and price-income bundle, the outcome of the de-
cision process is such that the utility of a is ūa (P, p, x, z), and that of b
is Υ (P, p, x, ūa (P, p, x, z)). Note that under strict quasi-concavity, these
utility levels are reached for only one consumption bundle.
If, in addition, we adopt a strictly concave cardinalization of individual
utilities, then the Pareto weight is also a well-defined function of prices,
income and possibly distribution factors, denoted μ (P, p, x, z). For analytic
tractability, we often add some structure to the problem by assuming that
the function μ has convenient properties such as continuous differentiability.
Such assumptions will be stated wherever they are needed.
In summary: under our two assumptions of stability and efficiency, us-
ing a strictly concave cardinalization of preferences, the behavior of the
household can be modelled in a simple way; that is, there exists a function
μ (P, p, x, z) such that the household solves:
¡ ¢ ¡ ¢
max μ (P, p, x, z) ua Q, qa , qb + ub Q, qa , qb (3.43)
under the budget constraint (3.38).
3.5.5 Pareto weights and ‘power’

A major advantage of the formulation in (3.40) or (3.43) is that the Pareto
weight has a natural interpretation in terms of respective decision powers.
The notion of ‘power’ may be difficult to define formally, even in a simplified
framework like ours. Still, it seems natural to expect that when two people
bargain, a person’s gain increases with the person’s power. This somewhat
hazy notion is captured very effectively by the Pareto weights. Clearly, if
μ in (3.40) is zero then it is as though b is a dictator, while if μ is large
then a effectively gets her way. A key property of (3.40) is precisely that
increasing μ will result in a move along the Pareto set, in the direction of
higher utility for a (and lower for b). If we restrict ourselves to economic
considerations, we may thus consider that the Pareto weight μ ‘reflects’ a’s
power, in the sense that a larger μ corresponds to more power (and better
outcomes) being enjoyed by a.
The empirical implications of this remark are quite interesting. For in-
stance, when a reform is known or predicted to improve the relative sit-
uation of a particular member (say, paying some family benefits to the
wife instead of the husband), we should find that the reform increases the
member’s Pareto weight. More generally, the intuitive idea that a specific
distribution factor z is favorable to member a can readily be translated by
the fact that μ is increasing in z. Conversely, we shall see later on that it

is sometimes possible to recover the Pareto weights from a careful analysis
of the behavior of the households at stake. Then one can find out which
factors increase or decrease the power of each member - quite a fascinating
outcome indeed.
Another important insight of the analysis is that, broadly speaking, co-
operation does not preclude conflict. In other words, the Pareto efficiency
assumption by no means implies that the members always agree on what
to do. On the contrary, each agent will plausibly try to obtain a favorable
Pareto efficient outcome. In other words, who gets what is a crucial but dif-
ficult and potentially conflictual issue, that the efficiency assumption leaves
totally open. It can be resolved in a number of different ways - bargaining,
legally binding contracts, tradition, social norms or less formal ways that
reflect the feelings of the two partners towards each other. Pareto efficiency
does not preclude any of these aspects; it just imposes that whichever so-
lution is found, no money is ultimately left on the bargaining table. In a
sense, the collective approach provides the tools needed to concentrate on
the interesting issue of who gets what - or, technically, what do the Pareto
weights look like as functions of prices, income and distribution factors.
3.5.6 Household utility

If the Pareto weight is not a function of prices and income, then we have a
unitary model and we can define a household utility function as a function
of household public and private goods. It turns out that for the collective
model we can also define a household utility function over household pur-
chases of public and private goods but this function has one extra argument
as compared to the unitary model. We define the household utility function
by:
uh (Q, q, μ (P, p, x, z)) =

© ¡ ¢ ¡ ¢ª
max μ (P, p, x, z) ua Q, qa , qb + ub Q, qa , qb
qa ,qb
subject to qa + qb = q (3.44)
With this definition of the household utility uh , program (3.40) is equiv-

alent to the maximization of uh under the budget constraint. This looks
a lot like standard utility maximization in a unitary model. However, the
critical feature of this household utility function is that it depends on the
Pareto weight μ (P, p, x, z). This remark is important for two reasons. First,
it explains why an efficient household needs not (and will not in general)
behave like an individual: since the utility uh is price-dependent, the de-
mand derived from its maximization under budget constraint needs not
(and will not in general) satisfy the standard conditions of consumer the-
ory. Secondly, while the idea of introducing prices into the utility function
is an old one, the important feature in our case is how it is done. Follow-
ing standard demand theory we do not allow prices to enter the individual
utility functions; prices and income can only affect the respective weights
given to individual utilities. As we shall see below, this gives very specific
predictions for household demands. Additionally, it makes analysis using
a collective model almost as easy as using a unitary model which is an
important consideration when considering non-unitary alternatives.
This approach allows us to decompose changes in the utility levels of the
two partners following a change in the environment into changes that would
follow in a unitary model and the additional effect due to the collective
framework. This is illustrated in figure 3.8, where we ignore distribution
factors. Here we consider a change in prices and incomes that moves the
UPF from U P F (P, p, x) to U P F (P0 , p0 , x0 ). Initially the point I is chosen
on U P F (P, p, x). If we hold μ constant when prices and income change
(the unitary assumption) then the utility levels move to point II at which
point the tangent to U P F (P0 , p0 , x0 ) is parallel to the tangent at point I
on U P F (P, p, x). However, a change in the economic environment may
also lead to a change in the Pareto weight. This is the ‘collective’ effect,
illustrated by the move around U P F (P0 , p0 , x0 ) from II to III.
Finally, the collective formalization provides a natural way of introducing
distribution factors within the framework of household decision process. If
some distribution factors z influence the process by shifting the individual
weights, then μ will depend on these variables. The fact that distribution
factors matter only through their impact on μ plays a key role in the results
of Chapter 4. As we shall show, efficiency can be tested using cross equa-
tion restrictions that arise from the fact that the same function μ (P, p, x, z)
appears in the demand for all goods. Moreover, there is an important dif-
ference between prices and total income, on the one hand, and distribution
factors on the other hand. A change in prices or total income will affect
not only the weight μ, but also the shape of the Pareto set; hence it final
impact on individual welfare may be difficult to assess. On the contrary, a
change in a distribution factor can by definition only influence the weight
μ. In general, its effect on welfare is not ambiguous. In terms of figure 3.8
a distribution factor shifts the tangent point but not the frontier itself.
As an illustration of this point, we may briefly come back to the exam-
ple discussed in subsection 3.4.2 of the impact of individual incomes Y a
and Y b on household behavior. From a collective perspective, this impact
should be decomposed into two components. One is the resulting change
in total income Y = Y a + Y b (hence of total expenditures x in our sta-
tic framework); this affects the shape of the Pareto frontier as well as the
weight μ, and its effect is a priori ambiguous. The second component is
the change in relative incomes, say z = Y a /Y b , keeping the sum constant.
The latter should be analyzed as a variation of a distribution factor, and
its consequences are much easier to assess. For instance, if we assume, as is
natural, that increasing the relative income of a increases a’s weight, then it
must increase a’s welfare. However, how this improvement in a’s situation
will be translated into observable household behavior (for example, which
consumptions will increase) is a difficult issue, for which a more precise
formalization is needed.
3.5.7 Caring
The way in which partners care about each other may also affect the Pareto
utility frontier. To take a simple example, consider the caring preferences
introduced in section 3.1:
¡ ¢ ¡ ¢
U a Q, qa , qb = ua (Q, qa ) + δ a ub Q, qb
¡ ¢ ¡ ¢
U b Q, qa , qb = ub Q, qb + δ b ua (Q, qa ) (3.45)
The maximand is now

£ ¡ ¢¤ ¡ ¢
μ ua (Q, qa ) + δ a ub Q, qb + ub Q, qb + δ b ua (Q, qa )
³ ´ ¡ ¢
= μ + δ b ua (Q, qa ) + (1 + μδ a ) ub Q, qb
Since (1 + μδ a ) > 0, we can then represent household preferences by:

¡ ¢
μ̃ua (Q, qa ) + ub Q, qb (3.46)
where
μ + δb
μ̃ =
1 + μδ a
Formally, (3.46) is identical to the egotistic case (δ a = δ b = 0), indicating
that any allocation that is Pareto efficient for the caring preferences is
also Pareto efficient for the egotistic ones. The argument underlying this
conclusion is quite general, and goes as follows: if an allocation fails to be
efficient for egotistic preferences, there exist another allocation that entails
higher values of both ua and ub . But then it also entails higher values of both
U a and U b , showing that the initial allocation was not efficient for caring
preferences as well. In other words, the Pareto set for caring preferences
is a subset of the Pareto set for egotistic preferences. Note, however, that
the two sets do not coincide: some allocations may be efficient for egotistic
preferences, but not for caring ones. Indeed, an allocation that gives all
resources to one member may be efficient for egotistic agents, but not for
caring persons - a redistribution in favor of the ‘poor’ member would then
typically be Pareto improving. Technically, when μ varies from 0 to infinity,
μ0 only varies from δ b to 1/δ a , and the new Pareto set is a strict interior
subset of the initial one.
One important feature of (3.46) is that if b’s caring for a increases (giving
an increase in δ b ) then it is as though a’s Pareto weight increases (and
conversely). This is entirely reasonable: increased caring gives the other

partner a greater weight with respect to egoistic preferences.
A variant of this is if the two partners care for each other in the following
way:
¡ ¢ © ¡ ¢ª
U a Q, qa , qb = min ua (Q, qa ) , ub Q, qb
¡ ¢ © ¡ ¢ª
U b Q, qa , qb = min ua (Q, qa ) , ub Q, qb (3.47)
This formalizes the maxim that ‘no man can be happier than his wife’. In
this very special case the utility possibility frontier shrinks to a single point
at which ua = ub .
3.5.8 Children
Finally, let us briefly come back to the distinction sketched above between
children being modeled as public goods or genuine decision makers. In the
first case, using parental utilities of the form us + κs uk described above,
the maximand in (3.40) becomes
¡ ¢ ¡ ¢
μ ua + κa uk + ub + κb uk
which is equivalent to
1 £ a ¡ ¢ ¤
a b
μu + ub + μκa + κb uk (3.48)
1 + μ + μκ + κ
the initial fraction in (3.48) gives a normalization that the weights sum to
unity.
Alternatively, we may model the child as a decision maker. Then (s)he
is characterized by an additional Pareto weight, say μk , and the household
maximizes the weighted sum:
¡ ¢ ¡ ¢
μua Q, qa , qb + ub Q, qa , qb + μk uk (3.49)
Although the two forms (3.48) and (3.49) look similar, they are, in fact,
quite different. Recall that the key insight of collective models is that Pareto
weights may depend on prices, wages, incomes and distribution factors, and
that this fact explains why collective households do not generally behave as
unitary ones. In (3.48) all Pareto weights are defined by the knowledge of
the function μ ; in (3.49), however, μ and μk can be defined independently,
and the location of the final outcome on the Pareto frontier now depends on
two parameters. Broadly speaking, the deviation from the unitary model
is one dimensional in the first case (it is summarized by a unique function
μ) whereas it is two-dimensional in the second case. As it turns out, this
distinction has testable implication; that is, we shall see later on that a
household with three decision makers does not generally behave as a couple,
pretty much in the same way as couples do not generally behave as singles.
Another fascinating implication is that, in principle, one can assess the
number of actual decision makers in a household from the sole examination
of the household’s behavior, even in a fairly general context!
3.5.9 The unitary model as a special case

It is clear, from the discussion above, that the unitary model is a special
case of the collective framework. An obvious example obtains when the
household utility (or equivalently the Pareto weight μ) does not depend
directly on prices, incomes and distribution factors. As a matter of fact,
the unitary assumption is far and away the most common assumption in
the modelling of household decisions. This, however, is certainly due to its
very great convenience, rather than any intuitive plausibility. If one is to
take seriously the idea of a decision process actually taking place between
the members, it hard to believe that neither prices (including respective
wages), nor respective incomes, nor any exterior factor will influence the
‘weights’ of individual agents in the decision process.
Nonetheless there are circumstances under which the household will act
as though it has a single utility function. One obvious example is if custom
or strong social traditions give all the power to one person (usually the
husband) in the household.
An alternative is given in Samuelson (1956). Samuelson considers the
family to be the fundamental unit on the demand side of the economy.
However, because such a unit consists of several members, we cannot expect
a consensus (that is, consistent family indifference curves). He recognizes
that preferences within a family are interrelated and that external con-
sumption effects (a la Veblen and Dusenberry ) are the “essence of family
life”. Nevertheless, if such external effects are put aside, and a restricted
form of altruism is assumed, families may behave as if they maximize a
single social utility. In particular Samuelson considers a common social
welfare function for the family that is restricted to depend on the individ-
ual consumptions of family members only through the preferences of these
members. This restriction, together with the assumption of no external
consumption effects and no public goods, implies that all family decisions
can be decentralized via a distribution of income.14 The important point
is the distribution of income depends on prices and income and will not be
constant. Thus schemes such as ‘a receives 60% of income and b receives
40%’ are generally not consistent with the maximization of a family SWF.15
1 4 If we have public goods and externalities then we also need Lindahl prices and
Pigouvian taxes to decentralize.

1 5 Although it may in special cases (e.g., Cobb Douglas preferences).
The main result that Samuelson provides is that if income is redistributed

so as to maximize a given social welfare function, then the family aggregate
consumptions will satisfy the Slutsky conditions. That is a family will act
in the same manner as a single person.
Becker (1991) criticizes Samuelson for not explaining how a social welfare
function arises. In the context of moral judgements, each person can have
a private utility that is defined on outcomes affecting them directly, and a
social utility function that reflects preferences on the outcomes for all family
members. So it is unclear how partners agree on a single common social
welfare function. One mechanism suggested by Becker is that one person,
the ‘head’, has most of the family resources and is sufficiently altruistic
that they will transfer resources to the other member. If the dependents’
consumption is a normal good for the head, all family members will align
their actions with the head’s preferences, as any improvement in the income
under the command of the head raises their utilities. It is then the case
that the family as a group acts as if a single objective is being maximized.
This is the Rotten Kid Theorem mechanism outlined in subsection 3.4.3.
In that noncooperative voluntary contributions model, one of the partners
may effectively be a dictator if they control most (but not necessarily all)
of household resources. In that case a unitary model obtains locally if one
partner is wealthier and they are the sole contributor to the public good.
Another important case is when the preferences display transferable util-
ity (TU); see subsection 3.1. Indeed, under TU, program (3.40) becomes
¡ ¡ ¢ ¢ ¡ ¡ ¢ ¢
max μ f a qa−1 , Q + G (Q) q1a + f b qb−1 , Q + G (Q) q1b (3.50)
(where qs−1 denotes the quantity of s’s private goods, except the first one)
under the budget constraint. The first surprising feature of the TU as-
sumption is that if the optimum has q1a and q1b both positive, then μ is
necessarily equal to unity. To see this, set the price of the first good to
unity and substitute for q1b using the budget constraint:
¡ ¡ ¢ ¢
max μ f a qa−1 , Q + G (Q) q1a
¡ ¢ ¡ ¡ ¢ ¢
+f b qb−1 , Q + G (Q) x − P0 Q − p0−1 qa−1 + qb−1 − q1a (3.51)
Taking the derivative with respect to q1a we see that:
μG (Q) − G (Q) = 0 (3.52)
which implies μ = 1 so that the UPF is a line with a constant slope of −1.
Thus the Pareto weight cannot depend on prices, income or any distribution
factors. Therefore the partners will always agree to act in¡ a manner which
¢
shifts the frontier out as far as possible by the choice of Q, qa−1 , qb−1 . In
fact they will agree to maximize the sum of their individual utilities given
by:
¡ ¢ ¡ ¢ ¡ ¡ ¢¢
f a qa−1 , Q + f b qb−1 , Q + G (Q) x − P0 Q + p0−1 qa−1 + qb−1 (3.53)
Thus, under transferable utility and assuming efficiency, married partners

will agree on almost all consumption choices. The only conflict will be in
how to divide the private good q1 which is often referred to as ‘money’ but
in the family context it may be interpreted more broadly as a medium of
exchange.
3.5.10 The Rotten Kid Theorem revisited

As we have just seen, under transferable utility and efficiency, a couple
acts as a single decision unit in the sense that both partners would agree
on the set of actions which maximizes the joint marital output, defined
as the sum of the partners’s individual utilities. In contrast to the case of
dictatorship, where the issue of implementation does not arise, for the case
of transferable utility we also need to ask how the actions that maximize the
joint output are actually enforced. One possibility is that bargaining takes
place at the outset of marriage, and some sort of binding agreement is signed
and then carried out. However, if the partners are altruistic towards each
other, these emotional ties generate commitments that can replace legally
binding contracts. In particular, commitments that arise from altruistic
preferences can be exploited in the design of a mechanism that implements
the maximization of the total output (sum of utilities) and is self enforcing.
One such scheme (see Becker, 1974) is to select a principal (a family head)
who is given control over family resources and can make transfers as she
or he sees fit. The only requirement is that the principal should care about
all family members in the sense that their utilities enter his or her own
preferences as normal goods. Once this scheme is put in place, each person
is allowed to choose their own actions selfishly. It had been observed by
Becker that such a mechanism is efficient and each participant voluntarily
acts in the interest of the group. The reason is that any productive action
which increases total output is rewarded by an increased transfer from
the principal. Conversely, any destructive action is punished by reduced
transfers. In this way the interests of the group are internalized by every
member. Although the allocation of income depends on who is the head,
family decisions will be invariant to his or her preferences. The crucial
aspect is that every partner should trust the principal to truly care about
all family members and that the principal should be able to fully control
the distribution of income (in the sense that her/his resources are such
that she/he gives everyone a positive transfer that can be reduced or
increased).16
1 6 Becker has two slightly different versions of the Rotten Kid Theorem. The early one
stated in Becker (1974, page 1080) is "If a head exists, other family members are also
motivated to maximize family income and consumption, even if their utility depend on
their consumption alone " The later version in Becker (1991, p288) is set in context of
mutual altruism where each person is a potential contributor to the other and states that
To illustrate the working of the ‘family head mechanism’, let each spouse
have two private actions: consumption and work. Time not spent at work is
used to produce a household good which is a public good (for example, child
quality). Let us assume transferable utility and write the person specific
utility as
U s (Q, q s , ls ) = Qq s + vs (ls ), s = a, b (3.54)
where where, q s denotes private consumption, ls is leisure time and Q is a
public good produced at home. The household production function is
Q = φ(ta , tb ) (3.55)
where ts denotes time spent by s on the production of the public good. The
family budget constraint is
q a + q b = (1 − ta − la )wa + (1 − tb − lb )wb , (3.56)
where ws is the wage of person s. Applying the results on transferable

utility, it is easy to verify that any Pareto efficient allocation must maximize
the sum of private utilities given by:
π = [(1 − ta − la )wa + (1 − tb − lb )wb ]φ(ta , tb ) + va (la ) + vb (lb ) (3.57)
To show that this is an equilibrium outcome of the ‘family head mechanism’,

we consider a two stage game, such that in the first stage each agent s
chooses independently the amount of work at home, ts and in the market 1−
ts − ls . In the second stage, the head, say partner a, chooses the level of the
private good, q s that each partner receives based on a’s social preferences,
W a (U a (Q, q a , la ), U b (Q, q b , lb )). We can solve this problem backwards. In
the last stage, the levels of work at home, ts and ls are given to a and she can
only transfer private goods. This means that the head faces a linear Pareto
frontier (see Figure ) and will select the best point for her on this frontier.
Now assume that the two private utilities appear as ‘normal goods’ in a’s
social welfare function so that whenever the Pareto frontier shifts up the
head reallocates private goods to raise the private utilities of both agents.
Anticipating that, each agent who chooses actions selfishly in the first stage
will realize that their private utility is a monotone increasing function of
the total resources available for the head for redistribution (π in equation
(3.57)). Therefore, each agent will choose the actions under their control
to maximize the pie and the outcome is the same efficient outcome that
would arise if the head could directly control all family decisions.
"Each beneficiary, no-matter how selfish, maximizes the family income of his benefactor
and thereby internalizes all effects of his actions on other beneficiaries." In both versions,
there is only one good that is distributed. Following Bergstrom (1989) we consider here
a problem with two goods and show that under transferable utility similar results are
obtained.
The family head mechanism was first proposed by Becker and is discussed
in detail in Becker (1991, chapter 8). One application of the analysis is par-
ent child relationship and the main result is that selfish children can act
in a manner that internalizes the consequences of their actions, yielding an
efficient outcome. This result was coined by Becker as ‘the rotten kid the-
orem’. His analysis, however, was much more general, dealing with various
forms of altruism and preference dependence. The subsequent literature ad-
dressed the generality of the efficiency head mechanism. Bergstrom (1991)
shows that this result generally fails in the absence of transferable utility,
because agents can then affect not only the location of the Pareto frontier
but also its slope, destroying the monotonicity result required for the theo-
rem to hold. Another issue is the precise sequence of events. Suppose that
the children can consume in both periods 1 and 2. Then, efficiency requires
that, for each child, the ratio of the marginal utilities of consumption in
the two periods is equated to the cost of transferring goods over time that
is facing the household, 1 + r. However, in choosing consumption, the child
will take into account that his first period consumption also influences the
transfer from the head. Being poor in the second period causes the parent
to transfer more, causing the child to under-save. This pattern of behav-
ior, where giving leads to under provision, is referred to as the Samaritan
Dilemma (Bruce and Waldman, 1990). This example shows that altruism
can also be a constraint on mechanism design. The parent could in prin-
ciple impose the efficient outcome by conditioning the payment on past
performance. However, an altruistic parent may not be able to commit to
punish a deviating child - a restriction that is captured in modeling the
stages of game and seeking a subgame perfect equilibrium.
3.5.11 Bargaining models

Throughout the chapter, we have stressed that the collective model, in its
fully general version, is agnostic about the specific decision process provided
that the latter generates Pareto efficient outcomes. Because of this gener-
ality, it is thus compatible with a host of more specific models that further
specify the way a particular point on the Pareto frontier is selected. For in-
stance, we shall show in detail in chapter 8 that under some conditions, this
choice can be fully determined by the competition in the marriage market,
where considerations such as what are the individual characteristics that
generate marital surplus, what is the matching process and does a person
have a close substitute play a crucial role. However, much of the literature
pursues a more partial view and concentrates on the relative strengths of
two individuals who are already matched and use tools from cooperative
game theory to derive the bargaining outcome.17

Any bargaining model requires a specific setting: in addition to the frame-
work described above (two agents, with specific utility functions), one has to
define a threat point T s for each individual s. Intuitively, a person’s threat
point describes the utility level this person could reach in the absence of an
agreement with the partner. Then resources are allocated between public
and private consumption, resulting in two utility levels ūa and ūb . Typi-
cally, bargaining models assume that the outcome of the decision process
is Pareto efficient. Bargaining theory is used to determine how the threat
points influence the location ¡ of the¢ chosen point on the Pareto frontier.
Clearly, if the point T = T a , T b is outside of the Pareto set, then no
agreement can be reached, since at least one member would lose by agree-
ing. However, if T belongs to the interior of the Pareto set so that both
agents can gain from the relationship, the model picks a particular point
on the Pareto utility frontier.
Before describing in more detail some of the standard solutions to the
bargaining problem, however, it is important to note that the crucial role
played by threat points - a common feature of all bargaining models - has
a very natural interpretation in terms of distribution factors. Indeed, any
variable that is relevant for threat points only is a potential distribution fac-
tor. For instance, the nature of divorce settlements, the generosity of single
parent benefits or the probability of re-marriage do not directly change a
household’s budget constraint (as long as it does not dissolve), but may
affect the respective threat points of individuals within it. Then bargain-
ing theory implies that they will influence the intrahousehold distribution
of power and, ultimately, household behavior. This intuition is perfectly
captured in the collective framework by the idea that the Pareto weight
depends on distribution factors. Moreover, it provides a clear idea of the
direction of these effects. That is, a change in a variable that increases the
wife’s threat point should always positively affect her Pareto weight. These
notions potentially provide a number of powerful tests, which are moreover
independent of the particular bargaining concept at stake.
Nash bargaining
The most commonly used bargaining solution was proposed by John Nash
in the early 1950’s. Nash derived this solution as the unique outcome of a set
of axioms that any ‘reasonable’ solution must satisfy. Some of the axioms
are uncontroversial. One is individual rationality: an agent will never accept
an agreement that is less favorable than her threat point. Another is Pareto
efficiency, as discussed above. A third mild requirement is invariance with
1 7 Bargaining approaches to household decision making were first introduced by Manser
and Brown (1980) and McElroy and Horney (1981). For a more complete discussion of
two person bargaining, see Myerson (1991, ch.8).
respect to affine transformations18 : if both the utility and the threat point
of an agent are transformed by the same increasing, affine mapping the
prediction about the equilibrium outcome of cooperation does not change.
Note, however, that a non linear transform will change the outcome; that
is, Nash bargaining requires a cardinal representation of preferences.
The last two axioms are more specific. One is symmetry; it states that
if utilities and threat points are permuted between members (ua and T a
are replaced with ub and T b , and conversely) then the outcomes are simply
switched (ūa is replaced with ūb and conversely). Natural as it may sound,
this assumption may still sometimes be too strong. In many socioeconomic
contexts, for instance, male and female roles are by no means symmetric.
Fortunately, Nash bargaining can easily be extended to avoid the symmetry
assumption.
The last and crucial axiom is independence. It can be stated as follows.
Assume that the set of available opportunities (the Pareto set) shrinks, so
that the new Pareto set is within the old one, but the initial equilibrium
outcome is still feasible; then the new equilibrium outcome will be the
same as before. In other words, the fact that one member misses some
opportunities that he had before does not affect his bargaining position
towards the other member. This requirement alone implies that the Nash
solution maximizes some function of the utilities of the two partners.
If one accepts these axioms, then only¡ a one ¢ outcome is possible. It given
b
by the following rule: find the pair ¡ b ū , ū
¢ on the Pareto frontier that
a a b 19
maximizes ¡the product¢ (u − T ) u − T . That is, the Nash bargaining
allocation Q, qa , qb solves
¡ ¡ ¢ ¢¡ ¡ ¢ ¢
max ua Q, qa , qb − T a ub Q, qa , qb − T b (3.58)
Q,qa ,qb
¡ ¢
under the budget constraint 6.18. Thus the product (ua − T a ) ub − T b
can be considered as a household utility function, that is maximized on the
Pareto set. Note that (us − T s ) is the surplus derived from the relationship
by agent s. The main implication of Nash bargaining is that the product
of surpluses should be maximized.20
Clearly, if the threat points do not depend on prices, incomes and dis-
tribution factors, the maximand can be seen as a standard, unitary utility
and the Nash bargaining solution boils down to a unitary model; the out-
come satisfies in particular the properties of a regular consumer demand.
This case, however, is of little interest. Typically, threat point depends on
a number of parameters, and the previous formalization allows us to study
how these effects translate into behavioral patterns. An important result
1 8 An affine mapping is of the form f (x) = ax +

1 9 Or
b.
equivalently the sum log (ua − T a ) + log ub − T b .
2 0 Note that simply maximizing the sum of surpluses, uA − T A + uB − T B , would
violate the invariance axiom.

¡ ¢
is that at the Nash bargaining equilibrium ūa , ūb , ūa is increasing in Ta
and decreasing in Tb (while, obviously, ūb is decreasing in Ta and increasing
in Tb ). Hence, any change that increases a member’s threat point without
changing the Pareto frontier (the typical impact of a distribution factor)
will ameliorate this member’s situation.
Finally, the symmetry axiom can be relaxed. Then the general form is
a straightforward generalization of the previous one: instead of maximiz-
ing the sum of log surpluses,¡ one¢ maximizes a weighted sum of the form
γ a log (ua − T a ) + γ b log ub − T b . In this form, the weights γ s introduce
an asymmetry between the members’ situations.
Kalai-Smorodinsky
An alternative concept has been proposed by Kalai and Smorodinsky (1975).
It relies on the following, monotonicity property. Consider two bargaining
problems such that (i) the range of individually rational payoffs that player
a can get is the same in the two problems, and (ii) for any given, individu-
ally rational utility level for player a, the maximum utility that player b can
achieve (given the Pareto frontier) is never smaller in the second problem
than in the first. Then player b should do at least as well in the second
problem than in the first. In other words, if one enlarges the Pareto set by
inflating b’s opportunities while keeping a’s constant, this change cannot
harm b.
Kalai and Smorodinsky prove that there exists a unique bargaining solu-
tion that satisfies all the previous axioms except for independence, which is
replaced with monotonicity. Moreover, the solution has an interesting inter-
pretation. Define the aspiration level As of player s as the maximum utility
he/she can get that is compatible with feasibility and the partner’s indi-
vidual rationality constraint; this corresponds to the point on the Pareto
0
frontier that leaves the partner, say s0 , at¡their threat
¢ point utility T s .
Define, now, the ideal point as the point Aa , Ab ; obviously, this point
lies outside
¡ ¢of the Pareto frontier. The solution, now, is to chose a point
U = ua , ub on the Pareto frontier so that
ua − T a Aa − T a
b b
= b
u −T A − Tb
In words, the bargaining is here influenced, in addition to the threat points,
by the players’ aspirations about what they might receive within marriage.
The surplus share received by player s, us − T s , is directly proportional to
the maximum gain s could aspire to, As − T s .
Non cooperative foundations

Finally, an on going research agenda, initially proposed by Nash, is to pro-
vide noncooperative foundations to the bargaining solutions derived from
axioms. The most influential framework is the model of Rubinstein (1982),
in which players make alternating offers until one is accepted. When time
matters through a constant discount factor, there exists a unique, subgame-
perfect equilibrium of this noncooperative game, which is characterized by
the requirement that each player should be indifferent between accepting
the current offer and waiting to an additional round and making an offer
that the opponent would accept. Binmore, Rubinstein and Wolinsky (1987)
have analyzed the link between these non cooperative formulations and the
axiomatic approaches. Specifically, they study a model in which the bar-
gaining process may, with some probability, be exogeneouly interrupted at
each period. This model has a unique, subgame perfect equilibirum; more-
over, if one allows the time interval between successive offers in both models
to decrease to zero, then the equilibirum converges to the Nash bargaining
solution.
Empirical content of bargaining models

Since the bargaining models just described all assume (or imply) Pareto
efficiency, their solutions will satisfy the general properties generated by
the collective model; these will be detailed in the next Chapter. But do
these models allow us to go one step further? That is, which additional
insights (if any) can we obtain from the use of bargaining concepts?
The answer to that question depends on what is known on the threat
points. Indeed, a first result (Chiappori, Donni and Komunjer 2010) is
that any Pareto efficient allocation can be derived as the Nash bargaining
solution for an ad hoc definition of the threat points. Hence the additional
information provided by the bargaining concepts (with respect to the sole
efficiency assumption) must come from specific hypotheses on the threat
points - that is, on what is meant by the sentence: ‘no agreement is reached’.
Several ideas have been used in the literature. One is to refer to divorce as
the ‘no agreement’ situation. Then the threat point is defined as the max-
imum utility a person could reach after divorce. Such an idea seems well
adapted when one is interested in the effects of laws governing divorce on
intrahousehold allocation. Another interesting illustration would be public
policies such as single parent benefits, or the guaranteed employment pro-
grams that exist in some Indian states; Kanbur and Haddad convincingly
argue that the main impact of the program was to change the opportunities
available to the wife outside marriage (or cohabitation). It is probably less
natural when minor decisions are at stake: choosing who will walk the dog
is unlikely to involve threats of divorce.21
A second idea relies on the presence of public goods, and the fact that
2 1 An additional difficulty is empirical. The estimation of utility in case of divorce is
delicate, since most data sets allow us to estimate (at best) an ordinal representation of
preferences, whereas Nash bargaining requires a cardinal representation. See Chiappori
(1991)
non-cooperative behavior typically leads to inefficient outcomes. The idea,

then, is to take the non-cooperative outcome as the threat point: in the
absence of an agreement, both members provide the public good(s) egotis-
tically, not taking into account the impact of their decision on the other
member’s welfare. This version captures the idea that the person who would
suffer more from this lack of cooperation (the person who has the higher
valuation for the public good) is likely to be more willing to compromise
in order to reach an agreement. Interestingly, in this context some of the
results derived in the non-cooperative case extend to the cooperative con-
text as well. For instance, the income pooling result for interior solutions,
derived in subsection 3.4.4, applies here as well: total income being kept
constant, a change in respective incomes that does not affect the noncoop-
erative consumption pattern leaves the threat point unchanged and hence
has no impact on the bargaining outcome. Thus local income pooling is
inherited by the bargaining solution.
Finally, it must be remarked that assumptions on threat points tend to
be strong, somewhat ad hoc, and often not independently testable.22 Given
this cost, models based on bargaining should be used parsimoniously, and
preferably when there is good evidence that the actual structure of the de-
cision process is close to what is implicitly assumed by the concept referred
to. An alternative approach is to concentrate on more general assumptions,
the implications of which hold for a large class of models. Efficiency is one
natural example. Another is that some distribution factors, whatever the
distribution process, can only be favorable to one partner (hence unfa-
vorable to the other) - an intuition that can often be documented using
sociological or ethnographic studies. This point should be kept in mind for
the next chapters.
3.5.12 Other approaches

Finally, we may briefly review three approaches that have been proposed
for analyzing household and family behavior. Two of them (the equilib-
rium models of Grossbard-Shechtman and Haller, and the ‘separate spheres’
model of Lundberg and Pollak) lead to efficient outcomes, therefore are con-
sistent with the cooperative/collective model; the third (Basu’s ‘inefficient
bargaining’) is not, although it relies on a bargaining framework.
Equilibrium models
Following the seminal contributions of Becker23 , several papers by Grossbard-
Schechtman24 analyze marriage in a general equilibrium framework, in
2 2 See Chiappori, Donni and Komunjer (2010) for a formal analysis.

2 3 See Becker (1991) for a general overview.
2 4 See Grossbard-Schechtman (1993) for a unified presentation.
which intrahousehold allocations are directly driven by the competitive

constraints that exist on the marriage market. In some of these models, the
women’s role is essentially to produce domestic commodities. Men employ
women to produce for them, and compensate them with transfers (which, in
developing societies, may take the form of provision of basic needs) and/or
non-pecuniary benefits. From this perspective, marriage can essentially be
analyzed as an employment relationship, which allows to apply the stan-
dard concepts of labor economics. The framework is then generalized to
situations where both men and women engage in household production
work. In all these models, the emphasis is put on a general equilibrium
analysis, and specifically on the impact of the economy on intrahousehold
decisions. One may remark, at this stage, that the outcome of the deci-
sion process thus described is efficient; therefore these models belong to
the cooperative/collective family.25
In a related line, Gerbach and Haller (1999) and Haller (2000) study
the general equilibrium implications of competitive exchange among mul-
timember households, in a context in which consumptions are exclusively
private but consumption externalities may exist within the household. They
compare two benchmark cases: one in which decision making within house-
holds is always efficient (therefore households can be described using the
collective representation), and one (‘individual decentralization’) in which
each household member ‘goes shopping on his or her own, following his
or her own interests, after receiving a share of household income’ (Haller,
(2000), page 835). They first analyze whether competitive exchange among
efficient households leads to a Pareto-optimal allocation at the global level.
The answer is positive as long as each household’s demand exhausts its
budget.26 They then ask whether such an optimal allocation can be ‘indi-
vidually decentralized’ in the sense just defined. They show that, generically
on preferences, the answer is now negative; they conclude that some specific
household decision processes are needed to internalize the externalities.
Separate spheres
The ‘separate sphere’ approach of Lundberg and Pollack (1993) considers
a model with two public goods and assumes that each partner is assigned
a public good to which they alone can contribute; this is their ‘sphere’ of
responsibility or expertise. These spheres are determined by social norms.
The question Lundberg and Pollak address is how the contributions to
the individual spheres are determined. If the partners cooperate, they pool
2 5 The relationship between intrahousehold decision processes in a collective framework
and equilibrium on the market for marriage will be the main topic of the second part of
the present book.
2 6 That the household should spend its entire budget may seem an obvious implication
of efficiency, et least in the static context under consideration here. However, the authors
show that the property may be violated in the presence of negative externalities.
their incomes and set the levels of all goods at the Nash bargaining solution,
which is efficient. The Nash solution is enforced by a binding agreement.
The resulting allocation then depends on the respective threat points of
the husband and wife. They consider the threats of continued marriage in
which the partners act non-cooperatively and each chooses independently
the level of public good under their domain. In this case, the outcome is
inefficient. Specifically, if the partners’ individual utilities are additively
separable in the two public goods (implying no strategic interactions) each
partner will choose the level desired by him\her given their respective in-
comes. If the wife is poor and the child is under her sphere, the outcome
will be under provision of child services. This solution can be modified,
however, by transfers that the husband voluntarily commits to pay his wife
(before incomes are known) or by a direct purchase of child services in the
market.
Inefficient bargaining
Basu (2006) considers a model in which agents bargain in a cooperative
way, but the respective threat points depend in part on endogenous de-
cisions. For instance, when deciding on labor supply and consumption, a
spouse’s bargaining position may depend not only on her wage and non la-
bor income, but also on the labor income she generates. Basu analyzes the
corresponding model, and shows in particular that multiple equilibria may
coexist; moreover, decisions may not be monotonic in a member’s power
(for instance, child labor can first decline, then rise as the wife’s power
increases). It is important to note, here, that although it uses a bargaining
framework, Basu’s model leads to Pareto inefficient decisions, because of
the noncooperative ingredient implicit in the framework. Typically, linking
a person’s weight to that person’s labor income leads to oversupply of la-
bor: once an efficient allocation has been reached, it is individually rational
for each spouse to marginally boost their Pareto weight through additional
labor supply. Both members could benefit from a simultaneous reduction of
their labor supply that would leave Pareto weights unchanged, but strategic
incentives prevent this Pareto improvement from taking place.
A similar intuition had actually been proposed earlier by Brossolet (1993)
and Konrad and Lommerud (1995). In the two-period model of Konrad and
Lommerud, individuals first invest in education, then marry; when married,
their decisions are derived from a Nash bargaining framework. Since invest-
ments in human capital are made noncooperatively and current investments
will serve to improve future bargaining power, there is again inefficient over-
investment in human capital. Unlike Basu, the second period outcome is
efficient in the static sense (that is, labor supply choices, conditional on
education, are ex post Pareto efficient); the inefficiency, here, is dynamic,
and can be seen in the initial overinvestment.
In both cases, efficiency could be restored through adequate commitment
devices. In practice, such devices are likely to exist in Basu’s setting (since
the Pareto improvement could be reached during marriage) but not in Kon-
rad and Lommerud’s framework (because investments are made before the
spouses meet). All in all, these models emphasize the key role of commit-
ment, a point that has been evoked earlier and that will be extensively
discussed in Chapter 6. They also indicate that the interaction between ex
ante investements and ex post matching on the marriage markets are both
important and complex; we shall analyze them in full detail in the second
part of the book.
3.6 References
[1] Apps, Patricia F. and Ray Rees, ‘Taxation and the Household’, Journal
of Public Economics, 35 (1988), 355-369.
[2] Basu, Kaushik, ‘Gender and Say: A Model of Household Behaviour
With Endogenously Determined Balance of Power’, Economic Jour-
nal, 116 (2006), 558-580.
[3] Becker, Gary S. and Kevin M. Murphy, Accounting for Tastes,
(Cambridge Mass: Harvard University Press, 1996).
[4] Becker, Gary S., ‘A Theory of the Allocation of Time’, The Economic
Journal, 75 (1965), 493-517.
[5] Becker, Gary S., ‘A Theory of Social Interactions’, Journal of Polit-
ical Economy, 82 (1974), 1063-93.
[6] Becker, Gary S., Treatise on the family, (Cambridge Mass: Harvard
[7] Becker, Gary S. and Robert J. Barro, ‘A Reformulation of the Eco-
nomic Theory of Fertility’, The Quarterly Journal of Economics,
103 (1988), 1-25.
[8] Bergstrom, Theodore C., ‘Puzzles: Love and Spaghetti, The Oppor-
tunity Cost of Virtue’, The Journal of Economic Perspectives, 3
(1989), 165-173.
[9] Bergstrom, Theodore C., Blume, Larry and Hal R. Varian, ‘On the
Private Provision of Public Goods’, Journal of Public Economics,
29 (1986), 25-49.
[10] Bergstrom, Theodore C., ‘A Fresh Look at the Rotten Kid Theorem—
and Other Household Mysteries’, Journal of Political Economy, 97
(1989), 1138-1159.
[11] Bernheim, B. Douglas and Kyle Bagwell, ‘Is Everything Neutral?’,
Journal of Political Economy, 96 (1988), 308-338.
[12] Bernheim, B. Douglas and Oded Stark, ‘Altruism within the Family
Reconsidered: Do Nice Guys Finish Last?’, The American Economic
Review, 78 (1988), 1034-1045.
[13] Binmore, Ken, Rubinstein, Ariel and Wolinsky, Asher (1986), ‘The
Nash Bargaining Solution in Economic Modelling’, RAND Journal of
Economics, v. 17, iss. 2, pp. 176-88
[14] Brossolet Cecile, Femmes au travail : l’introuvable égalité, (Paris

: L’Harmattan, 1993).
[15] Browning, Martin, Chiappori, Pierre-André and Valérie Lechene, “Dis-
tributional effects in household models: separate spheres and income
pooling”, Economic Journal, 120 (545) (2010).
[16] Browning, Martin, Chiappori, Pierre-André and Arthur Lewbel, ‘Esti-
mating Consumption Economies of Scale, Adult Equivalence Scales, and
Household Bargaining Power’, CAM Working Paper No. 2003-12, (2006).
[17] Bruce, Neil and Michael Waldman, ‘The Rotten-Kid Theorem Meets
the Samaritan’s Dilemma’, The Quarterly Journal of Economics,
105 (1990), 155-165.
[18] Chen, Zhiqi and Frances Woolley, ‘A Cournot-Nash Model of Family
Decision Making’, The Economic Journal, 111 (2001), 722-748.
[19] Chiappori, Pierre-André, ‘Rational Household Labor Supply’, Econo-
metrica, 56 (1988), 63-90.
[20] Chiappori, Pierre-André, ‘Collective Labor Supply and Welfare’,
[21] Chiappori, Pierre-André, ‘Nash-Bargained Household Decisions: A Re-
joinder’, International Economic Review, 32 (1991), 761-62.
[22] Chiappori, Pierre-André, ‘Random Labor Supply’, Columbia Univer-
sity, (2009).
[23] Chiappori, Pierre-André, Olivier Donni and Ivana Komunjer (2010),
‘Learning From a Piece of Pie’, Mimeo, Columbia University.
[24] Chiappori, Pierre-André and Ivan Werning, ‘Comment on "Rotten
Kids, Purity, and Perfection"’, Journal of Political Economy, 110
(2002), 475-504.
[25] Cornes, Richard C. and Emilson C.D. Silva, ‘Rotten Kids, Purity and
Perfection’, Journal of Political Economy, 107 (1999), 1034-1040.
[26] Gersbach, Hans and Hans Haller, ‘Allocation among Multi-Member
Households: Issues, Cores and Equilibria’, in Alkan, Ahmet (eds.),
Aliprantis, Charalambos D. (eds.) and Yannelis, Nicholas C. (eds.), Cur-
rent Trends in Economics: Theory and Applications (Studies in
Economic Theory, 8), (New York: Springer-verlag Berlin Heidelberg,
1999).
[27] Gronau, Reuben, ‘Home Production and the Macro Economy-Some
Lessons from Pollak and Wachter and from Transition Russia’, NBER
Working Paper No. 12287, (2006).
[28] Grossbard-Shechtman, Shoshana, On the Economics of Marriage:

A Theory of Marriage, Labor, and Divorce, (Boulder: Westview
Press, 1993).
[29] Haddad, Lawrence and Ravi Kanbur, ‘Intrahousehold Inequality and

the Theory of Targeting’, European Economic Review, 36 (1992),
372-378.
[30] Haller, Hans, ‘Household Decisions and Equilibrium Efficiency’, In-

ternational Economic Review, 41 (2000), 835-847.
[31] Konrad, Kai A. and Kjell Erik Lommerud, ‘Family Policy with Non-
Cooperative Families’, The Scandinavian Journal of Economics, 97
(1995), 581-601.
[32] Lundberg, Shelly and Robert A. Pollak, ‘Separate Spheres Bargaining

988-1010.
[33] Lundberg, Shelly and Robert A. Pollak, ‘Efficiency in Marriage’, Re-

view of Economics of the Household, 1 (2003), 153-167.
[34] Manser, Marilyn and Murray Brown, ‘Marriage and Household

Decision-Making: A Bargaining Analysis’, International Economic
Review, 21 (1980), 31-44.
[35] Mas-Colell, Andreu, Whinston, Michael D. and Jerry R. Green, Mi-

croeconomic Theory, (New York: Oxford University Press, 1995).
[36] Matthews, Steven A., ‘Smooth Monotone Contribution Games’, PIER

Working Paper Archive 06-018, Penn Institute for Economic Research,
Department of Economics, University of Pennsylvania, (2006).
[37] McElroy, Marjorie B. and Mary Jean Horney, ‘Nash-Bargained House-

hold Decisions: Toward a Generalization of the Theory of Demand’, In-
ternational Economic Review, 22 (1981), 333-349.
[38] Myerson, Roger B., Game Theory: Analysis of Conflict , (Cam-

bridge Mass: Harvard University Press, 1991).
[39] Pollak, Robert A. and Michael L. Wachter, ‘The Relevance of the

Household Production Function and Its Implications for the Allocation
of Time’, The Journal of Political Economy, 83 (1975), 255-278.
[40] Samuelson, Paul A., ‘Social Indifference Curves’, The Quaterly

Journal of Economics, 70 (1956), 1-22.
[41] Samuelson, Paul A., ‘The Pure Theory of Public Expenditure’, The
Review of Economics and Statistics, 36 (1954), 387-389.
[42] Udry, Christopher, ‘Gender, Agricultural Production, and the Theory

of the Household’, The Journal of Political Economy, 104 (1996),
1010-1046.
[43] Warr, Peter G., ‘The Private Provision of a Public Good is Indepen-
dent of the Distribution of Income’, Economics Letters, 13 (1983),
207-211.
FIGURE 3.1. Demand for public good

FIGURE 3.2. The demand for public goods with altruism.

FIGURE 3.3. Household demands for public goods.

ub
Low μ
UPF(P,p,x)
High μ
ua
FIGURE 3.4. The utility possibility frontier.

ub
∂W/∂u a
 
∂W/∂u b
UPF(P,p,x)
W’s indifference
curve
ua
FIGURE 3.5. Linear and non linear generalized Samuelson index
Ub
Ua
FIGURE 3.6. All utilities in the shaded area correspong to the same μ
Ub
Ua
FIGURE 3.7. All μ in the shaded cone correspond to the same ua
FIGURE 3.8. The effects of changes in prices

ub
I Maximal UPF
II
ua
FIGURE 3.9. The RKT utility possibility frontier.

This is page 157
4
The collective model: a formal
analysis
4.1 Collective demand functions: a general
characterization
4.1.1 The collective household utility function.
The basic aspects of the collective model have been described in the pre-
vious chapter. As stated earlier, the particular form adopted has testable
implications for demand functions. We now describe these implications in
detail. We start with the most
¡ general
¢ version of the model with individual
preferences of the form us Q, qa , qb . This allows for any type of consump-
tion externalities between agents. We define the collective household utility
function by
© a ª
uf (q, Q, μ) = max
a
μu ((Q, qa , q − qa )) + ub (Q, qa , q − qa ) (4.1)
q
where μ may be a function of (P, p, x, z) where z is a vector of distribution

factors. We shall always assume that μ (.) is zero homogeneous in (P, p, x)
and any elements of z that are denominated in monetary terms.
At this level of generality, the distinction between public and private
goods is somewhat blurred, and we can leave it aside for the moment. We
thus adopt a general notation with g = (q, Q) denoting the quantities con-
sumed by the household and r = (p, P) denoting the corresponding price
vector. Then the household’s behavior is described by the maximization of
uf (g,μ) under the household budget constraint r0 g =x.
4.1.2 Structural and observable demand.

The household’s program is:
max uf (g,μ) subject to r0 g = x (4.2)

g
which generates collective demand functions, g̃ (r, x, μ). It is important to

emphasize that this program is not equivalent to standard utility maximiza-
tion (the unitary model) because uf varies with μ, which in turn depends
on prices, income and distribution factors. Yet, for any fixed μ, g̃ (., μ) is a
158 4. The collective model: a formal analysis
standard demand function. From standard consumer theory, we therefore

know that it satisfies Slutsky symmetry and negativeness. This property is
crucial in what follows; it can be exploited in a more formal way. Define
the generic Slutsky matrix element of g̃ (., μ), always holding μ constant,
as:
∂g̃i ∂g̃i
σ ij = + g̃j (4.3)
∂rj ∂x
and denote its Slutsky matrix by Σ = [σ ij ]i,j . We then have that Σ is sym-
metric and negative1 . Rearranging (4.3), we get the standard interpretation
of a Slutsky matrix; namely, the Marshallian response of the demand for
∂g̃i
good i to changes in the price of good j ( ∂r j
) can be decomposed into the
difference between a substitution effect (σ ij ) and an income effect (g̃j ∂g̃ ∂x ).
i
The intuition is that a marginal increase in the price of any good i affects,
among other things, the real income (the purchasing power) of all agents.
The substitution term σ ij represents the effect of the infinitesimal variation
if it was fully compensated in income (that is, accompanied by a variation
in income sufficient to exactly offset the loss in purchasing power); for that
reason, we often talk of compensated demand. The income effect, on the
other hand, reflects the fact that the price increase decreases the agent’s
purchasing power in proportion to the quantity purchased, which in turn
influences the demand.
Although the analysis of g̃ (r, x, μ), holding μ constant is conceptually
useful, it is crucial to realize that g̃ cannot be observed directly; indeed, such
an observation would require changing prices and income without modifying
μ. Since, in general, μ does depend on (r, x) this can be, at best, a thought
experiment. What we do observe is the household demand, in which price
and income variations affect both g̃ and μ. Thus the empirically relevant
concept is the demand function defined by:
ĝ (r, x) = g̃ (r, x, μ (r, x)) (4.4)
where we have, for notational economy, dropped the distribution factors.2

Thus, we make a distinction between the ‘structural’ demand function,
g̃ (r, x, μ), and the observable demand function, ĝ (r, x). Again, the dif-
ference between these collective demand functions and the unitary model
(Marshallian) demand functions is the presence of the Pareto weight in the
demands.
1 Throughout the book we use ‘negative’ as shorthand for negative semidefinite; it
does not imply that all the elements of the matrix are negative.
2 We shall maintain the ˆ notation for an observable function and ˜ for structural
throughout the book. Think of the ˆ as denoting a function that could be estimated.
4. The collective model: a formal analysis 159
For the observable demand function we have:

∂ĝi ∂g̃i ∂g̃i ∂μ
= +
∂rj ∂rj ∂μ ∂rj
∂ĝi ∂g̃i ∂g̃i ∂μ
= + (4.5)
∂x ∂x ∂μ ∂x
Thus we can decompose the price effect into a Marshallian response (the
first term on the right hand side) and a collective effect (the second term),
which operates through variations of the Pareto weight μ. Figure 4.1 il-
lustrates for two goods. We start with prices and income (r, x) and the
demand at point I. We then change prices so that good 1 is cheaper; de-
note the new environment (r0 , x). The substitution effect is given by the
move from I to II and the income effect is II to III. The collective effect
associated with the change in μ is represented by the final term in (4.5)
which is shown as the move from III to IV .
4.1.3 The Slutsky matrix for collective demands.

Using the observable functions ĝ (.), we can define the observable or quasi-
Slutsky matrix S = [sij ]i,j by its general term:
∂ĝi ∂ĝi
sij = + ĝj (4.6)
∂rj ∂x
From (4.5) this can be written as:

∙ ¸ ∙ ¸
∂g̃i ∂g̃i ∂g̃i ∂μ ∂μ
sij = + g̃j + + g̃j (4.7)
∂rj ∂x ∂μ ∂rj ∂x
From (4.3), the first term between brackets is the substitution term σij
with associated matrix Σ. We adopt the following notation:
∙ ¸
∂g̃i
Dμ g̃ =
∂μ i
∙ ¸
∂μ ∂μ
v = + g̃j (4.8)
∂rj ∂x j
This gives:
S = Σ + (Dμ g̃) .v0 = Σ + R (4.9)
so that the Slutsky matrix of the observable collective demand ĝ (r, x) is the
sum of a conventional Slutsky matrix Σ, which is symmetric and negative,
and an additional matrix R. The latter is the product of a column vector
(Dμ g̃) and a row vector (v0 ). Note that such an outer product has rank of
at most one; indeed, for any vector w such that v0 .w = 0 we have that
R.w = 0. Finally, this analysis and the homogeneity assumption on μ (.)

yields that the necessary conditions for the collective model demands are
the generalized Slutsky conditions:
ĝ (r, x) is zero homogeneous (4.10)

r0 ĝ (r, x) ≡ x (4.11)
S is the sum of a symmetric, negative matrix and a rank 1 matrix
(4.12)
(Browning and Chiappori (1998)). We denote the third property SNR1.

One can readily see that these conditions generalize the conventional
Slutsky conditions in the unitary setting. In the particular case of R = 0,
indeed, we are back to the predictions of the unitary model. This is the
case, in particular, when either μ is constant (so that v = 0) or when g̃
does not depend on μ (so that Dμ g̃ = 0). The latter case corresponds to the
two partners having the same cardinal preferences; ub (g) = k0 + k1 ua (g)
with k1 > 0. In general, however, R is not zero, and the predictions of
the model deviate from those of the unitary model; in a sense, matrix R
summarizes this deviation. The main result is that this deviation is only
one-dimensional - which formally translates into the rank of R being at
most one. This is a strong result because the size of matrix R can be quite
large - as many as goods the household buys.3
The result has a simple, geometric intuition, given by Figure 3.8 in chap-
ter 3. The move from I to II represents the variation that would obtain
if μ was kept constant; as such, it does not violate Slutsky symmetry. The
violation comes from the second component,that is, the move from II to
III which reflects the impact of changes in μ. This change takes place along
the Pareto frontier. But this frontier is one dimensional, independently of
the number of commodities in the economy. Consequently the matrix R
has at most rank 1.
Finally, it can be shown that these conditions are also (locally) sufficient
for the existence of a collective model. Chiappori and Ekeland (2006) show
that any ‘smooth’4 demand function satisfying the three properties above
(homogeneity, adding-up and SNR1) can locally be constructed as the col-
lective demand of a well chosen household. This is a very difficult result,
that requires complex mathematical tools; it constitutes a generalization
of the classical ‘integrability’ result in standard consumer theory.
3 In general, R has (n + N) eigenvalues (possibly complex); the rank condition means
that all of them, but maybe one, are equal to zero. Equivalently, one can find a basis in
which all of the (n + N) columns of R but one are identically zero.
4 Technically, the result has been proved for twice continuously differentiable demand
functions.
4.1.4 Distribution factors

We may now reintroduce distribution factors. An interesting feature is that
such factors do not change the Pareto frontier, but only the Pareto weight.
In geometrical terms, thus, they can only generate moves along the Pareto
frontier (from II to III in Figure 3.8). This suggests that analyzing the
impact of distribution factors may help understanding the nature and the
form of such moves. This intuition can be given a formal translation. Equa-
tion (4.4) above can now be rewritten as:
ĝ (r, x, z) = g̃ (r, x, μ (r, x, z)) (4.13)

Because the same μ (.) function appears in all goods the collective model
yields cross-equation restrictions. To see this, consider the consequences of
a marginal change in distribution factor zk on the collective demand for
commodity i:
∂ĝi ∂g̃i ∂μ
= (4.14)
∂zk ∂μ ∂zk
Comparing the effect of different distribution factors, say zk and zl , we find
that (assuming ∂gi /∂zl 6= 0):
∂ĝi /∂zk ∂μ/∂zk

= (4.15)
∂ĝi /∂zl ∂μ/∂zl
The right hand side term is independent of the good we are considering.
Hence we have the proportionality property that the ratio of derivatives
with respect to two sharing factors is the same for all goods. The result
that the impact of zk and zl must be proportional across commodities is
very important empirically, and can be given various equivalent forms; for
instance, we can write that5
∂ĝi ∂μ/∂zk ∂ĝi

= . (4.16)
∂zk ∂μ/∂zl ∂zl
If the impact of a change in zk on household demand for good i is, say, twice
as large as that of zl , then the same must be true for all commodities; and
we can actually conclude that the impact of zk on the Pareto weight μ is
twice as large as that of zl . Intuitively, whatever the number of distribution
factors, they only operate through their impact on μ; hence their impact
is one-dimensional. In a sense, it is as if there was one distribution factor
only. This prediction is empirically testable (subject to having at least two
distribution factors); possible tests will be discussed in the next chapter.
Another interesting feature of (4.14) is that it provides additional in-
formation about the structure of price and income effects in the collective
5 Equivalently, ∂gi
the matrix Dz g with general terms ∂zk
is of rank (at most) one.
demand. From (4.14), we have that:
∂g̃i 1 ∂ĝi
= for all i, k
∂μ ∂μ/∂zk ∂zk
∂ĝi
= λk for all i, k (4.17)
∂zk
so that (4.9) becomes
S = Σ + R = Σ + λk . (Dzk ĝ) .v0 for any k (4.18)

Thus regarding price and income effects, not only is the deviation from
the unitary model (the ‘collective effect’) one-dimensional, but it is closely
related to the impact of distribution factors on demand. This is a surprising
property, since it establishes links between the impact of purely economic
factors - prices and incomes - and that of variables of a different type (say,
divorce laws or sex ratios). Again, empirical tests of this property will be
discussed in the next chapter.
4.1.5 Larger households

The analysis developed above can be extended to larger households. Sup-
pose there are T agents in the household. We continue to assume efficiency
so that the collective household utility function is defined as:
( T )
X ¡ ¢
f s 1 T
u (q, Q, μ) = max μs u Q, q , ..., q
g
s=1
T
X
subject to qs = q (4.19)
s=1
where the vector μ = (μ1 , ..., μT ) of Pareto weights is normalized by μT =

1. Again, the μt are functions of prices, income and distribution factors.
The household maximizes this utility under budget constraint. With the
same
¡ notations as ¢above, we can define a ‘structural’ demand function,
g r, x, μ1 , ..., μT −1 as the solution to (4.19); note that it now depends on
T − 1 Pareto weights. As before, the empirically relevant concept is the
observable demand function, defined by:
¡ ¢
ĝ (r, x, z) = g̃ r, x, μ1 (r, x, z) , ..., μT −1 (r, x, z) (4.20)
Similar computations to the two person case yield:
∙ ¸ TX−1 ∙ ¸
∂g̃i ∂g̃i ∂g̃i ∂μt ∂μt
sij = + g̃j + + g̃j (4.21)
∂rj ∂x t=1
∂μt ∂rj ∂x
Again, the collective Slutsky matrix is the sum of a symmetric, negative

matrix Σ and of a ¡‘deviation’
¢ 0 R. However, R is now the sum of T − 1
terms
h of the iform Dμ s
g̃ .vt , in which the vector vt is of general term
∂μt
∂rj + ξ̂ j ∂μ
∂x ; indeed, the deviations now come from the T − 1 functions
t
μt . In particular, its rank is at most T − 1.

The generalized Slutsky conditions for a T person household are given
by:
g (r, x, z) is zero homogeneous

r0 ĝ (r, x, z) ≡ x
T is the sum of a symmetric, negative matrix and a rank T − 1 matrix
(4.22)
These conditions are sometimes called the SNR(T − 1) conditions. They

have a nicely nested structure, in the sense that SNR(k) is a special case of
SNR(k + 1). They are more restrictive, the larger the number of goods and
the smaller the size of the household. Note, in particular, that when the
number of persons in the household is equal to (or larger than) the number
of commodities, the SNR(T − 1) conditions are not restrictive at all: any
(n + N ) × (n + N ) matrix satisfies them (just take Σ = 0). This is by no
means a problem in real life, since the number of commodities available is
very large. However, it may be an issue in econometric estimation, which
typically use a small number of aggregate ‘commodities’.
4.1.6 Children
Finally, we may briefly come back to the issue of children. We described
in the previous chapters two different ways of modelling children: either as
a ‘public good’ that enters parents’ utility or as a genuine decision maker.
The previous analysis sheds light on the respective implications of these
options. In the first case the household has two decision makers, whereas it
has three in the second. According to the generalized Slutsky conditions,
the demand function should satisfy SNR1 in the first case, but not in the
second (it only satisfies SNR2). In words: one can devise a test allowing to
find out how many decision makers there are in the household (the precise
implementation of the test will be described in the next chapter).
Clearly, one has to keep in mind the limits of this exercise. What the
theory predicts is that the rank of the R matrix is at most T − 1. Still, it
can be less. For instance, if μs and μs0 have a similar impact on household
demand (in the sense that Dμs g̃ and Dμs0 g̃ are colinear) then matrix R will
be of rank T − 2. In other words, if a household demand is found to satisfy
SNRk, the conclusion is that there are at least k decision makers; there may
be more, but there cannot be less. Or, in the case of children: a demand
satisfying SNR1 is consistent with children being decision makers; however,
if it satisfies SNR2 and not SNR1, then the hypothesis that children are
not decision makers is rejected.
4.2 Duality in the collective model

4.2.1 The collective expenditure function.
The standard tools of duality theory which have been developed in con-
sumer theory can readily be extended to collective models. They provide
useful ways of analyzing welfare issues in the collective setting. We in-
troduce these notions for a two-person household; the extension to larger
units is straightforward. The first concept is that of collective expenditure
function, denoted E, which is defined by:
¡ ¢
E r, ua , ub = min r0 g
g,qa qb ,Q
¡ ¢
subject to us qa , qb , Q ≥ ūs , s = a, b.
¡ ¢
and g = qa + qb , Q (4.23)
The collective
¡ ¢ expenditure function depends on prices and on two utility
levels ūa , ūb ; it represents the minimum level of expenditures needed at
these prices to achieve these utilities.
¡ One
¢ can then define the compensated
collective demand function, ğ r, ūa , ūb , as a solution to program (4.23). A
key remark is that the definition of household collective expenditure and
demand functions depends only on individual preferences and not on the
household’s decision process.
The properties of the functions just defined are analogous to those of
their standard counterpart. The basic one is the following. Consider the
‘primal’ model stated in Chapter 3:
¡ ¢
max ub Q, qa , qb (4.24)
Q,qa ,qb
subject to r0 g ≤ x
¡ ¢
ua Q, qa , qb ≥ ūa
¡ ¢
and g = qa + qb , Q (4.25)
¡ ¢
The two programs (4.23) and (4.24) are closely ¡ related.¢ Indeed, let Q̄, q̄a , q̄b
denote the solution to (4.24) and let ūb = ub Q̄, q̄a , q̄b . Then:
¡ ¢
E r, ūa , ūb = x
¡ ¢ ¡ ¢
and Q̄, q̄a , q̄b solves (4.23). Conversely, ¡if Q̄, q̄a ,¢q̄b denotes the solution
to (4.23) for some ūa , ūb , then for x = E r, ūa , ūb we have that
¡ ¢
ub Q̄, q̄a , q̄b = ūb
¡ ¢
and Q̄, q̄a , q̄b solves (4.24). The intuition is simply that if a particular
bundle maximizes b’s utility subject to constraints on a’s utility and total
expenditures - this is program (4.24) - then one cannot reach the same
utilities at a lower total cost than this bundle (if that was possible, the dif-
ference in costs could be used to buy extra public commodities and increase
both members’ utilities, a contradiction). Conversely, if a bundle minimizes
total cost for two given utility levels - and therefore solves Program (4.23)
- then one cannot increase b’s utility without either reducing a’s utility or
spending more.
The notion of collective expenditure function - and the duality property
just described - is a direct generalization of the standard expenditure func-
tion of consumer theory; the only difference is that, now, there are two
utility levels that should be reached. Many results follow that generalize
standard theorems of consumer theory; in particular:
Proposition 4.1 We have:
¡ ¢ ¡ ¢
ğ r, ua , ub = Or E r, ua , ub (4.26)
where Or E denotes the gradient of E with respect to r (that is, the vector
of partial derivatives ∂E/∂rj ).
The result is a consequence of the envelope theorem applied to program
(4.23).
In the case of egotistic preferences of the form us (qs , Q), we¡have further¢
results. Define the compensated demand for public goods by Q̆ p, P, ua , ub .
Then:
Proposition 4.2 If us only depends on (qs , Q) , s = a, b, then:
¡ ¢ ¡ ¢
E p, P, ua , ub ≤ ea (p, P, ua ) + eb p, P, ub
¡ ¢
E p, P, ua , ub ≥ ea (p, P, ua )
¡ ¢ ¡ ¢
+eb p, P, ub − P0 Q̆ p, P, ua , ub (4.27)
where es (p, P, us ) denotes the (individual) expenditure function of member

s.
Proof. The last inequality stems from the definition of individual expendi-
ture functions, since
¡ ¢ ¡ ¢
es (p, P, us ) ≤ p0 qs p, P, ua , ub + P0 Q p, P, ua , ub (4.28)
¡ s s¢
For the first inequality, let q̄ , Q̄ denote the individual compensated de-
mand of s (corresponding to prices p, P and utility us ). If Q̄a = Q̄b the
conclusion follows. If not, say Q̄a > Q̄b , then
¡ ¢
ua q̄a , Q̄a = ua
¡ ¢
ub q̄b , Q̄a > ub (4.29)
therefore
¡ ¢ ¡ ¢
E p, P, ua , ub ≤ p0 q̄a + q̄b + P0 Q̄a
¡ ¢
≤ p0 q̄a + q̄b + P0 Q̄a + P0 Q̄b
¡ ¢
= ea (p, P, ua ) + eb p, P, ub (4.30)
4.2.2 Indirect utilities

We can also define indirect utility functions. Consider first the program
¡ ¢
max μua (qa , Q) + ub qb , Q
(qa ,qb ,Q)
¡ ¢
subject to r0 qa + qb , Q = x (4.31)
¡ ∗a ∗b ∗ ¢
Let q , q , Q denote its solution. Then the function ω s , defined for
s = a, b by:
ω s (r, x, μ) = us (q∗s , Q∗ )
is the direct equivalent, in the collective setting, of the indirect utility
concept in standard consumer theory. In particular, ω s only depends on
preferences, not on the decision process; technically, ω s is a function of the
Pareto weight μ, and a change in the decision process would result in the
same function ω s being applied to a different μ.
A second, and more important definition is obtained by plugging the par-
ticular Pareto weight adopted by the household into the previous definition.
In this case, the collective indirect utility of a member is the level of utility
ultimately reached by this member as a function of prices and income and
distribution factors. Formally, if the decision process is characterized by a
function μ (r,x, z), the collective indirect utility of member s is defined for
s = a, b by:
V s (r,x, z) = ω s (r, x, μ (r,x, z))
Now, the definition of s’s collective indirect utility depends not only
on s’s preferences, but also on the whole decision process. In other words,
collective indirect utilities are specific to a particular match between agents
and a particular decision rule (summarized by the function μ). This is
in sharp contrast with the unitary case, where there exists a one-to-one
correspondence between direct and indirect utility at the individual level.
Also, a key remark, here, is that if one is interested in welfare analysis,
then the collective indirect utility is the appropriate concept. Indeed, it
preserves the basic interpretation of standard, indirect utilities in consumer
theory - namely, it characterizes each agent’s final welfare once all aspects
of the decision process have been taken into account.
4.2.3 Welfare
An important application of consumer theory relates to welfare issues, such
as the cost-benefit evaluation of economic reforms. A standard tool is the
notion of compensating variation. Consider a reform that changes the price
vector from r to r0 . For an agent with initial income x, the compensating
variation (CV) is defined as the change in income that would be needed to
exactly compensate the agent. That is, the income that would allow her to
remain on the same indifference curve. For a single person this is defined
by:
CV = e (r0 , v (r, x)) − x
where e and v respectively denote the agent’s expenditure and indirect util-
ity functions. This concept can directly be extended to a collective setting.
This leads to the following definition:
Definition 4.1 The potentially compensating variation is the function
Γ1 (.) such that:
¡ ¢
Γ1 (r, r0 , x, z) = E r0 , V a (r,x, z) , V b (r,x, z) − x
In words, consider a household in which, before the reform, total income

is x and member s’s utility is us = V s (r,x, z). The potentially compensat-
ing variation measures the change in income that has to be given to the
household for the previous utility levels to be affordable at the new prices
r0 . Natural as this extension may seem, it nevertheless raises problems that
are specific to a multi-person setting. The variation is potentially com-
pensating, in the sense that the additional income thus measured could,
if allocated appropriately within the household, enable both members to
reach their pre-reform utility levels.¡That is, the income x ¢+ Γ1 (r, r0 , x, z)
has the property that the utilities V a (r,x, z) , V b (r,x, z) belong to the
Pareto¡ frontier at prices r0 . What
¢ is not guaranteed, however, is that the
point V a (r,x, z) , V b (r,x, z) will still be chosen on the new frontier. In
other words, the compensation is such that the welfare level of each mem-
bers could be maintained despite the reform. Whether the household will
choose to do so is a different story.
The idea is illustrated in Figure 4.2. The potentially compensating vari-
ation ¡is such that the new ¢frontier (the dashed frontier) goes through
uu = V a (r,x, z) , V b (r,x, z) . However, the reform changes both the fron-
tier and the Pareto weights. While the initial allocation uu is still affordable
(it belongs to the new frontier), the household may instead choose the allo-
cation uu0 . It follows that although both members could have been exactly
compensated, in practice one partner will strictly gain from the reform (a
in Figure 4.2), whilst the other will strictly lose. This is despite the fact
that, as drawn, the Pareto weight for a has actually gone down.
This suggests an alternative definition of the compensation, which is the
following:
Definition 4.2 The actually compensating variation is the function Γ2

such that:
Γ2 (r, r0 , x, z) = min
0
{(x0 − x) subject to V s (r0 ,x0 , z) ≥ V s (r,x, z) , s = a, b}
x
(4.32)
Thus Γ2 (r, r0 , x, z) is the minimum amount to be paid to the household for

each agent to be actually compensated for the reform, taking into account
the intrahousehold allocation of additional income. This is illustrated in
figure 4.3. The actually compensating change moves the Pareto frontier
out until b is no worse off. On the new frontier uu00 is the chosen allocation.
Note, still, that while b is then exactly compensated for the reform, a gains
strictly; the initial point uu lies strictly within the new frontier.
Clearly, both concepts raise specific difficulties. The concept of potential
compensation disregards actual decision processes, and ignores intrahouse-
hold inequality. In a fully compensated household, the reform may worsen
the situation of one of the members. This may have a social cost, at least if
we accept that the actual intrahousehold decision process need not always
be optimal from a normative, social viewpoint. On the other hand, the no-
tion of actual compensation may lead to costly compensations, resulting in
a bias in favor of the status quo. Moreover, it de facto rewards (marginal)
unfairness, since the amount paid to the household has to be larger when
most of the additional transfers goes to the dominant member. These issues
are still largely open. We may simply make two remarks. First, these issues
are inherent to any context in which the social planner cannot fully control
intragroup redistribution; they are by no means specific to the collective
approach, or for that matter to cooperative models. The obvious conclu-
sion is that welfare economics can hardly do without a precise analysis of
intrafamily decision processes.
Secondly, the notion of distribution factors suggests an additional direc-
tion for public intervention. Some of these factors can indeed be controlled
by the planner. For example, a benefit can be paid to the husband or to the
wife, in cash or in kind. The benefit should then be designed taking into
account the planner’s ability to influence the decision process; technically,
the maximization in (4.32) should be over x0 and z. For instance, several
authors have suggested that a benefit aimed at improving the welfare of
children should be paid to the mother, because such a shift may increase
her weight in the decision process. Again, we may conclude that a theoret-
ical and empirical analysis of intrahousehold allocation is a key step in any
policy design.
4.3 The case of purely private consumptions

4.3.1 The sharing rule.
Although the Pareto weight captures very clearly our intuitive idea about
power, it turns out that there is an equivalent concept which is easier to
work with and to think about, if preferences are egotistic and we ignore
public goods:
¡ ¢
ua qa , qb = ua (qa )
¡ ¢ ¡ ¢
ub qa , qb = ub qb (4.33)
It is a very familiar idea in convex economies with independent agents that

if there are no externalities, then any efficient outcome can be decentral-
ized by a choice of prices and the (re)distribution of income. This is the
Second Fundamental Theorem of Welfare Economics. In collective models
we can exploit a similar idea. The efficiency assumption has a very simple
and natural translation. With preferences of this kind, the economic inter-
actions within the household are minimal: neither externalities, nor public
goods are involved - agents essentially live side by side and consume inde-
pendently.6 Efficiency simply means that for each agent, the consumption
bundle is optimal, in the sense that no other bundle could provide more util-
ity at the same cost. In other words, take any particular (re)distribution of
total income between members, and assume each member chooses his/her
preferred consumption bundle subject to the constraint that the corre-
sponding expenditures cannot exceed his/her share of total income. Then
the resulting consumption will be Pareto efficient. Conversely, when pref-
erences are quasi-concave, any Pareto efficient allocation can be obtained
in this way.
Suppose a household faces prices p and has decided on¡ a level ¢ of total
a b
expenditure
¡ ¢ x. Let the resulting allocation be denoted q̂ , q̂ so that
p0 q̂a + q̂b = x. The decentralization procedure is simple: each person is
given a share of total expenditure and allowed to spend it on their own
private goods, using their own private sub-utility function us (qs ). In what
follows, let xs denote s’s total expenditures; then xa = p0 q̂a , xb = p0 q̂b ,
and xa + xb = x. Traditionally, a’s part of total expenditures xa is denoted
ρ (so that xb = x − ρ), and called the sharing rule.7 Hence the following
statement:
Proposition 4.3 Define ρ = p0 q̂a so that x − ρ = p0 q̂b . We have:
6 This claim should be qualified. One could easily introduce additional, non monetary
benefits of marriage (love, sex, companionship etc.).

7 The terminology is not completely tied down with some authors referring to the
fraction of expenditures going to A (that is, xA /x) as the sharing rule.

• q̂a solves
max ua (qa ) subject to p0 qa = ρ (4.34)
• q̂b solves ¡ ¢
max ub qb subject to p0 qb = x−ρ (4.35)
a b
Conversely,
¡ a for ¢ any ρ, if q̂ and q̂ solve (4.34) and (4.35) then the
b
allocation q̂ , q̂ is Pareto efficient.
The demands functions q̃a and q̃b , as functions of (p,ρ) and (p,x − ρ), are
conventional demand functions and have all of the usual (Slutsky) proper-
ties.
In other words, when all commodities are privately consumed, the deci-
sion process can be decomposed into two phases: a sharing phase in which
agents determine the sharing rule and a consumption phase, in which agents
allocate their share between the various commodities available. In this con-
text, efficiency only relates to the second phase: whatever the sharing rule,
the resulting allocation will be efficient provided that agents maximize their
utility during the consumption phase. On the other hand, the collective part
of the process (whether it entails bargaining, formal rules or others) takes
part in the first stage.
Also, note that a sharing rule can be defined for any decision process
(one can always consider the outcome and compute the amount privately
spent by member a). However, Proposition (4.3) is satisfied (that is, the
outcome maximizes a’s utility under a’s budget constraint) if and only if
the process is efficient. Clearly, there exists a close connection (actually, if
ua and ub are strictly concave, a one-to-one, increasing mapping) between
a’s share ρ and a’s Pareto weight; both reflect a’s power in the bargaining
phase of the relationship. This implies that the sharing rule depends not
only on prices and total expenditures but also on distribution factors.8
An advantage of the sharing rule is that, unlike the Pareto weight, it is
easy to interpret. In particular, it is independent of the cardinal represen-
tation of individual utilities. For this reason, it is often more convenient to
use the sharing rule as an indicator of the agent a’s ‘weight’ in the decision
process: any change in, say, a distribution factor that increases ρ makes
a better off. Of course, this quality comes at a price: the sharing rule in-
terpretation, as presented above, is valid only when all goods are privately
consumed. We will see in Section 5 to what extent it can be generalized to
public goods
Finally, one should keep in mind that the functions q̃a (p,ρ) and q̃b (p,x − ρ),
although ‘structural’ in the previous sense, cannot be observed, for two rea-
8 The sharing rule depends on prices and income even if the Pareto weight is inde-
pendent of the latter. Thus even in a unitary model with egotistic preferences we have a
sharing rule and it depends on prices and total expenditure. However, the sharing rule
cannot depend on distribution factors unless the Pareto weight does.
sons. One is that, in general, one cannot change prices without changing
the sharing rule as well; what can be observed, at best, are the functions
q̂a (p,x, z) and q̂b (p,x, z), which are related to the previous ones by the
relationships:
q̂a (p,x, z) = q̃a (p, ρ (p,x, z)) (4.36)

q̂b (p,x, z) = q̃b (p, x − ρ (p,x, z))
However, even these functions are in general unknown, because most of

the time the intrahousehold allocation of purchases is not observed. Expen-
diture surveys invariably collect information about expenditures that are
aggregated at the household level; but who consumes what remains largely
unknown, except, maybe, for some specific commodities (for example, ex-
penditure surveys typically distinguish between male and female clothing).
In general what we observe is the household demand which is equal to the
sum of the individual demands:
q̂ (p,x, z) = q̂a (p,x, z) + q̂b (p,x, z)

= q̃a (p, ρ (p,x, z)) + q̃b (p, x − ρ (p,x, z)) (4.37)
As we shall see below, one can often use this relationship to derive the
properties of collective demand functions.
4.3.2 Caring preferences

Let us now consider the case of preferences of the ‘caring’ type, namely
¡ ¢ ¡ ¢
U a qa , qb = ua (qa ) + δ a ub qb
¡ ¢ ¡ ¢
U b qa , qb = ub qb + δ b ua (qa ) (4.38)
Here, the Welfare Theorems do not directly apply, since caring involves an
externality component. Two points should however be remembered. First,
any allocation that is Pareto efficient for caring preferences is also Pareto
efficient for the egotistic preferences ua and ub . This implies that the first
part of Proposition 4.3 still applies: whenever an allocation is efficient, it
can be decentralized through a sharing rule. The converse, however, no
longer holds in general. We know that some allocations may be efficient
for egotistic preferences, but not so for caring ones. It follows that only
a subset of possible sharing rules generate efficient allocations for caring
preferences. For instance, a sharing rule such as ρ ' 0 typically generates
inefficient allocations since a redistribution of the resulting allocation in
favor of a may increase both agents’ welfare (if δ b > 0 and ∂ua /∂q a is
sufficiently large when qa is very small).
4.3.3 Indirect utilities

In the private good case, there exists a simple link between the collective in-
direct utilities defined above and the standard, individual indirect utilities.
Denote the indirect utility corresponding to us (for s = a, b):
v s (p, xs ) = max us (q)

subject to p.q = xs (4.39)
Thus v s (.) denotes the (maximum) utility level reached by s when facing
prices p and consuming a total amount xs . This is the standard, unitary
concept, which makes no reference to the intrahousehold decision process.
Now, in the case of private goods, the decision process is fully summarized
by the sharing rule. It follows that:
V a (p,x, z) = v a (p, ρ (p,x, z)) (4.40)

V b (p,x, z) = v b (p, x − ρ (p,x, z)) (4.41)
where V s is the collective indirect utility of member s, according to the

definition of the previous section. In particular, the first phase of the deci-
sion process (deciding over the sharing rule) can readily be modeled using
indirect utilities: whenever some ρ is chosen, a receives va (p, ρ) and b gets
v b (p, x − ρ). The program would therefore become:
max μva (p, ρ) + v b (p, x − ρ) (4.42)

ρ
More specific processes can also be considered. For instance, Nash bargain-
ing with respective threat points T a and T b would solve:
£ ¤
max [v a (p, ρ) − T a ] v b (p, x − ρ) − T b (4.43)
ρ
It is important to note that, in general, many different structures (that

is, individual preferences and a sharing rule) generate ¡ the ¢same ¡collective¢
indirect utilities V a , V b . Indeed, for any given pair V a , V b , let va , v b , ρ
be such that (4.40) and (4.41) are satisfied, and assume that va and vb are
strictly increasing¡and strictly ¢ quasi-concave. Pick up an arbitrary function
φ (p), and define vεa , vεb , ρε by:
vεa (p,r) = v a (p, r − εφ (p))

vεb (p,r) = v b (p, r + εφ (p))
ρε (p,x, z) = ρ (p,x, z) + εφ (p)
then one can readily check that
V a (p,x, z) = vεa (p, ρε (p,x, z)) (4.44)

V b (p,x, z) = vεb (p, x − ρε (p,x, z)) (4.45)
¡ ¢ ¡ ¢
In other words, the structures v a , vb , ρ and vεa , vεb , ρε , although different,
generate the same collective indirect utilities. It follows that the welfare
conclusions reached by the two structures are always identical. For instance,
if a given reform is found to increase his welfare and decrease her welfare
when the evaluation is made using the first structure, using the second
instead will lead to the same conclusion. We say that different structures
that generate the same collective indirect utilities are welfare equivalent.
The notion of welfare equivalence plays an important role, notably in
the discussion of identification in Chapter 5. In many situations, welfare
equivalent structures are hard to empirically distinguish; in some cases, only
the collective indirect utilities can actually be recovered. The key remark
is that as far are welfare judgment are concerned, identifying collective
indirect utilities is sufficient.
4.4 Application: labor supply with private

consumption
4.4.1 The general setting
An example that has been widely analyzed in the literature concerns labor
supply. In the most stripped down model without household production,
labor supply is modelled as a trade off between leisure and consumption:
people derive utility from leisure, but also from the consumption purchased
with labor income. In a couple, however, an additional issue is the division
of labor and of labor income: who works how much, and how is the result-
ing income distributed between members? As we now see, the collective
approach provides a simple but powerful way of analyzing these questions.
Let ls denote member s’s leisure (with 0 ≤ ls ≤ 1) and q s the con-
sumption by s of a private Hicksian composite good whose price is set to
unity. We start from the most general version of the model, in which mem-
ber s’s welfare can depend on his or her spouse’s consumption and labor
supply in a very general way, including for instance altruism, public con-
sumption of leisure, positive or negative externalities, etc. In this general
framework,
¡ member
¢ s’s preferences are represented by a utility function
U s la , q a , lb , q b . Let wa , wb , y denote respectively real wage rates and
household non-labor income. Finally, let z denote a K-vector of distribu-
tion factors. The efficiency assumption generates the program:
max μU a + U b
{la ,lb ,q a ,q b }
subject to q a + q b + wa la + wb lb ≤ wa + wb + y
0 ≤ ls ≤ 1, s = a, b (4.46)
¡ ¢
where μ is a function of wa , wb , y, z , assumed continuously differentiable
in its arguments.
In practically all empirical applications we observe only q = q a + q b .
Consequently our statement of implications will involve only derivatives
of q, la and lb . In this general setting and assuming interior solutions,
the collective model generates one set of testable restrictions, given by the
following result:
Proposition 4.4 Let ˆls (wa , wb , y,z), s = a, b be solutions to program (4.46).
Then
∂ ˆla /∂zk ∂ ˆlb /∂zk
= , ∀k = 2, ..., K. (4.47)
∂ l̂a /∂z1 ∂ l̂b /∂z1
This result is by no means surprising, since it is just a restatement of the
proportionality conditions (4.15). The conditions are not sufficient, even in
this general case, because of the SNR1 condition (4.12). Namely, one can
readily check that the Slutsky matrix (dropping the equation for q because
of adding up) takes the following form:
⎛ ³ ´ a ³ ´ a ⎞
∂ l̂a ∂ l̂a
∂w a − 1 − ˆla ∂∂yl̂
∂w b − 1 − l̂b ∂∂y
l̂
S= ⎝ ³ ´ ³ ´ ⎠
∂ l̂b ˆla ∂ l̂b ∂ l̂b ˆlb ∂ l̂b
∂w a − 1 − ∂y ∂w b − 1 − ∂y
As above, S must be the sum of a symmetric negative matrix and a matrix

of rank one. With three commodities, the symmetry requirement is not
restrictive: any 2×2 matrix can be written as the sum of a symmetric matrix
and a matrix of rank one. Negativeness, however, has a bite; in practice,
it requires that there exists at least one vector w such that w0 Sw < 0.
With distribution factors, the necessary and sufficient condition is actually
slightly
³ stronger,
´ For K = 1, there must exist a vector w such that S −
0
∂ l̂1 ∂ l̂2 w0 is symmetric and negative.
∂z ∂z
4.4.2 Egoistic preferences and private consumption

Much stronger predictions obtain if we add some structure. One way to do
that is to assume private consumption and egotistic (or caring) preferences,
that is utilities of the form us (ls , q s ). Then there exists a sharing rule ρ,
and efficiency is equivalent to the two individual programs:9
max ua (la , q a )
{la ,q a }
subject to q a + wa la ≤ w + ρa
0 ≤ la ≤ 1 (4.48)
9 In what follows, we shall assume for simplicity that only one distribution factor is
available; if not, the argument is similar but additional, proportionality conditions must
be introduced.
and
max ub (lb , q b )
{lb ,q b }
subject to q b + wb lb ≤ wb + (y − ρ)
0 ≤ lb ≤ 1
Note that now ρ may be negative or larger than y, since one member
may receive all non-labor income plus part of the spouse’s labor income.
Two remarks can be made at this point. First, ρ is the part of total non-
labor income allocated to member a as an outcome of the decision process.
This should be carefully distinguished from a’s contribution to household
non-labor income (although the latter may be a distribution factor if it
influences the allocation process). That is, if non-labor income comes either
from a (denoted y a , representing for instance return on a’s capital) or from
b (denoted y b , representing, say, a benefit paid exclusively to b), so that
y = y a + y b , then a’s part of total expenditures, denoted ρ, may depend
(among other things) on y a or on the ratio y a /y - just as it may depend
on any relevant distribution factor. But it is not equal to y a in general.
The second point is that ρ may be an arbitrary function of wages, non-
labor income and distribution factors. However, our assumptions imply that
ρ cannot depend on the agents’ total labor income, ws (1 − ls ). Indeed, effi-
ciency precludes a person’s allocation to depend on an endogenous variable
such as the labor supply of this person. The intuition is that such a link
would act as a subsidy that would distort the price of leisure faced by the
agents, as in Basu’s (2006) model of inefficient bargaining described in the
previous chapter.
4.4.3 Collective labor supply

In turn, these programs shed light on various aspects of household labor
supply. First, we have that
la = ˜la (wa , ρ) (4.49)

¡ ¢
lb = ˜lb wb , y − ρ (4.50)
where ˜ls denotes the Marshallian demand for leisure corresponding to us .

The function ˜ls is structural (in the sense that it depends on preferences),
but only ls is observed. The first implication of this model is that the
spouse’s wage matters for an individual’s demand for leisure, but only
through its impact on the sharing rule; that is„ through an income effect.
The same is true of non-labor income and of distribution factors:

∂ ˆla ∂ ˜la ∂ρ
= ,
∂wb ∂ρ ∂wb
∂ l̂a ∂ ˜la ∂ρ
=
∂y ∂ρ ∂y
∂ l̂a ∂ ˜la ∂ρ
= (4.51)
∂zk ∂ρ ∂zk
The second equation can be rewritten in elasticity terms:
Ã !µ ¶
y ∂ ˆla ρ ∂ ˜la y ∂ρ
= (4.52)
ˆla ∂y ˜la ∂ρ ρ ∂y
Thus the income elasticity of a’s observed demand for leisure is the product
of two terms. The first is the structural income elasticity which character-
izes a’s preferences - what would be observed if a’s fraction of total non-
labor income could be independently monitored. The second term is the
income elasticity of ρ, reflecting the change (in percentage) of a’s allocation
resulting from a given percentage change in household non-labor income.
Hence if a member’s allocation is elastic, then the elasticity of this per-
son’s demands for leisure, as computed as the household level, will exceed
(in absolute value) the ‘true’ value (as observed for instance on singles,
assuming that preferences are not changed by marriage). Conversely, if the
allocation is inelastic (< 1), then her income elasticity will be found to be
smaller than the ‘true’ value.
The same argument applies to own wage elasticities. From (4.49), we
have that: Ã !µ ¶
wa ∂ ˆla wa ∂ ˜la ρ ∂ ˜la wa ∂ρ
= + (4.53)
l̂a ∂wa la ∂wa ˜la ∂ρ ρ ∂wa
Thus the own wage elasticity observed at the household level is the sum
of two terms. The first is the ‘structural’ elasticity, corresponding to the
agent’s preferences; the second is the product of the person’s structural
income elasticity by the wage elasticity of the sharing rule. To discuss the
sign of the latter, consider the consequences for intrahousehold allocation
of an increase in a’s wage. If leisure is a normal good, then the observed
own wage elasticity (the left hand side) is smaller than the structural value
(the first expression on the right hand side) if and only if ρ is increasing
in wa . This will be case if the wage increase dramatically improves a’s
bargaining position, so that a is able to keep all the direct gains and to
extract in addition a larger fraction of household non-labor income. Most
of the time, we expect the opposite; that is, part of a’s gain in labor income
is transferred to b, so that ρ is decreasing in wa . Then the observed own
wage elasticity (the left hand side) will be larger than the structural value.
The impact of distribution factors is in principle much easier to assess,

because they leave the budget set unchanged and can only shift the distri-
bution of power. Assuming that leisure is normal we have that if a change
in a distribution factor favors member a, then a’s share of household re-
sources will increase which will reduce labor supply through a standard
income effect. This simple mechanism has been repeatedly tested, using
distribution factors such as sex ratios and ‘natural experiments’ such as
the legalization of divorce (in Ireland) or abortion (in the United States).
Interestingly enough, all existing studies tend to confirm the theory. The
effects are found to be significant and of the predicted sign; moreover, they
are specific to married people and are typically not significant when singles
are considered (see the discussion in the next Chapter).
4.5 Public goods

4.5.1 Lindahl prices
We now consider a more general version of the model with egotistic pref-
erences in which we allow for public goods. Hence individual utilities are
of the form us (qs , Q), s = a, b . While the general form of the Pareto
program remains unchanged, its decentralization is trickier, because the
welfare theorems do not apply in an economy with public goods.10 One
solution, which generalizes the previous intuitions, is to use individual (or
‘Lindahl’) prices. It relies on an old idea in public economics, namely that
decisions regarding public commodities can be decentralized using agent-
specific prices; see, for example, Mas-Colell, Whinston and Green (1995).
In a sense, this is part of the standard duality between private and public
consumptions. When a good is private, all agents face the same price and
choose different quantities; with public goods, they all consume the same
quantity but would be willing to pay different marginal prices for it.
A precise statement is the following:
Proposition
³ ´ 4.5 For any (P, p, x, z), assume that the consumption vector
Q̂, q̂ , q̂ is efficient. Then there exists a ρ and 2N personal prices Pa =
a b
¡ ¢
(P1a , ..., PNa ) and Pb = P1b , ..., PNb , with Pja + Pjb = Pj , j = 1, ..., N , such
³ ´
that q̂a , Q̂ solves:
max ua (qa , Q)
subject to p0 qa + (Pa )0 Q = ρ (4.54)
1 0 Private contributions to the public goods are ruled out, since they generate inefficient
outcomes (see Chapter 3).

³ ´
and q̂b , Q̂ solves:
¡ ¢
max ub qb , Q
¡ ¢0
subject to p0 qb + Pb Q = y − ρ (4.55)
Note that both the function ρ and the personal prices Pa and Pb will in
general depend on (P, p, x, z).
These programs correspond to a decentralization of the efficient alloca-
tion in the sense that each agent is faced with their own budget constraint,
and maximizes their utility accordingly. There is however a clear difference
with the private good case, in which all relevant information was read-
ily available to each agent as soon as the sharing rule has been decided
upon. Here, individuals need to know not only the ‘resources’ devoted to
them, as described by ρ, but also their personal prices. Computing the per-
sonal prices is a difficult task, that is basically equivalent to solving for the
efficient allocation; hence the ‘decentralization’ only obtains in a specific
sense.11
Still, the Lindahl approach generates interesting insights on the outcome
of the model. Assuming an interior solution, the first order conditions of
(4.54) give:
∂ua /∂Qj
Pja = pi (4.56)
∂ua /∂qi
The right hand side of this equation is often called a’s marginal willingness
to pay (or MWP) for commodity j; indeed, it is the maximum amount a
would be willing to pay to acquire an additional unit of public good j, if
the amount was to be withdrawn from a’s consumption of private good i.
Note that this amount does not depend on the private good at stake since
the marginal utility of any private good divided by its price is equalized
across private goods. Intuitively Pja increases with a’s preference for the
public good; the intuition of Lindahl prices is precisely that agents with a
higher private valuation of the public good should pay more for it. This is
required for an efficient allocation of the family income between alternative
uses.
Let us now compare the budget constraint the agent is facing in (4.54)
with what the same agent would face if she was a single: p0 qa + P0 Q = y a ,
where y a denotes a’s income as single. An obvious difference is that the
amount of resources has changed - from y a to ρ; this is similar to the
private goods case. However, another difference, which is specific to the
public good case, is that the relative prices of the public commodities have
been changed, from Pj /pi to Pja /pi . Since Pja +Pjb = Pj and Pjb > 0, we have
that Pja < Pj . Intuitively, the publicness of good j makes it less expensive
1 1 The literature on planning has developed several procedures through which infor-
mation exchanges may lead to the determination of Lindahl prices.

relatively to any private good, precisely because the other spouse will also
contribute to the purchase of the public good.
4.5.2 The conditional sharing rule.

An alternative approach
³ relies
´ on the notion of the conditional sharing
a b
rule. Again, let Q̂, q̂ , q̂ denote an efficient consumption vector. The
total expenditure of a and b on private goods only are xa = p0 q̂a and
xb = p0 q̂b . This implies that xa + xb = x − P0 Q̂. Then:
Proposition 4.6 For s = a, b, q̂s solves:

³ ´
max us q, Q̂ subject to p0 q = xs (4.57)
q
Note that, in the two programs above for s = a, b, individuals maximize

over private consumptions taking public consumptions as given. The value
xa is called the conditional sharing rule precisely because its definition is
conditional to the level of public expenditures. The proof is clear: if a could,
through a different choice of her private consumption bundle, reach a higher
utility level while spending the same amount, then the initial allocation had
to be inefficient, a contradiction.
Again, the decision process can be interpreted as operating in two phases,
although the precise definition of the phases differs from the private good
case. Specifically, during the first phase agents determine both the level
of public expenditures and the conditional sharing rule; then comes the
consumption phase, when agents allocate their conditional share between
the various private commodities available. It is important to note that in
sharp contrast with the private good case, the existence of a conditional
sharing rule is necessary for efficiency, but by no means sufficient. The
reason for that is that, in general, efficiency introduces a strong relationship
between the level of public expenditures and the conditional sharing rule.
Broadly speaking, for any given level of public expenditures, most (actually,
almost all) sharing rules would be incompatible with efficiency.
Before analyzing in more detail the first phase, it is useful to define a’s
indirect conditional utility ṽ a as the value of program (4.57) above:
ṽa (p, xa ; Q) = max

a
ua (qa , Q)
q
0 a a
subject to p q = x (4.58)
That is, ṽa denotes the maximum utility a can ultimately reach given pri-
vate prices and conditional on the outcomes (xa , Q) of the first phase de-
cision. We may now consider the first phase, which determines the pub-
lic consumption, Q, and the disposable income allocated to each spouse,
¡ ¢
xa , xb . Efficiency leads to the following program:
© ¡ ¢ª
max μṽ a (p, xa ; Q) + ṽb p, xb ; Q
xa ;xb ;Q
subject to xa + xb + P0 Q = x (4.59)
The first order conditions give:
∂ṽa ∂ṽb
μ =
∂xa ∂xb
a b
∂ṽ /∂Qj ∂ṽ /∂Qj
+ = Pj , j = 1, ..., N (4.60)
∂ṽa /∂xa ∂ṽb /∂xb
The second set of conditions are aoften called the Bowen-Lindahl-Samuelson
∂ ṽ /∂Q
(BLS) conditions. The ratio ∂ ṽa /∂xaj is exactly a’s willingness to pay for
public good j. To see this, note that the first order conditions of (4.57)
a a a
imply that ∂u∂qia = λ pi , where λ is the Lagrange multiplier of a’s budget
constraint; and the envelope theorem applied to the definition of ṽa gives
∂ ṽa a ∂ ṽ a 1 ∂ua
that ∂x a = λ , hence ∂xa = p ∂q a . Thus the conditions simply state that
i i
MWP’s (or private prices) must add up to the market price of the public
good, as argued above. The BLS conditions (the second set of (4.60)) are
necessary and sufficient for efficiency. The choice of a particular allocation
on the Pareto frontier is driven by the first condition in (4.60).
As an application, consider the model of collective labor supply proposed
by Donni (2007), who assumes individual preferences of the form:
Us (T − hs , Q),
where Q is a Hicksian good which represents public consumption. Under
this hypothesis, and taking into account the property of homogeneity, labor
supplies can be written as:
µ ¶
ws ρ (y, wa , wb )
hs = hs , i ,
π s (y, wa , wb ) π i (y, wa , wb )
where
hi wi + ρi (y, wa , wb )
π i (y, wa , wb ) =
y + ha wa + hb wb
denotes member i’s Lindahl price for the public good. In this context, Donni
shows that the utility functions are identified, up to a positive transforma-
tion, from individual labor supplies.
4.5.3 Application: labor supply, female empowerment and

expenditures on public good
While the previous concepts may seem somewhat esoteric, they have im-
portant practical applications. For instance, a widely discussed issue in
development economics and welfare policy in general is the impact of in-

trahousehold redistribution on the structure of household consumption, and
in particular on household demand for public goods. The notion of ‘public
goods’ should be understood here in a very general sense - any expenditure
that benefits both partners. A typical and normatively important example
is expenditures on children, at least if we assume that both parents care
about the well being of their children. The crucial question, then, is the
following: if a given policy ‘empowers’ women, in the sense that it increases
their weight in the household decision process, what will be the impact on
household expenditures on children? For instance, by paying a given ben-
efit to the wife instead of the husband, can we expect children health or
education to be improved? A large and growing body of empirical evidence
suggests that such redistributive effects do exist and can actually be quite
large, at least in some countries. As an instance, Duflo (2003), studying
elderly benefits in South Africa, concludes that the same transfer has dras-
tically different impact of the health of female grandchildren depending on
whether it is paid to the grandmother or the grandfather.
The collective framework provides a very adequate framework for study-
ing these effects. The basic intuition is that while the amount received has
a direct impact on the household’s budget constraint, the gender of the
recipient does not. It can only affect the respective Pareto weights; as such,
it is a perfect example of a distribution factor. We therefore want to in-
vestigate the impact of distribution factors (or equivalently of exogenous
changes in the Pareto weights) on household demand. Two questions are
of particular interest. First, is it possible to predict, from the knowledge of
preferences, which public consumptions will increase when the wife’s weight
raises? Second, is it always the case that female empowerment also results
in more spending on the wife’s private consumption - or could it be the case
that she puts so much emphasis on public consumption that her private
consumption actually declines when she has more power?
To investigate these issues, we start with a very simple example. Assume
individual preferences are Cobb-Douglas:
X X
U s (xs , Q) = αsk log qks + δ sj log Qj (4.61)
k j
P P
where the coefficients are positive and normalized by k αsk + j δ sj = 1.
As above, let μ denote a’s Pareto weight. Prices are normalized to 1, so
that the budget constraint is simply
X¡ ¢ X
qka + qkb + Qj = x
k j
Straightforward computations give household demands:

μαak
qka = x
1+μ
αbk
qkb = x
1+μ
μδ aj + δ bj
Qj = x
1+μ
and the following conclusions follow:
1. The private consumptions of a are all increasing in μ

2. The private consumptions of b are all decreasing in μ
3. Since
∂Qj δ aj − δ bj
= 2x
∂μ (μ + 1)
household consumption in public commodity j increases if and only
if a ‘cares more’ about that commodity than b does, in the sense that
δ aj > δ bj .
As above, it is natural to interpret these results in terms of marginal

willingness to pay. These are given for any public good j by:
xs
M W Pjs = δ sj , s = a, b
Qj
P
where xs = k qks is the conditional sharing rule of member s. Interestingly
enough, the condition δ aj > δ bj is not equivalent to her MWP being larger
than his; rather, it implies that
∂M W Pja ∂M W Pjb
> (4.62)
∂xb ∂xa
In words, the MWP of a must be more income sensitive than that of b.
Still, it may be the case that M W P a < M W P b (particularly if xb is large
with respect to xa ): the absolute magnitude of the respective MWP plays
no role in the result.
The interpretation of these findings is quite intuitive. First, one may
think of the wife’s empowerment (as resulting from an increase in μ) in
purely economic terms: she now receives a higher fraction of household
resources. With Cobb-Douglas preferences, all commodities are normal,
therefore more income always results in more consumption for her; con-
versely, his share has been reduced and he consumes less. Regarding public
goods, however, things are more complex, because a transfer from the hus-
band to the wife typically increases her MWP for each public good but
reduces his. The question, here, is whether her increase is sufficient to com-
pensate his reduction - which is exactly what is implied by equation (4.62).
If the condition is satisfied, the impact of the change over total MWP for
the public good is positive, and consumption grows; in the opposite situa-
tion, it is reduced.
The previous results, natural as they sound, are still dependent on the
very specific functional form chosen for utilities. Whether they extend to
non homothetic preferences, for instance, is not clear. In full generality,
the comparative statics of the model just described are somewhat complex,
if only because, unlike the Cobb-Douglas case, the MWP for a particu-
lar commodity depends in an a priori arbitrary way on the quantities of
the other public goods. However, a clearer picture obtains when there is
only one public good, a case considered by Blundell, Chiappori and Meghir
(2005). They show that if preferences are such that both private expendi-
tures and the public good are normal (in the usual sense that an increase in
income would raise the corresponding, individual demands for these goods),
then a marginal improvement in a member’s Pareto weight increases the
household’s expenditures on the public good if and only if the marginal
willingness to pay of this member is more sensitive to changes in his/her
share than that of the other member. Again, it is not the magnitude of the
MWP’s that matters, but their income sensitivity. Moreover, the private
consumptions of the beneficiary member are always increased.
Coming back to the initial motivation, consider the model discussed in
Chapter 3 in which children’s well being is modeled as a public good that
enters the parents’ utility. Assume that some policy measure may increase
the relative weight of the wife within the household. It is often argued that
children should benefit from such a change, the (somewhat hazy) intuition
being that ‘mothers care more about children than do fathers’. What is the
exact meaning of such a statement, and what exactly does it assume about
preferences? The answer is given by the previous result. She ‘cares more’
means, in this context, that her MWP for children is more income-sensitive:
should she receive an additional dollar to be spent either on children or on
her private consumption, she would spend a larger fraction of it on children
than her husband would.
4.6 Household production in the collective model

Becker (1965) put forward a generalized approach of consumption and time
use in which final consumption is produced within the household by inter-
mediate goods purchased in the market and personal time withdrawn from
market work. Although house production is important for singles, it is par-
ticular relevant for married (or cohabiting) couples. Household production
generates several of the gains from marriage that we mentioned in chapter
3, including increasing returns, specialization and sharing (home produced)

public goods. At a global level, household production represents, according
to several estimates, up to 20% of total production in developed countries,
although it is usually not explicitly taken into account in aggregate mea-
sures such as GDP, and much more in developing economies. At the house-
hold level, domestic production represents a significant fraction of resources
(and especially of time) used and consumed. Finally, at an individual level,
utility depends on leisure, which can be defined as time not spent working
either at home or on the market (although such a definition raises delicate
problems) and also on the consumption of internally produced commodity.
The analysis of household production raises several important issues. One
is the choice of the commodities produced at home and their quantity. In
many cases, a trade-off exists between home production and market trade.
For instance, I can clean my apartment or hire a cleaning person; and in
the opposite direction, the vegetables I grow in my garden can be consumed
internally by my family or sold on the market.12 The commodity is then
said to be marketable. Alternatively, some commodities have to be at least
partly internally ‘produced’; for instance, a nanny cannot, in many cases,
be a perfect substitute for parental care. Another issue is whether and how
these decisions depend on the partners’ respective ‘powers’. Is it the case,
for instance, that the allocation of work by each spouses to the domestic
production process reflects the bargaining positions of the spouses - or is
it exclusively determined by the production technology?
Finally, these issues must be analyzed in an equilibrium context, in which
many key factors have drastically evolved over time. In particular, the
division of labor within households has changed as married women have
dramatically increased their labor force participation. Becker’s framework
allows one to conceptualize the distinct roles of technological advance in
home production and in industrial production in explaining the observed
changes in allocation of time. There is extensive research that applies the
household production approach and tries to sort out the roles of technolog-
ical advance and changes in norms that have made this revolution possible
(Greenwood et al., 2005, Fernandez, 2007). Mulligan and Rubinstein (2007)
emphasize the role of higher rewards for ability (reflected in the general in-
crease in wage inequality) in drawing married women of high ability into
the labor market. See also Albanesi and Olivetti (2009), who emphasize the
role of medical progress in child feeding that enabled women to stay out of
home.13
1 2 This issue is particularly important in development economics, since a majority of
the population of a developing economy typically work in agriculture, often producing

marketable commodities at the household level.
1 3 Another application is De Vries (1994, 2008) who applied this framework to iden-
tify an "industrious revolution", characterized by an increased production of marketable

goods within households, which "preceded and prepared the way for the Industrial Rev-
Another crucial determinant of the time spent on household production

is its opportunity cost, which is directly related to the wage the person
could receive by working on the market. Over the last decades, a striking
phenomenon is the global increase in female education, an evolution that
has deeply modified the trade-off between domestic and market work by
raising female market wages. Of course, education is not exogenous; it is
the outcome of an investment decision based on future (expected) returns,
therefore on (among other things) the fraction of time that individuals
expect to spend working on the market. In other words, education and
current wages affect current decisions regarding household production, but
are themselves the outcomes of past expectations about future domestic
work. The general equilibrium aspects will be left for the second part of
the book; here, we concentrate on a providing a conceptual framework
for analyzing the respective impacts of wages, technology and powers on
domestic production.
4.6.1 The basic model

We have already discussed home production in section 3.2 in chapter 3;
here we focus on the novel aspects that arise in a collective model. Let cs
denote the vector of private consumption of the home produced commodity
by s and let C denote public home produced goods. For the time being,
we ignore time inputs and let q denote the purchases of market goods that
are used in home production. Assuming for the moment that household
commodities are not marketable, the Pareto program thus becomes
¡ ¢ ¡ ¢
max μU a C, ca , cb + U b C, ca , cb (4.63)
subject to
¡ ¢
F C, ca + cb , q = 0
p0 q = x (4.64)
where F is the production function. As above, what is observed is the the

household’s demand function q = q̂ (p, x, z). Note that the model implic-
itly assumes that all commodities are input for household production. This
is without loss of generality: if commodity i is directly consumed, the cor-
responding row of the production equation simple reads cai + cbi = qi for
private consumption, or Ci = qi if the consumption if public.
When compared with the household production model in the unitary
framework, (4.63) exhibits some original features. For instance, the outcome
of the intrahousehold production process can be consumed either privately
or publicly; the two situations will lead to different conclusions, in particular
olution".
in terms of identification. On the other hand, two main issues - whether the
goods produced within the household are marketable or not, and whether
the output is observable - remain largely similar between the collective and
the unitary frameworks.
4.6.2 Domestic production and time use

Of particular interest are the various versions of the collective model with
production involving labor supply. For simplicity, we present one version
of the model, initially analyzed by Apps and Rees (1997) and Chiappori
(1997), in which the two partners supply labor and consume two private
consumption goods, one (denoted q and taken as numeraire) purchased on
a market and the other ¡(denoted ¢ c) produced domestically, according to
some concave function F ta , tb , where ts is member s’s household work.14
Market and domestic labor supplies for person s, hs and ts , are assumed
observed as functions of wages wa , wb , non-labor income y and a distrib-
ution factor z. For simplicity, we ignore the tax system and assume that
budget sets are linear;15 similarly, we exclude joint production.16 Finally,
we assume that preferences are ‘egoistic’, so that s’s are represented by
U s (q s , cs , ls ), where ls denotes leisure and total time is normalized to unity
so that
ls + ts + hs = 1 for s = a, b (4.65)
When the domestic good is not marketable, the previous model therefore
becomes: ¡ ¢
max μU a (q a , ca , la ) + U b q b , cb , lb (4.66)
subject to
¡ ¢
ca + cb = F ta , tb (4.67)
a b a b
q +q = y + wa h + wb h (4.68)
and the time constraint (4.65).17 Conversely, if the commodity is mar-

ketable - that is, if good c can be bought and sold on a market, we let cM
denote the quantity sold (or bought if negative) on the market and p its
market price, which the household takes as given. Then total production of
the good is c = ca + cb + cM ; if cM > 0 then the household produces more
than it consumes (ca + cb ) and sell the difference, if cM < 0 the household
1 4 The model can easily be generalized by adding other inputs to the production
process; the main conclusions below do not change.

1 5 For a comprehensive analysis of taxation with household production, the reader is
referred to Apps and Rees (2009).

1 6 See Pollak and Wachter (1975), and Apps and Rees (2009) for a general presentation.
1 7 Note that utility depends only on consumption and leisure and that, by assumption,
time spent at work either at home or in the market do not enter utility directly.
produces only a fraction of the amount it consumes and purchases the rest.
The production equation is now:
¡ ¢
ca + cb + cM = F ta , tb
and the budget constraint at the household level becomes:
q a + q b = wa ha + wb hb + y + pcM . (4.69)
In our analysis of household production models, we shall first consider

the benchmark situation in which both spouses are working outside the
family, and their working time is flexible enough to allow for marginal
variations. Then the opportunity cost of a person’s time is determined by
the person’s wage, which is taken as given for the family decision process.
We later consider ‘corner’ solutions, in which one spouse works exclusively
at home.
Marketable production
Cost minimization
Let us first assume that good c is marketable. In this context, efficiency
has an immediate implication, namely profit maximization. Specifically, ta
and tb must solve:
¡ ¢
max pF ta , tb − wa ta − wb tb (4.70)
(ta ,tb )
implying the first order conditions:

∂F ¡ a b ¢ ws
t ,t = , s = a, b (4.71)
∂ts p
The economic interpretation of these equations is clear. The opportunity
cost of an additional unit of time spent on domestic production is the per-
son’s wage. If this is not equated to the marginal productivity of domestic
labor, efficiency is violated. For instance, if this marginal productivity is
smaller than the wage, then the person should spend less time working at
home and more working for a wage, keeping total leisure constant. Intra-
household production would decline, but household income would increase
by more than the amount needed to purchase the missing production on
the relevant market. To put it differently, the condition reflects cost min-
imization; if it not satisfied, then the household could achieve the same
level of leisure and domestic consumption while saving money that could
be used to purchase more of the consumption goods - clearly an inefficient
outcome.
The same argument can be presented in a more formal way. Consider
the household as a small economy, defined by preferences ua and ub and
by two ‘production’¡ constraints

¢ - namely, the production of the household
good (here c = F ta , tb ) and the budget constraint. By the second welfare
theorem, any Pareto efficient allocation can be decentralized as a market
equilibrium. On the production side, the second constraint (the budget
constraint) implies that the intrahousehold prices of the consumption
¡ goods
¢
q and c and the leisures la and lb are proportional to 1, p,¡wa , wb ; we¢
can normalize the proportionality factor to be one, and keep 1, p, wa , wb
as intrahousehold prices as well. Then market equilibrium requires profit
maximization, which does not depend on individual preferences. This is
the well-known separation principle, according to which the production
side is fully determined by profit maximization, irrespective of individual
preferences.
Choosing domestic work
The first order conditions of the profit maximization program give
∂F ¡ a b ¢ ws
t ,t = , s = a, b (4.72)
∂ts p
If F is strictly concave (that is, if the domestic technology exhibits decreas-
ing returns to scale), these relations can be inverted to give:
µ a b¶
w w
ts = f s , , s = a, b (4.73)
p p
Knowing the f s (.) functions is strictly equivalent to knowing F . The rela-
tionships (4.73) can in principle be econometrically estimated, leading to
a complete characterization of the production side. It is important to note
that, in this logic, the time spent by each spouse on domestic production
is totally determined by ‘technological’ consideration: it depends only on
wages and on the household production function F , but neither on prefer-
ences nor on ‘powers’ (as measured by Pareto weights). The model predicts,
for instance, that when a change in a distribution factor redistributes power
in favor of the wife (say, a benefit that used to be paid to the husband is
now paid to the wife), the result will be a different consumption pattern
(as discussed above, the household now consumes more of the commodities
preferred by the wife), but the times spent on domestic production by the
husband and the wife remain unchanged. On the contrary, an exogenous
increase in female wage reduces her domestic labor; the impact on his do-
mestic work then depends on the domestic production technology (that is,
are male and female housework complements or substitutes?).
It should be stressed that the marketability assumption is demanding.
Strictly speaking, it requires that households can freely buy or sell the
domestic good. Selling the domestic good is natural in some contexts (for
example, agricultural production in developing countries), but less so in
others (many people clean their own house but would not think of selling
their cleaning services to a third party). If domestic goods can only be

purchased but not sold, our analysis still applies whenever wages and tech-
nology are such that they always consume more than what they produce -
that is, the household as a positive net demand of the domestic good. How-
ever, some households may reach a corner solution, in which the market
purchase of domestic goods is nil, and the normalized marginal productiv-
ity of a person’s domestic work exceeds the person’s wage. In practice, this
is equivalent to the domestic good not being marketable, a case we consider
below.
Finally, the model above assumes that all forms of labor are equally costly
- that is, that the subjective disutility of one hour of labor is the same,
whether it is spent working in a factory or taking care of children. This
assumption, however, can readily be relaxed. One may posit, for instance,
that for some activities (say domestic work), one hour of work ‘costs’ to
spouse s only a fraction αs of one hour of leisure (intuitively, the remaining
fraction (1 − αs ) is leisure). Under this extension, the time constraint (4.65)
should be replaced with:
ls + αs ts + hs = 1 for s = a, b (4.74)
and the first order conditions become:
∂F ¡ a b ¢ ws
t , t = αs , s = a, b (4.75)
∂ts p
In words, the opportunity cost of domestic work should be adjusted for the
associated amenities. Note, however, that the same logic applies; that is, the
time spent by each spouse on domestic production is fully determined by
wages, technology and the individual preferences captured here by amenity
parameter αs . However, they do not depend on the power of the spouses
as measured by μ.
The demand side
The separability principle implies that the demand side is totally divorced
from production. Indeed, the household’s total ‘potential’ income is
¡ ¢
Y = wa (1 − ta ) + wb 1 − tb + y + pc (4.76)
This potential income has to be split between the members and spent on
individual leisures and consumptions of the two goods. Since all commodi-
ties are private, efficiency is equivalent to¡the existence¢ of a sharing
¡ rule. As
¢
above, thus, there exists two functions ρa wa , wb , y, p and ρb wa , wb , y, p ,
with ρa + ρb = Y , such that each member s solves:
max U s (q s , cs , ls )
under the member-specific budget constraint
q s + pcs + ws ls = ρs
At this stage, we are back to the standard collective model of labor supply.
Non-marketable production
The other polar case obtains when no market for the domestic good ex-
ists (then cM = 0 ). Then we are back to maximizing (4.66) under the
constraints (4.67), (4.68) and (4.65). One can still define a price p for the
domestic good, equal to the marginal rate of substitution between the do-
mestic and the market goods for each of the members (the MRS’s are
equalized across members as a consequence of the efficiency assumption).
The difference, however, is that p is now endogenous to the model - that
is, it is determined by the maximization program.
A particularly interesting case obtains when the domestic production
function exhibits constant returns to scale (CRS). Then:
µ a¶
¡ a b¢ b t
F t ,t = t Φ b (4.77)
t
for some function Φ (.). First order conditions imply that:
∂F/∂ta wa
b
= b,
∂F/∂t w
which give in this case:
µ ¶ µ ¶
ta ta ta wa
Φ − b Φ0 = ,
tb t tb wb
This relationship, which is a direct consequence of³the éfficiency assump-
a a
tion, pins down the ratio ttb to be some function φ wp . In other words,
it is now the case that the ratio of male to female domestic work depends
only on wages and household production technology - a natural consequence
of cost minimization. On the other hand, the scale of production - that is,
the quantity eventually produced - is indeterminate from the production
perspective; it depends on preferences and the decision process. We con-
clude that preferences and powers determine the total quantity of household
goods produced; however, conditional on that quantity, the particular com-
bination of male and female time is determined by respective wages and the
production technology, and does not depend on preferences or powers.18
The (household-specific) price of the domestic good can readily be re-
covered. Indeed, an interior solution under constant returns require zero
profits, therefore it must be the case that:
¡ ¢
wa ta /tb + wb
p=
Φ (ta /tb )
1 8 Pollak and Wachter (1975) discuss the roles of constant returns to scale and joint
production. They show that with joint production (i.e., activities that generate more
than one final good), it is generally impossible to separate household technology from
preferences, even under a constant return to scale technology.
Again, this price depends only on wages and on the technology. It is house-
hold specific in the sense that two households with different wages will
price the household good differently, even if they have access to the same
domestic technology. However, for given wages and domestic technology, it
depends neither on preferences nor on respective powers. Finally, the sepa-
ration result still holds. That is, each member’s decision can be modeled as
if they were maximizing their own utility under the member specific bud-
get constraint defined by a sharing rule; this mechanism determines all the
components of consumption, including The only difference with the mar-
ketable case is that p is no longer a market price; instead, it is determined
by the wages and the technological constraints.
Power and domestic work

While the previous conclusions are not really surprising, at least from a
general equilibrium perspective - they basically illustrate standard results
in welfare economics - their implications can be somewhat unexpected.
Consider, for instance, a change in Pareto weights that benefits women -
say, through the impact of a distribution factor - while wages and incomes
are unaffected. As discussed in the previous subsection, a first consequence
is that the structure of consumption will change; intuitively, the household
will now consume more of the commodities that the wife ‘likes more’. If,
as it is often argued, women generally care more about the goods that
are domestically produced (child care being a primary example), the total
consumption of these commodities should increase. If the commodity is
marketable and initially (partly) purchased on the market, the result will be
higher market purchases of these goods, with no impact on domestic labor
by the partners. In all other cases, domestic labor will increase, and the
distribution of the additional effort between spouses is completely driven by
the technology. For instance, under a standard, Cobb-Douglas production
function, inputs are complements; at constant prices (here wages), more
production requires increasing both inputs. We conclude that more power
to the wife may actually imply more domestic work for both spouses. Note,
however, that because a transfer of income to the wife does not affect
her time input into home production, the income effect will induce her to
reallocate the remaining time so that her market work should decline and
her leisure increase. This conclusion should be contrasted with the impact
of an increase in the wife’s market wage, which always affect her domestic
labor supply. When the commodity is marketable, her domestic work is
always reduced. In the alternative situation, her domestic work decreases
with respect to her husband’s, but the absolute impact also depends on the
structure of consumption - especially if her Pareto weight is boosted by her
higher wage.
Extensions
Public goods
In the previous analysis, the internally produced commodity was privately
consumed. What if, instead, the commodity is public within the household
- as it is the case for childcare, for instance? Interestingly, not much is
changed, because the separation principle still applies. If the commodity,
although public within the household, is marketable, then its production is
driven by profit maximization; the only change regards the demand side,
where the decision process can no longer be decentralized using a sharing
rule. Even in the non marketable case, the logic of cost minimization pre-
vails. In particular, under constant returns to scale, it still the case that the
level of production is determined by preferences and the decision process,
while for any given level the time allocation of domestic work between
spouses stems from technological considerations.19
Specialization
Another special (but empirically relevant) case obtains when one of the
spouses - say b - does not enter the labor market, and specializes instead
in home production. This happens when, for the chosen allocation of time
and consumption, b’s potential wage, w̄b , is smaller than both b’s marginal
productivity in household production and b’s marginal rate of substitution
between leisure and consumption. In words: the marginal hour can indiffer-
ently be spent in leisure or household production, and both uses dominate
market work.20
The situation, here, is more complex, because the opportunity cost of
labor for b is no longer exogenously given; instead, it is now endogenous to
the program. Still, if we keep the assumption of constant return to scale
domestic technology, some of the previous conclusions remain valid. Indeed,
in the marketable case, efficiency in a’s allocation of time still requires that:
∂F ¡ a b ¢ wa
t ,t =
∂ta p
while the CRS condition (4.77) implies that

µ a¶
∂F ¡ a b ¢ 0 t
t , t = Φ
∂ta tb
It follows that, again, the ratio ta /tb is pinned down by technological con-
1 9 The reader is referred to Blundell, Chiappori and Meghir (2005) for a more detailed
investigation.
2 0 Technically, this result is true at the marginal level only in the absence of non
convexities. In the presence of fixed costs of work or constraints on the number of hours
worked, the same constraint must be redefined at a more global level.
straints - namely, it must be such that

µ a¶
0 t wa
Φ =
tb p
In words: the volume of domestic production is now determined by prefer-

ences, but the distribution (between spouses) of effort needed to produce
that amount is fixed by the technology.
Finally, in the case of specialization into the production of a non mar-
ketable good, both the price of the domestic good and b’s opportunity
cost of labor are endogenous. Then all aspects of household production are
potentially affected by the distribution of powers within the couple.
4.6.3 Empirical issues

To what extent can the previous analysis generate testable restrictions?
Note first that, as discussed in section 2 of chapter 3, when the outcome is
observable, efficiency can directly be tested empirically. Indeed, a straight-
forward implication of efficiency is cost minimization: whatever the value
of the output, it cannot be the case that the same value of output could be
produced with a cheaper input combination. Udry (1996) provides a test
of this sort on data from Burkina-Faso. Also, it is in general possible to di-
rectly estimate the production function; then one can refer to the standard,
collective setting, using the methods presented above. Usually, however, the
output of the intrahousehold production process is not observable. Still,
some of the techniques described for models without home production can
be extended to the case of production. For instance, distribution factor pro-
portionality should still hold in that case; the basic intuition (distribution
factors matter only through the one-dimensional Pareto weight μ) remains
perfectly valid in Program (4.63). The same is true for the various versions
of the SNR conditions, with and without distribution factors, which rely
on the same ideas.
Moreover, if time use data are available, then the previous models gen-
erate several, testable restrictions regarding the impact of wages, income
and power on domestic production. If we consider the benchmark case of
CRS technology, the basic prediction is that the proportion of total domes-
tic time spent by each member only depends on wages and the technol-
ogy. Therefore any variable that does not affect the production side of the
household (but only, say, preferences or the decision process) should not
be relevant for the determination of the ratio ta /tb . On the other hand,
changes in wages do affect the ratio; as expected, a (proportionally) higher
female wage reduces the ratio of her domestic work to his.
Regarding identification, note first that if the internally produced com-
modity is marketable (as will often be the case for, say, agricultural produc-
tion in developing countries), then conditions (4.73) above can in principle
be econometrically estimated, leading to a complete characterization of the

production side. In the opposite case, however, the separability property
no longer applies; the price p has to be estimated as well. As discussed by
Chiappori (1997), identifiability does not obtain in general; however, it can
still be achieved under additional assumptions.
Finally, a much stronger result obtains when the produced good is pub-
licly consumed. Blundell, Chiappori and Meghir (2005) consider a model
which is formally similar to the previous one, except that the second com-
modity is public and its production requires labor and some specific in-
s s s
put, Q. Technically, individual utilities
¡ take
¢ the form u (q , C, l ), and the
a b
production constraint is C = F Q, t , t . A natural (but not exclusive)
interpretation of C is in terms of children’s welfare, which enters both util-
ities and is ‘produced’ from parental time and children expenditures Q.
Blundell, Chiappori and Meghir show that strong testable restrictions are
generated. Moreover, the structure (that is, utilities and the Pareto weights)
are identifiable from labor supplies (both domestic and on the market) and
children’s expenditures, provided that one distribution factor (at least) is
available.
4.7 References
[1] Albanesi, Stefania and Claudia Olivetti, ‘Home Production, Market
Production and the Gender Wage Gap: Incentives and Expectations’,
Review of Economic Dynamics, 12 (2009), 80—107.
[2] Apps, Patricia F. and Ray Rees, ‘Collective Labor Supply and House-
hold Production’, Journal of Political Economy, 105 (1997), 178-90.
[3] Apps, Patricia F. and Ray Rees, Public Economics and the House-
hold, (Cambridge, Cambridge University Press, 2009).

nal, 116 (2006), 558-580.
[5] Becker, Gary S., ‘A Theory of the Allocation of Time’, The Economic
Journal, 75 (1965), 493-517.
[6] Blundell, Richard, Chiappori, Pierre-André and Costas Meghir, ‘Col-

lective Labor Supply with Children’, Journal of Political Economy,
113 (2005), 1277-1306.
[7] Browning, Martin and Pierre-André Chiappori, ‘Efficient Intra-

Household Allocations: A General Characterization and Empirical
Tests’, Econometrica, 66 (1998), 1241-1278.
[8] Chiappori, Pierre-André, ‘Introducing Household Production in Col-

lective Models of Labor Supply’, Journal of Political Economy, 105
(1997), 191-209.
[9] Chiappori, Pierre-André and I. Ekeland, ‘The Micro Economics of
Group Behavior: General Characterization’, Journal of Economic
Theory, 130 (2006), 1-26.
[10] De Vries, Jan, ‘The Industrial Revolution and the Industrious Revo-
lution’, The Journal of Economic History, 54 (1994), 249-270.
[11] De Vries, Jan, The Industrious Revolution: Consumer Behavior
and the Household Economy, 1650 to the Present, (Cambridge:
Cambridge University Press, 2008).
[12] Duflo, Esther, ‘Grandmothers and Granddaughters: Old-Age Pensions
and Intrahousehold Allocation in South Africa’, World Bank Eco-
nomic Review, 17 (2003), 1-25.
[13] Fernandez, Raquel, ‘Alfred Marshall Lecture Women, Work, and Cul-
ture’, Journal of the European Economic Association, 5 (2007),
305-332.
[14] Greenwood, Jeremy, Seshadri, Ananth and Mehmet Yorukoglu, ‘En-
gines of Liberation’, Review of Economic Studies, 72 (2005), 109-133.
[15] Mas-Colell, Andreu, Whinston, Michael D. and Jerry R. Green, Mi-
croeconomic Theory, (New York: Oxford University Press, 1995).
Women’s Relative Wages Over Time’, The Quarterly Journal of Eco-
nomics, 123 (2008), 1061-1110.
[17] Pollak, Robert A. and Michael L. Wachter, ‘The Relevance of the
Household Production Function and Its Implications for the Allocation
of Time’, Journal of Political Economy, 83 (1975), 255-277.
of the Household’, Journal of Political Economy, 104 (1996), 1010-
1046.
Q2
III
I
II
IV
q1
FIGURE 4.1. Collective price responses
ub
uu
uu '
ua
FIGURE 4.2. A potentially compensating variation.

ub
uu uu ''
uu '
ua
FIGURE 4.3. An actually compensating variation.

This is page 199
5
Empirical issues for the
collective model
5.1 What are the objects of interest?
We have seen above that various approaches can be used to describe house-
hold behavior, from the unitary setting to noncooperative approaches and
the collective model. Ultimately, the choice between these various frame-
works will rely on particular considerations. First, general methodological
principles may favor one approach over the others. For instance, one can
argue that the unitary framework is not totally faithful to methodological
individualism, a cornerstone of micro theory that postulates that individu-
als, not groups, are the ultimate decision makers. A second requirement is
the model’s ability to generate testable predictions for observable behavior,
that can be taken to data using standard techniques. Standard consumer
theory fares pretty well in this respect. Utility maximization under a lin-
ear budget constraint yields strong predictions (adding-up, homogeneity,
Slutsky symmetry and negative semidefiniteness and income pooling) and
adequate methodologies have been developed for testing these properties.
Finally, a crucial criterion is the fruitfulness of the approach, particularly
in terms of normative analysis and policy recommendations. A remark-
able feature of standard consumer theory is that individual preferences can
be uniquely recovered from demand functions (if these satisfy the Slutsky
conditions); it is therefore possible to analyze welfare issues from the sole
knowledge of observed behavior. This is a particular case of the general re-
quirement that the model be identifiable, that is, that it should be possible
to recover the underlying structure from observed behavior.
The first line of argument, concerning methodological individualism, has
been evoked earlier. In this chapter, we concentrate on the remaining two
aspects, namely testability and identifiability of preferences and processes
from observed behavior. Most of the existing knowledge for non-unitary
models concerns the cooperative framework, and especially the collective
model. The testability requirement, per se, is not problematic. The idea that
a model should generate predictions that can be taken to data belongs to
the foundations of economics (or any other science!). Identifiability is more
complex and it is useful to define more precisely what is meant by ‘recov-
ering the underlying structure’. The structure, in our case, is the (strictly
convex) preferences of individuals in the group and the decision process.
In the collective setting, because of the efficiency assumption, the decision
200 5. Empirical issues for the collective model
process is fully summarized (for any particular cardinalization of individual

utilities) by the Pareto weight corresponding to the outcome at stake. The
structure thus consists of a set of individual preferences (with a particular
cardinalization) and a Pareto weight - which, as we should remember, can
be (and generally is) a function of prices, incomes and distribution factors.
The structure cannot be directly observed; instead we observe the out-
comes of the interactions between preferences, constraints and the decision
process. Often we observe only aggregate outcomes and not individual out-
comes. In addition, the ‘observation’ of, say, a demand function is a complex
process, that entails specific difficulties. For instance, one never observes a
(continuous) function, but only a finite number of values on the function’s
graph. These values are measured with some errors, which raises problems
of statistical inference. In some cases, the data are cross-sectional, in the
sense that different groups are observed in different situations; specific as-
sumptions have to be made on the nature and the form of (observed and
unobserved) heterogeneity between the groups. Even when the same group
is observed in different contexts (panel data), other assumptions are needed
on the dynamics of the situation - for example, on the way past behavior
influences present choices. All these issues lay at the core of what is usually
called the inference problem.1
A second and different aspect relates to what has been called the iden-
tifiability problem, which can be defined as follows: when is it the case
that the (hypothetically) perfect knowledge of a smooth demand function
uniquely defines the underlying structure within a given class? This ab-
stracts from the econometrician’s inability to exactly recover the form of
demand functions - say, because only noisy estimates of the parameters can
be obtained, or even because the functional form itself (and the stochastic
structure added to it) have been arbitrarily chosen. These econometric ques-
tions have, at least to some extent, econometric or statistical answers. For
instance, confidence intervals can be computed for the parameters (and be-
come negligible when the sample size grows); the relevance of the functional
form can be checked using specification tests; etc. The non-identifiability
problem has a different nature: even if a perfect fit to ideal data was fea-
sible, it might still be impossible to recover the underlying structure from
this ideal data.
In the case of individual behavior, as analyzed by standard consumer the-
ory, identifiability is an old but crucial result. Indeed, it has been known for
more than a century that an individual demand function uniquely identifies
the underlying preferences. Familiar as this property may have become, it
remains one of the strongest results in microeconomic theory. It implies, for
1 In the original Koopmans discussion of identification, the step from sample informa-
tion to inferences about population objects (such as demand functions) is referred to as

identification. Here we follow modern terminology and refer to it as the inference step.
5. Empirical issues for the collective model 201
instance, that assessments about individual well-being can unambiguously

be made based only on the observation of demand behavior with suffi-
ciently rich (and ‘exogenous’) variation in prices and total expenditures; a
fact that opens the way to all of applied welfare economics. It is thus nat-
ural to ask whether this classical identifiability property can be extended
to more general approaches.2
Finally, it should be remembered that identifiability is only a neces-
sary condition for identification. If different structures are observationally
equivalent, there is no hope that observed behavior will help to distinguish
between them; only ad hoc functional form restrictions can do that. Since
observationally equivalent models may have very different welfare impli-
cations, non-identifiability severely limits our ability to formulate reliable
normative judgments: any normative recommendation based on a particu-
lar structural model is unreliable, since it is ultimately based on the purely
arbitrary choice of one underlying structural model among many. Still,
whether an identifiable model is econometrically identified depends on the
stochastic structure representing the various statistical issues (measure-
ment errors, unobserved heterogeneity,...) discussed above. After all, the
abundant empirical literature on consumer behavior, while dealing with
a model that is always identifiable, has convinced us that identification
crucially depends on the nature of available data.
The main properties of the collective model have been described in the
previous chapter. However, which empirical test can actually be performed
obviously depends on the nature of available data. Three different con-
texts can be distinguished. In the first context, individual demand can be
estimated as a function of income and possibly distribution factors; this
approach is relevant when no price variation is observed, for instance be-
cause data are cross-sectional and prices are constant over the sample. We
then allow that we also observe price variation so that we can estimate a
complete demand system. The analysis of labor supply raises specific is-
sues that are considered in the third section. The final half of this chapter
presents a review of empirical analysis using non-unitary models (includ-
ing the results of applying the tests of the first half of the chapter). We
conclude the chapter with an account of intra-household allocation based
on two Danish data sets that were specifically designed to address research
issues concerning intrahousehold allocation.
2 Note, however, that only one utility function is identifiable in the standard case. In
a ‘unitary’ framework in which agents are characterized by their own utility function
(see chapter 3, subsection 3.5.9) but the household behaves as a single decision unit, it
is typically not possible to identify the individual utility functions.
5.2 Data without price variation.

5.2.1 Necessary and sufficient conditions for a collective
model.
In this section we consider testing and identification in the absence of price
variation as is often the case with cross-sectional data. We begin with the
case in which we observe only household (aggregate) demand of each good.
Let x denote the household’s total expenditures and let z be a K-vector of
distribution factors. Recall that distribution factors, by definition, influence
neither preferences nor the budget constraint. In a unitary setting, they
have no impact on demand. In the collective framework, on the contrary,
household behavior can be described by a program of the following form:
max μ (x, z) ua (g) + ub (g) (5.1)
0
subject to e g ≤ x
¡ a b ¢
where g is the vector q , q , Q and quantities are normalized so that the
price vector is a vector of ones, e. The resulting vector of collective demand
functions can be written g =g̃(x, μ (x, z)) with a corresponding observable
demand functions ĝ (x, z).
An alternative demand formulation which is useful for empirical work
(see below) can be formulated if there is at least one good (good j, say)
that is strictly monotone in one distribution factor (z1 , say); that is, gj (x, z)
is strictly monotone in z1 . This demand function can be inverted on the
first factor to give:
z1 = ζ(x, z−1 , gj )
where z−1 is the vector of distribution factors without the first element.
Now substitute this into the demand for good i:
gi = ĝi (x, z1 , z−1 ) = ĝi [x, ζ(x, z−1 , gj ), z−1 ] = θji (x, z−1 , gj ) .
Thus the demand for good i can be written as a function of total ex-
penditure, all distribution factors but the first and the demand for good j.
To distinguish this conditioning from the more conventional conditional de-
mands used in the demand literature, we shall refer to them as z-conditional
demands.3
We now address the issue of what restrictions a collective model imposes
upon observable demands. Bourguignon, Browning and Chiappori (2009)
provide a complete characterization of these conditions. Specifically they
prove that the following equivalent conditions are necessary consequences
of the collective model:
3 In the unitary setting, distribution factors cannot influence demand, so that z-
conditional demands are not defined in this case.

1. there exist real valued functions g̃1 , ....., g̃n and μ such that :
ĝi (x, z) = g̃i [x,μ(x, z)] ∀i = 1, ....., n (5.2)
2. household demand functions satisfy the proportionality condition:
∂ĝi /∂zk ∂ĝi /∂z1

= ∀i = 1, .., n; j = 1, .., n; k = 2, .., K (5.3)
∂ĝj /∂zk ∂ĝj /∂z1
3. for any good j such that ∂ĝj /∂z1 = 6 0, the z-conditional demands
satisfy:
j
∂ θ̂i (x, z−1 , gj )
=0 ∀i 6= j, k = 2, .., K (5.4)
∂zk
The intuition for this result relates to the discussion provided in earlier
chapters. Again, the basic idea is that, by definition, distribution factors
do not influence the Pareto set. They may affect consumption, but only
through their effect upon the location of the final outcome on the Pareto
frontier or, equivalently, upon the respective weighting of each member’s
utility that is implicit in this location. The key point is that this effect is
one-dimensional (see chapter 4, subsection 1.3). This explains why restric-
tions appear only in the case where there is more than one distribution
factor. Whatever the number of such factors, they can only influence con-
sumption through a single, real-valued function μ. Conditions (5.2) and
(5.3) are direct translations of this remark. By the same token, if we com-
pute qi as a z-conditional function of (x, z−1 , gj ), it should not depend
on z−1 . The reason is that, for any given value of x, whenever distribu-
tion factors (z1 , z−1 ) contain some information that is relevant for intra-
household allocation (hence for household behavior), this information is
one-dimensional and can be fully summarized by the value of gj . Once we
condition on gj , z−1 becomes irrelevant. This is the meaning of condition
(5.4).
The conditions (5.2)-(5.4) are also sufficient for the collective model: if
they are satisfied for the observable demands ĝ(x, z), then one can find util-
ity functions and Pareto weights which rationalize the observed demands
(see Bourguignon et al (2009)). An important implication of these condi-
tions is that in the absence of price variation, proportionality is the only
testable implication of the collective model. This means that if we have
only one distribution factor, then we can never reject the hypothesis of col-
lective rationality. Any extra restrictions for a collective model require that
additional assumptions be made on the form of individual preferences. For
instance, restrictions exist even for a single distribution factor when some
goods are private and/or are consumed exclusively by one member of the
household. It may surprise readers that in the absence of price variation,
proportionality is the full empirical content of the collective model. Recall,
however, that in the unitary model, without price variation, any demands
as a function of total expenditure are compatible with utility maximization.
This result provides two distinct ways of testing for efficiency. Condition
(5.3) leads to tests of cross-equation restrictions in a system of uncondi-
tional demand equations. An alternative method, implied by (5.4), tests for
exclusion restrictions in a conditional demand framework. Empirically, the
latter is likely to be more powerful for at least two reasons. First we can em-
ploy single equation methods (or even non-parametric methods). Second,
single equation exclusion tests are more robust than tests of the equality of
parameters across equations. Both tests generalize easily to a framework in
which domestic goods are produced by the household. Adding a domestic
production function that relates market inputs and domestic labor to goods
actually consumed by household members does not modify the above tests
on household demands for market goods.
As discussed in Chapter 3, the bargaining version of the collective model
has attracted lot of attention. A bargaining framework should be expected
to impose additional restrictions to those discussed above. Indeed, an easy
test can be described as follows. Assume that some distribution factors,
which are part of a K 0 -sub-vector z0 , are known to be positively correlated
with member b’s threat point, while others, constituting a K 00 -sub-vector
z00 , are known to favor a. Then in program (5.1) μ should decrease with
distribution factors in z00 and increase with those in z0 . This property can
readily be tested; it implies that,
∂ĝi /∂zk0 ∂ĝj /∂zk0

00
= 00
≤ 0 for i, j = 1, .., n; k = 1, ..., K 0 ; m = 1, ..., K 00
∂ĝi /∂zm ∂ĝj /∂zm
Should one be willing to go further and assume, for instance, that only the
ratio z10 /z200 of distribution factors matters, then we have in addition:
∂ĝi ∂ĝi
+ =0 ∀i = 1, .., n
∂ ln(z1 ) ∂ ln(z200 )
0
This is simple to test and easy to interpret.
5.2.2 Identifiability.
A more difficult issue arises when we consider identifiability. That is, when
is it possible to recover the underlying structure from the sole observation of
household behavior? Note, first, that the nature of the data strongly limits
what can be recovered. For instance, one cannot hope to identify utility
functions in the absence of price variations. ‘Identifiability’, in this context,
essentially means recovering individual Engel curves (that is, demand as a
function of income) and the decision process, as summarized by the Pareto
weights or (in the private good case) by the sharing rule, again as functions
of income and distribution factors only.
With these precautions in mind, we start with some mathematical results

concerning integrability that are useful in the current context. Suppose we
have a smooth unknown function f (x, y) with non-zero partials fx and fy .
Suppose first that we observe:
h1 (x, y) = fx (x, y) and h2 (x, y) = fy (x, y) (5.5)
If f (.) is twice continuously differentiable, these two functions must satisfy

the cross derivative restriction h1y (x, y) = h2x (x, y). In general, these condi-
tions can be translated into empirical tests of the hypothesis that h1 (.) and
h2 (.) are indeed partials of the same function. Moreover, if this symmetry
condition is satisfied, then f (.) is identifiable up to an additive constant.
That is, if f¯ is a solution of (5.5), then any alternative solution must be of
the form f¯ (x, y) + k where k is an arbitrary constant.
Suppose now that rather than observing the partials themselves we only
observe their ratio:
fy
h (x, y) = (5.6)
fx
Given h (x, y), f (x, y) is identifiable ‘up to a strictly monotone transforma-
tion’. That is, we can ¯
¡ recover ¢ some f (x, y) such that any solution is of the
form f (x, y) = G f¯ (x, y) where G (.) is an arbitrary strictly monotone
function.
In general, when f (.) has more than two arguments, f (x1 , ..., xn ), as-
sume that we observe m < n − 1 ratios of partials, say those involving
the m + 1 first partials of f : ff21 , ..., fm+1
f1 . Then f is identifiable up to a
function of the other variables. That is, we can identify some f¯ (x1 , ..., xn )
such that any solution is of the form
¡ ¢
f (x1 , ..., xn ) = G f¯ (x1 , ..., xn ) , xm+2 ..., xn
where G (.) is an arbitrary function. In particular:
• if we observe only one ratio of partials, say h (x1 , ..., xn ) = f1 /f2 , then
f (.) is identifiable up to a function of the other variables (x3 , ..., xn ).
• if we observe all ratios of partials, then f (.) is identifiable up to an
arbitrary, strictly monotone transformation. Note, as well, that when-
ever we observe more than one ratio of partials, testable restrictions
are generated. These generalize the previous cross-derivative condi-
tions.
• Finally, if in addition the m+1 first ratios ff21 , ..., fm+1
f1 only depend on
(x1 , ..., xm+1 ) then f¯ can be chosen to only depend on (x1 , ..., xm+1 )
- this is an usual separability property.
We can now return to the identifiability problem for the collective model.
Even in the most general case (no identifying restriction beyond efficiency),
some (but by no means all) of the structure can be recovered from the
observation of demand functions. To see why, note that by equation (5.3)
we have:
∂ĝi /∂zk ∂μ/∂zk μ
= = k = κk for all i and k (5.7)
∂ĝi /∂z1 ∂μ/∂z1 μ1
The left hand side expression is potentially observable so that we can iden-
tify the ratio of partials of μ (x, z) with respect to distribution factors. Since
the right hand side does not depend on the good, the ratio on the left hand
side must be the same for all goods; this is the proportionality condition.
Given the ratio of partials of the Pareto weight, we can recover μ (.) up to
some function of x. That is, we can recover a particular Pareto weight μ̄
such the true Pareto weight μ must be of the form:
μ (x, z) = m (x, μ̄ (x, z)) (5.8)
for some unobserved function m (.).

The ratio κk in equation (5.7) has a natural interpretation in terms of
power compensation. Assume, for instance, that μ1 > 0 and μk < 0 so
that z1 favors b while zk serves a. If zk is increased by some infinitesimal
quantity dzk then κk dzk is the increase in z1 required to offset the change
and maintain the same balance of power. Power compensations may be
important for welfare analysis, whenever a ‘shift of power’ has to be com-
pensated. The good news is that even in the most general version of the
collective model, they can be directly recovered from observed demands.
Furthermore, the proportionality condition (5.3) imposes that the estima-
tion of the power compensation ratio does not depend on the particular
commodity chosen. An alternative and important interpretation of this re-
sult is that the model always behaves ‘as if’ there were only one factor, μ̄,
influencing the individual’s relative powers. Whatever the actual number
of distribution factors, they always operate through the index μ̄. Moreover,
this index is identifiable. What is not identifiable in the general case is the
exact impact of the index on the actual Pareto weight; an impact that will
in general depend on the level of total expenditures.
5.2.3 Private consumption.

Although useful, recovery of the Pareto weight up to a strictly monotone
function that also depends on total expenditure is far short of what is
needed for some important purposes. Is it possible to recover more? To
achieve this, we need either better data or more theory restrictions. As
an example of the latter, consider the particular but useful case in which
all commodities are privately consumed and preferences are either egoistic
or caring. As we have seen in chapter 4, efficiency is then equivalent to
the existence of a sharing rule in which a receives ρ (x, z) and b receives
(x − ρ (x, z)). Individual a solves:
max υa (qa ) subject to e0 qa = ρ (x, z) (5.9)
and similarly for b. It follows that the household aggregate demand for
commodity i takes the form:
qi (x, z) = qia (ρ (x, z)) + qib (x − ρ (x, z))
where qis is s’s demand for good i. The question is: what can be said about
qia , qib and ρ from the observation of household demands qi for i = 1, ...n.
Equation (5.7) has an equivalent in this context:
∂qi /∂zk ∂ρ/∂zk

= for all k (5.10)
∂qi /∂z1 ∂ρ/∂z1
This result remains valid in the presence of public goods, provided that
the sharing rule is taken to be conditional on public goods (as described in
subsection 5.2 of Chapter 4). The potential observability of the left hand
side of equation (5.10) means that we can recover the sharing rule up to an
arbitrary monotone function of total expenditures x. In other words, we can
recover some ρ̄ (x, z) such that the true sharing rule must be of the form
ρ (x, z) = G (ρ̄ (x, z) , x) for some mapping G. And, as above, instead of
analyzing the impact of each distribution factor independently, we may just
consider the impact of the ‘index’ ρ̄. Consequently we can always consider
the case of a unique distribution factor; no loss of generality results.
5.2.4 Assignability.
Up until now we have considered the case where we only observe aggregate
household demands. In some cases, we can observe the consumption of a
particular good by each partner. That is, for some goods we observe qia
and qib . We refer to such a good as being assignable. The most widely used
example of an assignable good is clothing: in expenditure surveys we always
see a distinction made between men and women’s clothing. An alternative
terminology is that each of the clothing commodities is an exclusive good.4
That is, an exclusive good is one that is consumed by a unique person in
the household.
Suppose that we observe the individual consumption of the first good
and estimate q̂1a (x, z) and q̂1b (x, z). Assuming, without loss of generality,
that there is only one distribution factor, the collective demands q̃1s are
4 In general, individual consumptions of an assignable good have the same price,
whereas exclusive goods have different prices. The distinction is ineffective in the present
context, but will become important when price variations are considered.
related to the observable demands q̂1s by:

q̂1a (x, z) = q̃1a (ρ (x, z)) (5.11)
q̂1b (x, z) = q̃1a (x − ρ (x, z)) (5.12)
Thus:
∂ q̂1a /∂x ρx
=
∂ q̂1a /∂z ρz
∂ q̂1b /∂x 1 − ρx
= − (5.13)
∂ q̂1b /∂z ρz
Thus the two ratios ρx /ρz and (1 − ρx ) /ρz are identifiable. There is a
unique solution to these two equations for (ρx , ρz ) if and only if:
∂ q̂1a ∂ q̂1b ∂ q̂ b ∂ q̂ a
Γ= − 1 1 6= 0 (5.14)
∂x ∂z ∂x ∂z
If this condition holds, we can identify the partials of ρ:
1 ∂ q̂1a ∂ q̂1b
ρx =
Γ ∂x ∂z
1 ∂ q̂1a ∂ q̂1b
ρz = (5.15)
Γ ∂z ∂z
By the result before (5.6), knowing the partials allows us to identify the
function itself, up to an additive constant: ρ = ρ (x, z) + k. Thus we can
learn everything about the sharing rule from observing the assignment of
a single good, except its location. One good is sufficient because the same
Pareto weight function appears in all goods; see equation (5.2). Moreover,
new restrictions are generated, since
µ ¶ µ ¶
∂ ∂ρ ∂ ∂ρ
=
∂z ∂x ∂x ∂z
This provides a test for assignability of any particular good within the
collective setting.
Finally, what about the individual Engel curves of the two spouses? First,
for any value of the constant k, (5.11) and (5.12) identify individual de-
mands for commodity 1. Consider, now, commodity i; remember that, in
general, i is neither exclusive nor assignable. Still, from:
q̂i (x, z) = q̃ia (ρ (x, z)) + q̃ib (x − ρ (x, z)) (5.16)
we have:
∂ q̂i dq̃ia dq̃ b
= ρx + i (1 − ρx )
∂x dρ dρ
µ a ¶
∂ q̂i dq̃i dq̃ b
= − i ρz (5.17)
∂z dρ dρ
Since the left hand side is observed and we have (ρx , ρz ) we invert (so
long as ρz 6= 0) and identify q̃ai and q̃bi up to an additive constant. We
conclude that the presence of an assignable good is sufficient to identify
(up to additive constant) the sharing rule and individual demands for each
commodity, including the non assignable ones.
We thus get a great deal of mileage from the presence of one assignable
(or two exclusive) goods. Can we do without? Surprisingly enough, the an-
swer is positive. Bourguignon, Browning and Chiappori (2009) prove the
following strong result: if we observe household demand (as a function of
total expenditures x and a distribution factor z) for at least three commodi-
ties, then in general we can recover individual demands and the sharing rule
up to the same additive constants as before and (this is the only twist) up
to a permutation of a and b.5 This result arises from equation (5.2) and
follows since we have three demands that depend on the one Pareto weight
function. For the technical details, see Bourguignon et al (2009). The re-
sult requires observation of cross partial terms involving x and z; since
these are are often difficult to pin down in empirical work, this route for
identifying the sharing rule is less robust than using assignability. It is im-
portant to note that the identification here does require the existence of at
least one distribution factor. Without a distribution factor no information
concerning the preferences or the sharing rule can be recovered.
5.3 Observing price responses.

5.3.1 Testing the collective model
The basic result
We now turn to the situation in which we observe variation in prices as well
as in income and distribution factors. This would be the case, for instance,
if we have panel data, or if the cross sectional data exhibit important and
exogenous fluctuations in prices. Then strong tests are available. Moreover,
the model can be proved to be identifiable under reasonably mild exclusion
conditions.
Again, we consider a two person household for expositional convenience.
Tests of the most general form of the collective model are based on the fun-
damental SNR1 condition demonstrated in Chapter 4. Namely, the Slutsky
matrix S (which can be derived from estimated demand functions) must
be of the form:
5 Identifiability, here, is only ‘generic’. It is indeed possible to construct examples in
which it does not hold, but these examples are not robust. For instance, if individual
demands and the sharing rule are all linear, identification does not obtain. However,
adding quadratic terms is sufficient to guarantee identification except maybe for very
specific values of the coefficients.
S = Σ+R (5.18)
where Σ is symmetric, negative and R is of rank at most one.

Direct tests of (5.18) are not straightforward, because the theorem sim-
ply says that there exists such a decomposition. To construct a testable
implication of the symmetry of Σ, consider the matrix M defined by:
M = S − S0
where S 0 is the transpose of S. Since Σ is symmetric:
M = R − R0
and since R is of rank (at most) 1, M is of rank (at most) 2. This property
is easy to test, using either standard rank tests or more specific approaches.
Note, however, that five commodities (at least) are needed for that purpose.
The reason is that neither M nor S are of full rank. Indeed, a standard
result of consumer theory, stemming from homogeneity and adding up,
states that
π 0 S = Sπ = 0
where π denotes the price vector. It follows that M π = 0, and M cannot

be invertible. Moreover, M is antisymmetric (equal to minus its transpose);
hence its rank must be even. With four commodities, M is a 4×4, antisym-
metric, non-invertible matrix, so that its rank can never exceed 2 anyway.
Negative semidefiniteness of Σ, on the other hand, can be directly tested
on the Slutsky matrix. Indeed, among the eigenvalues of S, one is zero
(reflecting non invertibility); among the others, one (at most) can be posi-
tive. Therefore, while symmetry of Σ cannot be tested from less than five
goods, three are sufficient to test negativeness. In practice, such a test
may however not be very powerful. An alternative approach is use revealed
preference techniques; following an early discussion in Chiappori (1988),
Cherchye, De Rock and Vermeulen (2007) and (2008) provide a complete
characterization of the revealed preference approach to collective models.
Distribution factors
Distribution factors can be readily introduced for parametric approaches.
Using equation (4.18) in Chapter 4, Browning and Chiappori (1998) prove
the following
³ result.´ Take any distribution factor k, and compute the vector
0 ∂ q̂1 ∂ q̂n
v = ∂zk , ..., ∂zk . Then replacing any column (or any row) of M with v
should not increase the rank. It is relatively simple to devise an empirical
test for this; see Browning and Chiappori (1998) for details.
Some extensions
Finally, a similar investigation has been conducted for other, non-unitary
models of household behavior. Lechène and Preston (2009) analyze the
demand function stemming from a non cooperative model (involving pri-
vate provision of the public goods) similar to that discussed in Chapter 4.
They show that, again, a decomposition of the type (5.18) holds. However,
the rank conditions on the ‘deviation’ matrix R are different; specifically,
Lechène and Preston show that the rank of R can take any value between 1
and the number of public goods in the model. Recently, d’Aspremont and
Dos Santos Fereira (2009) have introduced a general framework that pro-
vides a continuous link between the cooperative and the non cooperative
solutions. In their setting, couples are characterized by a pair of parameters
that indicate how ‘cooperatively’ each agent behaves. Again, they derive a
(5.18) decomposition; however, the rank of matrix R can now take values
between 1 and twice the number of public goods. On the empirical front,
Del Bocca and Flinn (2009) have proposed models in which agents may
cooperate at some coordination cost; the decision to cooperate (or not) is
then endogeneously derived from the model.
5.3.2 Identifying the collective model

In the presence of price variation, the identifiability problem can be stated
in full generality; indeed, when price effects are observable it may be pos-
sible to recover individual preferences and demand functions (not only the
Engel curves). Clearly, identifying assumptions ¡ are necessary;
¢ ¡ in its most
¢
general version (with general preferences ua qa , qb , Q and ub qa , qb , Q ),
there exists a continuum of different structural models generating the same
demand function. For instance, Chiappori and Ekeland (2006) show that
any function satisfying SNR1 (see equation (5.18)) can be generated as the
Pareto efficient demand of a household in which all consumption is public,
and also of an (obviously different) household in which all consumption is
private. Therefore, we ¡assume ¢ in this subsection that preferences are egois-
tic (ua (qa , Q) and ub qb , Q ) , although our results have implications for
caring preferences as well. We also assume that the econometrician knows
which goods are private and which are public.
Even with egoistic preferences, however, the collective structure cannot
in general be fully identified from demand data. To give a simple counterex-
ample, assume for a moment that all¡goods are publicly ¢ consumed
¡ and con-¢
sider two pairs of utility functions, ua (Q) , ub (Q) and ũa (Q) , ũb (Q)
with
¡ ¢
ũa = F ua , ub
¡ ¢
ũb = G ua , ub
for two arbitrary, increasing functions F and G. It is easy to check that any
¡ ¢
¡allocation
¢ that is Pareto efficient for ũa , ũb must be Pareto efficient for
ua , ub as well; otherwise one could increase ua and ub without violating
the budget constraint, but this would increase ũa and ũb ,¡a contradiction.
¢
It follows that any ¡demand ¢ that can be rationalized by ũa , ũb can also
be rationalized by ua , ub (of course, with different Pareto weights), so
that the two structures are empirically indistinguishable. Since F and G
are arbitrary, we are facing a large degree of indeterminacy.
A negative result of this type has a simple meaning: additional identi-
fying hypotheses are required. If there are at least four commodities, then
Chiappori and Ekeland (2009) prove the following results.
• If for each household member there is a commodity that this member
does not consume and is consumed by at least one other member,
then generically one can exactly recover the collective indirect util-
ity function6 of each member (up to an increasing transform). For
any cardinalization of these utility functions, Pareto weights can be
recovered. If there are only two persons in the household then this ex-
clusion restriction is equivalent to an exclusivity condition that each
member has one good that only they consume; with at least three
members, exclusion is weaker than exclusivity.
• If all commodities are publicly consumed, identifying collective indi-
rect utility functions is equivalent to identifying individual utilities.
With private consumptions, on the contrary, any given pair of collec-
tive indirect utilities is compatible with a continuum of combinations
of individual utilities and (conditional) sharing rules. However, all
these combinations are welfare equivalent, in the sense that they gen-
erate the same welfare conclusions. For instance, if a given reform is
found to increase the welfare of a while decreasing that of b under a
specific combination of individual utilities, the same conclusion will
hold for all combinations.
• Finally, if there is at least one distribution factor, the exclusivity re-
striction can be relaxed and identifiability obtains with one assignable
good only.
In the literature the traditional choice for exclusive goods for husband
and wife is men and women’s clothing respectively. There is a subtle but im-
portant difference between the notion of exclusivity and that of assignabil-
ity. In both cases, we observe consumptions at the individual level. But
exclusive goods have different prices, whereas under assignability we ob-
serve individual consumptions of the same good - so there is only one price.
Therefore, when considering clothing as two exclusive goods we have to as-
sume they have different prices. In practice prices for men and women’s
6 See section 2.2 of chapter 4 for the definition of the collective indirect utility function.
clothing tend to be very colinear and we have to treat clothing as an

assignable good.
Two remarks are in order at that point. First, the identifiability result
just presented is, by nature, non parametric, in the sense that it does not
rely on the choice of a specific functional form for either preferences or
Pareto weights.7 Under an explicitly parametric approach, stronger iden-
tification results may obtain; for instance, it may be the case that one
exclusive good only is sufficient to identify all the relevant parameters.
Clearly, these additional properties are due to the specific functional form
under consideration. Second, the result is generic, in the sense that it holds
for ‘almost all’ structures. An interesting remark is that (non-generic) ex-
ceptions include the case in which Pareto weights are constant; in such a
case, the collective indirect utilities are not identifiable in general.8 To see
why, simply note that, in that case, the household maximizes a collective
utility of the form:
¡ ¢ ¡ ¢
U qa , qb , Q = μua (qa , Q) + ub qb , Q (5.19)
under budget constraint; remember that here μ is a constant. Standard

results in consumer theory guarantee that we can recover U from observed
(household) demand. However, for any given U there exists a continuum of
ua and ub such that (5.19) is satisfied. For instance, take any such ua and
ub that are strongly increasing and concave, pick up any smooth function
φ, and define ūa and ūb by:
ūa (qa , Q) = ua (qa , Q) + εφ (Q)

¡ ¢ ¡ ¢
ūb qb , Q = ub qb , Q − μεφ (Q)
Then μūa + ūb = U and (5.19) is satisfied; moreover, on any compact set,
ūa and ūb are concave and increasing for ε small enough.
Ironically, the case of a constant Pareto weight corresponds to the Samuel-
son justification of the unitary setting, in which a single, price-independent
welfare index is maximized. From an identification viewpoint, adopting a
unitary framework is thus a very inappropriate choice, since it rules out the
identification of individual welfares.
Our general conclusion is that welfare relevant structure is indeed identi-
fiable in general, provided that one can observe one exclusive consumption
7 This notion of ‘non parametric’, which is used for instance by econometricians,
should be carefully distinguished from the perspective based on revealed preferences -

which, unfortunately, is also often called ‘non parametric’. In a nutshell, the revealed
preferences approach does not require the observability of a demand function, but only
of a finite number of points; it then describes relationship that must be satisfied for the
points to be compatible with the model under consideration. This view will be described
in Subsection 5.3.4.
8 This case is ‘non generic’ in the sense that in the set of continuous functions, constant
functions are non-generic.

per member (or one overall with a distribution factor). However, identifia-
bility fails to obtain in a context in which the household behaves as a single
decision maker.
5.3.3 A simple example

The previous results can be illustrated by the following example, directly
borrowed from Chiappori and Ekeland (2009).Consider individual prefer-
ences of the LES type:
n
X N
X
U s (q s , Q) = αsi log (qis − csi ) + αsj log (Qj − Cj ) , s = a, b
i=1 j=n+1
PN
where the parameters αsi are normalized by the condition s
i=1 αi =
s
1 for all s, whereas the parameters ci and Cj are unconstrained. Here,
commodities 1 to n are private while commodities n + 1 to N are public.
Also, given the LES form, it is convenient to assume that the household
maximizes the weighted sum μU a + (1 − μ) U b , where the Pareto weight μ
has the simple, linear form:
μ = μ0 + μx x + μz z, s = a, b
Household demand
The group solves the program:
⎛ ⎞
Xn N
X
¡ 0 ¢
max μ + μx x + μz z ⎝ αai log (qia − cai ) + αaj log (Qj − Cj )⎠
i=1 j=n+1
⎛ ⎞
Xn N
X
¡ ¡ 0 ¢¢ ¡ ¢
+ 1 − μ + μx x + μz z ⎝ abi log qib − cbi + αbj log (Qj − Cj )⎠
i=1 j=n+1
under the budget constraint:

¡ ¢
p0 qa + qb + P0 Q = x
where one price has been normalized to 1. Individual demands for private
goods are given by:
⎛ ⎞
¡ ¢ X X
pi qia = pi cai + αai μ0 + μx x + μz z ⎝x − pi csi − Pj Cj ⎠
i,s j
⎛ ⎞
£ ¡ ¢¤ X X
pi qib = pi cbi + αbi 1 − μ0 + μx x + μz z ⎝x − pi csi − Pj Cj ⎠
i,s j
generating the aggregate demand:
£ ¡ ¢ ¡ ¡ ¢¢¤
pi qi = pi ci + αai μ0 + μx x + μz z + αbi 1 − μ0 + μx x + μz z Y (5.20)
and for public goods:
£ ¡ ¢ ¡ ¡ ¢¢¤
Pj Qj = Pj Cj + αaj μ0 + μx x + μz z + αbj 1 − μ0 + μx x + μz z Y
³ P P ´
Here, ci = cai + cbi and Y = x − i,s pi csi − j Pj Cj . The household
demand is thus a direct generalization of the standard LES, with additional
quadratic terms in x2 and cross terms in xpi and xPj , plus terms involving
the distribution factor z; one can readily check that it does not satisfy
Slutsky symmetry in general, although it does satisfy SNR1.
A first remark is that cai and cbi cannot be individually identified from
group demand, since the latter only involves their sum ci . As discussed
above, this indeterminacy is however welfare irrelevant, because the collec-
tive indirect utilities of the wife and the husband are, up to an additive
constant:
¡ ¢ X a X
W a (p, P, x, z) = log Y + log μ0 + μx x + μz z − αi log pi − αaj log Pj
i j
¡ ¡ ¢¢ X b X
W b (p, P, x, z) = log Y + log 1 − μ0 + μx x + μz z − αi log pi − αbj log Pj
i j
which does not depend on the csi .

Secondly, the form of aggregate demands
is such that private and public goods have exactly the same structure. We
therefore simplify our notations by defining
ξ i = qi for i ≤ n, ξ i = Qi for n < i ≤ N
and similarly
γ i = ci for i ≤ n, γ i = Ci for n < i ≤ N
π i = pi for i ≤ n, π i = Pi for n < i ≤ N
so that the group demand has the simple form:
£ ¡ ¢ ¡ ¡ ¢¢¤
πi ξ i = πi γ i + αai μ0 + μx x + μz z + αbi 1 − μ0 + μx x + μz z Y
(5.21)
leading to collective indirect utilities of the form:
¡ ¢ X a
W a (p, P, x, z) = log Y + log μ0 + μx x + μz z − αi log π i
i
¡ ¡ ¢¢ X b
W (p, P, x, z) = log Y + log 1 − μ0 + μx x + μz z −
b
αi log π i
i
It is clear, on this form, that the distinction between private and public
goods can be ignored. This illustrates an important remark: while the ex
ante knowledge of the public versus private nature of each good is necessary
for the identifiability result to hold in general, for many parametric forms
it is actually not needed.
Identifiability
The general case
The question, now, is whether the empirical estimation of the form (5.21)
allows to recover the relevant parameters - namely, the αsi , the γ i , and the
μα . We start by rewriting (5.21) as:
Ã !
¡ b ¡ a ¢ 0 ¡ a ¢ x ¢ X
b b z m
π i ξ i = π i γ i + αi + αi − αi μ + αi − αi (μ x + μ z) x − πmγ
m
(5.22)
The right hand side of (5.22) can in principle be econometrically identi-
fied; we can thus recover the coefficients of the variables, namely x, x2 , xz,
the π m and the products xπ m and zπm . For any i and any m 6= i, the ratio
of the coefficient of x by that of π m gives γ m ; the γ m are therefore vastly
overidentified. However, the remaining coefficients are identifiable only up
to an arbitrary choice of two of them. Indeed, an empirical estimation of
the right hand side of (Ei0 ) can only recover for each j the respective
¡ coeffi-
¢
cients of x, x2 and xz, that is the three expressions Kxj = αbj + αaj − αbj μ0 ,
j
¡ ¢ ¡ ¢
Kxx = αaj − αbj μx and Kxz j
= αaj − αbj μz . Now, pick up two arbitrary
¡ ¢
values for μ0 and μx , with μx 6= 0. The last two expressions give αaj − αbj
and μz ; the first gives αbj therefore αaj .
As expected, a continuum of different models generate the same aggre-
gate demand. Moreover, these differences are welfare relevant, in the sense
that the individual welfare gains of a given reform (say, a change in prices
and incomes) will be evaluated differently by different models; in practice,
the collective indirect utilities recovered above are not invariant across the
various structural models compatible with a given aggregate demand.
A unitary version of the model obtains when the Pareto weights are
constant: μx = μz = 0. Then Kxz j
= 0 for all j (since distribution factors
cannot matter9 ), and Kxx j
= 0 for all j (demand must be linear in x,
since ¡a quadratic
¢ term would violate Slutsky). We are left with Kxj =
αbj + αaj − αbj μ0 , and it is obviously impossible to identify independently
αaj , αbj and μ0 ; as expected, the unitary framework is not identifiable.
Identification under exclusion
We now show that in the non-unitary version of the collective framework,
an exclusion assumption per member is sufficient to exactly recover all the
9 For a discussion of the role of distribution factor in a unitary context, see Browning,
Chiappori and Lechene (2006).

coefficients. Assume, indeed, that member a does not consume commodity

1 and member b does not consume commodity 2; that is, αa1 = αb2 = 0.
Then equation (E10 ) gives:
¡ ¢
αb1 1 − μ0 = Kx1 , −αb1 μx = Kxx
1
, −αb1 μz = Kxz
1
while (E20 ) gives:
αa2 μ0 = Kx2 , αa2 μx = Kxx

2
, αa2 μz = Kxz
2
Combining the first two equations of each block and assuming μx 6= 0, we

get:
1 − μ0 K1 μ0 K2
x
= − 1x and x
= 2x
μ Kxx μ Kxx
therefore, assuming Kx2 Kxx
1
− Kx1 Kxx
2
6= 0
1 − μ0 Kx1 Kxx
2
0 Kx2 Kxx
1
= − and μ =
μ0 Kx2 Kxx
1 Kx2 Kxx
1 − K 1K 2
x xx
It follows that
2
Kxx K2 K1
μx = 2
μ0 = 2 1 xx xx1 2
Kx Kx Kxx − Kx Kxx
and all other coefficients can be computed as above. It follows that the
collective indirect utility of each member can be exactly recovered, which
allows for unambiguous welfare statements. As mentioned above, identifi-
ability is only generic in the sense that it requires Kx2 Kxx
1
− Kx1 Kxx2
6= 0.
Clearly, the set of parameters values violating this condition is of zero mea-
sure.
Finally, it is important to note that this conclusion requires μx 6= 0; in
particular, it does not hold true in the unitary version, in which μx = μz =
0. Indeed,
¡ ¢the same exclusion restrictions as above only allow to recover
αb1 1 − μ0 = Kx1 and αa2 μ0 = Kx2 ; this is not sufficient to identify μ0 , let
alone the αij for j ≥ 3. This confirms that the unitary version of the model
is not identified even under the exclusivity assumptions that guarantee
generic identifiability in the general version.
5.3.4 The revealed preference approach

Up until now we have considered analysis that posits that we can estimate
smooth demands and test for the generalized Slutsky conditions for inte-
grability. An alternative approach to empirical demand analysis that has
gained ground in the last few years is the revealed preference (RP) approach
that derives from Afriat (1967) and Varian (1982). This style of analysis
explicitly recognizes that we only ever have a finite set of observations on
prices and quantities which cannot be used to directly construct smooth
demand functions without auxiliary assumptions. The revealed preference

approach instead identifies linear inequality conditions on the finite data
set that characterize rational behavior. The most attractive feature of the
Afriat-Varian approach is that no functional form assumptions are imposed.
Moreover powerful numerical methods are available to implement the RP
tests. The drawback of the RP approach is that even when the data satisfy
the RP conditions, we can only set identify preferences; see Blundell et al
(2008`).
Generalizing the unitary model RP conditions to the collective setting
was first achieved in Chiappori (1988) for a specific version of the collec-
tive model. The conditions for the general model have been established in
Cherchye, De Rock and Vermeulen (2007), (2009a) and Cherchye, De Rock,
Sabbe, and Vermeulen (2008); these papers provide a complete characteri-
zation of the collective model in a revealed preference context. This requires
several significant extensions to the RP approach for the unitary model. In
particular, these authors allow for non-convex preferences and develop novel
(integer programming) methods since the linear programming techniques
that work for the unitary model are not applicable for the collective model.
The tests for ‘collective rationality’ require finding individual utility levels,
individual marginal utilities of money (implying Pareto weights) and indi-
vidual assignments for private goods and Lindahl prices for public goods.
As in the unitary model, these methods can only set identify the preferences
of the household members and the Pareto weight. Cherchye, De Rock and
Vermeulen (2009b) apply these methods to a Russian expenditure panel.
5.4 The case of labor supply

5.4.1 Egoistic preferences and private consumption
A large part of the empirical literature on household behavior is devoted
to labor supply. The theory has been presented in Section 4 of Chapter 4;
here we concentrate on the empirical implications. Most empirical works
consider the simple setting with egoistic preferences and private consump-
tion; see subsection 4.2 of Chapter 4. In this framework, results have been
established by Chiappori (1988, 1992) and Chiappori, Fortin and Lacroix
(2002). Regarding testability, strong implications can be derived, even in
this simple setting. Even more remarkable is the fact that the observation
of individual labor supplies, as functions of wages, non labor income and
distribution factors, allows us to identify the sharing rule up to an additive
constant. We start from the two leisure demand equations:
¡ ¢ ¡ ¡ ¢¢
la wa , wb , y, z = ˜la wa , ρ wa , wb , y, z (5.23)
¡ ¢ ¡ ¡ ¢¢
lb wa , wb , y, z = ˜lb wb , y − ρ wa , wb , y, z (5.24)
where ˜la denotes the Marshallian demand for leisure by person a, y is

full income and ρ (.) is a’s share of full income. We assume that both
partners shares are increasing in full income, 0 < ∂ρ/∂y < 1, and that the
distribution factor is ‘meaningful’, ∂ρ/∂z 6= 0.
Taking derivatives through (5.23):
∂la ∂ ˜la ∂ρ
=
∂wb ∂ρ ∂wb
= (5.25)
∂y ∂ρ ∂y
=
∂z ∂ρ ∂z
so that:
∂la /∂z ∂ρ/∂z
= (5.26)
∂la /∂y ∂ρ/∂y
Similarly for b:
∂lb ∂ ˜lb ∂ρ
= −
∂wa ∂y b ∂wa
µ ¶
∂lb ∂ ˜lb ∂ρ
= 1− (5.27)
∂y ∂y b ∂y
∂lb ˜ b
∂ l ∂ρ
= − b
∂z ∂y ∂z
so that:
∂lb /∂z ∂ρ/∂z
b
=− (5.28)
∂l /∂y 1 − ∂ρ/∂y
s
∂l /∂z
For notational simplicity, let F s denote the fraction ∂l s /∂y for s = a, b;
s
note
¡ a bthat F ¢ can in principle be observed (or estimated) as a function of
w , w , y, z , and that F a = F b would imply
∂ρ/∂z ∂ρ/∂z
=−
∂ρ/∂y 1 − ∂ρ/∂y
which is impossible if ∂ρ/∂z 6= 0.

Now, (5.26) and (5.28) can be solved in ∂ρ/∂z and ∂ρ/∂y (since F b 6=
a
F ):
∂ρ Fb
=
∂y Fb − Fa
∂ρ F aF b
=
∂z Fb − Fa
We thus conclude that the partials of ρ with respect to income and distri-
bution factor are identifiable.
Finally, the first two equations of (5.25) and of (5.27) give respectively:
∂ρ ∂la /∂wb ∂ρ ∂la /∂wb Fb

= = and
∂wb ∂la /∂y ∂y ∂la /∂y F b − F a
µ ¶
∂ρ ∂lb /∂wa ∂ρ ∂la /∂wb F a
= 1− =− a (5.29)
∂wa ∂lb /∂y ∂y ∂l /∂y F b − F a
The conclusion is thus that all partial derivatives of the sharing rule can be
exactly recovered from the observation of the two labor supply functions.
From the sole observation of labor supplies, one can recover the impact
of wages, non labor income and distribution factors on the sharing rule.
Finally, the cross derivative restrictions generate additional testable pre-
dictions. The reader may realize that this conclusion is by no means sur-
prising; indeed, it stems from the general results presented in the previous
subsection.
The sharing rule itself is identified up to an additive constant; that con-
stant cannot be identified unless either all commodities are assignable or
individual preferences are known (for instance, from data on singles). To
see why, take labor supply functions la and lb that satisfy (5.23) and (5.24)
for some sharing rule ρ and some Marshallian demands l̃s derived from
individual utilities us , s = a, b. Now, for some constant K, define ρK , uaK
and ubK by:
¡ ¢ ¡ ¢
ρK wa , wb , y, z = ρ wa , wb , y, z + K
uaK (la , C a ) = ua (la , C a − K)
¡ ¢ ¡ ¢
ubK lb , C b = ub lb , C b + K
It is easy to check that the Marshallian demands derived from ρK , uaK and
ubK satisfy (5.23) and (5.24). The intuition is illustrated in Figure 5.1 in the
case of a. Switching from ρ and ua to ρK and uaK does two things. First,
the sharing rule, therefore the intercept of the budget constraint, is shifted
downward by K; second, all indifference curves are also shifted downward
by the same amount. When only labor supply (on the horizontal axis) is
observable, these models are empirically indistinguishable.
Note, however, that the models are also welfare equivalent (that is, the
constant is ‘irrelevant’), in the sense defined in section 3.3 of chapter 4:
changing the constant affects neither the comparative statics nor the wel-
fare analysis derived from the model. Technically, the collective indirect
utility of each member is the same in both models; one can readily check
that the two models generate the same level of utility for each spouse. In
the end, the optimal identification strategy depends on the question under
consideration. If one want to formulate welfare judgments, collective indi-
rect utilities are sufficient, and they can be recovered without additional
FIGURE 5.1. A Shift in the Sharing Rule
assumptions. If, on the other hand, the focus is on intrahousehold inequal-

ity, the basic model can identify the changes affecting intrahousehold in-
equality, but not its initial level ; therefore additional assumptions may be
needed. For instance, some empirical works assume that preferences are
unchanged by marriage, therefore can be identified from the labor supply
of singles; then the constant can also be recovered.
Finally, one should not conclude from the previous derivation that the
presence of a distribution factor is needed for identifiability. This is actually
not the case. The observation of individual labor supplies, as functions
of wages and non labor income, are ‘generically’ sufficient to recover the
sharing rule up to an additive constant (Chiappori 1988, 1992). However,
identification is only generic in that case; moreover, it is arguably less
robust, since it involves second derivatives of the labor supply functions.
5.4.2 Extensions
The model has been extended in various directions. First, while the assump-
tion of a unique, Hicksian composite consumption good is standard in the
labor supply literature, the model can address a more general framework.
Chiappori (2011) consider a model with two leisures and many consumption
goods that are privately (but not exclusively) consumed by the members.
The context is cross-sectional, in the sense that there is variation in wages
but not in prices. He shows that if one distribution factor (at least) is
available, then it is possible to identify (again up to additive constants)
not only the sharing rule but also the individual demands for all private
commodities, as functions of wages and non-labor income. It follows that
in a collective model of consumption and labor supply estimated on cross
sectional data, it is possible to recover the income and wage elasticities of
individual demands for each good.
Secondly, the computations above rely on the assumption that labor

supply is a continuous variable. This may fail to hold for two reasons.
First, in some households one member may elect not to participate; in
that case, the person’s labor supply is at a corner solution equal to zero.
Secondly, the structure of labor markets may put constraints on the number
of hours supplied by individuals. For instance, the choice may be only
between working part time, full time or not at all; then labor supply should
rather be modelled as a discrete variable. Extensions of the previous model
to such situations have been studied by Blundell et al (2007) and Donni
(2007).
Although very convenient, this framework has its limitations. The pri-
vateness assumption has been criticized on two grounds. First, while some
consumptions are indeed private, others are not. Children expenditures are
a typical example of public goods within the household. Blundell, Chiappori
and Meghir (2005) analyze a model similar to the previous one but for the
consumption good, which is taken to be public. They show that, again, the
model is identifiable from the observation of labor supply behavior. They
show how their approach can be extended to household production under
various specifications. A second criticism concerns the private nature of
individual leisure. It could indeed be argued that leisure is, to some extent,
publicly consumed; after all, the utility I derive from my own free time may
be higher when my spouse is available as well. The general insight, here, is
that a model in which both members’ leisure enter each individual utility is
still identifiable, provided that some other commodities are exclusive (this
is a consequence of the general identifiability results described in Section
2). Fong and Zhang (2001) analyze a framework in which leisure is partly
private and partly public; they show that one assignable good is sufficient
for identification in the presence of a distribution factor.
Finally, a standard problem with traditional models of labor supply is the
implicit assumption that time is divided between market work and leisure
- so that any moment not spend working of a wage tends to be assimilated
with leisure. This, of course, disregards domestic production, and may re-
sult in misleading evaluations. For instance, if a given reform is found to
reduce female market labor supply, we may conclude that it increases her
leisure, hence her utility, whereas the actual outcome is more domestic work
(and ultimately less leisure) for the wife. Donni (2008) shows, however, that
the direction of the mistake depends on the properties of the domestic pro-
duction function. To take an extreme example, consider the case in which
the latter is additively separable; that is, when ts denotes the time spend
on domestic production by agent s, then the outcome is:
¡ ¢
C = f a (ta ) + f b tb
Assuming that the domestic good is marketable with price p, first order
conditions require that:

ws
f 0s (ts ) =
p
which implies that the time spent on domestic production by s only depends
on their wage (and on the price of the domestic good). It follows that any
welfare judgment that ignores domestic production is in fact unbiased -
that is, a reform that is found to increase the wife’s welfare when ignoring
domestic production has the same impact even when domestic production
is taken into account and conversely. This conclusion, however, does not
hold when the productivity of the wife’s domestic work depends on the
husband’s. In the latter case, it becomes necessary to estimate a model
that explicitly allows for domestic production - which requires in general
time use data; for a short list of such works, see subsection 5.5.3.
5.5 Empirical evidence.

5.5.1 Evidence against the unitary model.
As we have seen, there are two strands to testing for the unitary model: the
Slutsky conditions and independence of behavior from distribution factors.
Regarding the former, Slutsky symmetry is often rejected on household
expenditure survey data. Rejections of Slutsky symmetry may be due to
many factors other than a failure of the unitary assumption. For example,
we might have the wrong functional form or an inappropriate grouping of
goods or be wrongly assuming separability from housing and durables or
accounting for latent heterogeneity inappropriately and so on. A widely
cited piece of evidence that the unitary assumption itself is problematic is
from Browning and Chiappori (1998) who model commodity demands using
Canadian data. Using a QAIDS formulation, they test for symmetry for
three sub-samples: single women, single men and couples with no children.
An important finding is that Slutsky symmetry is not rejected for single
women or single men, while it is (very strongly) for couples. Since most
of underlying modelling assumptions are the same across the three strata,
this suggests that it is the unitary assumption that is the problem. These
findings have been replicated by Kapan (2009) using Turkish data; and
Vermeulen (2005) obtains similar results for labor supply.
Although suggestive, the rejection of Slutsky symmetry would not, by
itself, warrant abandoning the unitary assumption. Much more convincing
are the next set of tests we discuss. The second principle implication of
the unitary model is that possible distribution factors do not have any
significant impact on the household choice variable being considered. Unlike
the test for the Slutsky conditions, such tests can be conducted whether or
not we have price variation. Table 5.1 gives a partial listing of distribution
factors that have been considered in the literature. Below we discuss the
validity of these factors. The most widely used distribution factor for this
is some measure of relative incomes, earnings or wages. Such tests are
often called tests of ‘income pooling’: only household income matters for
choice outcomes and not the source of the income.10 As we have seen,
Becker explicitly introduced the RKT to justify income pooling. Tests for
the exclusion of other distribution factors constitute a generalization of
income pooling.
Distribution factor
1 Relative income
2 Relative wages
3 Relative unearned income
4 Relative age
5 Relative education
6 Local sex ratio
7 Household income
8 Background family factors
9 Control of land
10 Previous children
11 Reported influence within household
12 Married or cohabiting
13 Divorce laws
14 Alimonies
15 Single parent benefits
16 Gender of a benefit’s recipient
TABLE 5.1. Distribution factors
Bruce (1989) provides a listing of the research on low income countries

documenting tensions within households about the use of household re-
sources. Strauss et al (2000) present an exhaustive list of tests for income
pooling for low income countries up to their publication date. Table 5.2
lists some of the studies that have considered non-unitary models.11 As
can be seen from this Table, a wide variety of outcomes and distribution
factors have been considered for many different countries. The most widely
used distribution factor is relative income (the ‘income pooling’ test). All
of the cited papers find a significant role for the distribution factors that
should not affect the outcomes in a unitary model. For instance, an early
and influential paper by Thomas (1990), based on Brazilian cross-sectional
1 0 Income pooling is a necessary condition for a unitary model but not a sufficient
condition. In particular, income pooling can hold locally if we have a noncooperative

voluntary contributions game; see section 4 of chapter 3.
1 1 This listing is by no means exhaustive and tends to focus on results from high income
countries.
data, finds that the relative share of non labor income coming from the
wife has a very significant impact on the health status of children within
the household.
This unanimity may be somewhat misleading; our impression is that
there is a strong publication bias against not finding an effect. That is, ed-
itors may not be interested in papers that confirm a conventional view by
finding an insignificant effect. Nonetheless, the evidence seems overwhelm-
ing: a principal implication of the unitary model is rejected on a wide set
of data sets for a wide range of outcomes.
Reference Outcome Country Df’s

Anderson & Baland (2002) Participation in a rosca Kenya 1
Aronsson et al (2001) Leisure demand Sweden 2,3,4,5,6
Attanasio & Lechene (2002) Commodity demands; Mexico 1
influence on various decisions
Barmby & Smith (2001) labor supplies Denmark, UK 2
Bayudan (2006) Female labor supply Philippines 2, 11
Bourguignon et al (1993) Commodity demands France 1
Browning (1995) Saving Canada 1
Browning & Bonke (2006) Commodity demands Denmark 1,7,8,10
Browning & Gørtz (2006) Commodity demands, leisures Denmark 2,4,7
Browning et al (1994) Demand for clothing Canada 1,4,7
Browning & Chiappori (1998) Commodity demands Canada 1,4
Chiappori et al (2002) labor supplies US 6
Couprie (2007) labor supply and leisure UK 2,3
Donni (2007) labor supplies, demands France 1,7
Duflo (2003) Child health South Africa 1
Ermisch & Pronzato (2006) Child support payments UK 1
Fortin & Lacroix (1997) Joint labor supply Canada 1,2
Haddad & Hoddinott (1994) Child health Cote D’Ivoire 1
Hoddinott & Haddad (1995) Food, alcohol and tobacco Cote D’Ivoire 1
Lundberg, et al (1997) Clothing demands UK 3
Oreffice (2008) Labor supply US 1
Phipps & Burton (1998) Commodity demands Canada 1
Schultz (1990) labor supplies and fertility Thailand 3
Thomas (1990) Child health Brazil 3
Udry (1996) Farm production Burkina Faso 9
Vermeulen (2005) labor supplies Netherlands 3,4,12
Ward-Batts (2008) Household demands UK 3
TABLE 5.2. Empirical collective studies
Even these results may not be fully conclusive, however, because these
rejections may in many cases have other explanations than a failure of
the unitary assumption. For example, consider a unitary demand model in
which the relative (labor or non labor) earnings of the two partners do not
affect demand behavior directly. Suppose, however, that there is unobserved
heterogeneity in tastes between husbands and wives and this heterogene-
ity is correlated with heterogeneity in earnings. For example, suppose the
relative preference for clothing between a husband and wife is correlated
with their relative tastes for work. Then we would find that the demand for
clothing (conditional on prices, total expenditure and preference factors)
will be correlated with relative earnings, with higher earners having rela-
tively more clothing expenditure than their partner. In this case, a finding
that relative clothing demands are partially correlated with relative earn-
ings is spurious in the sense that it is due to inadequate control for hetero-
geneity rather than a failure of the unitary assumption. Attempts to find
instruments to wash out this spurious correlation have not been notably
successful: it has proven impossible to find observables that are correlated
with, say, relative earnings but not with demand heterogeneity.12 Similarly,
Thomas’s findings might simply reflect the fact that some women are more
willing to invest over the long term than others; such women would be
likely to spend more on children, and also to have saved more in the past,
hence to receive more non labor income today. Such a mechanism does not
rely on a shift in powers triggered by the wife’s larger relative contribution
to total income, but only on unobserved heterogeneity between women; as
such, it is fully compatible with a unitary representation.
However, several recent papers provide strong evidence concerning in-
come pooling that can hardly be attributed to heterogeneity biases. Lund-
berg et al (1997) present quasi-experimental evidence based on a reform
of the UK child public support system in April 1977. Prior to that time
families with children received a child tax allowance and a taxable child
allowance. This effectively meant that the child benefits were paid to the
higher earner, mostly the father. After April 1977, the old scheme was
dropped in favor of a non-taxable child benefit which is paid directly to the
mother. This re-allocation of income within the household can reasonably
be treated as exogenous to the affected households. Moreover, the child
benefit was a sizable transfer (equal to 8% of male earnings for a two child
household). Thus we have a large, exogenous ‘treatment’ which can be used
to assess the importance of the distribution of income within the household.
The major confounding factor is that the reform was not revenue neutral
for all households with children and some saw a substantial rise in net
household income. LPW use UK Family Expenditure Survey cross-section
1 2 Luo (2002) estimates a demand system explicitly allowing for uncorrelated hetero-
geneity and finds that the BC results for Slutsky symmetry hold up
data from before and after the change to gauge the effect of the reform on
assignable expenditures. They focus attention on the ratio of expenditures
on children’s clothing and women’s clothing, both relative to men’s cloth-
ing. Their findings are unequivocal: both ratios rose significantly after the
reform.13 Another strong rejection is provided by Duflo (2003), who ana-
lyzes a reform of the South African social pension program for elderly that
extended the benefits to a large, previously not covered black population.
Due to eligibility criteria, the coverage is not universal; in some households,
in particular one only of the grand parents receives the benefit. Duflo uses
a difference of difference approach based on the demographics of the sib-
lings to control for selection in eligibility. She shows that the recipient’s
gender - a typical distribution factor - is of considerable importance for the
impact of the transfers on children’s health: a payment to the grandfather
has no significant effect, whereas the same amount paid to the grandmother
results in a huge improvement in the health status of girls in the family.
These contributions and several others (including a subsequent analysis on
micro data for all goods by Ward-Batts (2008)) very convincingly suggest
that income pooling is indeed strongly rejected on real data.
5.5.2 Evidence on the collective model.

Although the evidence against the unitary model in specific contexts is
not as robust as widely believed, it does add up and most researchers in
the field now seem to agree that any reasonable model should account for
spouses having different preferences and for the intrahousehold distribution
of ‘powers’ to matter for behavior. Evidence against the unitary model does
not, however, necessarily constitute evidence for the collective model. Un-
fortunately it has turned out to be difficult to devise powerful tests for
the collective model. This is because such tests must rely either on a test
of the quasi-Slutsky condition or the proportionality restriction on distri-
bution factors; see subsection 5.2.1. As regards the SN R1 restriction (see
equation (5.18)), we need price variation and at least five goods to reject
symmetry. This largely restricts our ability to test for SRN 1 in the labor
supply context, although tests based on more specific assumptions - for ex-
ample, exclusivity of leisure - or on different approaches - typically revealed
preferences - are indeed feasible (see below). Tests based on proportional-
ity are in general easier to implement, but they still require at least two
unequivocal distribution factors.
Among the few attempts to take SRN 1 to the data are Browning and
1 3 A re-analysis of the Lundberg et al episode by Hotchkiss (2005) suggests that it
may not be valid. The point at issue is that women in childless couples also appeared to
increase their clothing expenditure in the same period. Ward-Batts (2008) convincingly
contests this finding: the Hotchkiss timing is not consistent and Ward-Batts uses micro
data rather than the grouped data of Lundberg et al and Hotchkiss.
Chiappori (1998), Dauphin et al. (2009) and Kapan (2009). These works
share common features: they both estimate a demand system, using a well
known and flexible functional form (QUAIDS) that nests both the uni-
tary and the collective settings as specific cases (the former being itself
nested within the latter). While the data sets are different (a specific fea-
ture of the Turkish data considered by Kapan is the presence of important
and largely exogenous variations in relative prices, due to high inflation
over the period), they reach similar conclusions; for instance, when testing
Slutsky and SNR1 on three subsamples - single males, single females, and
couples, they all fail to reject the unitary version on singles; it on couples,
they very strongly reject the unitary version, but not SNR1. In addition,
the contributions provide interesting insights on various specific aspects
of intrahousehold decision processes. Both Browning and Chiappori and
Dauphin et al. provide additional tests using distribution factors, which
tend to support the collective model. Kapan finds that while most Turkish
families do not behave as if there was a single decision maker, a notable
exception is provided by traditional, rural households, for whom the uni-
tary version is not rejected. Finally, both Kapan and Dauphin et al. find
that older children (above 16) do play a role in the decision process.
The validity of proportionality tests, on the other hand, depends crucially
on an a priori division of demographic and environmental factors between
preference factors and distribution factors (a variable can be both). Typical
candidate preference factors include household composition, the age of one
of the spouses, the ownership of a car or a house, region of residence etc..
Typical distribution factors are listed in Table 5.1. A general concern is
that the household specific variables could be correlated with constraints
or preferences which would invalidate them as distribution factors; soci-
etal variables are less susceptible to this problem. Fortunately, as we have
shown above (see subsection 5.2.1) we only need one unequivocal distrib-
ution factor to credibly test for proportionality. To illustrate, suppose we
construct an index quantifying the extent to which laws governing divorce
favor women, and we take that index as a unequivocal distribution fac-
tor. If the index is ‘significant’ in the choice equations, we can then test
for proportionality for other candidate distribution factors. In theory, we
could simply take all of the factors that satisfy the proportionality tests
as distribution factors and assign other ‘significant’ variables as preference
factors. In practice, this may not be appealing if the factor that fails the
proportionality test is unlikely to be a preference factor. For example, if
the situation on the marriage market (as measured for instance by the lo-
cal sex ratio) impacts on demand behavior but fails the proportionality
test, we would be very reluctant to designate it a preference factor. Rather,
this would cast doubt on our original choice of an unequivocal distribution
factor (or the collective model itself!).
There is no evidence against the collective model in the papers listed
in Table 5.1. There is, however, alternative evidence against the efficiency
assumption of a different sort. The most convincing evidence of inefficient

outcomes is Udry (1996). This is a different style of test than SNR1 and
the relevance of distribution factors. Udry uses information on household
production.
To sum up: there is considerable evidence against the unitary model
and some evidence in favor of the collective model.14 What is singularly
lacking in the literature are tests for the collective model against other
non-unitary models for high income countries. This is part reflects the lack
of non-collective models that can be taken to the data.
5.5.3 Estimating the collective model

Many of the works mentioned above go beyond testing the collective model;
insofar as the predictions are not rejected, they often propose an estimation
of the structural components of the model. Although this field is still largely
in construction, we may briefly summarize some findings obtained so far.
Demand studies
Some of the works mentioned above go beyond testing the collective model;
when the predictions are not rejected, they often propose estimation of the
structural components of the model. Although this field is still largely in
construction, we may briefly summarize some findings obtained so far.
Many of the papers listed in Table 5.2 use demand data alone to test for
the collective model. Only three of them go beyond testing and impose the
collective model restrictions and then estimate the sharing rule and how it
depends on distribution factors. The first paper to do this was Browning
et al (1994). These authors use Canadian Family Expenditure Survey data
on men and women’s clothing to test for the collective model restrictions
and to identify the determinants of the sharing rule. Although they have
price data they absorb prices into year/region dummies and treat the data
as cross-sectional. Thus the ‘no price variation’ analysis of section 5.2 is
appropriate. They only consider singles and married couples who are in
fulltime employment. The distribution factors they find significant are the
difference in ages and the relative earnings of the two partners; they also
allow that total expenditure on nondurables and services enters the sharing
rule. They address directly the problem that variations in relative earnings
may be spuriously correlated with spending on clothing (higher paid jobs
might require relatively more expensive clothing) by testing whether singles
1 4 A notable exception to the latter are the results for efficient risk sharing in low in-
come countries; see, for example, Dercon and Krishnan (2000), Dubois and Ligon (2005),
Duflo and Udry (2003), Goldstein (2002), Ligon (2002). These tests, however, are based
on specific models that crucially involve specific asusmptions regarding commitment;
their discussion is therefore postponed until chapter 6 which deals with dynamic issues.
have clothing demands that depend on earnings. They find that for single
men and single women, earnings do not impact on clothing demand once
we take account of total expenditure. It is important to note that this
does not imply that clothing demand is separable from labor supply (it
is not) since they condition on both partners being in fulltime work and
effectively test for whether wages affect preferences. Given the finding for
singles, relative earnings are a reasonable candidate for being a distribution
factor for couples. As discussed in section 5.2 we cannot generally identify
the location of the sharing rule, so Browning et al simply set it equal to
one half (at the median of total expenditure) if the two partners have
the same age and earnings. They find that differences in earnings have a
highly significant but quantitatively small impact on sharing: going from
the wife having 25% of total earnings to 75% of total earnings shifts the
sharing rule by 2.3 percentage points. Differences in age are similar with
significant but small effects: going from being 10 years younger than her
husband to being 10 years older raises the wife’s share by two percentage
points. Conversely, total expenditure (taken as a proxy for lifetime wealth)
is less statistically significant but with a large effect: a 60% increase in total
expenditure increases the wife’s share by 12%. This suggests that wives in
high wealth households have a higher share of nondurable expenditure.
Browning and Bonke (2009) use a supplement to the Danish Household
Expenditure Surveys for 1999 to 2005. This supplement (designed by the
authors) takes the form of respondents recording for every expenditure in a
conventional expenditure diary for whom the item was bought: ‘mainly for
the household’, ‘for the husband’, ‘for the wife’, ‘for the children’ and ‘out-
side the household’. This is the first time that such information has been
collected in a representative survey in a high income country. Another no-
table feature of these data is that they contain a richer set of potential
distribution factors than most expenditure data sets. For example, ques-
tions were asked on the length of the current partnership; the labor force
participation of the mothers of the husband and wife when they were 14
and the marital and fertility histories of the two partners. Since all expendi-
tures are allocated in these data, a sharing rule can be constructed for each
household. This allows for the identification of the location of the sharing
rule as well as its dependence on distribution factors. These authors find
that the mean of the sharing rule is very close to one half (at the mean
of the data).15 This equality of the mean total expenditures for the two
partners masks that the sharing rule in different households varies widely.
For example the first and third quantiles for the wife’s share are 0.31 and
0.68 so that close to half of households have one partner receiving twice as
1 5 This equality of total assigned expenditures is not reflected in the expenditures on
individual goods. For example, the individual allocations show that, in mean, wives
spend more on clothing but less on alcohol and tobacco than their husbands.
much as the other. Some of this variation can be attributed to observable

differences in distribution factors but most of it is ‘latent’ heterogeneity.
Some of the significant distribution factors in Browning and Bonke (2009)
are familiar from earlier studies; for example, if the wife has a higher share
of gross income then she has a higher share of total expenditure. On the
other hand, these authors do not find a significant role for the difference in
age nor for total expenditure. Of more interest (because they have never
been used in this context before) are the family and individual background
variables. The two highly significant variables here are on whether the hus-
band’s mother was in full-time employment when he was 14 and whether
the partners have children from before the partnership. A husband having
grown up in a household in which his mother was in full-time employ-
ment increases his share of expenditure. This is consistent with the theory
model in which such men make desirable husbands (perhaps because they
contribute more in housework) and hence do better in any match than
an otherwise similar male who does not have this background. The other
finding is less easy to rationalize. If either the husband or the wife has a
previous child then the wife’s share is lower. Thus a women who has had a
previous child and is married to a man who has also had a previous child
receives a share of total expenditure that is about nine percentage points
lower than an otherwise comparable women in which neither partner has
children from before the marriage. This is a very large effect which defies
easy rationalization.
Browning, Chiappori and Lewbel (2009) also present identification re-
sults and estimates of the location of the sharing rule. These are based on
making the strong assumptions that the preferences of singles and married
people are the same and that only the household technology changes at
marriage. This allows them to identify the location of the sharing rule as
well as its dependence on distribution factors. Differences between the de-
mands of singles and couples are picked up by a Barten style technology
(see section 2 of chapter 2). For example, ‘transport’ is largely a public good
whereas ‘food at home’ is largely private. The data used is the same as in
Browning et al (1994) with the important difference that explicit account
is taken of price variations across time and over regions. The distribution
factors are very similar to those used in Browning et al (1994): the wife’s
share in total gross income, the difference in age between husband and wife,
a home-ownership dummy and household total expenditure. The point es-
timate for the sharing rule (at the mean of the distribution factors) is 0.65;
this is much higher than found in any other study. Mechanically it arises
since the budget shares of couples are more similar to those of single women
than to the budget shares of single men; this suggests that some relaxing of
the unchanging preferences assumption is called for in future work. Having
the allocations of total expenditure to each partner allows us to calculate
budget shares for husbands and wives; see Table 5.3. Wives have higher
budget shares for clothing, personal services and recreation whereas hus-
Model Budget shares (×100)

Wife Husband
Food at home 13.9 20.7
Restaurants 9.9 12.8
Clothing 16.3 7.1
Alcohol and tobacco 6.3 11.8
Transport 22.0 27.6
Personal services 15.2 12.1
Recreation 16.4 7.8
TABLE 5.3. Budget shares for husbands and wives
bands have higher budget shares for food inside and outside the home,
alcohol and tobacco and transport. Where comparisons can be made, this
is similar to the Danish data discussed in the previous paragraph.
The results presented here on the location and determinants of the shar-
ing rule do not sit together comfortably. This partly reflects the fact that
potential distribution factors differ widely across different data sets and
the excluded distribution factors are correlated with the included ones. For
example, only one study can take account of the impact of previous chil-
dren but this is correlated with the difference in age between the partners.
More fundamentally, there is no coherent theory of the sharing rule. With-
out such a theory a ‘kitchen sink’ approach is adopted in which whatever
variables are available in a particular data set are included as distribution
factors (if they are not obviously preference or constraint factors) with lim-
ited explicit concern for biases due to endogeneity (a particular worry for
income shares), omitted distribution factors or correlated latent heterogene-
ity. Equally worrying is the widespread assumption that private assignable
goods are separable from public goods (see Donni (2009)). It is clear that
much remains to be done and that ‘much’ probably requires better data
than we have had available until now.
Labor supply
The first empirical estimations of a collective model of labor supply are due
to Fortin and Lacroix (1997) and Chiappori, Fortin and Lacroix (2002). Us-
ing data from the 1988 PSID, the latter analyze the total number of hours
worked each year by single males, single females and couples, concentrat-
ing exclusively on couples without children in which both spouses work.
They consider two distribution factors, namely the state of the market for
marriage, as summarized by the sex ratio computed by age and race at
the state level, and the legislation governing divorce, summarized by an
aggregate index with the convention that a larger value indicates laws that
are more favorable to women. Their main findings can be summarized as
follows:
• The distribution factors have a significant impact on both labor sup-

plies. The signs are as predicted by the theory; that is, a higher sex
ratio (denoting a smaller percentage of women on the marriage mar-
ket), as well as divorce laws more favorable to women, reduce the
wife’s labor supply and increase the husband’s, suggesting a transfer
of resources to the wife. Interestingly, these effects are not present for
singles; divorce laws do not impact singles’ labor supplies in a signif-
icant way, whereas the sex ratio has no effect on the labor supply of
single males and increases the labor supply of single women. Finally,
the authors do not reject the prediction from the collective model
that the impacts of the two factors on the two labor supplies should
proportional.
• The corresponding transfers can be evaluated, since the sharing rule is
identified up to an additive constant. A one percentage point increase
in the sex ratio (representing roughly one standard deviation from
the mean) is found to result in an annual transfer to the wife of more
than $2,000, or about 5% of the average household income. Likewise,
a one point increase in the Divorce Laws Index (which varies from 1
to 4, with a mean at 2.8) induces husbands to transfer and additional
$4,300 to their wives. Both estimates are statistically significant at
conventional levels.
• In addition, one can recover the impact of wages and non labor in-
comes on the sharing rule. For instance, a one dollar increase in the
wife’s wage rate (which is equivalent to an annual increase of about
$1,750 in her labor income, at the mean of hours worked by women)
translates into more income being transferred to her husband. At
sample mean, the transfer amounts to more than $1,500, although
this effect is not precisely estimated. Also, a one dollar increase in
the husband’s wage rate (equivalent to an annual increase of $2,240
in his labor income) translates into $600 being transferred to his wife,
although again this effect is imprecisely estimated. Finally, a one dol-
lar increase in household nonlabor income will increase the wife’s
nonlabor income by 70 cents; that non labor income goes mostly to
the wife on average is actually a common finding of most empirical
studies based on the collective framework.
• Finally, wage elasticities can be computed in two ways. A direct es-
timation gives a positive, significant elasticity for women, close to
0.2, while men’s wage elasticities are very small and not statisti-
cally significant. The structural model also allows us to estimate the
‘true’ own-wage elasticities of individual labor supplies, taking into
account the impact of wages on the sharing of nonlabor income. Both
women’s and men’s elasticities are significant but smaller than those
reported previously - reflecting the fact that a marginal increase in
either spouse’s wage rate reduces their share of the nonlabor income,
which in turn increases their labor supply through an income effect.
Indeed, both men’s and women’s labor supply elasticities with respect
to nonlabor income are negative and significant.
Recent empirical developments involving cooperative models of labor

supply include Donni (2003), which generalizes the standard approach to
corner solutions and non-linear budget constraints, and Blundell, Chiap-
pori, Magnac and Meghir (2007), who consider a model in which female
labor supply is continuous whereas male labor supply is discrete; they show
that the sharing rule can equally be recovered in this case. Moreau and
Donni (2002) also introduce distribution factors, applied to French data,
and take into account the non-linearity of taxation. Other empirical ana-
lyzes include Bloemen (2009), Clark, Couprie and Sofer (2004) and Ver-
meulen (2005) on Dutch, British and Belgian data respectively.
In a series of recently published papers, several authors apply the col-
lective model to welfare issues, including the impact of changes in the
tax/benefit system, in different European countries. The basic method-
ology, as described in Vermeulen et al (2006), presents interesting features.
One is its scope: the approach addresses standard problems of welfare analy-
sis of labor supply, such as non linear taxation, non convex budget sets
and discrete participation decisions, within a collective framework. In ad-
dition, individual preferences are more general than in the standard col-
lective model of labor supply (Chiappori 1988, 1992) in the sense that
they allow for interactions between individual leisures (that is, the mar-
ginal utility of a spouse’s leisure is a function of the other spouse’s labor
supply). Since individual leisures are treated as public goods, the standard
identification results do not apply. The identification strategy relies on a
different assumption - namely, that the ‘direct’ trade-off between individual
leisure and consumption (disregarding the impact of the spouse’s leisure)
is identical for singles and married individuals, and can therefore be di-
rectly estimated from the labor supply of singles; of course the additional,
‘external’ effect of one spouse’s leisure on the other’s utility can only be
estimated from the sample of married couples. This approach allows to cal-
ibrate a collective model that can then be used for welfare analysis. Myck
et al. (2006) uses this framework to analyze the impact of a recent welfare
reform in the UK, namely the introduction of the Working Families’ Tax
Credit (WFTC). In particular, they consider two hypothetical versions of
the reform: one in which the recipient remains the main carer (as for the
previous Family Credit), and another in which the benefit is paid to the
main earner. The model allows to predict the impact of each version on the
spouses’ respective Pareto weights, and the corresponding labor supply re-
sponses; they conclude that, indeed, the two versions have different impact
on individual labor supplies and ultimately welfares. Similar studies have
been undertaken in various countries, including Belgium, France, Germany,
Italy and Spain; the findings are summarized in Myck et al. (2006). Finally,
Beninger et al. (2006) provide a systematic comparison of the evaluations
of tax policy reforms made within the unitary or the collective approaches
respectively. They show, in particular, that the unitary version tends to
overestimate male (and underestimate female) labor supply responses vis a
vis the collective counterpart; moreover, for a significant fraction of house-
holds, a tax reform that appears to be Pareto improving in the collective
setting is found to reduce household utility in the unitary version - a possi-
bility that had already be mentioned by the theoretical literature but had
not received an empirical confirmation so far.
Another interesting analysis is provided by Lise and Seitz (2009), who
study consumption inequality in the UK from 1968 to 2001. The main
findings of the paper is that ignoring consumption inequality within the
household produces misleading estimates of inequality. Using a rich version
of the collective model that allows for public consumption and caring pref-
erences, they reach to important conclusions. First, the standard analysis
of inequality, based on adult equivalence scales and the implicit assumption
of equal sharing of consumption within the household, underestimates the
level of cross sectional consumption inequality in 1968 by 50%; the reason
being that large differences in the earnings of husbands and wives translate
into large intrahousehold inequality in consumption. Second, the consider-
able and well known rise in inequality between household during the 80s
was largely offset by a drastic reduction in intrahousehold inequality, due to
changes in female labor supply. As a result, inequality between individual,
once (properly) computed by taking into account changes in intrahouse-
hold allocation, turns out to be practically the same in 2000 as in 1970 -
a conclusion that sharply contrasts with standard studies. Other works on
intrahousehold inequality include Kalugina, Radchenko and Sofer (2009a,
b) and Lacroix and Radchenko (2009).
Natural experiments can provide a rich source of applications for the
collective approach to labor supply. Kapan (2009) studies the impact of
a change in UK divorce laws in 2000, whereby the allocation of wealth,
initially based on a principle of separate ownership of assets, shifted to ‘the
yardstick of equal division’. A change of this kind is a typical distribution
factor; however, because of its discrete nature, the analysis cannot rely
on the same technique as Chiappori, Fortin and Lacroix (2002). Kapan
shows how the estimation strategy can be adapted to take advantage of
discrete distribution factors. He finds that, indeed, the shift resulted in an
additional transfer to women, at least when their wealth was smaller than
their husband’s; in turn, this reallocation had a significant impact of labor
supplies and individual welfares.
Finally, models involving domestic productions have been empirically an-
alyzed in a number of contributions. For example, Apps and Rees (1996),
Rapoport, Sofer and Solaz (2004) estimate the canonical model with Aus-
tralian, French and Dutch data, respectively, whereas Couprie (2007) and
van Klaveren, van Praag and Maassen van den Brink (2008) consider mod-
els where the domestic good is public and present empirical results on
various data sets.
5.5.4 Concluding remarks.

The empirical evidence reviewed in this chapter employs a largely static
framework. Although intertemporal separability gives a justification for
such an approach, dynamic issues do arise. One such issue is the stability
over time of the Pareto weight (or sharing rule) which requires a coherent
theory of the evolution of the Pareto weight. Another issue is human capi-
tal formation which necessarily introduces dynamics into the labor supply
decisions. A third issue is accounting for the formation and dissolution of
partnerships which is tightly bound up with the individual gains from a
marriage. These are discussed in the next chapter and, in greater depth, in
the second half of the book. For example, chapter 8 presents the implica-
tions for the within household distribution of the gains from marriage if we
embed the couple in a society in which agents choose to match partly on the
share they receive in equilibrium. Chapter 9 takes up the issue of human
capital formation. Chapters 10 and 11 considers models with marriage and
divorce.
5.6 References
[1] Anderson, Siwan and Jean-Marie Baland, ‘The Economics of Roscas
and Intrahousehold Resource Allocation’, The Quarterly Journal of
Economics, 117 (2002), 963-995.
[2] Apps, Patricia F. and Ray Rees, ‘Taxation and the Household’, Journal
of Public Economics, 35 (1988), 355-369.
[3] Aronsson, Thomas, Daunfeldt, Sven-Olov and Magnus Wikstrom, ‘Esti-
mating Intrahousehold Allocation in a Collective Model with Household
Production’, Journal of Population Economics, 14 (2001), 569-584.
[4] Attanasio, Orazio and Valérie Lechene, ‘Tests of Income Pooling in
Household Decisions’, Review of Economic Dynamics, 5 (2002), 720-
748.
[5] Bargain, Olivier, Beblo, Miriam, Beninger, Denis, Blundell, Richard,
Carrasco, Raquel, Chiuri, Maria-Concetta, Laisney, Francois, Lechene,
Valérie, Moreau, Nicolas, Myck, Michal, Ruiz-Castillo, Javier and Fred-
eric Vermeulen, ‘Does the Representation of Household Behavior Matter
for Welfare Analysis of Tax-benefit Policies? An Introduction’, Review
of Economics of the Household, 4 (2006), 99-111.
[6] Barmby, Tim and Nina Smith, ‘Household labor Supply in Britain and
Denmark: Some Interpretations Using a Model of Pareto Optimal Be-
haviour’, Applied Economics, 33 (2001), 1109-1116.
[7] Bayudan, Connie G., ‘Wives’ Time Allocation and Intrahousehold
Power: Evidence from the Philippines’, Applied Economics, 38 (2006),
789-804.
[8] Beninger, Denis, Bargain, Olivier, Beblo, Miriam, Blundell, Richard,
Carrasco, Raquel, Chiuri, Maria-Concetta, Laisney, Francois, Lechene,
Valérie, Longobardi, Ernesto, Moreau, Nicolas, Myck, Michal, Ruiz-
Castillo, Javier and Frederic Vermeulen, ‘Evaluating the Move to a Lin-
ear Tax System in Germany and Other European Countries’, Review
of Economics of the Household, 4 (2006), 159-180.
[9] Bloemen, Hans G.,’An Empirical Model of Collective Household labor
Supply with Non-Participation’, Economic Journal, 120 (2010), 183-
214.
[10] Blundell, Richard, Chiappori, Pierre-André and Costas Meghir, ‘Col-
lective Labor Supply with Children’, Journal of Political Economy,
113 (2005), 1277-1306.
[11] Blundell, Richard, Chiappori, Pierre-André, Magnac, Thierry and
Costas Meghir, ‘Collective labor Supply: Heterogeneity and Non-
Participation’, Review of Economic Studies, 74 (2007), 417-445.
[12] Bourguignon, Francois, Browning, Martin and Pierre-André Chiap-

pori, ‘Efficient Intra-Household Allocations and Distribution Factors:
Implications and Identification’, Review of Economic Studies, 76
(2009), 503-528.
[13] Bourguignon, Francois, Browning, Martin, Chiappori, Pierre-André

and Valérie Lechene, ‘Intra Household Allocation of Consumption: A
Model and Some Evidence from French Data’, Annales d’Economie
et de Statistique, 29 (1993), 137-156.
[14] Browning, Martin, ‘Saving and the Intra-Household Distribution of In-

come: An Empirical Investigation’, Ricerche Economiche, 49 (1995),
277-292.
[15] Browning, Martin and Jens Bonke, ‘Allocation within the Household:
Direct Survey Evidence’, University of Oxford, Working Paper No. 429,
(2009).
[16] Browning, Martin, Bourguignon, Francois and Valérie Lechene, ‘In-

come and Outcomes: A Structural Model of Intrahousehold Allocation’,
[17] Browning, Martin and Pierre-André Chiappori, ‘Efficient Intra-

Household Allocations: A General Characterization and Empirical
Tests’, Econometrica, 66 (1998), 1241-1278.
[18] Browning, Martin and Mette Gørtz, ‘Spending Time and Money
within the Household’, University of Oxford, Working Paper No. 288,
(2006).
[19] Bruce, Judith, ‘Homes divided’, World Development, 17 (1989),

979-991.
[20] Burda, Michael C., Hamermesh, Daniel S. and Philippe Weil, ‘The
Distribution of Total Work in the EU and US’, IZA Discussion Paper
No. 2270, (2006).
[21] Cherchye, Laurens, De Rock, Bram, Sabbe, Jeroen and Frederic Ver-
meulen, ‘Nonparametric Tests of Collectively Rational Consumption Be-
havior: An Integer Programming Procedure’, Journal of Economet-
rics, 147 (2008), 258-265.
[22] Cherchye, Laurens, De Rock, Bram and Frederic Vermeulen, ‘The Col-
lective Model of Household Consumption: A Nonparametric Character-
ization’, Econometrica, 75 (2007), 553-574.
[23] Chiappori, Pierre-André, ‘Collective Labor Supply and Welfare’,

[24] Chiappori, Pierre-André, ‘Rational Household Labor Supply’, Econo-

metrica, 56 (1988), 63-90.
[25] Chiappori, Pierre-André, ‘Collective labor Supply with Many Con-
sumption Goods’, Review of Economics of the Household, forth-
coming (2011).
[26] Chiappori, Pierre-André and I. Ekeland, ‘The Micro Economics of
Group Behavior: General Characterization’, Journal of Economic
Theory, 130 (2006), 1-26.
[27] Chiappori, Pierre-André and I. Ekeland, ‘The Microeconomics of Ef-
ficient Group Behavior: Identification’, Econometrica, 77 (2009), 763-
799.
[28] Chiappori, Pierre-André, Fortin, Bernard and Guy Lacroix, ‘Marriage
Market, Divorce Legislation, and Household Labor Supply’, Journal of
[29] Clark, Andrew, Couprie, Helene and Catherine Sofer, ‘La modélisation
collective de l’offre de travail. Mise en perspective et application aux
données britanniques’, Revue économique, 55 (2004), 767-789.
[30] Couprie, Helene, ‘Time Allocation within the Family: Welfare Impli-
cations of Life in a Couple’, Economic Journal, 117 (2007), 287-305.
[31] Dauphin, Anyck, El Lahga, AbdelRahmen, Fortin, Bernard and Guy
Lacroix, ‘Are Children Decision-Makers Within the Household?’, IZA
Discussion Paper No. 3728, (2008).
[32] Dercon, Stefan and Pramila Krishnan, ‘In Sickness and in Health: Risk
Sharing within Households in Rural Ethiopia’, Journal of Political
Economy, 108 (2000), 688-727.
[33] Donni, Olivier, ‘Collective Household Labor Supply: Nonparticipation
and Income Taxation’, Journal of Public Economics, 87 (2003), 1179-
1198.
[34] Donni, Olivier, ‘Collective Consumption and Welfare’, Canadian
Journal of Economics, 39 (2006), 124-144.
[35] Donni, Olivier, ‘Les modèles non coopératifs d’offre de travail: théorie
et évidence’, L’Actualité économique, 82 (2006), 181-206.
[36] Donni, Olivier, ‘Collective Female labor Supply: Theory and Applica-
tion’, Economic Journal, 117 (2007), 94-119.
[37] Donni, Olivier, ‘Choix collectif du temps travaillé, consommation
publique, et équilibre de Lindahl’, Revue économique, 58 (2007), 535-
543.
[38] Donni, Olivier, ‘Labor Supply, Home Production, and Welfare Com-
parisons’, Journal of Public Economics, 92 (2008), 1720-1737.
[39] Donni, Olivier, ‘A Simple Approach to Investigate Intrahousehold Al-
location of Private and Public Goods’, The Review of Economics
and Statistics, 91 (2009), 617-628.
[40] Donni, Olivier and Nicolas Moreau, ‘Collective Labor Supply: A
Single-Equation Model and Some Evidence from French Data’, Jour-
nal of Human Resources, 42 (2007), 214-246.
[41] Dubois, Pierre and Ethan Ligon, ‘Incentives and Nutrition for Rotten
Kids: Intra household Food Allocation in the Philippines’, UC Berkeley,
(2005).
[42] Duflo, Esther, ‘Grandmothers and Granddaughters: Old-Age Pensions
and Intrahousehold Allocation in South Africa’, World Bank Eco-
nomic Review, 17 (2003), 1-25.
[43] Duflo, Esther and Christopher Udry, ‘Intrahousehold Resource Alloca-
tion in Cote d’Ivoire: Social Norms, Separate Accounts and Consumption
Choices’, NBER Working Paper No. 10498, (2004).
[44] Ermisch, John and Chiara Pronzato, ‘Intra-Household Allocation of
Resources: Inferences from Non-resident Fathers’ Child Support Pay-
ments’, Economic Journal, 118 (2008), 347-362.
[45] Fong, Yuk-fai and Junsen Zhang, ‘The Identification of Unobservable
Independent and Spousal Leisure’, The Journal of Political Econ-
omy, 109 (2001), 191-202.
[46] Fortin, Bernard and Guy Lacroix, ‘A Test of the Unitary and Collective
Models of Household labor Supply’, Economic Journal, 107 (1997),
933-955.
[47] Gimenez-Nadal, Jose Ignacio and Almudena Sevilla-Sanz, ‘The Time-
crunch Paradox’, Department of Economics, University of Oxford, WP
483, (2010).
[48] Goldstein, Markus P., ‘Intrahousehold Efficiency and Individual Insur-
ance in Ghana’, LSE STICERD Research Paper No. DEDPS38, (2004).
[49] Haddad, Lawrence and John Hoddinott, ‘Women’s Income and Boy-
Girl Anthropometric Status in the Cote d’Ivoire’, World Develop-
ment, 22 (1994), 543-553.
[50] Hoddinott, John and Lawrence Haddad, ‘Does Female Income Share
Influence Household Expenditures? Evidence from Cote d’Ivoire’, Ox-
ford Bulletin of Economics and Statistics, 57 (1995), 77-96.
[51] Hotchkiss, Julie L., ‘Do Husbands and Wives Pool Their Resources?:
Further Evidence’, Journal of Human Resources, 40 (2005), 519-531.
[52] Kalugina, Ekaterina, Radtchenko, Natalia and Catherine Sofer, ‘How

Do Spouses Share Their Full Income? Identification Of The Sharing Rule
Using Self-Reported Income’, Review of Income and Wealth, 55
(2009), 360-391.
[53] Kalugina, Ekaterina, Sofer, Catherine and Natalia Radtchenko, ‘Intra-

household Inequality in Transitional Russia’, Review of Economics of
the Household, 7 (2009), 447-471.
[54] Kapan, T., ‘Essays in Household behavior’, PhD Dissertation,

Columbia University, (2009).
[55] Klaveren, Chris, Praag, Bernard and Henriette Maassen van den
Brink, ‘A Public Good Version of the Collective Household Model: an
Empirical Approach with an Application to British Household Data’,
Review of Economics of the Household, 6 (2008), 169-191.
[56] Lacroix, Guy and Natalia Radtchenko, ‘The Changing Intra-household

Resource Allocation in Russia’, Journal of Population Economics,
DOI No. 10.1007/s00148-009-0275-2, (2009).
[57] Ligon, Ethan, ‘Dynamic Bargaining in Households (with application

to Bangladesh)’, UC Berkeley, Working Paper Series No. 972, (2002).
[58] Lise, Jeremy and Shannon Seitz, ‘Consumption Inequality and Intra-
Household Allocations’, Labor and Demography, EconWPA, No.
0504001, (2005).
[59] Lundberg, Shelly J., Pollak, Robert A. and Terence J. Wales, ‘Do Hus-
bands and Wives Pool Their Resources? Evidence from the United King-
dom Child Benefit’, The Journal of Human Resources, 32 (1997),
463-480.
[60] Luo, Guo Ying, ‘Collective Decision-Making and Heterogeneity in

Tastes’, Journal of Business & Economic Statistics, 20 (2002),
213-226.
[61] Myck, Michal, Bargain, Olivier, Beblo, Miriam, Beninger, Denis, Blun-
dell, Richard, Carrasco, Raquel, Chiuri, Maria-Concetta, Laisney, Fran-
cois, Lechene, Valérie, Longobardi, Ernesto and Nicolas Moreau, ‘The
Working Familiess’ Tax Credit and Some European Tax Reforms in
a Collective Setting’, Review of Economics of the Household, 4
(2006), 129-158.
[62] Phipps, Shelley A. and Peter S. Burton, ‘What’s Mine Is Yours? The
Influence of Male and Female Incomes on Patterns of Household Expen-
diture’, Economica, 65 (1998), 599-613.
[63] Rapoport, Benoit, Sofer, Catherine and Anne Solaz, ‘Household Pro-
duction in a Collective Model: Some New Results ‘, Journal of Popu-
lation Economics, DOI No. 10.1007/s00148-010-0308-x, (2009).
[64] Schultz, T. Paul, ‘Testing the Neoclassical Model of Family Labor
Supply and Fertility’, The Journal of Human Resources, 25 (1990),
599-634.
[65] Strauss, John, Mwabu, G. and Kathleen Beegle, ‘Intrahousehold Al-
locations: a Review of Theories and Empirical Evidence’, Journal of
African Economies, 9 (2000), 83-143.
[66] Thomas, Duncan, ‘Intra-Household Resource Allocation: An Inferen-
tial Approach’, The Journal of Human Resources, 25 (1990), 635-
664.
of the Household’, The Journal of Political Economy, 104 (1996),
1010-1046.
[68] Vermeulen, Frederic, ‘And the Winner is... An Empirical Evaluation of
Unitary and Collective labor Supply Models’, Empirical Economics,
30 (2005), 711-734.
[69] Vermeulen, Frederic, ‘A Collective Model for Female labor Supply with
Non-Participation and Taxation’, Journal of Population Economics,
19 (2006), 99-118.
[70] Vermeulen, Frederic, ‘Collective Household Models: Principles and
Main Results’, Journal of Economic Surveys, 16 (2002), 533-564.
[71] Vermeulen, Frederic, Bargain, Olivier, Beblo, Miriam, Beninger, Denis,
Blundell, Richard, Carrasco, Raquel, Chiuri, Maria-Concetta, Laisney
Francois, Lechene, Valérie, Moreau, Nicolas and Michal Myck, ‘Collec-
tive Models of Labor Supply with Nonconvex Budget Sets and Nonpar-
ticipation: A Calibration Approach’, Review of Economics of the
Household, 4 (2006), 113-127.
[72] Vermeulen, Frederic and Watteyne, André, ‘Quand un et un ne font
plus deux’, L’Actualité Economique, 82 (2006), 155-180.
[73] Ward-Batts, Jennifer, ‘Out of the Wallet and into the Purse: Using
Micro Data to Test Income Pooling’, Journal of Human Resources,
43 (2008), 325-351.
This is page 243
6
Uncertainty and Dynamics in
the Collective model
The models developed in the previous chapters were essentially static and
were constructed under the (implicit) assumption of perfect certainty. As
discussed in chapter 2, such a setting omits one of the most important
roles of marriage - namely, helping to palliate imperfections in the insur-
ance and credit markets by sharing various risks and more generally by
transferring resources both across periods and across states of the world.
Risk sharing is an important potential gain from marriage: individuals who
face idiosyncratic income risk have an obvious incentive to mutually pro-
vide insurance. In practice, a risk sharing scheme involves intrahousehold
transfers that alleviate the impact of shocks affecting spouses; as a result,
individual consumptions within a couple may be less responsive to idio-
syncratic income shocks than it would be if the persons were single. Not
only are such risk-sharing mechanisms between risk averse agents welfare
improving, but they allow the household to invest into higher risk/higher
return activities; as such, they may also increase total (expected) income
and wealth in the long run. For instance, the wife may be able to afford
the risk involved in creating her own business because of the insurance
implicitly offered by her husband’s less risky income stream.
Another, and closely related form of consumption smoothing stems from
intrafamily credit relationship: even in the absence of a perfect credit mar-
ket, a spouse can consume early a fraction of her future income thanks to
the resources coming from her partner. Again, intrahousehold credit may in
turn enable agents to take advantage of profitable investment opportunities
that would be out of the reach of a single person.
While intertemporal and risk sharing agreements play a key role in eco-
nomic life in general and in marriage in particular, they also raise specific
difficulties. The main issue relates to the agents’ ability to credibly com-
mit to specific future behavior. Both types of deals typically require that
some agents reduce their consumption in either some future period or some
possible states of the world. This ability to commit may however not be
guaranteed. In some case, it is even absent (or severely limited); these are
cases in which the final agreement typically fails to be fully efficient, at
244 6. Uncertainty and Dynamics in the Collective model
least in the ex ante sense.1

The theoretical analysis underpinning these issues leads to fascinating
empirical questions. Again, these can be formulated in terms of testability
and identifiability. When, and how, is it possible to test the assumption of
perfect commitment, and more generally of ex ante efficiency? And to what
extent is it possible to recover the underlying structure - namely individual
preferences (here, aversions to risk and/or fluctuations) and the decision
process (here, the Pareto weights) from observed behavior? These questions
- and others - are analyzed in the present chapter.
6.1 Is commitment possible?

We start with a brief discussion of the commitment issue. As discussed
above, credit implies repayment, and the very reason why a formal credit
market may fail to be available (say, non contractible investments) may
result in enforcement problems even between spouses. As the usual cliche
goes, a woman will be hesitant to support her husband through medical
school if she expects him to break the marriage and marry a young nurse
when he finishes (this is a standard example of the hold-up problem). Sim-
ilarly, risk sharing requires possibly important transfers between spouses;
which enforcement devices can guarantee that these transfers will actually
take place when needed is a natural question. In subsequent sections we
shall consider conventional economic analyzes of the commitment problem
as they relate to the family. In the remainder of this section we consider
possible commitment mechanisms that are specific to the family.
From a game-theoretic perspective, marriage is a typical example of re-
peated interactions between the same players; we know that cooperation
is easier to support in such contexts.2 This suggests that, in many case,
cooperation is a natural assumption. Still, the agents’ ability to commit is
probably not unbounded. Love may fade away; fidelity is not always lim-
itless; commitment is often constrained by specific legal restrictions (for
1A second problem is information: in general, efficient trade is much easier to im-

plement in a context of symmetric information. Asymmetric information, however, is
probably less problematic in households than in other types of relationship (say, be-
tween employers and employees or insurers and insurees), because the very nature of the
relationship often implies deep mutual knowledge and improved monitoring ability.
2 Del Boca and Flinn (2009) formulate a repeated game for time use that determines
the amount of market work and housework that husbands and wives perform. Their
preferred model is a cooperative model with a noncooperative breakdown point. They
have a repeated game with a trigger strategy for adopting the inefficient non-cooperative
outcome if the discount is too small. The value of the threshold discount factor they
estimate to trigger noncooperative behavior is 0.52which implies that 94% of households
behave cooperatively.
6. Uncertainty and Dynamics in the Collective model 245
instance, agents cannot legally commit not to divorce).3 And while the re-
peated interaction argument for efficiency is convincing in many contexts,
it may not apply to some important decisions that are made only excep-
tionally; moving to a different location and different jobs is a standard
example, as argued by Lundberg and Pollak (2003).
A crucial aspects of lack of commitment is that, beyond restraining ef-
ficiency in the ex ante sense, it may also imply ex post inefficiencies. The
intuition is that whenever the parties realize the current agreement will
be renegotiated in the future, they have strong incentives to invest now
into building up their future bargaining position. Such an investment is in
general inefficient from the family’s viewpoint, because it uses current re-
sources without increasing future (aggregate) income. For instance, spouses
may both invest in education, although specialization would be the efficient
choice, because a high reservation wage is a crucial asset for the bargaining
game that will be played later.4
Love and all these things

How can commitment be achieved when the repeated interaction argu-
ment does not hold? Many solutions can actually be observed. First, actual
contracts can be (and actually are) signed between spouses. Prenuptial
agreements typically specify the spouses’ obligations both during marriage
and in case of divorce; in particular, some provisions may directly address
the hold-up problem. To come back to the previous example, a woman will
less be hesitant fund her husband’s training if their prenuptial agreement
stipulates that she will receive, in case of divorce, a large fraction of his
(future) income. Contract theory actually suggests that even if long term
agreements are not feasible, efficiency can in general be reached through
a sequence of shorter contracts that are regularly renegotiated (see for in-
stance Rey and Salanié (1990).5 Still, even though a private, premarital
agreement may help alleviating the limits to commitments (say, by making
divorce very expensive for one of the parties), renegotiation proofness may
be an issue, especially if divorce has been made costly for both spouses;
furthermore, in some countries courts are free to alter ex post the terms of
premarital agreements. At any rate, some crucially important intrahouse-
hold issues may hardly be contractible.
3 Of course, moral or religious commitment not to divorce do exist, although they
may not be globally prevalent.

4 See Konrad and Lommerud (2000) and Brossolet (1993).
5 In practice, prenuptial agreements are not common (although they are more fre-
quently observed in second marriages). However, this may simply indicate that, although
easily feasible, they are rarely needed, possibly because existing enforcement mechanisms
(love, trust, repeated interactions) are in general sufficient. Indeed, writing an explicit
contract that lists all contingencies may in fact "crowd out" the emotional bonds and
diminish the role of the initial spark of blind trust that is associated with love.
Alternative enforcement mechanisms can however be implemented. Re-

ligious or ethical factors may be important; in many faiths (and in several
social groups), a person’s word should never be broken. Love, affection
and mutual respect are obviously present in most marriages, and provide
powerful incentives to honoring one’s pledge and keeping one’s promises.
Browning (2009) has recently provided a formalization of such a mecha-
nism. The model is developed in the specific context of the location deci-
sion model of Mincer (1978) and Lundberg and Pollak (2003) but has wider
application. In the location model a couple, a and b, are presented with an
opportunity to increase their joint income if they move to another location.
Either partner can veto the move. The problem arises when the move shifts
power within the household toward one partner (partner b, say); then the
other partner (a) will veto the move if she is worse off after the move.
Promises by partner b are not incentive compatible since a does not have
any credible punishment threat.6 Particular commitment mechanisms may
be available in this location decision model. For example, suppose there is
a large indivisible choice that can be taken at the time of moving; choosing
a new house is the obvious example. If this choice has a large element of
irreversibility then partner b can defer to a on this choice and make the
move more attractive. At some point, however, commitment devices such
as this may be exhausted without persuading a that the move is worth-
while. Now assume that spouses are caring, in the usual sense that their
partner’s utility enter their preferences. Browning (2009) suggests that if
one partner exercises too aggressively their new found bargaining power
then the other partner feels betrayed and loses some regard (or love) for
them. The important element is that this loss of love (by a in this case)
is out of the control of the affected partner; in this sense, this is betrayal.
Thus the threat is credible. In a model with mutual love, this ‘punishment’
is often sufficient to deter a partner from exercising their full bargaining
power if the move takes place.
To formalize, consider a married couple a and b. Income, which is nor-
malised to unity if they do not move, is divided between them so that a
receives x for private consumption and b receives 1 − x. There are no public
goods. Each person has the same strictly increasing, strictly concave felicity
function, so that:
ua = u (x) , ub = u (1 − x) (6.1)
Each person also cares for the other with individual utility functions given
6 We neglect the option in which they divorce and the husband moves to the new
location. Mincer (1978) explicitly considers this.

by:
W a = ua + δ a ub
= u (x) + δ a u (1 − x) (6.2)
b b a b
W =δ u +u
= δ b u (x) + u (1 − x) (6.3)
where δ s ≥ 0 is person s’s caring for the other person, with δ a δ b < 1 (see
chapter 3). We assume that the caring parameters are constant and outside
the control of either partner. Rather than choosing an explicit game form
to choose x, we simply assume that there is some (collective) procedure
that leads the household to behave as though it maximizes the function:
W = W a + μW b
³ ´
= 1 + μδ b u (x) + (δ a + μ) u (1 − x) (6.4)
As discussed in chapter 4, ³caring modifies

´ the Pareto weight for b to an
effective value of (δ a + μ) / 1 + μδ b .
Now suppose there is a (moving) decision that costlessly increases house-
hold income from unity to y > 1. If this is the only effect then, of course,
both partners would agree to move. However, we also assume that the de-
cision increases b’s Pareto weight to μ (1 + m) where m ≥ 0. In this case
there is a reservation income y ∗ (m) such that person a will veto the move
if and only if y < y ∗ (m). In such a case there will be unrealized potential
Pareto gains. Now allow that the husband can choose whether or not to
exercise his new found power if they do move. If he does not exercise his
new power then the household utility function is given by:
³ ´
W = 1 + μδ b u (x) + (δ a + μ) u (y − x) (6.5)
which obviously dominates (6.4). Of course, a simple statement that "I

promise to set m = 0" has no credibility. Suppose, however, that if such a
promise is made and then broken, then the wife feels betrayed. In this case
her love for her husband falls from δ a to δ a (1 − σ) where σ ∈ [0, 1]. The
fall in her caring for him is taken to be out of her control, so that a has an
automatic and hence credible punishment for b choosing to take advantage
of his improved position. If they move and the husband exercises his new
power the household utility function is given by:
³ ´
W = 1 + μ (1 + m) δ b u (x) + (δ a (1 − σ) + (1 + m) μ) u (y − x) (6.6)
If the husband’s implicit Pareto weight is less in this case than in (6.5) then
he will not betray his wife. In the simple case in which he does not care for
her (δ b = 0) this will be the case if:
(δ a + μ) ≥ (δ a (1 − σ) + (1 + m) μ)
⇔ δ a σ ≥ mμ (6.7)
That is, there will be a move with no betrayal if δ a and σ are sufficiently
large relative to m and μ. For example, a husband who lacks power (and
hence relies on his wife’s caring for resources) or has a small increase in
power (so that mμ is small) will be less likely to betray; and the same holds
if his wife cares a lot for him (δ a ) and she feels the betrayal strongly (σ
close to unity).
Psychological games
A different but related analysis is provided by Dufwenberg (2002), who
uses "psychological games" to discuss commitment in a family context. The
basic idea, due to Geanakoplos et al (1989), is that the utility payoffs of
married partners depend not only on their actions and the consequences in
terms of income or consumption but also on the beliefs that the spouses may
have on these actions and consequences. The basic assumption is that the
stronger is the belief of a spouse that their partner will act in a particular
manner, the more costly it is for that partner to deviate and disappoint their
spouse. This consideration can be interpreted as guilt. A crucial restriction
of the model is that, in equilibrium, beliefs should be consistent with the
actions. Dufwenberg (2002) uses this idea in a context in which one partner
(the wife) extends credit to the other spouse. For instance, the wife may
work when the husband is in school, expecting to be repaid in the form
of a share from the increase in family income (see Chapter 2). But such a
repayment will occur only if the husband stays in the marriage, which may
not be the case if he is unwilling to share the increase in his earning power
with his wife and walks away from the marriage.
Specifically, consider again the two period model discussed in Chapter
2. There is no borrowing or lending and investment in schooling is lumpy.
In the absence of investment in schooling, each spouse has labor income
of 1 each period. There is also a possibility to acquire some education;
if a person does so then their earnings are zero in the first period and 4
in the second period. We assume that preferences are such that in each
period each person requires a consumption of 12 for survival and utility
is linear in consumption otherwise. This implies that without borrowing,
no person alone can undertake the investment, while marriage enables the
couple to finance the schooling investment of one partner. We assume that
consumption in each period is divided equally between the two partners
if they are together and that if they are divorced then each receives their
own income. Finally, suppose that each partner receives a non monetary
gain from companionship of θ = 0.5 for each period they are together. The
lifetime payoff if neither educate is (2 + 2θ) = 3 for each of them. Since
both have the same return to education, for ease of exposition we shall
assume that they only consider the husband taking education.7 If he does
educate and they stay together then each receives a total of (3 + 2θ) = 4
over the two periods. There is thus a potential mutual gain for both of
them if the investment is undertaken and marriage continues. However, if
the husband educates and then divorces, he receives a payoff of 4 in the
second period and if he stays he receives only 3 (= 2.5 + θ). Thus, without
commitment, he would leave in the second period8 and the wife will then
be left with a lifetime utility of 2 which is less than she would have in the
absence of investment, 3. Therefore, the wife would not agree to finance her
husband’s education in the first period. The basic dilemma is illustrated in
Figure 6.1, where the payoffs for the wife are at the top of each final node
and the payoffs for the husband are at the bottom. The only equilibrium in
this case is that the wife does not support her husband, the husband does
not invest in schooling and stays in the marriage so that the family ends up
in an inefficient equilibrium. However, Dufwenberg (2002) then shows that
if one adds guilt as a consideration, an efficient equilibrium with consistent
beliefs can exist. In particular, suppose that the husband’s payoff following
divorce is 4 − γτ where τ is the belief of the husband at the beginning of
period 2 about the beliefs that his wife formed at the time of marriage,
about the probability that her husband will stay in the marriage following
her investment and γ is a fixed parameter. Then, if γ > 2, the husband
chooses to stay in the marriage, the wife agrees to support her husband to
invest and efficiency is attained. To show the existence of consistent beliefs
that support this equilibrium, consider the special case in which γ = 2.
Suppose that the wife actually invests, as we assume for this equilibrium.
Then, she reveals to her husband that she expects to get a life time utility
of at least 3 following this choice, which means that her belief, τ 0 about
the probability that the husband would stay is such that 1 + τ 0 4 > 3,
implying τ 0 ≥ 12 . Knowing that, the husband’s belief τ about her belief
that he stays exceeds 12 . Therefore, his payoff upon leaving in the second
period 4 − 2τ is less or equal to his payoff if he stays, 3. Thus for any γ
strictly above 2, he stays. In short, given that the wife has shown great
trust in him, as indicated by her choice to support him, and given that
he cares a great deal about that, as indicated by the large value of γ, the
husband will feel more guilty about disappointing her and will in fact stay
in the marriage, justifying his wife’s initial beliefs. The husband on, his
part, avoids all feelings of guilt and efficient investment will be attained. A
7 The issue of what happens if the two have different returns to education is one that
deserves more attention.

8 Note that if the match quality is high enough then he will not divorce even if he
educates. In the numerical example this will be the case if θ > 1.5. In this case there is
no need for commitment. This is analogous to the result concerning match quality and
children.
happy marriage indeed.
wife
no yes
husband
3
3
divorce stay
2 4
5 4
FIGURE 6.1. Game tree for investment in education
Somewhat different considerations arise when we look at ‘end game’ situ-

ations in which the spouse has no chance to reciprocate. A sad real example
of this sort is when the husband has Altzheimer’s and his wife takes care of
him for several (long) years, expecting no repayment from him whatsoever
as he does not even know her. Here, the proper assumption appears to be
that she believes that he would have done for her the same thing had the
roles been reversed. Unfortunately, the consistency of such beliefs is impos-
sible to verify. Another possibility is that she cares about him and about her
children that care about him to the extent that caring for the sick husband
in fact gives her satisfaction. In either case, some emotional considerations
must be introduced to justify such cases of unselfish behavior in families.
The commitment issue is complex. In the end, whether agents are able to
implement and enforce a sufficient level of commitment to achieve ex ante
efficiency is an empirical issue. Our task, therefore, is to develop concep-
tual tools that allow a precise modeling of these problems, and empirical
tests that enable us to decide whether, and to what extent, the lack of
commitment is an important problem for household economics.
6.2 Modeling commitment

6.2.1 Full commitment
Fortunately enough, the tools developed in the previous chapters can read-
ily be extended to modeling the commitment issues. We start from the full
commitment benchmark. The formal translation is very simple: under full
commitment, Pareto weights remain constant, either over periods or over
states of the world (or both). To see why, consider for instance the risk
sharing framework with two agents. Assume that thereP exists S states of
the world, with respective probabilities π1 , ..., πS (with s π s = 1); let ysa
denote member a’s income is state s. Similarly, let ps (resp. Ps ) be the
price vector for private (public) goods in state s, and qas (resp. Qs ) the
vector of private consumption by member a (the vector of household pub-
lic consumption). An allocation is ex ante efficient if it solves a program
of the type:
X ¡ ¢
max π s ua Qs , qas , qbs
Qs ,qa b
s ,qs s
¡ ¢ ¡ ¢
subject to P0s Qs +p0s qas + qbs ≤ ysa + ysb for all s (6.8)
X ¡ ¢
and πs ub Qs , qas , qbs ≥ ūb
s
for some ūb . As in chapter 3, if μ denotes the Lagrange multiplier of the

last constraint, this program is equivalent to:
X ¡ ¢ X ¡ ¢
max π s ua Qs , qas , qbs + μ π s ub Qs , qas , qbs
Qs ,qa b
s ,qs s s
¡ a ¢ ¡ ¢
subject to P0s Qs +p0s qs + qbs ≤ ysa + ysb for all s
or:
X £ ¡ ¢ ¡ ¢¤
max πs ua Qs , qas , qbs + μub Qs , qas , qbs (6.9)
Qs ,qa b
s ,qs s
¡ ¢ ¡ ¢
subject to P0s Qs +p0s qas + qbs ≤ ysa + ysb for all s
This form shows two things. First, ¡for any state ¢ s, the allocation contin-
a b
gent on the realization ¡ of this state,
¢ Q¡ s , qs , qs , ¢maximizes the weighted
sum of utilities ua Qs , qas , qbs + μub Qs , qas , qbs under a resource con-
straint. As¡such, it is ¢efficient in the ex post sense: there is no alternative
allocation Q̄s , q̄as , q̄bs that would improve both agents’ welfare in state s.
Secondly, the weight μ is the same across states of the world. This guaran-
tees ex ante efficiency: there is no alternative allocation
£¡ ¢ ¡ ¢¤
Q̄1 , q̄a1 , q̄b1 , ..., Q̄S , q̄aS , q̄bS
that would improve both agents’ welfare in expected utility terms - which
is exactly the meaning of programs (6.8) and (6.9).
Finally, note that the intertemporal version of the problem obtains simply
by replacing the state of the world index s by a time index t and the
probability π s of state s with a discount factor - say, δ t .
6.2.2 Constraints on commitment

Limits to commitment can generally be translated into additional con-
straints in the previous programs. To take a simple example, assume that
in each state of the world, one member - say b - has some alternative op-
tion that he cannot commit not to use. Technically, in each state s, there
is some lower bound ūbs for b’s utility; here, ūbs is simply the utility that
b would derive from his fallback option. This constraint obviously reduces
the couple’s ability to share risk. Indeed, it may well be the case that, in
some states, efficient risk sharing would require b’s welfare to go below this
limit. However, a contract involving such a low utility level in some states
is not implementable, because it would require from b more commitment
than what is actually available.
The technical translation of these ideas is straightforward. Introducing
the new constraint into program (6.8) gives:
X ¡ ¢
max π s ua Qs , qas , qbs
Qs ,qa b
s ,qs s
¡ ¢ ¡ ¢
subject to P0s Qs +p0s qas + qbs ≤ ysa + ysb for all s, (6.10)
X ¡ ¢
πs ub Qs , qas , qbs ≥ ūb
s
¡ ¢
and ub Qs , qas , qbs ≥ ūbs for all s (6.11)
Let μs denote the Lagrange multiplier of constraint (Cs ); the program

can be rewritten as:
X ¡ ¢ X ¡ ¢
max πs ua Qs , qas , qbs + (μπs + μs ) ub Qs , qas , qbs
Qs ,qa b
s ,qs s s
¡ a ¢ ¡ ¢
subject to P0s Qs +p0s qs + qbs ≤ ysa + ysb for all s
or equivalently:
X ∙ µ ¶ ¸
¡ ¢ μ ¡ ¢
max πs ua Qs , qas , qbs + μ + s ub Qs , qas , qbs (6.12)
Qs ,qa b
s ,qs s πs
¡ a ¢ ¡ a ¢
0 0
subject to Ps Qs +ps qs + qs ≤ ys + ysb for all s
b
Here, ex post efficiency still³obtains:´ in each state s, the household maxi-

mizes the weighted sum ua + μ + μπss ub . However, the weight is no longer
constant; in any state s in which constraint (6.11) is binding, implying that

μs > 0, b’s weight is increased by μs /π s . Intuitively, since b’s utility can-
not go below the fallback value ūbs , the constrained agreement inflates b’s
Pareto weight in these states by whichever amount is necessary to make
b just indifferent between the contract and his fallback option. Obviously,
this new contract is not efficient in the ex ante sense; it is only second best
efficient, in the sense that no alternative contract can do better for both
spouses without violating the constraints on commitment.
6.2.3 Endogenous Pareto weights

Finally, assume as in Basu (2006), that the fallback utility ūbs is endogenous,
in the sense that it is affected by some decision made by the agents. For
instance, ūbs depends on the wage b would receive on the labor market,
which itself is positively related to previous labor supply (say, because of
human capital accumulation via on the job training). Now, in the earlier
periods b works for two different reasons. One is the usual trade-off between
leisure and consumption: labor supply generates an income that can be
spent on consumption goods. The second motive is the impact of current
labor supply on future bargaining power; by working today, an agent can
improve her fallback option tomorrow, therefore be able to attract a larger
share of household resources during the renegotiation that will take place
then, to the expenses of her spouse. The first motive is fully compatible
with (static) efficiency; the second is not, and results in overprovision of
labor with respect to the optimum level.
We can capture this idea in a simple, intertemporal version of the previ-
ous framework. Namely, consider a two-period model with two agents and
two commodities, and assume for simplicity that agents are egoistic:
2
X
max δ t−1 ua (qat )
qa b
t ,qt t=1
¡ ¢ ¡ ¢
subject to p0t qat + qbt ≤ yta + ytb for t = 1, 2, (6.13)
2
X ¡ ¢
δ t−1 ub qbt ≥ ūb
t=1
¡ ¢
and ub qb2 ≥ ūb2 (6.14)
¡ X X¢
where qX t = q1,t , q2,t , X = a, b; note that we assume away external fi-
nancial markets by imposing a resource constraint at each period. Assume,
moreover, that the fallback option ūb2 of b in period 2 is a decreasing func-
b
tion of q1,t ; a natural interpretation, suggested above, is that commodity
1 is leisure, and that supplying labor at a given period increases future
potential wages, hence the person’s bargaining position. Now the Lagrange
b
multiplier of (6.14), denoted μ2 , is also a function of q1,t . The program
becomes:
2
X 2
X ¡ ¢ ¡ b ¢ b ¡ a b¢
max δ t−1 ua (qat ) + μ δ t−1 ub qbt + μ2 q1,t u q2 , q2
qa b
t ,qt t=1 t=1
¡ a ¢ ¡ ¢
subject to p0t qt + qbt ≤ yta + ytb for t = 1, 2
or equivalently:
£ ¡ ¢¤ £ ¡ ¢¤ ¡ b ¢ b ¡ b¢
max ua (qa1 ) + μub qb1 + δ ua (qa2 ) + μub qb2 + μ2 q1,t u q2
qa b
t ,qt
¡ ¢ ¡ ¢
subject to p0t qat + qbt ≤ yta + ytb for t = 1, 2.
b
The first order conditions for q1,1 are:
¡ ¢ ¡ b ¢
∂ub qb1 b
¡ b ¢ dμ2 q1,t
μ b
= λp1,t − u q2 b
∂q1,1 dq1,t
which does not coincide with the standard condition for static efficiency
because of the last term. Since the latter is positive, the marginal utility
of leisure is above the optimum, reflecting under-consumption of leisure
(or oversupply of labor). In other words, both spouses would benefit from
an agreement to reduce both labor supplies while leaving Pareto weights
unchanged.
6.3 Efficient risk sharing in a static context

6.3.1 The collective model under uncertainty
Ex ante and ex post efficiency
We can now discuss in a more precise way the theoretical and empirical is-
sues linked with uncertainty and risk sharing. For that purpose, we special-
ize the general framework sketched above by assuming that consumptions
are private, and agents have egoistic preferences. We first analyze a one-
commodity model; then we consider an extension to a multi-commodity
world.
We consider a model in which two risk averse agents, a and b, share
income risks through specific agreements. There are N commodities and S
states of the world, which realize with respective probabilities (π1 , ..., πS ).
Agent X (X =¡a, b) receives¢ in each state s some income ysX , and consumes
a vector cX X X
s = cs,1 , ..., cs,N ; let ps = (ps,1 , ..., ps,N ) denote the price vector
in state s. Agents are expected utility maximizers, and we assume that their
respective Von Neumann-Morgenstern utilities are strictly concave, that is
that agents are strictly risk averse.
The efficiency assumption can now take two forms. Ex post efficiency
requires that, in each state s of the world, the allocation of consumption is
efficient in the usual, static sense: no alternative allocation
¡ could ¢ improve
both utilities at the same cost. That is, the vector cs = cas , cbs solves:
max ua (cas ) (6.15)

under the constraints:
¡ ¢
ub cbs ≥ ūbs
X ¡ ¢
pi,s cai,s + cbi,s = ysa + ysb = ys
i
As before, we may denote by μs the Lagrange multiplier of the first con-

straint; then the program is equivalent to:
¡ ¢
max ua (cas ) + μs ub cbs
under the resource constraint. The key remark is that, in this program, the
Pareto weight μs of member b may depend on s. Ex post efficiency requires
static efficiency in each state, but imposes no restrictions on behavior across
states.
Ex ante efficiency requires, in addition, that the allocation of resources
across states is efficient, in the sense that no state-contingent exchange can
improve both agents’ expected utilities. Note that, now, welfare is computed
ex ante, in expected utility terms. Formally, the vector c = (c1 , ..., cS ) is
efficient if it solves a program of the type:
X
max πs ua (cas ) (6.16)
s

X ¡ ¢
πs ub cbs ≥ ūb (6.17)
s
X ¡ ¢
pi,s cai,s + cbi,s = ysa + ysb = ys , s = 1, ..., S (6.18)
i
Equivalently, if μ denotes the Lagrange multiplier of the first constraint,

the program is equivalent to:
X X ¡ ¢ X £ a a ¡ ¢¤
max π s ua (cas ) + μ π s ub cbs = π s u (cs ) + μub cbs
s s s
under the resource constraint (6.18).

One can readily see that any solution to this program also solves (6.15) for
μs = μ. But ex ante efficiency generates an additional constraint - namely,
the Pareto weight μ should be the same across states. A consequence of this
requirement is precisely that risk is shared efficiently between agents.
The sharing rule as a risk sharing mechanism

We now further specify the model by assuming that prices do not vary:
ps = p, s = 1, ..., S
Let V X denote the indirect utility of agent X. For any ex post efficient
allocation, let ρX
s denote the total expenditure of agent X in state s:
X
ρX
s = pi cX
s,i
i
Here as above, ρX is the sharing rule that governs the allocation of house-
hold resources between members. Obviously, we have that ρas + ρbs = ysa +
ysb = ys . If we denote ρs = ρas , then ρbs = ys − ρs . Program (6.16) becomes:
X X
W (y1 , ..., yS ; μ) = max π s V a (ρs ) + μ π s V b (ys − ρs ) (6.19)
ρ1 ,...,ρS
s s
In particular, in the absence of price fluctuations, the risk sharing problem

is one-dimensional: agents transfer one ‘commodity’ (here dollars) across
states, since they are able to trade it for others commodities on markets
once the state of the world has been realized, in an ex post efficient manner.
When is a unitary representation acceptable?

The value of the previous program, W (y1 , ..., yS ; μ), describes the house-
hold’s attitude towards risk. For instance, an income profile (y1 , ..., yS ) is
preferred over some alternative (y10 , ..., yS0 ) if and only if W (y1 , ..., yS ; μ) ≥
W (y10 , ..., yS0 ; μ). Note, however, that preferences in general depend on the
Pareto weight μ. That is, it is usually the case that profile (y1 , ..., yS ) may
be preferred over (y10 , ..., yS0 ) for some values of μ but not for others. In that
sense, W cannot be seen as a unitary household utility: the ranking over
income profiles induced by W varies with the intrahousehold distribution
of powers (as summarized by μ), which in turns depends on other aspects
(ex ante distributions, individual reservation utilities,...).
A natural question is whether exceptions can be found, in which the
household’s preferences over income profiles would not depend on the mem-
ber’s respective powers. A simple example can convince us that, indeed,
such exceptions exist. Assume, for instance, that both VNM utilities are
logarithmic:
V a (x) = V b (x) = log x
Then (6.19) can be written as:
X X
max πs log (ρs ) + μ π s log (ys − ρs ) (6.20)
ρ1 ,...,ρS
s s
First order conditions give

πs μπ s
=
ρs ys − ρs
therefore
ys
ρs =
1+μ
Plugging into (6.20), we have that:
X µ ¶ X µ ¶
ys μys
W (y1 , ..., yS ; μ) = π s log +μ π s log
s
1+μ s
1+μ
X ∙µ 1
¶ µ
μ
¶¸
= π s log + log ys + μ log + log ys
s
1+μ 1+μ
X
= k (μ) + 2 π s log ys
s
where
1 μ
k (μ) = log + μ log
1+μ 1+μ
P
and we see that maximizing W is equivalent to maximizing s π s log ys ,
which does not depend on μ. In other words, the household’s behavior
under uncertainty is equivalent to that of a representative agent, whose
VNM utility, V (x) = log x, is moreover the same as that of the individual
members. Equivalently, the unitary approach - which assumes that the
household behaves as if there was a single decision maker - is actually valid
in that case.
How robust is this result? Under which general conditions is the uni-
tary approach, based on a representative agent, a valid representation of
household behavior under risk? Mazzocco (2004) shows that one condition
is necessary and sufficient; namely, individual utilities must belong to the
ISHARA class. Here, ISHARA stands for ‘Identically Shaped Harmonic
Absolute Risk Aversion’, which imposes two properties:
• individual VNM utilities are of the harmonic absolute risk aversion

(HARA) type, characterized by the fact that the index of absolute
risk aversion, −u00 (x) /u0 (x), is an harmonic function of income:
u00 (x) 1
− =
u0 (x) γx + c
For γ = 0, we have the standard, constant absolute risk aversion
(CARA). For γ = 1, we have an immediate generalization of the log
form just discussed:
¡ ¢
ui (x) = log ci + x
for some constants ci , i = a, b. Finally, for γ 6= 0 and γ 6= 1, we have:

¡ i ¢1−1/γ i
i c + γix
u (x) =
1 − 1/γ i
for some constants ci and γ i , i = a, b.

• moreover, the ‘shape’ coefficients γ must be equal:
γa = γb
The intuition of this result is that in the ISHARA case, the sharing
rule that solves (6.19) is an affine function of realized income. Note that
ISHARA is not simply a property of each utility independently: the second
requirement imposes a compatibility restriction between them. That said,
CARA utilities always belong to the ISHARA class, even if their coefficients
of absolute risk aversion are different (that’s because they correspond to
γ a = γ b = 0). On the other hand, constant relative risk aversion (CRRA)
utilities, which correspond to ca = cb = 0, are ISHARA if and only if the
coefficient of relative risk aversion, equal to the shape parameter γ i in that
case, is identical for all members (it was equal to one for both spouses in
our log example).
6.3.2 Efficient risk sharing in a one-commodity world

Characterizing efficient risk sharing
We now characterize ex ante efficient allocations. We start with the case
in which prices do not vary; as seen above, we can then model efficient risk
sharing in a one commodity context. A sharing rule ρ shares risk efficiently
if it solves a program of the form:
X £ ¡ ¡ ¢¢ ¡ ¡ ¢¢¤
max πs ua ρ ysa , ysb + μub ysa + ysb − ρ ysa , ysb
ρ
s
for some Pareto weight μ. The first order condition gives:

¡ ¡ ¢¢ ¡ ¡ ¢¢
u0a ρ ysa , ysb = μ.u0b ysa + ysb − ρ ysa , ysb
or equivalently:
u0a (ρs )
= μ for each s (6.21)
u0b (ys − ρs )
¡ ¢
where ys = ysa + ysb and ρs = ρ ysa , ysb .
This relationship has a striking property; namely, since μ is constant, the
left hand side does not depend on the state of the world. This is a standard
characterization of efficient risk sharing: the ratio of marginal utilities of
income of the agents remains constant across states of the world.
The intuition for this property is easy to grasp. Assume there exists two
states s and s0 such that the equality does not hold - say:
u0a (ρs ) /dρ u0a (ρs0 ) /dρ

<
u0b (ys − ρs ) /dρ u0b (ys0 − ρs0 ) /dρ
Then there exists some k such that
π s u0a (ρs ) /dρ πs u0b (ys − ρs ) /dρ

< k <
π s0 u0a (ρs0 ) /dρ π s0 u0b (ys0 − ρs0 ) /dρ
But now, both agents can marginally improve their welfare by some ad-
ditional trade. Indeed, if a pays some small amount ε to b in state s but
receives kε in state s0 , a’s welfare changes by
dW a = −πs u0a (ρs ) ε + π s0 u0a (ρs0 ) kε > 0
while for b
dW b = π s u0b (ys − ρs ) ε − πs0 u0b (ys0 − ρs0 ) kε > 0
and both parties gain from that trade, contradicting the fact that the initial
allocation was Pareto efficient.
The sharing rule ρ is thus a solution of equation (6.21), which can be
rewritten as:
u0a (ρ) = μu0b (ys − ρ) (6.22)
¡ a b¢
where ρ = ρ ys , ys . Since the equation depends on the weight μ, there
exists a continuum of efficient risk sharing rules, indexed by the parameter
μ; the larger this parameter, the more favorable the rule is to member b.
As an illustration, assume that agents have Constant Absolute Risk Aver-
sion (CARA) preferences with respective absolute risk aversions equal to
α and β for a and b respectively:
ua (x) = − exp (−αx) , ub (x) = − exp (−βx)
Then the previous equation becomes:

£ ¡ ¢¤ £ ¡ ¡ ¢¢¤
α exp −αρ ysa , ysb = μβ exp −β ysa + ysb − ρ ysa , ysb
which gives
µ ¶
¡ ¢ β ¡ a ¢ 1 μβ
ρ ysa , ysb = ys + ysb − log
α+β α+β α
We see that CARA preferences lead to a linear sharing rule, with slope
β/ (α + β); the intercept depends on the Pareto weight μ.
Similarly, if both spouses exhibit Constant Relative Risk Aversion (CRRA)

with identical relative risk aversion γ, then:
x1−γ
ua (x) = ub (x) =
1−γ
and the equation is:
£ ¡ a b ¢¤−γ £ ¡ ¢¤−γ
ρ ys , ys = μ ysa + ysb − ρ ysa , ysb
which gives ¡ ¢ ¡ ¢
ρ ysa , ysb = k ysa + ysb (6.23)
where 1
μ− γ
k= 1 (6.24)
1 + μ− γ
Therefore, with identical CRRA preferences, each spouse consumes a fixed
fraction of total consumption, the fraction depending on the Pareto weight
μ. Note that, in both examples, ρ only depends on the sum ys = ysa + ysb ,
and
0 ≤ ρ0 (ys ) ≤ 1
Properties of efficient sharing rules

While the previous forms are obviously specific to the CARA and CRRA
cases, the two properties just mentioned are actually general.
Proposition 6.1 For any efficient risk sharing agreement, the sharing rule
ρ is a function of aggregate income only:
¡ ¢ ¡ ¢
ρ ysa , ysb = ρ̄ ysa + ysb = ρ̄ (ys )
Moreover,
0 ≤ ρ̄0 ≤ 1
Proof. Note, first, that the right hand side of equation (6.22) is increasing
in ρ, while the left hand side is decreasing;
¡ ¢ therefore
¡ ¢the solution in ρ must
be unique. Now, take two pairs ysa , ysb and ȳsa , ȳsb such that ysa + ysb =
ȳsa + ȳsb . Equation (6.22) is the same for both pairs, therefore its solution
must be the same, which proves the first statement. Finally, differentiating
(6.22) with respect to ys gives:
u00a (ρ̄) 0 u00b (ys − ρ̄)

ρ̄ = 0b (1 − ρ̄0 ) (6.25)
u0a (ρ̄) u (ys − ρ̄)
and finally:
00b
0
−ρ̄)
− uu0 (y(yss−ρ̄)
ρ̄ (ys ) = 00a u00b (ys −ρ̄)
(6.26)
− uu0a (ρ̄)
(ρ̄)
− u0b (ys −ρ̄)
which belongs to the interval [0, 1]. Note, moreover, that 0 < ρ̄0 (ys ) < 1
unless one of the agents is (locally) risk neutral.
The first statement in Proposition 6.1 is often called the mutuality prin-
ciple. It states that when risk is shared efficiently, an agent’s consumption
is not affected by the idiosyncratic realization of her income; only shocks
affecting aggregate resources (here, total income ys ) matter. It has been
used to test for efficient risk sharing, although the precise test is much
more complex than it may seem - we shall come back to this aspect below.
Formula (6.26) is quite interesting in itself. It can be rewritten as:
0a
0
− uu00a(ρ̄)
(ρ̄)
ρ̄ (ys ) = 0a u0b (ys −ρ̄)
(6.27)
− uu00a(ρ̄)
(ρ̄) − u00b (ys −ρ̄)
0a
The ratio − uu00a(ρ̄)
(ρ̄) is called the risk tolerance of A; it is the inverse of A’s
risk aversion. Condition (6.27) states that the marginal risk is allocated
between the agents in proportion of their respective risk tolerances. To
put it differently, assume the household’s total income fluctuates by one
(additional) dollar. The fraction of this one dollar fluctuation born by agent
a is proportional to a’s risk tolerance. To take an extreme case, if a was
infinitely risk averse - that is, her risk tolerance was nil - then ρ̄0 = 0 and
her share would remain constant: all the risk would be born by b.
It can actually be showed that the two conditions expressed by Proposi-
tion 6.1 are also sufficient. That is, take any sharing rule ρ satisfying them.
Then one can find two utility functions ua and ub such that ρ shares risk
efficiently between a and b.9
6.3.3 Efficient risk sharing in a multi-commodity context: an

introduction
Regarding risk sharing, a multi commodity context is much more complex
than the one-dimensional world just described. The key insight is that con-
sumption decisions also depend on the relative prices of the various available
commodities, and that typically these prices fluctuate as well. Surprisingly
enough, sharing price risk is quite different from sharing income risk. A
precise investigation would be outside the scope of the present volume;
instead, we simply provide a short example.10
9 The exact result is even slightly stronger; it states that for any ρ satisfying the
conditions and any increasing, strictly concave utility uA , one can find some increasing,
strictly concave utility uB such that ρ shares risk efficiently between A and B (see Chi-
appori, Samphantharak, Schulhofer-Wohl and Townsend 2010 for a precise statement).
1 0 The reader is referred to Chiappori, Townsend and Yamada (2008) for a precise
analysis. The following example is also borrowed from this article.

Consider a two agent household, with two commodities - one labor supply
and an aggregate consumption good. Assume, moreover, that agent b is risk
neutral and only consumes, while agent a consumes, supplies labor and is
risk averse (with respect to income risk). Formally, using Cobb-Douglas
preferences:
1−γ
(la ca ) ¡ ¢
U a (ca , la ) = and U b cb = cb
1−γ
with γ > 1/2. Finally, the household faces a linear budget constraint; let
wa denote 2’s wages, and y (total) non labor income.
Since agent b is risk neutral, one may expect that she will bear all the
risk. However, in the presence of wage fluctuations, it is not the case that
agent a’s consumption, labor supply or even utility will remain constant.
Indeed, ex ante efficiency implies ex post efficiency, which in turn requires
that the labor supply and consumption of a vary with his wage:
ρ + wa T a ρ + wa T
la = ,c =
2wa 2
where ρ is the sharing rule. The indirect utility of a is therefore:
2γ−1 2−2γ −(1−γ)

V a (ρ, wa ) = (ρ + wa T ) wa
1−γ
while that of b is simply V b (y − ρ) = y − ρ.

Now, let’s see how ex ante efficiency restricts the sharing rule. Assume
there exists S states of the world, and let wa,s , ys and ρs denote wage, non
labor income and the sharing rule in state s. Efficient risk sharing requires
solving the program:
X £ ¤
max π s V a (ρs , wa,s ) + μV b (ys − ρs )
ρ
s
leading to the first order condition:
∂V a (ρs , wa,s ) ∂V b (ys − ρs )

=μ
∂ρs ∂ρs
In words, efficient risk sharing requires that the ratio of marginal utilities
of income remains constant - a direct generalization of the previous results.
Given the risk neutrality assumption for agent b, this boils down to the
marginal utility of income of agent a remaining constant:
∂V a
= 2γ (ρ + wa T )1−2γ wa−(1−γ) = K
∂ρ
which gives
1−γ
ρ = 2K 0 .wa1−2γ − wa T
where K 0 is a constant depending on the respective Pareto weights. In the

end:
−γ 1−γ
− 2γ−1
la = K 0 .wa2γ−1 , ca = K 0 .wa
and the indirect utility is of the form:
1−γ
− 2γ−1
V a = K 00 .wa
for some constant K 00 . As expected, a is sheltered from non labor income

risk by his risk sharing agreement with b. However, his consumption, labor
supply and welfare fluctuate with his wage. The intuition is that that agents
respond to price (or wage) variations by adjusting their demand (here labor
supply) behavior in an optimal way. The maximization implicit in this
process, in turn, introduces an element of convexity into the picture.11
6.3.4 Econometric issues

Distributions versus realizations
We now come back to the simpler, one-commodity framework. As expressed
by Proposition 6.1, efficient risk sharing schemes satisfy the mutuality prin-
ciple, which is a form of income pooling: the sharing rules depends only on
total income, not on the agent’s respective contributions y a and y b per se.
This result may sound surprising; after all, income pooling is a standard im-
plication of the unitary setting which is typically not valid in the collective
framework; moreover, it is regularly rejected empirically.
The answer to this apparent puzzle relies on the crucial distinction be-
tween the (ex post) realization and the (ex ante) distribution of income
shocks. When risk is shared efficiently, income realizations are pooled: my
consumption should not suffer from my own bad luck, insofar as it does
not affect aggregate resources. On the other hand, there exists a contin-
uum of efficient allocations of resources, indexed by some Pareto weights;
different weights correspond to different (contingent) consumptions. The
Pareto weights, in turn, depend on the ex ante situations of the agents;
for instance, if a has a much larger expected income, one can expect that
her Pareto weight will be larger than b’s, resulting in a higher level of con-
sumption. In other words, the pooling property does not apply to expected
incomes, and in general to any feature (variance, skewness,...) of the proba-
bility distributions of individual income streams. The main intuition of the
collective model is therefore maintained: power (as summarized by Pareto
weights) matters for behavior - the nuance being that under efficient risk
1 1 Generally, the ability of risk neutral agents to adjust actions after the state is ob-
served induces a "risk loving" ingredient, whereby higher price variation is preferred,
and which may counterweight the agent’s risk aversion.
sharing it is the distribution of income, instead of its realization, that (may)

affect individual powers.
In practice, however, this raises a difficult econometric issue. Testing for
efficient risk sharing requires checking whether observed behavior satisfies
the mutuality principle, that is pooling of income realization. However, by
the previous argument, this requires being able to control for distributions,
hence to distinguish between ex post realizations and ex ante distributions.
On cross-sectional data, this is impossible.
It follows that cross-sectional tests of efficient risk sharing are plagued
with misspecification problems. For instance, some (naive) tests of efficient
risk sharing that can be found in the literature rely on a simple idea: since
individual consumption should not respond to idiosyncratic income shocks
(but only to aggregate ones), one may, on cross sectional data, regress
individual consumption (or more specifically marginal utility of individual
consumption) on (i) indicators of aggregate shocks (for example, aggregate
income or consumption), and (ii) individual incomes. According to this
logic, a statistically significant impact of individual income on individual
consumption, controlling for aggregate shocks, should indicate inefficient
risk sharing.
Unfortunately, the previous argument suggests that in the presence of
heterogeneous income processes, a test of this type is just incorrect. To get
an intuitive grasp of the problem, assume that two agents a and b share risk
efficiently. However, the ex ante distributions of their respective incomes
are very different. a’s income is almost constant; on the contrary, b may be
hit by a strong, negative income shock. In practice, one may expect that
this asymmetry will be reflected in the respective Pareto weights; since b
desperately needs insurance against the negative shock, he will be willing
to accept a lower weight, resulting in lower expected consumption than a,
as a compensation for the coverage provided by a.
Consider, now, a large economy consisting of many independent clones of
a and b; assume for simplicity that, by the law of large numbers, aggregate
resources do not vary. By the mutuality principle, efficient risk sharing im-
plies that individual consumptions should be constant as well; and since a
agents have more weight, their consumption will always be larger than that
of b agents. Assume now than an econometrician analyzes a cross section
of this economy. The econometrician will observe two features. One is that
some agents (the ‘unlucky’ b’s) have a very low income, while others (the
lucky b’s and all the a’s) have a high one. Secondly, the low income agents
also exhibit, on average, lower consumption levels than the others (since
they consume as much as the lucky b’s but less than all the a’s). Tech-
nically, any cross sectional regression will find a positive and significant
correlation between individual incomes and consumptions, which seems to
reject efficient risk sharing - despite the fact that the mutuality principle
is in fact perfectly satisfied, and risk sharing is actually fully efficient. The
key remark, here, is that the rejection is spurious and due to a misspecifi-
cation of the model. Technically, income is found to matter only because

income realizations capture (or are proxies for) specific features of income
distributions that influence Pareto weights.
A simple solution
We now discuss a specific way of solving the problem. It relies on the
availability of (short) panel data, and on two additional assumptions. One is
that agent’s preferences exhibit Constant Relative Risk Aversion (CRRA),
a functional form that is standard in this literature. In practice:
x1−α b x1−β
ua (x) = , u (x) =
1−α 1−β
The second, much stronger assumption is that risk aversion is identical
across agents, implying α = β in the previous form.
We have seen above (in equations 6.23 and 6.24) that under these as-
sumptions, the efficiency condition (6.22) leads to a sharing rule that is
linear in income, the coefficient depending on the Pareto weights. Taking
logs:
Ã 1
!
a μ− α
log c = log ρ = log 1 + log y, and
1 + μ− α
µ ¶
1
log cb = log 1 + log y
1 + μ− α
Assume, now, that agents are observed for at least two periods. We can
compute the difference between log consumptions in two successive periods,
and thus eliminate the Pareto weights; we get:
∆ log ca = ∆ log cb = ∆ log y
In words, a given variation, in percentage, of aggregate income should gen-

erate equal percentage variations in all individual consumptions.12
Of course, this simplicity comes at a cost - namely, the assumption that
individuals have identical preferences: one can readily check that with dif-
ferent risk aversion, the sharing rule is not linear, and differencing log con-
sumptions does not eliminate Pareto weights. Assuming homogeneous risk
aversions is difficult for two reasons. First, all empirical studies suggest that
the cross sectional variance of risk aversion in the population is huge. Sec-
ond, even if we assume that agents match to share risk (so that a sample of
people belonging to the same risk sharing group is not representative of the
1 2 This prediction is easy to test even on short panels - see for instance Altonji et al
(1992) and Duflo and Udry (2004); incidentally, it is usually rejected. See Mazzocco and
Saini (2006) for a precise discussion.
general population), theory13 suggests that the matching should actually

be negative assortative (that is, more risk averse agents should be matched
with less risk averse ones) - so that heterogeneity should be, if anything,
larger within risk sharing groups than in the general population.14
Finally, can we test for efficient risk sharing without this assumption?
The answer is yes; such a test is developed for instance in Chiappori,
Townsend and Yamada (2008) and in Chiappori, Samphantharak, Schulhofer-
Wohl and Townsend (2010). However, it requires long panels - since one
must be able to disentangle the respective impacts of income distributions
and realizations.
6.4 Intertemporal Behavior

6.4.1 The unitary approach: Euler equations at the household
level
We now extend the model to take into account the dynamics of the re-
lationships under consideration. Throughout this section, we assume that
preferences are time separable and of the expected utility type. The first
contributions extending the collective model to an intertemporal setting
are due to Mazzocco (2004, 2007); our presentation follows his approach.
Throughout this section, the household consists of two egoistic agents who
live for T periods. In each period t ∈ {1, ..., T }, let yti denote the income of
member i.
We start with the case of a unique commodity which is privately con-
sumed; cit denotes member i’s consumption at date t and pt is the cor-
responding price. The household can save by using a risk-free asset; let
st denotes the net level of (aggregate) savings at date t, and Rt its gross
return. Note that, in general, yti , st and cit are random variables
We start with the standard representation of household dynamics, based
on a unitary
¡ ¢framework. Assume, therefore, that there exists a utility func-
tion u ca , cb representing the household’s preferences. The program de-
1 3 See,for instance Chiappori and Reny (2007).

1 4 An alternative test relies on the assumption that agents have CARA preferences.
Then, as seen above, the sharing rule is an affine function, in which only the intercept
depends on Pareto weights (the slope is determined by respective risk tolerances). It
follows that variations in levels of individual consumptions are proportional to varia-
tions in total income, the coefficient being independent of Pareto weights. The very nice
feature of this solution, adopted for instance by Townsend (1994), is that it is com-
patible with any level of heterogeneity in risk aversion. Its main drawback is that the
CARA assumption is largely counterfactual; empirical evidence suggests that absolute
risk aversion decreases with wealth.
scribing dynamic choices is:

Ã !
X ¡ ¢
t
max E0 βu cat , cbt
t
under the constraint

¡ ¢
pt cat + cbt + st = yta + ytb + Rt st−1 , t = 0, ..., T
Here, E0 denotes the expectation taken at date 0, and β is the house-

hold’s discount factor. Note that if borrowing is excluded, we must add the
constraint st ≥ 0.
Using a standard result by Hicks, we can define household utility as
a function of total household consumption; technically, the function U is
defined by:
© ¡ ¢ ª
U (c) = max u ca , cb such that ca + cb = c
and the program becomes:

Ã !
X t
max E0 β U (ct )
t
under the constraint
pt ct + st = yta + ytb + Rt st−1
The first order conditions give the well-known Euler equations:

∙ 0 ¸
U 0 (ct ) U (ct+1 )
= βEt Rt+1 (6.28)
pt pt+1
In words, the marginal utility of each dollar consumed today equals, in ex-
pectation, β times the marginal utility of Rt+1 dollars consumed tomorrow;
one cannot therefore increase utility by marginally altering the savings.
In practice, many articles test the empirical validity of these household
Euler equations using general samples, including both couples and singles
(see Browning and Lusardi 1995 for an early survey); most of the time, the
conditions are rejected. Interestingly, however, Mazzocco (2004) estimates
the same standard household Euler equations separately for couples and for
singles. Using the CEX and the Panel Study of Income Dynamics (PSID),
he finds that the conditions are rejected for couples, but not for singles. This
seems to suggest that the rejection obtained in most articles may not be
due to technical issues (for example, non separability of labor supply), but
more fundamentally to a misrepresentation of household decision processes.
6.4.2 Collective Euler equations under ex ante efficiency

Household consumption
We now consider a collective version of the model. Keeping for the moment
the single commodity assumption, we now assume that agents have their
own preferences and discount factors. The Pareto program is therefore:
Ã ! Ã !
X a t X ³ b ´t ¡ ¢
a a b b
max (1 − μ) E0 (β ) u (ct ) + μE0 β u ct
t t
under the same constraints as above. First order conditions give:

" ¡ a ¢ #
0a
u0a (cat ) u c t+1
= β a Et Rt+1 (6.29)
pt pt+1
¡ ¢ " ¡ ¢ #
u0b cbt b u0b cbt+1
= β Et Rt+1
pt pt+1
which are the individual Euler equations. In addition, individual consump-

tions at each period must be such that:
(β a )t u0a (cat ) μ
³ ´t 0b ¡ b ¢ = (6.30)
u c 1 − μ
βb t
The right hand side does not depend on t: the ratio of discounted marginal
utilities of income of the two spouses must be constant through time. This
implies, in particular, that
³ ´t
0a
u (cat ) μ βb
¡ ¢ =
u0b cbt 1 − μ (β a )t
If, for instance, a is more patient than b, in the sense that β a > β b , then
the ratio u0a /u0b declines with time, because a postpones a larger fraction
of her consumption than b.
An important remark is that if individual consumptions satisfy (6.29),

then typically the aggregate consumption process ct = cat + cbt does not
satisfy an individual Euler equation like (6.28), except in one particular
case, namely ISHARA utilities and identical discount factors. For instance,
assume, following Mazzocco (2004), that individuals have utilities of the
CRRA form:
X
X c1−γ
u (c) = , X = a, b
1 − γX
and that, moreover, β a = β b = β. Then (6.29) becomes:

½ ∙ ¸¾−1/γ a
Rt+1 ¡ a ¢−γ a
cat = βpt Et ct+1 (6.31)
pt+1
½ ∙ ¸¾−1/γ b
Rt+1 ¡ b ¢−γ b
cbt = βpt Et ct+1
pt+1
If γ a = γ b (the ISHARA case), one can readily see that the ratio cat+1 /cbt+1
is constant across states of the world; therefore
cat+1 = kct+1 , cbt+1 = (1 − k) ct+1
for some constant k. It follows that:

½ ∙ ¸¾−1/γ
a Rt+1 −γ
ct = βpt Et (kct+1 ) (6.32)
pt+1
½ ∙ ¸¾−1/γ
Rt+1 −γ
= k βpt Et (ct+1 )
pt+1
and by the same token
½ ∙ ¸¾−1/γ
Rt+1 −γ
cbt = (1 − k) βpt Et (ct+1 )
pt+1
so that finally:
½ ∙ ¸¾−1/γ
Rt+1 −γ
ct = cat + cbt = βpt Et (ct+1 ) (6.33)
pt+1
and aggregate consumption satisfies an individual Euler equation: the house-
hold behaves as a single.
However, in the (general) case γ a 6= γ b , Mazzocco shows that this result
no longer holds, and household aggregate consumption does not satisfy
a Euler equation even though each individual consumption does. In par-
ticular, testing the Euler conditions on aggregate household consumption
should lead to a rejection even when all the necessary assumptions (effi-
ciency, no credit constraints, ...) are fulfilled.
Individual consumption and labor supply

The previous, negative result is not really surprising: it simply stresses once
more than groups, in general, do not behave as single individuals. What
then? Well, if individual consumptions are observable, conditions (6.29) and
(6.30) are readily testable using the standard approach. Most of the time,
however, only aggregate consumption is observed. Then a less restrictive
framework is needed. In particular, one may relax the single commodity
assumption. Take, for instance, a standard model of labor supply, in which

each agent consumes two commodities, namely leisure and a consumption
good. The collective model suggests that individual consumptions can be
recovered (up to additive constants - see chapters 4 and 5). Then tests of
the Euler equation family can be performed.
As an illustration, Mazzocco (2007) studies a dynamic version of the
collective model introduced in chapter 4. The individual Euler equations
become, with obvious notations:
¡ ¢ " ¡ ¢ #
∂ui cit , lti /∂c i ∂ui cit+1 , lt+1
i
/∂c
= β Et Rt+1 (6.34)
pt pt+1
¡ ¢ " ¡ ¢ #
∂ui cit , lti /∂l i ∂ui cit+1 , lt+1
i
/∂l
= β Et Rt+1
wti i
wt+1
for i = a, b. In particular, since individual labor supplies are observable,

these equations can be estimated.
6.4.3 The ex ante inefficiency case

What, now, if the commitment assumption is not valid? We have seen above
that this case has a simple, technical translation in the collective frame-
work - namely, the Pareto weights are not constant. A first remark, due to
Mazzocco (2007), is that even in the ISHARA case, aggregate consumption
no longer satisfies the martingale property (6.33). Indeed, let μt denote the
Pareto weight of b in period t, and assume for the moment that μt does not
depend on the agent’s previous consumption decisions. We first have that
½ ∙ ¸¾−1/γ a ½ ∙ ¸¾−1/γ b
Rt+1 ¡ a ¢−γ a Rt+1 ¡ b ¢−γ b
cat +cbt
= βpt Et c + βpt Et c
pt+1 t+1 pt+1 t+1
(6.35)
Moreover,
Ã !t
u0a (cat ) μt βb
¡ b¢ = (6.36)
u0b ct 1 − μt β a
for all t, which for ISHARA (γ a = γ b = γ) preferences becomes
µ a ¶−γ Ã !t
ct μt βb
=
b
ct 1 − μt β a
If μt is not constant, neither is the ratio cat /cbt . A result by Hardy, Little-
wood and Polya (1952) implies that whenever the ratio x/y is not constant,
then for all probability distributions on x and y:
n h io−1/γ © £ ¤ª−1/γ © £ −γ ¤ª−1/γ
Et (x + y)−γ > Et x−γ + Et y
which directly implies that:

½ ∙ ¸¾−1/γ
Rt+1 −γ
βpt Et (ct+1 ) > ct
pt+1
In words, the (marginal utility of) aggregate consumption now follows a
supermartingale.
Regarding now individual consumptions, one can readily check that equa-
tions (6.29) become:
" ¡ ¢ #
u0a (cat ) ¡ ¢ a u0a cat+1
(1 − μt ) = 1 − μt+1 β Et Rt+1 (6.37)
pt pt+1
¡ ¢ " ¡ ¢ #
u0b cbt u0b
cb
t+1
μt = μt+1 β b Et Rt+1
pt pt+1
or equivalently:
" ¡ ¢ #
u0a cat+1 pt Rt+1 1 1 − μt
Et = (6.38)
u0a (cat ) pt+1 β a 1 − μt+1
" ¡ ¢ #
u0b cbt+1 pt Rt+1 1 μt
Et ¡ ¢ =
u0b cbt pt+1 β b μt+1
In words: under full commitment, the left hand side expressions should
be constant, while they may vary in the general case. A first implication,
therefore, is that whenever individual consumptions are observable, then
the commitment assumption is testable. Moreover, we know that (6.36)
holds for each t. These relations imply that μt is identifiable from the data.
That is, if Pareto weights vary, it is possible to identify their variations,
which can help characterizing the type of additional constraint that ham-
pers full commitment.
Finally, individual consumptions are not observed in general, but individ-
ual labor supplies typically are; the same tests can therefore be performed
using labor supplies as indicated above. Again, the reader is referred to
Mazzocco (2007) for precise statements and empirical implementations. In
particular, Mazzocco finds that both the unitary and the collective model
with commitment are rejected, whereas the collective model without com-
mitment is not. This finding suggests that while static efficiency may be
expected to hold in general, dynamic (ex ante) efficiency may be more
problematic.
6.4.4 Conclusion
The previous results suggest several conclusions. One is that the collec-
tive approach provides a simple generalization of the standard, ‘unitary’
approach to dynamic household behavior. Empirically, this generalization

seems to work significantly better than the unitary framework. For instance,
a well-known result in the consumption literature is that household Euler
equations display excess sensitivity to income shocks. The two main ex-
planations are the existence of borrowing constraints and non-separability
between consumption and leisure. However, the findings in Mazzocco (2007)
indicate that cross-sectional and longitudinal variations in relative decision
power explain a significant part of the excess sensitivity of consumption
growth to income shocks. Such variations, besides being interesting per
se, are therefore crucial to understanding the dynamics of household con-
sumption. A second conclusion is that the commitment issue is a crucial
dimension of this dynamics; a couple in which agents can credibly commit
on the long run will exhibit behavioral patterns that are highly specific.
Thirdly, it is possible to develop models that, in their most general form,
can capture both the ‘collective’ dimensions of household relationships and
the limits affecting the spouse’s ability to commit. The unitary model and
the full efficiency version of the collective approach are nested within this
general framework, and can be tested against it.
6.5 Divorce
6.5.1 The basic model
Among the limits affecting the spouses’ ability, an obvious one is the pos-
sibility of divorce. Although divorce is, in many respects, an ancient insti-
tution, it is now more widespread than ever, at least in Western countries.
Chiappori, Iyigun and Weiss (2008) indicate for instance that in 2001,
among American women then in their 50s, no less than 39% had divorced
at least once (and 26% had married at least twice); the numbers for men
are slightly higher (respectively 41% and 31%). Similar patterns can be
observed in Europe (see chapter 1). Moreover, in most developed countries
unilateral divorce has been adopted as the legal norm. This implies that
any spouse may divorce if (s)he will. In practice, therefore, divorce intro-
duces a constraint on intertemporal allocations within the couple; that is,
at any period, spouses must receive each within marriage at least as much
as they would get if they were divorced.
Clearly, modeling divorce - and more generally household formation and
dissolution - is an important aspect of family economics. For that purpose,
a unitary representation is probably not the best tool, because it is es-
sential to distinguish individual utilities within the couple. If each spouse
is characterized, both before and after marriage, by a single utility, while
the couple itself is represented by a third utility with little or no link with
the previous ones, modeling divorce (or marriage for that matter) becomes
very difficult and largely ad hoc . Even if the couple’s preferences are closely
related to individual utilities, for instance through a welfare function a la

Samuelson, one would like to investigate the impact of external conditions
(such as wages, the tax-benefit system or the situation on the marriage
market) on the decision process leading to divorce; again, embedding the
analysis within the black box of a unitary setting does not help clarifying
these issues.
In what follows, we show how the collective approach provides a useful
framework for modeling household formation and dissolution. Two ingre-
dients are crucial for this task. One is the presence of economic gains from
marriage. A typical example is the presence of public goods, as we have
extensively discussed in the previous chapters. Alternative sources of mar-
ital gains include risk sharing or intertemporal consumption smoothing,
along the lines sketched in the previous sections. At any rate, we must first
recognize that forming a couple is often efficient from the pure economic
perspective.
A second ingredient is the existence of non-pecuniary benefits to mar-
riage. These ‘benefits’ can be interpreted in various ways: they may repre-
sent love, companionship, or other aspects. The key feature, in any case,
is that these benefits are match-specific (in that sense, they are an indi-
cator of the ‘quality’ of the match under consideration) and they cannot
be exactly predicted ex ante; on the contrary, we shall assume that they
are revealed with some lag (and may in general be different for the two
spouses). The basic mechanism is that a poor realization of the non pecu-
niary benefits may trigger divorce, either because agents hope to remarry
(and, so to speak, ‘take a new draw’ from the distribution of match qual-
ity), or because the match is so unsatisfactory that the spouses would be
better off as singles, even at the cost of forgoing the economic gains from
marriage. The existence of a trade-off between the economic surplus gener-
ated by marriage and the poor realization of non economic benefits plays
a central role in most models of divorce.
More specifically, we shall consider a collective framework in which cou-
ples may consume both private and public goods, and marriage generates
a non-pecuniary benefit. In principle, this benefit can enter individual util-
ities in an arbitrary manner. In what follows, however, we concentrate on
a particular and especially tractable version of the model, initially due to
Weiss and Willis (1993, 1997), in which the non monetary gain is additive;
that is, the utility of each spouse is of the form
¡ ¢
U i = ui q i , Q + θi , i = a, b
¡ ¢
where q i = q1i , ..., qni is the vector of private consumption of agent i,
Q = (Q1 , ..., QN ) is the vector of household public consumption, and θi is
the non monetary gain of i. In particular, while the total utility does depend
on the non monetary components θi , the marginal rates of substitution
between consumption goods does not, which simplifies the analysis.
³ ´
For any couple, the pair θa , θb of match qualities is drawn from a
given distribution Φ. In general, any correlation between θa and θb is pos-
sible. Some models introduce an additional restriction by assuming that
the quality of the match is the same for both spouses - that is, θa = θb .
To keep things simple, we present the model in a two periods framework.
In period one, agents marry and consume. At the end of the period, the
quality of the match is revealed, and agents decide whether to remain
married or split. If they do not divorce, they consume during the second
period, and in addition enjoy the same non monetary gain as before. If they
split, we assume for the moment that they remain single for the rest of the
period, and that they privately consume the (previously) public goods.15
The prices of the commodities will not play a role in what follows; we may,
for simplicity, normalize them to unity.
Finally, let y a and y b denote the agents’ respective initial incomes, which
they receive at the beginning of each period; and to simplify, we assume no
savings and borrowing. In case of divorce, the couple’s total income, y a +y b ,
is split between the ex-spouses. The rule ¡ governing
¢ this division leads
¡ a to ¢an
a a b b b
allocation in which
¡ a b¢a receives some D y , y and b receives D y , y =
a b a
y + y − D y , y . For instance,¡ if incomes ¢ are considered to be private
property of each spouse, then Di y¡a , y b ¢= y i , i =¡ a, b, ¢whereas ¡ a an equal
¢
distribution rule would lead to Da y a , y b = D b
y
¡ a b¢
a b
, y = y + y b /2.
A natural interpretation is that the rule D = D , D is exogenous and
imposed by law; however, while an agent cannot be forced to transfer to
the ex-spouse more than the legal amount D, he may freely elect to do
so, and will in some cases (see next subsection). An alternative approach
considers divorce contracts as endogenous, for instance in a risk sharing
perspective.16
We may now analyze the couple’s divorce decision. ¡ ¡ First, ¢¢ the second
period utility of agent i if divorced is simply V i Di y a , y b (where, as
before, V i is agent i’s indirect utility). If, on the other hand, the spouses
remain married, then they choose some efficient allocation; as usual, their
consumption plan therefore solves a program of the type:
max ua (q a , Q) + θa

X¡ ¢ X
qja + qjb + Qk = ya + yb
j k
¡ ¢
u q b , Q + θb
b
≥ ūb
1 5 Some commodities may remain public even after divorce; children expenditures are
a typical example. For a detailed investigation, see Chiappori et al (2007).

1 6 See for instance Chiappori and Weiss (2009).
¡ ¢
where ūb is a constant. Let q̄ a , q̄ b , Q̄ denote the solution to this program,
and ūa = ua (q a , Q) + θa the corresponding ¡ ¢utility for a. Note that both
are functions of ūb ; we note therefore ūa ¡ūb . Let ¢ PM denote the Pareto ¡ ¢
set if married, that is the set of utilities ua , ub such that ua ≤ ūa ub ;
in words, any pair of utilities in PM can be reached by the couple if they
remain married.
Then we are in one of the following two situations:
¡ ¡ ¡ ¢¢ ¡ ¡ ¢¢¢
• either the reservation point V a Da y a , y b , V b Db y a , y b , rep-
resenting the pair of individual utilities reachable through divorce,
belongs to the Pareto set if married, PM . Then there exists a second
period distribution of income which is preferred over divorce by both
spouses. The efficiency assumption implies that this opportunity will
be taken, and the model predicts that the marriage will continue.
¡ ¡ ¡ ¢¢ ¡ ¡ ¢¢¢
• or, alternatively, V a Da y a , y b , V b Db y a , y b is outside PM .
Then the marriage cannot continue, because any second period al-
location of resources the spouses may choose will be such that one
spouse at least would be better off as a single; therefore, divorce must
follow.
The model thus provides a precise description of the divorce decision;

namely, divorce takes place whenever it is the efficient decision under the
constraint
¡ ¡ that ¢¢ agents cannot receive less than their reservation utility
V i Di y a , y b , i = a, b.
Some remarks are in order at this point. First, the argument presented
above assumes that divorce is unilateral, in the sense that each partner is
free to terminate the marriage and obtain divorce, even if the spouse does
not agree. An alternative setting requires mutual consent - that is, divorce
cannot occur unless both spouses agree. An old question of family economics
is whether a shift from mutual consent to unilateral has an impact on
divorce rates; we shall consider that question in the next subsection.
Secondly, the fact that spouses may disagree about divorce - that is,
a spouse may ask for divorce against the partner’s will - does not imply
that they will. In the setting just presented, a partner who would con-
sider divorce may sometimes be ‘bribed back’ into marriage by her spouse,
through an adequate redistribution of income. Only when such a redistrib-
ution cannot take place, because the cost to the other partner would exceed
the benefits of remaining married, will divorce occur. In that sense, there
is not disagreement about divorce in this model; simply, divorce sometimes
comes out as the best solution available.
A third remark is that, ultimately,
³ divorce
´ is triggered by the realization
a b
of the match quality parameters θ , θ . Large values of the θs inflate the
Pareto frontier, making it more likely to contain the divorce threat point;
conversely, poor realizations contract it, and divorce becomes probable.
Formally, it is easy to check that the divorce decision is monotonic in³ the θ’s,
´
a b
in the sense that if a couple remains married for some realization θ̄ , θ̄ ,
³ ´
i
then they also do for any θa , θb such that θi ≥ θ̄ , i = a, b; and conversely,
³ ´ ³ ´
a b
if they divorce for some θ̄ , θ̄ , so do they for any θa , θb such that θi ≤
i
θ̄ , i = a, b. In general, there exists a divorce frontier, namely a³decreasing
´
function φ such that the coupe divorce if and only if θa < φ θb . Note,
however, that for a ‘neutral’ realization θa = θb = 0, the couples always
remains married, because of the marital gains arising from the presence of
public consumption; negative shocks are required for a marriage to end.
Finally, how is the model modified when divorced agents are allowed to
remarry? The basic principle remains valid - that is, agents (efficiently)
divorce if no point within the Pareto frontier if married can provide both
agents with the same expected utility as if single. The latter value is however
more difficult to compute, because it now includes the probability of finding
a new mate multiplied by the utility the ex spouse will get in their new
marriage. In other words, one need to predict which particular allocation
of resources and welfare will prevail in newly formed couples - a task that
requires a more complete investigation of the equilibrium forces governing
the (re)marriage market. We shall come back to this issue in the second
part of the book.
6.5.2 Divorce under transferable utility and the Becker-Coase

theorem
The TU framework
We now further investigate the divorce model under an additional assump-
tion - namely, that utility is transferable between spouses, both during and
after marriage.17 Technically, we first assume that preferences of married
individuals are of the generalized quasi-linear (GQL) form (see Bergstrom,
1989). ¡ ¢ ¡ ¢
uim q i , Q = F (Q) q1i + Gim Q, q−1i
+ θi , i = a, b (6.39)
i
¡ i i
¢ i
where q−1 = q2 , ...qn . Here, the functions F and Gm , i = a, b, are posi-
tive, increasing, concave functions such that F (0) = 1 and Gim (0) = 0.
Secondly, we assume that preferences if single take the strictly quasi-
linear form: ¡ ¢ ¡ ¢
uis q i , Q = qi1 + Gis Q, q−1
i
, i = a, b, (6.40)
where again the Gis , i = a, b, are increasing concave functions, with Gis (0) =
0. Because of quasi-linearity, the optimal consumptions of public goods and
1 7 The material presented in this subsection is borrowed from Chiappori, Iyigun and
Weiss (2007).
private goods other than good 1 are given by the conditions:

¡ ¢ ¡ ¢
∂Gis Q, q−1
i
∂Gis Q, q−1
i
= 1, 1 ≤ j ≤ N and i
= 1, 2 ≤ k ≤ n .
∂Qj ∂qki
Neither these conditions nor the optimal levels of all private and public
consumptions (except¡ for¢ good 1) depend on income. Let these optimal
i
levels be denoted Q̄, q̄−1 . To simplify notations, we choose units such that
¡ ¢ PN P
Gs Q̄, q̄−1 = j=1 Q̄j + nk=2 q̄ki , i = a, b. Then, the indirect utility of a
i i
single person equals his or her income.

Now, consider a man with income y b married with a woman with income
a
y . There is a unique efficient level for the consumption of each of the public
goods and each of the private goods 2 to n. Moreover, these levels depend
only on the total income of the partners, y = y a + y b . If we define
( h P Pn ¡ a ¢i )
F (Q) y − N Q j + q k + q b
η (y) = max ¡ j=1a ¢ k=2
¡ ¢ k
(Q,q−1 −1 )
a ,q b
+Gam Q, q−1 + Gbm Q, q−1
b
then the Pareto frontier is given by
uam + ubm = η (y) + θa + θb , (6.41)
Here, uam and ubm are the attainable utility levels that can be implemented
by the allocations of the private good q1 between the two spouses, given
the efficient consumption levels of all other goods. The Pareto frontier is
a straight line with slope -1: utility is transferable between spouses (see
chapter 3). Assuming, as is standard, that the optimal public consumptions
are such that F (Q) is increasing in Q, we see that η(y) is increasing and
convex in y.18 Moreover, η (0) = 0 and η 0 (0) = F (0) = 1. Since η is convex,
this implies that η (y) > y and η0 (y) > 1 for all y > 0.
Finally, if divorce takes place, the post-divorce utility of agent i is:
¡ ¡ ¢¢ ¡ ¢
Vsi Di y a , y b = Di y a , y b (6.42)
In particular, we see that

¡ ¢ ¡ ¢
Vsa + Vsb = Da y a , y b + Db y a , y b = y a + y b = y (6.43)
In this framework, the divorce decision takes¡ a particularly

¢ simple form.
Indeed, agents divorce if and only if the point Vsa , Vsb is outside the Pareto
set when married. Given (6.41), this occurs when the sum Vsa + Vsb is larger
1 8 By the envelope theorem, the derivative η 0 (y) is equal to F (Q). Therefore, η is
increasing in y and, if F (Q) is increasing in y as well, then η is convex. Note that a

sufficient (but by no means necessary) condition is that public consumptions are all
normal.
than η (y) + θa + θb . Using (6.43), we conclude that divorce takes place

whenever:
η(y) + θa + θb < y
or equivalently:
θa + θb < y − η (y) (6.44)
Condition (6.44) has a simple, probabilistic translation; namely, the prob-

ability that a couple with total income y divorces is simply
³ ´
P = Pr θa + θb < y − η (y) = Φ (y − η (y))
where Φ is the cumulative distribution function of θa +θb . As expected, the

threshold θ̄ = y − η (y) is negative and decreases with income: wealthier
couples are less likely to divorce, because they receive larger economic gains
from marriage. Note also that the divorce decision only depends on the
realization of the sum θa + θb : under transferable utility, a poor realization
of θ for one spouse can always be compensated by a transfer from the
partner.
The Becker-Coase theorem

This result has several consequences. One is that the divorce decision does
not depend on the law governing post divorce income allocation; indeed,
condition (6.44) above is independent of the rule D.
Moreover, let us compare the two dominant legal systems governing di-
vorce, namely unilateral divorce and mutual consent. One can readily see
that in both cases, agents divorce if and only if condition (6.44) is satisfied.
The result is obvious under unilateral divorce, because condition (6.44) im-
plies that no intrahousehold resource allocation can provide both agents
with at least as much as their utility if single. The case of mutual consent
is slightly more complex, because even when condition (6.44) is satisfied,
the post divorce allocation D may be such that one member, say a, strictly
loses from divorce (of course, (6.44) then requires that her spouse, b, strictly
gains). But then b may bribe a into divorcing by offering a post divorce
allocation that is more favorable to a than Da . Of course, the price for b is
that he will receive less than Db . But condition (6.44) precisely state that
this is still better for b than remaining married.
We can therefore conclude that the laws governing divorce have no impact
on divorce probability. This neutrality result, initially established by Becker
in a slightly less general framework, is in fact a natural consequence of the
well-known Coase (1960) theorem, stating that under transferable utility,
the allocation of the surplus stemming from a decision has no impact on
the decision taken. This does not mean that divorce laws are irrelevant,
but simply that they only influence the distribution of welfare between
Uw
Utility Frontier in Divorce
Utility Frontier in
Marriage θ h + θ w < 0
0 Uh
FIGURE 6.2. Pareto frontiers in marriage and divorce, no public goods
the spouses, both in marriage and after divorce - not the divorce decision
itself.19
The corresponding intuition is easy to grasp from Figure 6.2. Under
transferable utility, both the Pareto frontier when married and the Pareto
frontier when divorced are straight line with slope −1. Therefore, they
cannot intersect; one Pareto set must be included within the other. The
optimal divorce decision simply picks up the larger Pareto set. What legal
dispositions can do is vary the post divorce allocation along the post divorce
Pareto frontier. But if the latter is located within the Pareto set when
married, there always exist a particular redistribution of marital surplus
that will make both spouses better off than divorce; if, conversely, it is
located outside, then whatever the planned allocation of resources within
the couple, it is always possible to redistribute income after divorce in such
a way that both agents prefer separation.
Finally, it is important to understand the assumptions that are needed for
the Becker-Coase theorem to hold. Chiappori, Iyigun and Weiss 2007 (from
now on CIW) show that there are three. One is that utility is transferable
within marriage (which, in our setting, justifies the GQL form taken for
19 A recent attempt to test this theoretical prediction is Wolfers (2006).

utilities when married). A second requirement is that utility be transferable

after divorce; here, we have therefore assumed quasi-linear preferences for
singles. Finally, the slopes of the two Pareto frontiers (before and after
divorce) must be equal.
While these requirements are indeed satisfied in the example just given,
they are in fact quite unlikely to hold in reality. For instance, the assump-
tion of quasilinear preferences if single is totally ad hoc . Assume, on the
contrary, that preferences if single have the same general form as when
married - that is, that:
¡ ¢ ¡ ¢
uis q i , Q = Fsi (Q) q1i + Gis Q, q−1
i
+ θi , i = a, b
The question, now, is whether commodity Q, which was publicly consumed

when the couple was married, remains public after divorce. In many cases, it
does not; for instance, housing typically stops being jointly consumed after
the separation. CIW show that, in that case, the second requirement is not
satisfied in general. In other situations, the commodity remains public, in
the sense that it still enters both ex-spouse’s utilities; this is the case for
children consumption, for instance. However, the utility adults derive from
children’s well being may well change after divorce, especially for the parent
who does not have full custody. Technically, the Fsi function is now different
between spouses, which violates either the second or the third requirement.
All in all, CIW argue that, in general, these requirements are unlikely to
be fulfilled - therefore that the Becker-Coase result is unlikely to hold.
An important implication is that the claim, frequently encountered in the
literature, that the Becker-Coase theorem is a consequence of the efficiency
assumption is incorrect. Whenever any of the CIW requirements is violated,
the neutrality result does not hold true. Then the general model developed
in the previous subsection, which only assumes efficient behavior (including
for divorce decisions), remains valid; but one can find situations in which
couples would split under unilateral divorce but not under mutual consent
- and also, more surprisingly, cases in which this intuition is reversed, in the
sense that divorce occurs under mutual consent but not under unilateral
divorce. Figure 6.3, borrowed from Clark (1999) and CIW, illustrates the
latter case. With mutual consent, each partner has a "property right" on
the allocation within marriage, represented by M . This point is contained
in the divorce frontier and both partners can be made better of by renego-
tiating the divorce settlement and leaving the marriage. In contrast, with
unilateral divorce partners have property rights on their divorce allocation,
represented by D. This point, however, is contained within the marriage
frontier and the partners can find an allocation within marriage that will
sustain the marriage.
Uw
Utility Frontier in Divorce
Utility Frontier in
Marriage θ h + θ w < 0
0 Uh
FIGURE 6.3. Divorce without transferable utility
6.5.3 Divorce and investment in children: a non transferable

utility example
Endogenous divorce raises some particular contracting issues that do not
arise when divorce is exogenous. This is particularly true when we take
into account marriage specific investments, such as children - who are (at
least partially) ‘specific’ in the sense that the welfare derived by the parents
from the presence of children is often reduced upon divorce (that is, parents
suffer a ‘capital loss’ upon divorce). This kind of problem usually motivates
post divorce transfers in the form of child support that will be discussed
at length in Chapter 11. Here we wish to examine the role of these post
divorce transfers on the investment in children when they are young. To
highlight their role, we shall now discuss an extreme case in which such
transfers are not possible within marriage, because all goods that a couple
consumes are public; therefore post divorce transfers are the only feasible
transfers between the spouses.
Agents live two periods. Marriage takes place at the beginning of the first
period and each marriage produces one child. Caring for the child requires
an investment of time by both parents in the first period and the outcome
(child quality) is enjoyed in the second period. The household production
function for child quality is

q
Q= (1 + ta )(1 + tb ) (6.45)
where ta and tb are are the proportions of available time spent on child
care by a and b, respectively. The time constrains are
0 ≤ ta ≤ 1 (6.46)
0 ≤ tb ≤ 1
The opportunity cost of the time spent with children in the first period
is market work. In the second period there is no need to spend time on
children and both spouses work full time. However the wage in the second
period of life depends on the amount of market work in the first period.
We normalize the first period wage of a to 1 and assume that wb < 1.
We further assume that the second period wages are directly proportional
to the first period labor supply - that is, they are equal to γ(1 − ta ) and
γwb (1−tb ) for a and b respectively, where γ > 1. Effectively, this means that
incomes in the two period are proportional, which simplifies the analysis
considerably.
The utility that parents derive from the child (or child quality) depends
on whether or not the parents live together. If the parents stay married,
their utility from quality is α ln Q, but if the parents separate, their utility
from child quality is reduced to (1 − δ)α ln Q, where 0 < δ < 1. The utility
parents depends on the child quality, on their consumption of goods q and
if married, the quality of their match, θ, that is revealed only after one
period of marriage.
If the partners are married, the utility of both partners is
um = ln q + α ln Q + θ (6.47)
Divorce may occur if the realized value (revealed at the beginning of the
second period) is sufficiently low. Following divorce, the utilities of the
former spouses are
ud = ln qdi + (1 − δ)α ln Q, i = a, b (6.48)
where qdidenotes the post divorce consumption of the two spouses. Note
that we assume here that when a couple is married all good are public. The
only way to influence the division of the gains from marriage is through
transfer in the aftermath of divorce. As we shall show, such transfers can
influence the investment in children during marriage and probability of
divorce.
As in the previous subsection, we continue to assume no borrowing or
lending. Then,
q1 = wa (1 − ta ) + wb (1 − tb ) (6.49)
q2 = γwa (1 − ta ) + γwb (1 − tb ) = γq1
where q1 denotes the joint consumption in the first period, while q2 is

the joint consumption if the partners remain married or the sum of their
private consumptions if they separate. Thus, the allocation of time in the
first period determines the consumption available to the parents each period
as well as the quality of the child that they enjoy in the second period. The
only issue then is how is this allocation determined.
A necessary condition for an efficient allocation of time is that the cost
of producing child quality, in terms of the foregone earnings of the couple
during the two periods of life, should be minimized. In the this example,
these costs are
C(Q) = (1 + γ)(wa ta + wb tb ) (6.50)
and cost minimization takes a simple form. In particular if there is an
interior solution and both partners contribute time to the child20 then we
must have
wb (1 + tb ) = 1 + ta (6.51)
Whether or not an interior solution arrises, efficiency requires that the low
wage person, b, should contribute more time to the child and the question
is if and how such unequal contribution can be implemented. The answer
depends on the contracting options that the couple have. We shall assume
here that the partners can always commit, at the time of marriage, on some
post divorce allocation of resources, provided that it falls within some legal
bounds. The justification for this assumption is that the event of separation
and the resources available upon separation can be verified so that contracts
contingent on these variables can be enforced by law. Denoting by β the
share received by the low wage person, b, the post divorce consumption
levels are
qda = (1 − β)[γwa (1 − ta ) + γwb (1 − tb )] (6.52)
qdb = β[γwa (1 − ta ) + γwb (1 − tb )].
It is more difficult, however, to verify the time allocation and in particular
time spent on children, and we shall allow for the possibility that partners
cannot commit at the time of marriage on how much time they will spend
with the child.
Following the realization of θ at the beginning of the second period, and
given the predetermined quality of children and divorce contract, marriage
will continue if
α ln Q + ln q2 + θ ≥ (1 − δ)α ln Q + max{(α ln qda , α ln qdb ) (6.53)
2 0 The efficiency requirements include regions in which only one person contributes.
These regions depend on the desire for children relative the wages of the two spouse. If
α < 1 the mother will work only at home and the father only in the market. To allow
for an interior solution, we assume that 2 > α > 1. Then for αwb > 1, both partners
work part time at home and part time in the market.
and dissolve otherwise. This rule holds because, by assumption, utility is

not transferable within marriage and each partner is free to walk away
from the marriage. Clearly, the person who can attain higher consumption
outside marriage will trigger the divorce.
Examining equation (6.52), we see that if b receives a higher share of
family resources upon divorce, β > 12 , he will trigger the divorce and divorce
occurs if
θ < −δα ln Q + a ln β)
If a obtains the larger share, β < 12 , she will trigger the divorce an divorce
occurs if
θ < −δ ln αQ + α ln(1 − β)
Finally, with equal sharing divorce occurs if
θ < −δα ln Q − α ln 2
The probability of divorce is, therefore,

⎧ 1
⎨ F (−δα ln Q + ln(1 − β)) if β≤ 2
1
Prob(divorce) = F (−δα ln Q − ln 2) if β= 2 (6.54)
⎩ 1
F (−δα ln Q + ln β) if β≥ 2
where F (.) is the cumulative distribution of θ. We assume that this distri-

bution is symmetric with zero mean. We see that a high child quality, Q,
and high loss of child quality upon divorce, δ, generate higher gains from
continued marriage and reduce the probability of divorce. A negative shock
to θ is required to initiate a divorce, because of the cost associated with
reduced child quality, represented here by the term δα ln Q, and loss of the
utility gains from joint consumption, which depends on the allocation of
resources upon divorce (ln(1 − β)) if β ≤ 12 or ln β if β ≥ 12 ).
At this point we can already make three observations:
• An increase in child quality reduces the probability of divorce.
• For a given child quality, Q, the lowest probability of divorce is at-

tained when β = 12 .
• For β 6= 12 , divorce is inefficient in the sense that the spouse who

triggers the divorce does not internalize the reduced welfare of the
spouse who is left behind and would rather stay married for at least
some range of θ’s below the trigger. Note that the contrast to the
results in the previous section, where divorce was efficient and the
probability;of divorce was independent of the division of income in
the aftermath of divorce. The Becker-Coase theorem does not hold
when transfers within marriage are not feasible.
We now turn to the determination of the investment in children in the

first period. We first consider the benchmark case of equal sharing, with
β = 12 . Defining the trigger value for divorce as
θ∗ = −δα ln Q − ln 2, (6.55)
the expected utility of each of the two partners is then

Z∞
∗
E(u) = ln q1 + (1 − F (θ )[α ln Q + ln q2 ] + θf (θ)dθ
θ∗
q2
+F (θ∗ [(1 − δ)α ln Q + ln ]
2
Z∞
= ln q1 + ln q2 + α ln Q + θf (θ)dθ + F (θ∗ )θ∗
θ∗
Maximizing E(u) with respect to ta and tb , respectively, we obtain the first

order conditions for an interior solution
1 γ α
+ = [1 − δF (θ∗ ] (6.56)
q1 q2 2(1 + ta )
wb wb γ α
+ = [1 − δF (θ∗ )] (6.57)
q1 q2 2(1 + tb )
The interpretation of these two conditions is transparent. For each spouse,

the couple equates the expected marginal gain in terms of child quality,
associated with an increase in the time investment, to the marginal costs
in terms of forgone consumption of the parents in the two periods. The
two conditions together imply condition (6.55) which means that, under
equal division, efficiency is maintained. Importantly, there is no need for
the partners to commit on the time spent with the child because the Nash
equilibrium that arises under non cooperation satisfies exactly the same
conditions. That is, in equilibrium, each spouse, including the low wage
person who is called upon to supply more hours, would do it from selfish
reasons, provided that the other spouse supplies the efficient quantity of
time.
The situation is quite different if the partners choose ex-ante an unequal
division but cannot commit on the allocation of time. For concreteness,
consider the case that in which the low wage person, b, is the husband and
he receives a lower share of family resources, β < 12 . Now each spouse will
maximize their own payoff functions. Let the new trigger function be
θ̂ = −δα ln Q + ln(1 − β). (6.58)
Then, the choice of ta as a function tb is determined by the maximization

with respect to ta of
Z∞
a
E(u ) = ln q1 + ln q2 + α ln Q + θf (θ)dθ + F (θ̂)(θ̂), (6.59)
θ̂
with the first order condition

1 γ α
+ = [1 − δF (θ̂] . (6.60)
q1 q2 2(1 + ta )
Similarly, the choice of tb as a function ta is determined by the maximization

with respect to tb of
Z∞
b
E(u ) = ln q1 + ln q2 + α ln Q + θf (θ)dθ + F (θ̂)θ̂ + F (θ̂)[ln β − ln(1 − β],
θ̂
(6.61)
with the first order condition
wb γwb β α
+ = [1 − δF (θ̂) + f (θ̂) ln ] . (6.62)
q1 q2 1 − β 2(1 + tb )
We see that the expected marginal reward from exerting effort is smaller
β
to the husband (note that for β < 12 , ln 1−β < 0). The husband takes
into account her lower consumption, and thus higher marginal utility from
consumption, following divorce. He responds by shifting additional time
in the first period into work so that his future wage will be higher. This
defensive investment in market work by the husband causes an inefficient
time allocation. Examining conditions (6.60) and (6.61), we see that the
requirement for cost minimization is not satisfied.
When partners cannot commit on the allocation of time, commitments
made at the time of marriage should adjust. One may assume that the
husband has a higher bargaining power at the time of marriage, because of
his higher wage and thus higher consumption as single. However, it makes
sense for the husband to give up some of his power, which will raise the
"pie" available during marriage that he and his the wife enjoy equally.
Returning now to the case of equal division and efficient allocation of
time, we can provide some further analysis of the investment decision. Using
the efficiency conditions (and constant returns to scale) we have that, in
an interior solution, √
Q = wb (1 + tb ). (6.63)
We also have that
q1 = 1 + wb − ta − wb tb (6.64)
√
= 2(1 + wb ) − 2 wb Q
We can, therefore, rewrite condition (6.60) in the form

√
2 wb α
√ = [1 − δF (−δα ln Q − ln 2] . (6.65)
b
1+w − w Q b Q
Condition (6.65) then determines the desired child quality and we can then
use the efficiency conditions to trace back the implied allocation of time.
The left hand side of (6.65) represents the marginal disutility (associated
with lost consumption) and unambiguously rises with Q. However, the right
hand side of (6.65), which represents the expected marginal utility from
having children in the second period, involves two conflicting effects: A
higher level of child quality reduces the marginal utility from children and
also reduces the probability of divorce. Therefore, the marginal expected
utility can either rise or fall and the outcome depends on the shape of the
hazard associated with the distribution of quality match F (θ) . Specifically,
d 1 − δF (−δ ln Q − ln 2)
α[ ]
dQ Q
(δ 2 f (−δ ln Q − ln 2) − (1 − δF (−δ ln Q − ln 2))
= α[ ]
Q2
which is negative if.
f (−δ ln Q − ln 2)
δ2 < 1.
1 − F (−δ ln Q − ln 2)
This condition is satisfied, for instance, for the normal distribution if σ ≥ 1,
because
q then the hazard is an increasing function and its value at zero is
2 1
π σ.
Assuming that the expected marginal utility from children declines with
Q, it is easy to see that the investment in children is reduced in response
to increasing risk, represented here by a mean preserving increase in the
spread of the shocks to match quality. However, that the expected mar-
ginal utility from children declines with Q, the investment in children may
rise in order to stabilize the marriage. In either case, the "efficient" fam-
ily responds to such change in circumstances in a way which is optimal
for both spouses, without having equality in action. What is required, of
course, is for both partners to have equal interest in the outcome and, given
that all goods in marriage are public, such harmony can be achieved by a
binding commitments of an equal division of resources upon divorce. This
considerations can go part of the way in explaining the prevalence of equal
divisions following divorce. A binding commitment is required, because, ex
post, after divorce has occurred and the investments have been made, there
is no incentive for further transfers. However at the time of marriage, a
spouse with a higher income may be willing to commit on a post divorce
transfer in order to induce the seemingly less powerful spouse to invest in
children.
6.6 References
[1] Altonji, Joseph G., Hayashi, Fumio and Laurence J. Kotlikoff, ‘Is the
Extended Family Altruistically Linked? Direct Tests Using Micro Data’,
American Economic Review, 82 (1992), 1177-1198.
nal, 116 (2006), 558-580.
[4] Brossolet, C., ‘Fondement de la division du travail dans les modéles
économiques du ménage’, Paris, Editions Arguments, (1992).
[5] Bourguignon, Francois, Browning, Martin, Chiappori, Pierre-André and
Valérie Lechene, ‘Intra Household Allocation of Consumption: A Model
and Some Evidence from French Data’, Annales d’Economie et de
Statistique, 29 (1993), 137-156.
[6] Browning, Martin, ‘Love Betrayal and Commitment’, University of Ox-
ford, (2009).
[7] Browning, Martin and Annamaria Lusardi, ‘Household Saving: Mi-
cro Theories and Micro Facts’, Journal of Economic Literature, 34
(1996), 1797-1855.
[8] Bergstrom, Theodore C., ‘A Fresh Look at the Rotten Kid Theorem—
and Other Household Mysteries’, Journal of Political Economy, 97
(1989), 1138-1159.
[9] Borch, Karl, ‘The Safety Loading of Reinsurance Premiums’, Skandi-
navisk Aktuarietidskrift, 43 (1960), 163—184.
[10] Chiappori, Pierre-André, ‘Characterizing the Properties of the Gener-
alized Quasi-linear Utilities’, Columbia University, (2007).
[11] Chiappori, Pierre-André and Philip J. Reny, ‘Matching to Share risk’,
Columbia University, (2005).
[12] Chiappori, Pierre-André, Krislert Samphantharak, Sam Schulhofer-
Wohl and Robert M. Townsend, ‘Heterogeneity and Risk Sharing in Thai
Villages’, Working paper, Columbia University (2010).
[13] Chiappori, Pierre-André, Robert M. Townsend and Hiro Yamada,
‘Sharing wage risk’, Columbia University (2008).
[14] Chiappori, Pierre-André and Yoram Weiss, ‘Marriage Contracts and
Divorce: an Equilibrium Analysis’, Tel Aviv University, (2009).
[15] Chiappori, Pierre-André and Yoram Weiss, ‘Divorce, Remarriage, and

Welfare: A General Equilibrium Approach’, Journal of the European
Economic Association, 4 (2006), 415-426.
[16] Chiappori, Pierre-André and Yoram Weiss, ‘Divorce, Remarriage, and
Child Support’, Journal of Labor Economics, 25 (2007), 37-74.
[17] Chiappori, Pierre-André, Iyigun, Murat, and Yoram Weiss, ‘Invest-
ment in Schooling and the Marriage Market’, American Economic
Review, 99 (2009), 1689-1713.
[18] Chiappori, Pierre-André, Iyigun, Murat, and Yoram Weiss, ‘Public
Goods, Transferable Utility and Divorce Laws’, IZA Discussion Paper
No. 2646, (2007).
[19] Chiappori, Pierre-André, Iyigun, Murat, and Yoram Weiss, ‘Spousal
Matching, Marriage Contracts and Property Division in Divorce’, Tel
Aviv University, (2006).
[20] Clark, Simon, ‘Law, Property, and Marital Dissolution’, The Eco-
nomic Journal, 109 (1999), C41-C54.
[21] Coase, Ronald H., ‘The Problem of Social Cost’, Journal of Law
and Economics, 3 (1960), 1-44.
[22] Del Boca, Daniela and Christopher Flinn, ‘Endogeneous Household In-
teraction’, CHILD Working Papers wp08_09, CHILD - Centre for House-
hold, Income, Labour and Demographic economics - ITALY, (2009).
[23] Duflo, Esther and Christopher Udry, ‘Intrahousehold Resource Alloca-
tion in Cote d’Ivoire: Social Norms, Separate Accounts and Consumption
Choices’, NBER Working Paper No. 10498, (2004).
[24] Dufwenberg, Martin, ‘Marital investments, time consistency and emo-
tions’, Journal of Economic Behavior & Organization, 48 (2002),
57-69.
[25] Geanakoplos, John, Pearce, David and Ennio Stacchetti, ‘Psychologi-
cal games and sequential rationality’, Games and Economic Behav-
ior, 1 (1989), 60-79.
[26] Hardy, G. H., Littlewood, J. E., and G. Polya, Inequalities, (Cam-
bridge Mathematical Library, 1952).
[27] Konrad, Kai A. and Kjell Erik Lommerud, ‘The bargaining family
revisited’, Canadian Journal of Economics, 33 (2000), 471-487.
[28] Lundberg, Shelly and Robert A. Pollak, ‘Separate Spheres Bargaining
988-1010.
[29] Lundberg, Shelly and Robert A. Pollak, ‘Efficiency in Marriage’, Re-

view of Economics of the Household, 1 (2003), 153-167.
[30] Mazzocco, Maurizio, ‘Saving, Risk Sharing, and Preferences for Risk’,
American Economic Review, 94 (2004), 1169-1182.
[31] Mazzocco, Maurizio and Shiv Saini, ‘Testing Efficient Risk Sharing
with Heterogeneous Risk Preferences: Semi-parametric Tests with an
Application to Village Economies’, 2006 Meeting Papers 108, Society
for Economic Dynamics, (2006).
[32] Mazzocco, Maurizio, ‘Household Intertemporal Behaviour: A Collec-
tive Characterization and a Test of Commitment’, Review of Eco-
nomic Studies, 74 (2007), 857-895.
[33] Mincer, Jacob, ‘Family Migration Decisions’, Journal of Political
Economy, 86 (1978), 749-773.
[34] Rey, Patrick, and Bernard Salanie, ‘Long-term, Short-term and Rene-
gotiation: On the Value of Commitment in Contracting’, Economet-
rica, 58 (1990), 597-619.
[35] Townsend, Robert M, ‘Risk and Insurance in Village India’, Econo-
metrica, 62 (1994), 539-591.
ples: Evidence and Interpretation’, Journal of Labor Economics, 11
(1993), 629-679.
[37] Weiss, Yoram and Robert J. Willis, ‘Match Quality, New Information,
and Marital Dissolution’, Journal of Labor Economics, 15 (1997),
S293-S329.
[38] Wolfers, Justin, ‘Did Unilateral Divorce Laws Raise Divorce Rates?
A Reconciliation and New Results’, American Economic Review, 96
(2006), 1802-1820.
Part II
Equilibrium Models of the

Marriage Market
291
292
This is page 293
7
Matching on the Marriage
Market: Theory
Individuals in society have many potential partners. This situation creates
competition over the potential gains from marriage. In modern societies,
explicit price mechanisms are not observed. Nevertheless, the assignment
of partners and the sharing of the gains from marriage can be analyzed
within a market framework. The main insight of this approach is that the
decision to form and maintain a particular union depends on the whole
range of opportunities and not only on the merits of the specific match.
However, the absence of explicit prices raises important informational is-
sues. There are two main issues distinguishing the approaches used in the
matching literature. The first issue concerns the information structure and
the second relates to the extent of transferability of resources among agents
with different attributes. Specifically, models based on frictionless match-
ing assume that perfect and costless information about potential matches
is available to all participants; the resulting choices exclusively reflect the
interaction of individual preferences. Such models may belong to several
classes, depending on whether or not compensating transfers are allowed
to take place between individuals and, if so, at what ‘exchange rate’. Still,
they all rely on a specific equilibrium concept, namely stability. Formally,
we say that a matching is stable if:
(i) There is no married person who would rather be single.
(ii) There are no two (married or unmarried) persons who prefer to form
a new union.
The interest in stable marriage assignments arises from the presumption
that in a frictionless world, a marriage structure which fails to satisfy (i)
and (ii) either will not form or will not survive.
Models based on frictionless matching are studied in the next three Sec-
tions. An alternative approach emphasizes the role of frictions in the match-
ing process; in these models, based on search theory, information is limited
and it takes time to find a suitable match. The corresponding framework
will be discussed in the last Section.
294 7. Matching on the Marriage Market: Theory
7.1 Stable Matching without transfers: the

Gale-Shapley Algorithm
We begin our analysis of the marriage market assuming that there are no
frictions - that is, that each man and woman knows the potential gains
from marrying any potential mate. Marriage can be viewed as a voluntary
matching of males and females, allowing for the possibility of staying single.
We consider here only monogamic marriages so that each person can have at
most one spouse of the opposite sex. These assignments can be presented
by matrices with 0/1 entries depending upon whether or not male i is
married to female j. Since we consider only monogamic marriages, there is
at most one nonzero entry in each column and row. An illustration of such
a representation is shown in Example 7.1. In this example there are 4 men
and 3 women, where man 1 is married to woman 3, man 2 is married to
woman 1, man 3 is single and man 4 is married to woman 2:
Example 7.1
1 2 3
1 0 0 1
2 1 0 0
3 0 0 0
4 0 1 0
We first study matching when agents cannot make transfers between
each other. We thus assume that a marriage generates an outcome for each
partner that is fully determined by the individual traits of the partners;
this outcome cannot be modified by one partner compensating the other
for his or her deficient traits. Although somewhat extreme, this assumption
captures situations where, because of public goods and social norms that
regulate within family allocations, there is limited scope for transfers, so
that the success of a marriage mainly depends on the attributes of the
partners. However, an undesired marriage can be avoided or replaced by a
better one. Although there is no scope for trade within marriage, there is
margin for trade across couples.
Let there be a given, finite number of men, M , and a given, finite number
of women, N . We designate a particular man by i and a particular woman
by j. Assume that each man has a preference ranking over all women and
each woman has a preference ordering over all men. Such preferences can
be represented by a M ×N bi-matrix with a pair of utility payoffs, (uij , vij )
in each cell. For a given j, the entries vij describe the preference ordering
of woman j over all feasible males, i = 1, 2...M . Similarly, for a given i,
the entries uij describe the preference ordering of man i over all feasible
women j = 1, 2...N . We may incorporate the ranking of the single state
by adding a column and a row to the matrix, denoting the utility levels
of single men and women by ui0 and v0j , respectively. The preferences of
7. Matching on the Marriage Market: Theory 295
men and women are datum for the analysis. However, the representations
of these preferences by the utility payoffs are only unique up to monotone
transformations. An illustration of such a representation with 4 men and 3
women is shown in Example 7.2:
Example 7.2
1 2 3 0
1 u11 , v11 u12 , v12 u13 , v13 u10
2 u21 , v21 u22 , v22 u23 , v23 u20
3 u31 , v31 u32 , v32 u33 , v33 u30
4 u41 , v41 u42 , v42 u43 , v43 u40
0 v01 v02 v03
Gale and Shapley (1962) were the first to demonstrate that a stable
matching always exists, and suggested an algorithm which generates a sta-
ble outcome in a finite number of steps. For simplicity, we assume here that
all rankings are strict. To begin, let each man propose marriage to his most
favored woman. A woman rejects any offer which is worse than the single
state, and if she gets more than one offer she rejects all the dominated of-
fers; the non rejected proposal is put on hold (‘engagement’). At the second
step, each man who is not currently engaged proposes to the woman that
he prefers most among those women who have not rejected him. Women
will reject all dominated offers, including the ones on hold. This mechanism
is repeated until no male is rejected; then the process stops. Convergence
is ensured by the fact that no woman is approached more than once by the
same man; since the number of men and women is finite, this requirement
implies that the process will stop in finite time. The process must yield a
stable assignment because women can hold all previous offers. So if there
is some pair not married to each other it is only because either the man
did not propose (implying that he found a better mate or preferred staying
single) or that he did and was rejected (implying that the potential wife
had found a better mate or preferred staying single).
The stable assignment that is realized in the way just described need not
be unique. For instance, a different stable assignment may be obtained if
women make the offers and men can reject or hold them. Comparing these
stable assignments, it can be shown that if all men and women have strict
preferences, the stable matching obtained when men (women) make the
proposal is weakly preferred by all men (women). This remarkable result
shows that social norms of courting can have a large impact on matching
patterns (see Roth and Sotomayor 1990, ch. 2).
As an example, let there be 3 men and 3 women and consider the matrix
of utility payoff in Example 7.3 (setting the value of being single to zero
for all agents):
Example 7.3
Women
1 2 3
1 3, 2 2, 6 1, 1
Men
2 4, 3 7, 2 2, 4
3 1, 1 2, 1 0, 0
Note that, in this case, preferences diverge among men; man 1 ranks woman
1 above women 2 and 3, while men 2 and 3 both put woman 2 at the top of
their ranking. Similarly, there is disagreement among women; man 1 is the
most attractive match for woman 2, while women 1 and 3 both consider
man 2 as the best match. There is also a lack of reciprocity; man 1 would
rather marry woman 1 but, alas, she would rather marry man 2.
As a consequence there are two possible stable assignments, depending
on whether men or women move first. If men move first, man 1 proposes
to woman 1, and men 2 and 3 both propose to woman 2, who rejects man
3, but keeps man 2. In the second round, man 3 proposes to woman 1
who rejects him. In the last round, man 3 proposes to woman 3 and is not
rejected so that the procedure ends up with the outcome emphasized in
bold letters in the matrix below:
Women
1 2 3
1 3, 2 2, 6 1, 1
Men
2 4, 3 7, 2 2, 4
3 1, 1 2, 1 0, 0
One can check directly that this assignment is stable. Men 1 and 2 obtain
their best option and do not wish to change spouse, while man 3 cannot
find a better match who is willing to marry him.
Now, if women move first, woman 2 proposes to man 1 and women 1
and 3 both propose to man 2, who rejects woman 3, but keeps woman
1. In the second round, woman 3 proposes to man 1 who rejects her. In
the last round, woman 3 proposes to man 3 and is not rejected so that
the procedure ends up with the outcome emphasized in bold letters in the
matrix below:
Women
1 2 3
1 3, 2 2, 6 1, 1
Men
2 4, 3 7, 2 2, 4
3 1, 1 2, 1 0, 0
Again, one can check directly that this assignment is stable. Women 1 and
2 obtain their best option and do not wish to change spouse, while woman
3 cannot find a better match who is willing to marry her. It is seen that
the first assignment, in which men move first is better for all men (except
for man 3 who is indifferent) and the second assignment, in which women
move first is better for all women (except for woman 3 who is indifferent).
A very special case arises if women and men can be ranked by a single
male trait x and a single female trait y. This assumption introduces a
strong commonality in preferences, whereby all men agree on the ranking
of all women and vice versa. Specifically, let us rank males and females
by their marital endowment (that is, xi+1 > xi and yj+1 > yj ), and
let us assume that there exists a “household output function” h(xi , yj )
that specifies the marital output as a function of the attributes of the
two partners.1 This output is then consumed jointly as a public good, or
shared between the partners in some rigid fashion (equally for instance) in
all marriages. A natural question is: Who marries whom? Would a stable
assignment associate a male with a high marital endowment to a female
with high marital endowment (what is called positive assortative mating)?
Or, to the contrary, will a highly endowed male be matched with a low
endowment female (negative assortative mating)? The answer obviously
depends on the properties of the function h(x, y). It is easy to show that
if h(x, y) is strictly increasing in both traits, the unique stable assignment
is one with perfect positive assortative mating. To see that, suppose that
men propose first. In the first round, all men will propose to the woman
with the highest female attribute and she will reject all offers but the one
from the best man. In the second round, all remaining men will propose to
the second best woman and she will reject all but the second best man and
so on. The situation when women propose first is identical. Symmetrically,
if the male and female traits have opposing effects on output, the unique
stable assignment is one with perfect negative assortative mating.
An interesting extension arises when the relevant features of spouses are
not immediately revealed, which may cause a delay in marriage. Bergstrom
and Bagnoli (1993) consider a matching with asymmetric information in
a two period model. They assume that the female trait is immediately
revealed but the male trait is revealed later. The equilibrium that emerges
is such all women marry in the first period. Men that know their high
quality will delay their marriage and low quality males will marry early
but to low quality women. The more desirable females marry successful
older males. Thus, the model can explain the prevalent pattern if matching
by age, whereby the bride is typically younger than the groom.
In addition to the identification of stable assignments, one can use the
Gale-Shapley algorithm to obtain simple comparative static results. Allow-
1 This “household output function” should be distinguished from the standard house-
hold production function described in the previous sections, which take the attributes
of the spouses as fixed. Here we are interested in a reduced form that depends only on
attributes after all relevant activities have been chosen so as to achieve intrahousehold
efficiency.
ing for unequal numbers of men and women, it can be shown that a change
in the sex ratio has the anticipated effect. An increase in the number of
women increases the welfare of men and harms some women. The same
result holds in many to one assignment.
7.2 Stable Matching with transferable utilities: the

Becker-Shapley-Shubik model
7.2.1 The basic framework
The properties of the previous model heavily depend on the assumption
that transfers are impossible, so that a person cannot ‘compensate’ a po-
tential partner for marrying him or her despite some negative traits. In
practice, this assumption is hard to maintain. Whenever one commodity
at least is privately consumed, a spouse can reduce her private consump-
tion to the partner’s benefit, which de facto implements a compensation.
We now consider the opposite polar case in which not only transfers are
feasible, but there is a medium of exchange that allows partners to transfer
resources between them at a fixed rate of exchange; that is, we assume that
utilities are transferable (see Chapter 3).
Instead of introducing two exogenous matrices u = (uij ) and v = (vij )
as in the case of non-transferable utility, we now consider a unique output
matrix with entries ζ ij which specifies the total output of each marriage.
Given the assumption of transferable utility, this total output can be di-
vided between the two partners. We denote the utility payoff of the husband
by uij and the utility payoff of the wife by vij . Thus, by definition, if i and
j form a match we have
uij + vij = ζ ij (7.1)
Note, however, the key difference with the previous section with no trans-
fers: while matrices u and v were then given (as part of the statement of
the problem), they are now endogenous (and part of its solution, since they
are determined at equilibrium - or, here, at the stable matching).
As before, we are ¡interested
¢ only in stable matching. The question is: for
a given matrix ζ = ζ ij , which are the stable assignments, and what are
the corresponding allocations of output (or imputations) within each mar-
riage. Note that the question is, in a sense, more difficult than in the case
with no transfers, since the distribution of output between members is now
endogenous and has to be determined in equilibrium. Still, it is relatively
easy to apply the criteria for stability in the case of transferable utility.
Specifically, one can show that a stable assignment must maximize total
output over all possible assignments. It is this simple and powerful result
that makes the assumption of transferable utility attractive in matching
models.
Two examples
To understand this result, consider first the simplest possible case: let there
be two people of each sex. Assuming that marriage dominates the single
state (that is if any two individuals remain unattached they can gain by
forming a union), there are two possible assignments: Man 1 marries woman
1 and man 2 marries woman 2, or man 1 is married to woman 2 and man
2 is married to woman 1. In testing for stability we treat the potential
marital outputs ζ ij as given and the divisions uij and vij as variables.
Suppose, now, that the assignment in which man 1 marries woman 2 and
man 2 marries woman 1 (the off diagonal assignment) is stable. Then, the
following inequalities must hold:
u12 + v21 ≥ ζ 11 (7.2)

u21 + v12 ≥ ζ 22 (7.3)
If the first inequality fails to hold then male 1 and female 1, who are
currently not married to each other, can form a union with a division of
utilities which will improve upon their current situations, defined by u12
and v21 . If the second inequality does not hold then man 2 and woman 2,
who are presently not married to each other, can form a union and divide
utilities so as to improve over the current values u21 and v12 . From equation
1 we have ζ 12 = u12 +v12 and ζ 21 = u21 +v21 so that equation (7.2) can
be rewritten as
ζ 12 − v12 + ζ 21 − u21 ≥ ζ 11 . (7.4)
Adding conditions (7.4) and (7.3) we obtain
ζ 12 + ζ 21 ≥ ζ 11 + ζ 22 (7.5)
By a similar argument, an assignment along the main diagonal will be
stable only if (7.5) is reversed. Condition (7.5) is not only necessary but also
sufficient for stability of the off diagonal assignment. For if it is satisfied we
can find values of u and v such that (7.2) and (7.3) hold. Such imputations
support the stability of the assignment since it is then impossible for both
partners to gain from reassignment.
To illustrate the implications of the transferable utility assumption and
the implied maximization of aggregate marital output, let us consider a
second example. There are 3 men and 3 women and consider the matrix of
marital outputs below:
Example 7.4
Women
1 2 3
1 5 8 2
Men
2 7 9 6
3 2 3 0
Notice that the entries in this matrix are just the sums of the two terms in
Example 7.3 discussed above. In this regard, non transferable utility can
be thought of as a special case of transferable utility, where the division
of the output in each marriage is predetermined and cannot be modified
by transfers between spouses. For instance, if each partner receives half of
the marital output in any potential marriage, the Gale Shapley algorithm
yields the unique stable outcome, which is on the diagonal of this matrix. In
contrast, with transferable utility, the unique assignment that maximizes
aggregate marital output, indicated by the bold numbers in the matrix
below, is not on the diagonal. This assignment yields aggregate output of
16, compared with an aggregate output of 14 on the diagonal.
Women
1 2 3
1 5 8 2
Men
2 7 9 6
3 2 3 0
Though all men would obtain the highest marital output with woman 2,
and all women would obtain the highest output with man 2 (implying that
ζ 22 is the largest entry in the marital output matrix 7.4), the best man
and the best woman are not married to each other. With transfers, the
assignment on the diagonal is no longer stable, because if couple 1, 1 and
couple 2, 2 exchange partners, there is an aggregate gain of 1 unit of the
transferable good. Then man 1 can, despite his lower contribution to the
marital output, bid away the best woman by offering her a larger amount
of private consumption and still be better off than in the initial match with
woman 1. Similarly, woman 1 can bid away the best man by offering him a
larger share of private consumption and still be better off than in the initial
match with man 1. The higher aggregate output achievable when man 2
and woman 2 are not married to each other implies that, for any division
of the marital output of 9 that these partners can obtain together, at least
one of the partners can be made better off in an alternative marriage.
Stable matching with a finite number of agents

Let us now consider the general assignment problem with M males and N
females. Let ζ ij denote the total output of a marriage between male i and
female j, and let ζ i0 (resp. ζ 0j ) be the utility that person i (resp. person
j) receives as single (with ζ 00 = 0 by notational convention). Then the
difference zij = ζ ij − ζ i0 − ζ 0j is the marital surplus that male i and female
j generate by marrying each other.
We define assignment indicators, aij , such that aij = 1 if and only if
i is married with j and aij = 0 otherwise. We also define ai0 = 1 if and
only if i is single, and similarly a0j = 1 if and only if j is single. Then,
following Gale (1960, chapters 1 and 5) and Shapley and Shubik (1972),
we may describe the stable assignment as a solution to an integer linear

programming problem:
M X
X N
max aij ζ ij (7.6)
aij
i=0 j=0
subject to aij ≥ 0 and

N
X
aij = 1, i = 1, 2., .M , (7.7)
j=0
M
X
aij = 1, j = 1, 2., .N. (7.8)
i=0
M
X N
X
A first remark is that since a0j = 1 − aij and ai0 = 1 − aij the
i=1 j=1
program can be rewritten as
M X
X N M X
X N
¡ ¢
max aij ζ ij − ζ i0 − ζ 0j + C = max aij zij + C (7.9)
aij aij
i=1 j=1 i=1 j=1
subject to
N
X
aij ≤ 1, i = 1, 2., .M, (7.10)
j=1
M
X
aij ≤ 1, .j = 1, 2., .N, (7.11)
i=1
M
X N
X
where C = ζ i0 + ζ 0j is the aggregate utility of singles. Therefore,
i=1 j=1
the maximization of aggregate marital output over all possible assignments
is equivalent to the maximization of aggregate surplus and, without loss of
generality, we can normalize the individual utilities by setting ζ i0 = ζ 0j = 0
for all i and j.
Secondly, one can actually assume that in the problem above, the aij
can be real numbers in the (M − 1)-dimensional simplex (instead of con-
straining them to be integers). Intuitively, aij can then be interpreted as
the probability that Mr. i marries Mrs. j. Note, however, that given the
linearity of the structure, one solution at least to this generalized problem
is anyway attained with all aij being either zero or one.
The basic remark, at that point, is that the program thus defined is a
standard, linear programming problem; that is, we want to find a vector
(aij ) that maximizes the linear objective (7.6) (or (7.9)) subject to the lin-
ear constraints (7.7) and (7.8) (resp. (7.10) and (7.11)). We can therefore
use the standard tools of linear programming - specifically, duality theory.
Associated with the maximization of aggregate surplus which determines
the assignment is a dual cost minimization problem that determines the set
of possible divisions of the surplus. Specifically, one can define a dual vari-
able ui for each constraint (7.10) and a dual variable vj for each constraint
(7.11); the dual program is then:
M
X N
X
min( ui + vj ) (7.12)
ui ,vj i=1 j=1
subject to
ui + vj ≥ zij , i ∈ {1, ..., M }, j ∈ {1, ..., N } (7.13)

ui ≥ 0, vj ≥ 0.
The optimal values of ui and vj can be interpreted as shadow prices of

the constraints in the original maximization problem (the primal). Thus,
ui +vj = zij if a marriage is formed and ui +vj ≥ zij otherwise.2 This result
is referred in the literature as the complementarity slackness condition, see
for instance Gale (1978). It has a very simple interpretation. Any man i is
a resource that be can allocated to any woman, but only one woman, in
society. Similarly woman j is a resource that can be allocated to any man
in society, but only one man. The shadow price of each constraint in (7.10)
describes the social cost of moving a particular man (woman) away from
the pool of singles, where he (she) is a potential match for others. The sum
of these costs ui + vj is the social cost of removing man i and woman j
from the pool while zij is the social gain. Thus if ui + vj > zij , the costs
exceed the gains and the particular marriage would not form. However, if
a marriage is formed then ui + vj = zij and each person’s share in the
resulting surplus equals their opportunity cost in alternative matches.
The crucial implication of all this is that the shadow price ui is simply
the share of the surplus that Mr.i will receive at the stable matching (and
similarly for vj ); consequently, conditions (7.13) are nothing else than the
stability conditions, stating that if i and j are not matched at the stable
matching, then it must be the case that the surplus they would generate
if matched together (that is zij ) is not sufficient to increase both utilities
above their current level!
These results have a nice interpretation in terms of decentralization of
the stable matching. Indeed, a stable assignment can be supported (imple-
mented) by a reservation utility vector, whereby male i enters the market
2 Conversely, a
ij can be seen as the dual variable for constraint (7.13). In particular,
if aij > 0, then the constraint must be binding, implying that ui + vj = zij .
with a reservation utility ui and is selected by the woman that gains the
highest surplus zij − ui from marrying him. Similarly, woman j enters with
a reservation utility vj and is selected by the man who has the highest gain
zij − vj from marrying her. In equilibrium, each agent receives a share in
marital surplus that equals his\her reservation utility. In a sense, ui and vj
can be thought of as the ‘price’ that must be paid to marry Mr. i or Mrs.
j; each agent maximizes his/her welfare taking as given this ‘price’ vector.
It is important to note that the informational requirements for imple-
menting a stable assignment with transferable utility is quite different than
for the Gale-Shapley no transfer case. For the latter, we only require that
each person can rank the members of the opposite sex. With transferable
utility, the planner needs to know the surplus values of all possible matches
and agents should each know the share of the surplus that they would
receive with any potential spouse.
In general, there is a whole set of values for ui , vj that support a stable
assignment. While the issues related to the distribution of surplus will be
discussed in the next Chapter, we present in the table below three (of
many) such imputations, denoted by a, b and c, for the stable assignment
in example 7.4.
Imputation a b c
W M W M W M
v1 2 u1 3 v1 2 u1 4 v1 1 u1 5
Individual shares
v2 5 u2 5 v2 4 u2 5.5 v2 3 u2 6
v3 1 u3 0 v3 0.5 u3 0 v3 0 u3 1
The reader can readily check that each of these imputations supports a
stable match.
Extension: continuum of agents

Finally, although the previous argument is presented in a finite setting, it
is fully general, and applies to continuous models as well. From a general
perspective, we only need that the set of men and the set of women, de-
noted X and Y , be complete, separable metric spaces equipped with Borel
probability measures F and G; note that no restriction is imposed on the
dimension of these spaces (it may even be infinite). The surplus function
h (x, y) is only assumed to be upper semi-continuous. The problem can be
stated as follows: find a measure Φ on X × Y such that:
• The marginals of Φ on X and Y are F and G, respectively.
R
• The measure Φ solves maxΦ X×Y h (x, y) dΦ (x, y), where the max is
taken over the set of measures satisfying the previous conditions.
A complete analysis of this problem is outside the scope of this book; the
reader is referred to Chiappori, McCann and Neishem (2010) or Ekeland
(2010) for recent presentations. Let us just mention that the existence of
a stable match obtains in general; this comes from the fact that the linear
optimization problem does have a solution under very general assumptions.
7.2.2 Assortative mating

The basic result
Suppose now that each male is endowed with a single characteristic, x,
and each female is endowed with a single characteristic, y, which positively
affects the family’s output. When can we expect the stable assignments to
exhibit either positive or negative assortative mating? Again, the answer
is quite different from the no transfer case. It follows in the present case
from the observation that a stable assignment must maximize the aggregate
marital output (or surplus) over all possible assignments.
Specifically, let, as above,
ζ ij = h(xi , yj ) (7.14)
be the household output function that specifies the marital output as a
function of the attributes of the two partners. We say that a function
h(xi , yj ) is super modular if x0 > x and y 0 > y always imply that
h(x0 , y 0 ) + h(x, y) ≥ h(x0 , y) + h(x, y 0 ), (7.15)
and it is sub modular if inequality (7.15) is always reversed. This definition
captures the idea of complementarity and substitution as usually under-
stood. Rewriting (7.15) in the form
h(x0 , y 0 ) − h(x0 , y) ≥ h(x, y 0 ) − h(x, y), (7.16)
we see that the requirement is that the contribution to marital output of
a given increase in the female attribute rises with the level at which the
male trait is held fixed. By a similar rearrangement, the impact of a given
increase in the male’s attribute rises in the female’s attribute. Note also
that if h is twice differentiable then h is super (sub) modular if the second
cross derivative hyx is always positive (negative).3 The condition that hyx
3 Indeed, for any given (x, y) define

H(x0 , y0 ) = h(x0 , y 0 ) + h(x, y) − h(x0 , y) − h(x, y0 ).
Then
Hx0 (x0 , y0 ) = hx0 (x0 , y0 ) − hx0 (x0 , y)
which is positive for y0 > y if hxy ≥ 0 (since hx is then increasing in f ). Similarly,
Hy0 (x0 , y0 ) = hy0 (x0 , y0 ) − hy0 (x, y0 ) ≥ 0
for x0 > x if hxy ≥ 0. Hence H is increasing in its arguments, and H(x, y) = 0; we
conclude that H(x0 , y 0 ) ≥ 0 whenever x0 > x and y0 > y and hxy ≥ 0.
is monotonic is sometimes called the single crossing or the Spence-Mirrlees

condition; indeed, a similar condition is crucial in contract theory, signalling
models (a la Spence) and optimal taxation (a la Mirrlees).
The basic result is that complementarity (substitution) in traits must
lead to a positive (negative) assortative mating; otherwise aggregate out-
put is not maximized. Assuming that h(x, y) is increasing in x and y, we
obtain that in the case of positive assortative mating, the best man marries
the best woman, and if there are more women than men the women with
low female quality remain single. If there is negative assortative mating,
the best man marries the worst woman among the married women but if
there are more women than men, it is the women with the lower female
attributes who remain single. (see Appendix) In other words, who mar-
ries whom depends on second order derivatives of h(x, y) but who remains
single depends on the first order derivatives of h(x, y). If there is no inter-
action in traits and the marginal contribution of each agent is the same
in all marriages, any assignment is (weakly) stable and it does not matter
who marries whom, because whichever way we arrange the marriages the
aggregate output of all marriages remains the same.
We may explain these results intuitively by referring again to the basic
idea of a stable assignment. Complementarity (substitution) implies that
males with high x will be willing to pay marginally more (less) for the
female attribute. Thus, if x stands for money and y stands for beauty,
the wealthy men will be matched with the pretty women if and only if
their (marginal) willingness to pay for beauty is higher. If there is negative
interaction between money and beauty, the most wealthy man will not
marry the most pretty woman, because whichever way they divide their
gains from marriage, either he is bid away by a less pretty woman or she
is bid away by a poorer man.
This result is in a sharp contrast to the non transferable case, where
monotonicity in traits is sufficient to determine the outcome.4 The conse-
quence is that assortative (negative or positive) mating is more prevalent
in the absence of transfers, because it is impossible for agents with less
desirable traits to compensate their spouses through a larger share of the
marital output (see Becker, 1991, ch. 4 and Becker and Murphy, ch. 12).
The sad message for the econometrician is that, based on the same informa-
tion, namely the household production function, one can get very different
outcomes depending on the ability to compensate within households, a fea-
ture that we usually cannot directly observe. But, conversely, it also means
that one can in principle test one model against the other (since they have
4 However, monotonicity may fail to hold when super modularity holds. A potentially
important case is when preferences are single peaked in the attribute of the spouse.
In such cases, we can have assortative mating in the sense that married partners have
similar traits, but individuals with extreme traits may fail to marry. The interested
reader may consider the case in which the marital surplus is given by g − (x − y)2 .
different implications); we will discuss such tests later on.

Finally, the impact of traits on the value of being single does not affect
these considerations, because the welfare of each person as single depends
only on his own traits. Therefore, in the aggregate, the output that indi-
viduals obtain as singles is independent of the assignment. Although the
value of being single does matter to the question who marries, it does not
affect who marries whom, in equilibrium.
Examples
In many models, the surplus function takes a specific form. Namely, the two
traits x and y can often be interpreted as the spouses’ respective incomes.
Following the collective approach described in the previous Chapters, we
may assume that a couple consisting of a husband with income x and a wife
with income y will make Pareto efficient decisions; then it behaves as if it
was maximizing a weighted sum of individual utilities, subject to a budget
constraint. The important remark is that the constraint only depends on
the sum of individual incomes. Then the Pareto frontier - or in our specific
case the value of the surplus function h (x, y) which defines it - only depends
on the sum (x + y);5 that is:
h (x, y) = h̄ (x + y)
The various properties described above take a particular form in this con-
text. For instance, the second cross derivative hxy is here equal to the
second derivative h̄00 . It follows that we have assortative matching if h̄ is
convex, and negative assortative matching if h̄ is concave. The interpreta-
tion is as above: a convex h̄ means that an additional dollar in income is
more profitable for wealthier people - meaning that wealthier husbands are
willing to bid more aggressively for a rich wife than their poorer competi-
tors. Conversely, if h̄ is concave then the marginal dollar has more value
for poorer husbands, who will outbid the richer ones.
In models of this type, the TU assumption actually tends to generate
convex output functions, hence assortative matching. To see why, consider
a simple model of transferable utility in the presence of a public good.
Preferences take the form
ui = ci g(q) + fi (q), (7.17)
where c and q denote private and public consumption, respectively. The
Pareto frontier is then
ua + ub = h(Y ) = max[(Y − q)g(q) + f (q)],
q
5 Of course, while the Pareto set only depends on total income, the location of the
point ultimately chosen on the Pareto frontier depends on individual incomes - or more
specifically on the location of each spouse’s income within the corresponding income
distribution. These issues will be analyzed in the next Chapter.
where f (q) = fa (q) + fb (q) and Y = x + y. By the envelope theorem,

h0 (Y ) = g(q) and therefore,
dq −(g 0 (q))2
h00 (Y ) = g 0 (q) = > 0.
dY (y − q)g (q) − 2g 0 (q) + f 00 (q)
00
since the denominator must be negative for the second order conditions
for a maximum to hold. Hence, if there is an interior solution for q, the
household production function is convex in family income, Y , implying
that the two incomes x and y must be complements.
As an illustration, recall the examples discussed in sections 2.1 and 2.2 of
Chapter 2. In section 2.1, we considered the case in which the spouses pool
their (fixed) incomes and share a public good and individual preferences
were of the form ui = ci .q, compatible with (7.17). If we now rank men
and women by their incomes we have a situation in which the household
2
production function is h(x, y) = (x+y)4 . This is a convex function of total
income; there is a positive interaction everywhere, leading to assortative
sorting.
In contrast, in section 2.2, we considered a case in which division of labor
has led to marital output given by max(wi , wj ), which is not a function
of total income. Here, we obtain negative assortative mating. This holds
because a high wage person is more useful to a low wage person, as indicated
by the submodularity of h(x, y) = max(x, y).6 For instance, if man i has
wage i and woman j has wage j, the output matrix for the 3 by 3 case is:
Example 7.5
Women
1 2 3
1 1 2 3
Men
2 2 2 3
3 3 3 3
implying three stable assignments; the opposite diagonal (in bold), one
close to it in which couples (1, 3) and (2, 2) exchange partners (emphasized),
and a symmetric one in which couples (3, 1) and (2, 2) exchange partners.
The assignment also depends on the location of the wage distribution for
each gender. As an extreme case, let the worst woman have a higher wage
6 For all x0 ≥ x and y0 ≥ y, we have max (x0 , y 0 ) + max (x, y) ≤ max (x0 , y) +
max (x, y0 ) .Going over the six possible orders of four numbers x,x0 , y, y0 satisfying x0 ≥ x
and y0 ≥ y , we see that
x0 ≥ x ≥ y 0 ≥ y ⇒ x0 + x ≤ x0 + x,
x0 ≥ y0 ≥ x ≥ f ⇒ x0 + x ≤ x0 + y0 ,
x0 ≥ y0 ≥ y ≥ x ⇒ x0 + y ≤ x0 + y0 ,
y 0 ≥ y ≥ x0 ≥ x ⇒ y0 + y ≤ y + y0 ,
y 0 ≥ x0 ≥ y ≥ x ⇒ y0 + y ≤ x0 + y0 ,
y 0 ≥ x0 ≥ x ≥ y ⇒ y0 + x ≤ x0 + y0 .
than the best man. Then in all marriages the female wage determines the
outcome and all assignments are equally good.
Note, finally, that in the absence of any interaction, we have h(x, y) =
x + y; this describes a situation where the two spouses simply pool their in-
comes and consume only private goods. Since the output is a linear function
of both incomes, any assignment of men to women is stable. It is interest-
ing that although the assignment is completely indeterminate, the set of
imputations shrinks substantially and is given by
vi = xi + p,
uj = yj − p, (7.18)
for some fixed p. Thus, in the absence of interaction in traits, the same
transfer p occurs in all marriages and we may interpret it as a common
bride price or dowry, depending on whether p is positive or negative in
equilibrium.7 As we shall show in the next Chapter, if there is interaction
in traits, this single price is replaced by an intrahousehold allocation rule
that depends on the attributes of both partners.
7.2.3 Matching with a continuum of agents

The discussion above shows that a crucial feature of the problem is the in-
teraction in the traits that the two partners bring into marriage. We shall
for the time being focus here on situations where income is the only mar-
ital trait and individual incomes are complement in the household output
function - that is, h(x, y) is super modular, or hxy (x, y) > 0. Moreover, we
assume here that there exists a continuum of men, with a total mass nor-
malized to 1, and a continuum of women, with a total mass denoted r. We
allow different income distributions for men and women; specifically, male
7 Consider any two couples, (i, j) and (r, s), in a stable assignment. Then, using the
duality results,
ui + vj = xi + yj
ur + vs = xr + ys ,
because the imputations for married couples exhaust the marital output. Also, because
couples (i, s) and (r, j) are not married to each other
ui + vs ≥ xi + ys
ur + vj ≥ xr + yj
But none of these inequalities can be strict, because their sum must equal to the sum of
the equalities above. It then follows that in all marriages on any stable assignment
ui − ur = xi − xr
vj − vs = yj − ys ,
which is equivalent to (7.18).
incomes x are distributed on [0, 1] according to some distribution F and

female incomes y are distributed on [0, 1] according to some distribution
G.
The assumed positive interaction implies a positive assortative matching.
Therefore, if a man with income x is married to a woman with income y,
then the set of men with incomes above x must have the same measure as
the set of women with incomes above y. Thus, for all x and y in the set of
married couples,
1 − F (x) = r (1 − G (y)) . (7.19)
Hence,
x = Φ [1 − r (1 − G (y))] =φ
¯ (y) , (7.20)
−1
where Φ = F , or equivalently,
∙ ¸
1
y = Ψ 1 − (1 − F (x)) =ψ
¯ (x) , (7.21)
r
where Ψ = G−1 and ψ = φ−1 ; note that both φ and ψ are increasing.
All men and women are married if there is an equal measure of men
and women, r = 1. All women are married if there is scarcity of women,
r < 1, implying that men with income x less than x0 = Φ(1 − r) remain
single. All men are married if there is scarcity of men, r > 1, implying that
women with income y less than y0 = Ψ (1 − 1/r) remain single. If r > 1,
then the function y = ψ (x) determines the income of the wife for each man
with income x in the interval [0, 1]. Similarly, if r < 1, then the function
x = φ (y) determines the husband’s income of each woman with income
y in the interval [0, 1]. We shall refer to these functions as the matching
functions and to the resulting assignment as the assignment profile.
In Figure 7.1 we show the matching function ψ (x) for the case in which x
is distributed uniformly on [0, 1], y is distributed uniformly on [0, σ], σ < 1
and r > 1. Applying (7.19) and solving
y
1 − x = r(1 − ),
σ
we obtain
σ
ψ (x) = (r − 1 + x).
r
We see that women with incomes y such that y ≤ y0 = σr (r − 1) remain
single. Women with incomes in the range [y0 , y 0 ] = [ σr (r − 1, σr (r − 1 + x0 )]
marry men with incomes in the range [0, x0]. Finally, women with incomes
in the range [y 00 , σ] = [ σr (r−1+x00 ), σ] marry men with incomes in the range
[x00 , 1]. Thus women with higher incomes marry men with higher incomes.
Note the equality in the measures of women and men in these intervals,
as indicated by the areas of the corresponding rectangles. For instance the
rectangular with base x0 and height 1 has the same area as the rectangular
FIGURE 7.1. Positive Assortative Mating

with base σr and height σr (r − 1 + x0 ) − σr (r − 1). Such equality of measures

must hold throughout the assignment profile.
The slope of each matching function is related to the local scarcity of
men relative to women. Men are locally scarce if there are more women
than men at the assigned incomes (φ (y) , y) = (x, ψ (x)). Or, equivalently,
if an increase in the husband’s income is associated with a smaller increase
in the income of the matched wife. That is,
dx g(y)
= φ0 (y) = r > 1, (7.22)
dy f (φ (y))
dy 1 f (x)
= ψ 0 (x) = < 1.
dx r g(ψ (x))
Men are locally abundant if these inequalities are reversed.
7.2.4 Multidimensional matching

The previous discussion explicitly refers to a one-dimensional framework.
Assortative matching is harder to define when several dimensions (or several
traits) are involved; moreover, conditions like supermodularity or single-
crossing do not have an obvious extension to a multidimensional setting.
Still, they can be generalized; again, the reader is referred to Chiappori,
McCann and Neishem (2010) or Ekeland (2010) for recent presentations.
The main insights can briefly be described as follows. Assume that X and
Y are finite dimensional. Then:
1. The Spence-Mirrlees condition generalizes as follows: if ∂ x h (x0 , y)

denotes the superdifferential of h in x at (x0 , y), then for almost all
x0 , ∂ x h (x0 , y1 ) is disjoint from ∂ x s(x0 , y2 ) for all y1 6= y2 in Y . This
is the ‘twisted buyer’ condition in Chiappori, McCann and Neishem
2010.
2. If the ‘twisted buyer’ condition is satisfied, then the optimal match
is unique; in addition, it is pure, in the sense that the support of the
optimal measure Φ is born by the graph of some function y = φ (x);
that is, for any x there exists exactly one y such that x is matched
with y with probability one.
3. There exists a relaxation of the ‘twisted buyer’ condition (called the
‘semi-twist’) that guarantees uniqueness but not purity.
The notion of ‘superdifferential’ generalizes the standard idea of a linear

tangent subspace to non differentiable functions. If h is differentiable, as
is the case is most economic applications, then ∂ x h (x0 , y) is simply the
linear tangent (in x) subspace to h at (x0 , y), and the condition states that
for almost all x0 , there exists a one to one correspondence between y and
∂ x h (x0 , y). Note that if X and Y are one-dimensional, then ∂ x h (x0 , y)

is fully defined by the partial ∂h/∂x (x0 , y), and the condition simply re-
quires that ∂h/∂x (x0 , y) be strictly monotonic in y - that is, the sign of
∂ 2 h/∂x∂y be constant, the standard single-crossing condition, Similarly, if
X and Y are one-dimensional, then purity imposes a one-to-one match-
ing relationship between x and y; if this matching is continuous, it has
to be monotonic, that is matching must be either positive or negative as-
sortative (in that sense, purity is a generalization of assortativeness to
multi-dimensional settings).
In general, purity rules out situations in which a subset of agents (with
a positive measure) randomize between several, equivalent matches. Such
situations may be frequent in practice; in particular, Chiappori, McCann
and Neishem (2010) show that they are likely to occur when agents are
located on an Hotelling-type circle. Finally, only recently have empirical
models of multidimensional matching been developed; the main reference,
here, is Galichon and Salanié (2009).
7.3 Matching with general utilities

In the previous two sections, the matching process is studied in specific and
somewhat extreme settings: either transfers cannot take place at all, or they
can be made at a constant exchange rate (so that reducing a member’s
utility by one ‘unit’ always increases the spouse’s utility by one unit as
well). We now consider the general case, in which although transfers are
feasible, there is no commodity that allows the partners to transfer utilities
at a fixed rate of exchange. Then the utility frontier is no longer linear and
it is impossible to summarize the marital output from a match by a single
number. In this more general framework, stability is defined in the same
manner as before, that is, an assignment is stable if no pair who is currently
not married can marry and choose an allocation of family resources that
yields a result which is better for both of them than under the existing
assignment and associated payoffs. Observe that the assignment and payoffs
are simultaneously restricted by this definition. However, it is no longer
true that aggregate marital output must be maximized - actually, such an
‘aggregate output’ is not even defined in that case. Mathematically, the
matching model is no longer equivalent to an optimization problem.
Still, it is in principle possible to simultaneously solve for the stable
assignment and the associated distribution(s) of surplus. The interested
reader is referred to Roth and Sotomayer (1990, ch. 6), Crawford (1991),
Chiappori and Reny (2006) and Legros and Newman (2007). To give a quick
idea of how the general problem can be approached with a continuum of
agents, let us assume, as above, that each agent is characterized by one
trait, and let’s assume that this trait is income (assumed to be exogenous).
Male income is denoted by x and female income is denoted by y. We no

longer assume transferable utility; hence the Pareto frontier for a couple
has the general form
u = H(x, y, v) (7.23)
with H(0, 0, v) = 0 for all v.
As above, if a man with income x remains single, his utility is given
by H(x, 0, 0) and if a woman of income y remains single her utility is
the solution to the equation H(0, y, v) = 0. By definition, H(x, y, v) is
decreasing in v; we assume that it is increasing in x and y, that is that a
higher income, be it male’s or female’s, tends to expand the Pareto frontier.
Also, we still consider a continuum of men, whose incomes x are distributed
on [0, 1] according to some distribution F , and a continuum of women,
whose incomes y are distributed on [0, 1] according to some distribution G;
let r denote the measure of women.
Finally, let us assume for the moment that an equilibrium matching ex-
ists and that it is assortative. Existence can be proved under mild conditions
using a variant of the Gale-Shapley algorithm; see Crawford (1991), Chiap-
pori and Reny (2006). Regarding assortativeness, necessary conditions will
be derived below. Under assortative matching, the ‘matching functions’ φ
and ψ are defined exactly as above (eq. 7.19 to 7.21).
Let u (x) (resp. v (y)) denote the utility level reached by Mr. x (Mrs. y)
at the stable assignment. Then it must be the case that
u (x) ≥ H (x, y, v (y))
for all y, with an equality for y = ψ (x). As above, this equation simply
translates stability: if it was violated for some x and y, a marriage between
these two persons would allow to strictly increase both utilities. Hence:
u (x) = max H(x, y, v (y))

y
and we know that the maximum is actually reached for y = ψ (x). First
order conditions imply that
∂H ∂H
(φ (y) , y, v (y)) + v 0 (y) (φ (y) , y, v (y)) = 0. (7.24)
∂y ∂v
while second order conditions for maximization are
µ ¶
∂ ∂H 0 ∂H
(φ (y) , y, v (y)) + v (y) (φ (y) , y, v (y)) ≤ 0 ∀y. (7.25)
∂y ∂y ∂v
This expression may be quite difficult to exploit. Fortunately, it can be

simplified using a standard trick. The first order condition can be written
as:
F (y, φ (y)) = 0 ∀y
where
∂H ∂H
F (y, x) = (x, y, v (y)) + v 0 (y) (x, y, v (y)) . (7.26)
∂y ∂v
Differentiating:
∂F ∂F 0
+ φ (y) = 0 ∀y,
∂y ∂x
which implies that
∂F ∂F 0
≤ 0 if and only if φ (y) ≥ 0.
∂y ∂x
The second order conditions can hence be written as:
µ 2 ¶
∂ H ∂2H
0
(φ (y) , y, v (y)) + v (y) (φ (y) , y, v (y)) φ0 (y) ≥ 0 ∀y.
∂x∂y ∂x∂v
(7.27)
Here, assortative matching is equivalent to φ0 (y) ≥ 0; this holds if
∂2H ∂2H
(φ (y) , y, v (y)) + v 0 (y) (φ (y) , y, v (y)) ≥ 0 ∀y. (7.28)
∂x∂y ∂x∂v
Since v0 (y) ≥ 0, a sufficient (although obviously not necessary) condition

is that
∂2H ∂2H
(φ (y) , y, v (y)) ≥ 0 and (φ (y) , y, v (y)) ≥ 0. (7.29)
∂x∂y ∂x∂v
One can readily see how this generalizes the transferable utility case. In-
∂2H
deed, TU implies that H (x, y, v (y)) = h (x, y) − v (y). Then ∂x∂v = 0 and
2
∂ H ∂2h
the condition boils down to the standard requirement that ∂x∂y = ∂x∂y ≥
0. General utilities introduces the additional requirement that the cross
∂2H
derivative ∂x∂v should also be positive (or at least ‘not too negative’).
Geometrically, take some point on the Pareto frontier, corresponding to
some female utility v, and increase x - which, by assumption, expands the
Pareto set, hence shifts the frontier to the North East (see Figure 7.2). The
condition then means that at the point corresponding to the same value
v on the new frontier, the slope is less steep than at the initial point. For
instance, a homothetic expansion of the Pareto set will typically satisfy this
requirement.
The intuition is that whether matching is assortative depends not only on

the way total surplus changes with individual traits (namely, the usual idea
that the marginal contribution of the husband’s income increases with the
∂2H
wife’s income, a property that is captured by the condition ∂x∂y ≥ 0), but
also on how the ‘compensation technology’ works at various income levels.
P’
FIGURE 7.2. The slope condition
With general utilities, while the technology for transferring income remains
obviously linear, the cost (in terms of husband’s utility) of transferring
utility to the wife varies with incomes. The second condition implies that,
keeping the wife’s utility level fixed, a larger income alleviates the cost
(in terms of husband’s utility) of providing an additional unit of utility
to the wife. Then wealthy males have a double motivation for bidding
aggressively for wealthy women: they benefit more from winning, and their
‘bidding costs’ are lower. They will thus systematically win. Note, however,
∂2H ∂2H
that when the two partials ∂x∂y and ∂x∂v have opposite signs, the two
aspects - benefits from winning and cost of bidding - vary with income in
∂2H ∂2H
opposite directions. Assume, for instance, that ∂x∂y ≥ 0 but ∂x∂v ≤ 0.
Then the outcome is uncertain because while wealthy males still value
wealthy females more than poor males do, they are handicapped by their
higher cost of bidding.
7.4 Search
We now turn to the alternative approach that stresses that in real life the
matching process is characterized by scarcity of information about potential
matches. The participants in the process must therefore spend time and
money to locate their best options, and the set of potential partners they
actually meet is partially random. The realized distribution of matches and
the division of the gains from each marriage are therefore determined in
an equilibrium which is influenced by the costs of search and the search
policies of other participants.
7.4.1 The basic framework

The main ingredients of the search model are as follows. There is a random
process which creates meetings between members of society of the opposite
sex. When a meeting occurs the partners compare their characteristics and
evaluate their potential gains from marriage. Each partner anticipates their
share in the joint marital output. If the gains for both partners from forming
the union exceed their expected gain from continued search then these
partners marry. Otherwise, they depart and wait for the next meeting to
occur (see Mortensen, 1988).
We assume that meetings occur according to a Poisson process. That is,
the waiting times between successive meetings are iid exponential variables
with mean 1/λ. Within a short period h, there is a probability of a meeting
given by λh + o(h) and a probability of no meeting given by 1 − λh + o(h),
where, o(h)/h converges to zero as h approaches zero. The arrival rate λ
is influenced by the actions of the participants in the marriage market.
Specifically, imagine an equal number of identical males and females, say
N , searching for a mate. Let si denote the ”search intensity” (that is num-
ber of meetings per period) initiated by a particular male. If all females
search at the same intensity sf , they will generate N sf contacts per pe-
riod distributed randomly across all males. In this case, the probability
that male i will make a contact with some female, during a short interval,
h, is (sf + si )h. If all males search at a rate sm and all females at a rate
sf then the rate of meetings between agents of opposite sex is
λ = sm + sf . (7.30)
The key aspect in (7.30) is that activities on both sides of the market
determine the occurrence of meetings. A limitation of the linear meeting
technology is that the number of searchers, N , has no effect on the arrival
rate λ. Each participant who searches actively and initiates meetings must
bear a monetary search cost given by ci (s), i = m, f, where we allow
the costs of search to differ by sex. The total and the marginal costs of
search increase as search intensity increases. When a meeting occurs the
marital output (quality of match) that the partners can generate together
is a random variable, z, drawn from some fixed distribution, F (z). Having
observed z, the couple decides whether to marry or not. With transferable
utility, the decision to marry is based on the total output that can be
generated by the couple within marriage relative to the expected total
output if search continues. Hence, a marriage occurs if and only if
z ≥ vm + vf , (7.31)
where, vm and vf denote the value of continued search for the male and
female partners, respectively. These values depend, in equilibrium, on the
search intensity that will be chosen if the marriage does not take place.
Specifically, for i = m, f ,
Z∞
rvi = M ax{(s + sj ) (wi (z) − vi )dF (z) − ci (s)}, (7.32)
s
vm +vf
where r is the instantaneous interest rate and wi (z) denote the shares of the
gains of marital output that male and female partners expect. By definition,
wm (z) + wf (z) = z. (7.33)

Equation (7.32) can be derived by using the following standard argument.
Let h be a short time interval. Then, the Bellman equation for dynamic
programming is
vi = M axe−rh {λh(s + sj )[p(z ≥ vm + vf )(Ez (M ax(vi , wi (z)|z ≥ vm + vf )]

s
+[1 − λh(s + sj )]vi } + o(h).
Note that, due to the stationarity of the Poisson process and the infinite
horizon, vi and and wi (z) do not depend on time. Approximating e−rh '
1−rh, cancelling terms that do not depend on h and rearranging, we obtain:
M ax{λh(s + sj )[p(z ≥ vm + vf )Ez (M ax(vi , wi (z)|z ≥ vm + vf ) − vi ]

s
+[1 − λh(s + sj )]vi − rh2 λ(s + sj )[(p(z ≥ vm + vf )Ez (M ax(vi , wi (z)|z ≥ vm + vf ) − vi ]}
+o(h) = rhvi .
Dividing both sides of this equation by h, we obtain (7.32) as the limit

when h approach zero.
Equation (7.32) states that the value of being an unattached player arises
from the option to sample from offers which arrive at a rate s + sf and
are accepted only if (7.31) holds. Each accepted offer yields a surplus of
wi (z)−vi for partner i. Integration over all acceptable offers yields expected
gain from search. Since each participant controls his own intensity of search,
he will choose the level of s that maximizes his value in the unattached
state. Therefore, with identical individuals in each gender,
Z∞
(wi (z) − vi )dF (z) = c0i (s), i = m, f. (7.34)
vm +vf
The marginal benefits from search, the left hand side of (7.34), depend
on the share that a person of type i expects in prospective marriages. As
wi (z) rises, holding z constant, he or she searches more intensely. Hence, the
equilibrium outcome depends on the allocation rules that are adopted. The
literature examined two types of allocation rules. One class of allocation
rules relies on Nash’s axioms and stipulates
wi (z) = vi + γ i (z − vm − vf ), (7.35)
where, γ i ≥ 0 and γ m + γ f = 1, i = m, f . The parameter γ i allows for

asymmetry in the bilateral bargaining between the sexes due to preferences
or social norms. The crucial aspect of this assumption, however, is that
outside options, reflected in the market determined values of vm and vf ,
influence the shares within marriage.
Wolinsky (1987) points out that a threat to walk out on a potentially
profitable partnership is not credible. Rather than walking away, the part-
ners exchange offers. When an offer is rejected, the partners search for an
outside opportunity that would provide more than the expected gains from
an agreement within the current marriage. Hence, during the bargaining
process the search intensity of each partner is determined by
Z∞
(wi (x) − wi (y))dF (x) = c0i (s), i = m, f, (7.36)
y
where, y is the quality of the current marriage and wi (y) is the expected
share in the current marriage if an agreement is reached. Since y ≥ vm + vf
and wi (y) ≥ vi , a person who searches for better alternatives during a
bargaining process will search less intensely and can expect lower gains than
an unattached person. The threat of each partner is now influenced by two
factors: The value of his outside opportunities (that is, the value of being
single), which enters only through the possibility that the other partner
will get a better offer and leave; The value of continued search during the
bargaining process, including the option of leaving when an outside offer
(whose value exceeds the value of potential agreement) arrives. Therefore,
the threat points, vi , in (7.35) must be replaced by a weighted average of
the value of remaining without a partner and the value of continued search
during the bargaining (the weights are the probabilities of these events).
Given these modified threat points, the parameters γ i that determine the
shares depend on the respective discount rates of the two partners and the
probabilities of their exit from the bargaining process. The logic behind
this type of formula, due to Rubinstein (1982), is that each person must be
indifferent between accepting the current offer of his partner or rejecting
it, searching for a better offer and, if none is received, return to make a
counter offer that the partner will accept.
Given a specification of the share formulae, one can solve for the equi-
librium levels of search intensities and the values of being unattached. For
instance, if the shares are determined by (7.35) and γ i is known, then
equations (7.34) and (7.35) determine unique values for sm , sf , vm , and

vf . Because of the linear meeting technology, these equilibrium values are
independent of the number of searchers. Observe that although the share
formulae depend on institutional considerations, the actual share of mar-
ital output that each partner receives depends on market forces and is
determined endogenously in equilibrium.
We can close the model by solving for the equilibrium number of unattached
participants relative to the population. Suppose that each period a new
flow of unattached persons is added to the population and the same flow of
married individuals exit. To maintain a steady state, this flow must equal
the flow of new attachments that are formed from the current stock of
unattached. The rate of transition into marriage is given by the product
of the meeting rate λ and the acceptance rate 1 − F (z0 ), where z0 is the
reservation quality of match. Using (7.30) and (7.31), we obtain
u(sm + sf )(1 − F (vm + vf )) = e (7.37)

where, u is the endogenous, steady state, rate of non-attachment and e is
the exogenous constant rate of entry and exit.
The meeting technology considered thus far has the unsatisfactory fea-
ture that attached persons "do not participate in the game". A possible
extension is to allow matched persons to consider offers from chance meet-
ings initiated by the unattached, while maintaining the assumption that
married people do not search. In this case divorce becomes an additional
option. If an unattached person finds a married person who belongs to
a marriage of quality z and together they can form a marriage of qual-
ity y then a divorce will be triggered if y > z. The search strategies will
now depend on the relative numbers of attached and unattached persons.
Specifically, (7.32) is replaced by
Z∞
rvi = M ax{u(s + sj ) (wi (z) − vi )dF (z) +
s
vm +vf
Z∞ Z∞
(1 − u)s (wi (z) − wi (y) − vi )dG(z)dF (y) − ci (s)}
(7.38)
vm +vf y
where G(z) is the distribution of quality of matched couples. The second

term in equation (7.38) is derived from the following argument. Suppose i
is a male and he meets a married woman who together with her current
husband has marital output y. Together with i, the marital output would
be z, where z ≥ y . The threat point of this woman in the bargaining
with man i is what she would receive from her current husband when she
threatens to leave him, which is y − vf . Thus, the total surplus of the new
marriage is z − (y − vm ) − vm . Hence, following bargaining, man i will
receive in the new marriage vm + γ m (z − y) = vm + wi (z) − wi (y).See

Mortensen (1988).
Observe that the expected returns from meeting an attached person are
lower than those of meeting with an unmarried one. Therefore, the higher
is the aggregate rate of non-attachment the higher are the private returns
for search.
Assuming that partners are ex-ante identical, the search models outlined
above do not address the question who shall marry whom. Instead, they
shift attention to the fact that in the process of searching for a mate there is
always a segment of the population which remains unmatched, not because
they prefer the single state but because matching takes time. A natural
follow up to this observation is the question whether or not there is "too
much" search. Clearly, the mere existence of waiting time for marriage
does not imply inefficiency since time is used productively to find superior
matches. However, the informational structure causes externalities which
may lead to inefficiency. One type of externality arises because in deciding
on search intensity participants ignore the higher chance for meetings that
others enjoy. This suggests that search is deficient. However, in the extended
model which allows for divorce there is an additional externality operating
in the opposite direction. When two unattached individuals reject a match
opportunity with z < vm + vf , they ignore the benefits that arise to other
couples from a higher non attachment rate. Thus, as in a related literature
on unemployment, it is not possible to determine whether there is too much
or too little non attachment.
An important aspect of equation (7.38) is the two way feedback be-
tween individual decisions and market outcomes. The larger is the propor-
tion of the unattached the more profitable is search and each unattached
person will be more choosy, further increasing the number of unattached.
As emphasized by Diamond (1982) such reinforcing feedbacks can lead to
multiplicity of equilibria. For instance, the higher is the aggregate divorce
rate the more likely it is that each couple will divorce. Therefore, some
societies can be locked into an equilibrium with a low aggregate divorce
rate while others will settle on a high divorce rate. There are some addi-
tional features which characterize search for a mate and can be incorpo-
rated into the analysis. First, as noted by Mortensen (1988), the quality
of marriage is revealed only gradually. Moreover, each partner may have
private information which is useful for predicting the future match quality
(see Bergstrom-Bagnoli, 1993). Second, as noted by Oppenheimer (1988),
the offer distribution of potential matches varies systematically with age,
as the number and quality of available matches change and the information
about a person’s suitability for marriage sharpens. Finally, meetings are not
really completely random. Unattached individuals select jobs, schools and
leisure activities so as to affect the chances of meeting a qualified person
of the opposite sex (see Goldin, 2006).
7.4.2 Search and Assortative Mating

Models of search add realism to the assignment model, because they provide
an explicit description of the sorting process that happens in real time.
Following Burdett and Coles (1999), consider the following model with non
transferable utility whereby if man m marries woman f, he gets f and she
gets m. Assume a continuum of men, whose traits m are distributed on
[0, m̄] according to some distribution
£ ¤ F , and a continuum of women, whose
traits f are distributed on 0, f¯ according to some distribution G. To bring
in the frictions, assume that men and women meet according to a Poisson
process with parameter λ. Upon meeting, each partner decides whether to
accept the match or to continue the search. Marriage occurs only if both
partners accept each other. A match that is formed cannot be broken. To
ensure the stationary of the decision problem, we assume a fixed and equal
number of infinitely lived men and women.
Each man chooses an acceptance policy that determines which women to
accept. Similarly, each woman chooses an acceptance policy that determines
which men to accept. These policies are characterized by reservation values,
R, such that all potential partners with a trait exceeding R are accepted
and all others are rejected. The reservation value that each person chooses
depends on his\her trait. In particular, agents at the top of the distribution
of each gender can be choosier because they know that they will be accepted
by most people on the other side of the market and hence continued search
is more valuable for them. Formally,
Zf¯
λμm
Rm = bm + (f − Rm )dGm (f ),
r
Rm
Zm̄
λμf
Rf = bf + (m − Rf )dFf (m), (7.39)
r
Rf
where, the flow of benefits as single, b, the proportion of meetings that

end in marriage, μ, and the distribution of "offers" if marriage occurs, all
depend on the trait of the person as indicated by the m and f subscripts.
The common discount factor, r, represents the costs of waiting.
In equilibrium, the reservation values of all agents must be a best re-
sponse against each other, yielding a (stationary) Nash equilibrium. The
equilibrium that emerges is an approximation of the perfect positive as-
sortative mating that would be reached without frictions. Using the Gale-
Shapley algorithm to identify the stable outcome, we recall that, in the
absence of frictions, this model generates a positive assortative mating.
Thus, if men move first, all men will propose to the best woman and she
will keep only the best man and reject all others. All rejected men will pro-
pose to the second best woman and she will accept the best of these and
reject all others and so on. This outcome will also emerge here if the cost
of waiting is low or frictions are not important, because λ is high. However,
if frictions are relevant and waiting is costly, agents will compromise. In
particular, the "best" woman and the "best" man will adopt the policies
Zf¯
λ
Rm̄ = bm̄ + (f − Rm̄ )dGm (f ),
r
Rm̄
Zm̄
λ
Rf¯ = bf¯ + (m − Rf¯)dFf (m). (7.40)
r
Rf¯
Thus, the best man accepts some women who are inferior to the best woman
and the best woman accepts some men who are inferior to the best man,
because one bird at hand is better than two birds on the tree.
The assumption that the rankings of men and women are based on a sin-
gle trait, introduces a strong commonality in preferences, whereby all men
agree on the ranking of all women and vice versa. Because all individuals
of the opposite sex accept the best woman and all women accept the best
man, μ is set to 1 in equation (7.40) and the distribution of offers equals
the distribution of types in the population. Moreover, if the best man ac-
cepts all women with f in the range [Rm̄ , f¯] then all men who are inferior
in quality will also accept such women. But this means that all women in
the range [Rm̄ , f¯] are sure that all men accept them and therefore will have
the same reservation value, Rf¯, which in turn implies that all men in the
range [Rf¯, m̄] will have the same reservation value, Rm̄ . These considera-
tions lead to a class structure with a finite number of distinct classes in
which individuals marry each other. Having identified the upper class we
can then examine the considerations of the top man and woman in the rest
of the population. These individuals will face μ < 1 and a truncated distri-
bution of offers that, in principle, can be calculated to yield the reservation
values for these two types and all other individuals in their group, forming
the second class. Proceeding in this manner to the bottom, it is possible to
determine all classes.
With frictions, there is still a tendency to positive (negative) assortative
mating based on the interactions in traits. If the traits are complements,
individuals of either sex with a higher endowment will adopt a more selec-
tive reservation policy and will be matched, on the average, with a highly
endowed person of the opposite sex. However, with sufficient friction, it is
also possible to have negative assortative mating under complementarity.
The reason for this result is that, because of the low frequency of meet-
ings and costs of waiting, agents in a search market tend to compromise.
Therefore, males with low m, expect some women with high f to accept
them, and if the gain from such a match is large enough, they will reject
all women with low f and wait until a high f woman arrives.
The class structure result reflects the strong assumption that the utility
that each partner obtains from the marriage depends only on the trait of
the other spouse, so that there is no interaction in the household production
function between the traits of the two spouses. In general, there will be some
mingling of low and high income individuals, but the pattern of a positive
assortative mating is sustained, provided that the complementarity in traits
is large enough to motivate continued search for the "right" spouse. Smith
(2006) provides a (symmetric) generalization of the problem where if man
m marries woman f he receives the utility payoff v = π(m, f ) and she
receives the utility payoff u = π(f, m). It is assumed that this function is
increasing in its second argument, π2 (x, y) > 0, so that all men prefer a
woman with a higher f and all women prefer a man with a higher m, but
individuals can differ in the intensity of their ordering.8 He then shows that
a sufficient condition for positive assortative mating, in the sense of a higher
likelihood that a rich person will have a rich spouse, is that log (π(m, f ))
be super modular. That is, m > m0 and f > f 0 imply that
π(m, f )π(m0 , f 0 ) > π(m, f 0 )π(m0 , f ). (7.41)
The reason for such a condition is that one needs sufficiently strong com-
plementarity to prevent the high types from accepting low types, due to
impatience.
Surprisingly, the assumption of transferable utility loses some of its edge
in the presence of frictions. In particular, it is no longer true that the
assignment is determined by the maximization of the aggregate marital
output of all potential marriages. To see why, consider the following output
matrix:
Example 7.6
Women
1 2 3
1 4 1 0
Men
2 1 0 1
3 0 1 4
where aggregate output is maximized on the main diagonal. With frictions,
this assignment is in general not stable, because man 2 and woman 2 will
prefer continued search to marriage that yield, 0, even if the value of being
single is 0. The reason is that they can marry other men and women with
whom they can obtain 1, who might be willing to marry them if the arrival
rate of offers is low or the cost of waiting is high.
Generally speaking, the nature of the assignment problem changes, be-
cause of the need to consider the cost of time spent in search, as well as the
8 Intensity is a meaningful concept because, given the risky environment, agents are
endowed with a Von Neumann Morgenstern utility function that is unique up to a linear
transformation.
benefits from matching. An additional complication, relative to the case

of non-transferable utility, is the presence of rents. As we have seen, when
meetings are random, and agents adopt reservation polices for accepted
matches, the realized match will generally exceed the outside options of
the married partner so that the rules for dividing the rents enter into the
analysis. As a consequence, one generally needs stronger conditions to guar-
antee assortative matching. Shimer and Smith (2000) provide an analysis of
the degree of complementarity that must hold to guarantee positive assorta-
tive mating if rents are divided equally in all marriages. Positive assortative
mating, in the sense that a high m male is more likely to match with a high
f female (on the average) requires, in addition to the supermodularity of
h(m, f ), the supermodularity of the logs of its partial derivatives and the
log of the cross derivative hmf (m, f ). This means that the simple predic-
tions of the frictionless model carry over only under restrictive assumptions.
)2
For instance, h(m, f ) = (m+f4 , which, as we have shown, arises naturally
in the presence of public goods, does not satisfy these requirements.9
7.5 Bargaining In Marriage (BIM)

As we have just seen, search models with random and intermittent meetings
provide a natural framework to deal with rents and bargaining over rents in
the marriage market. However, if marriage specific capital, such as children,
is generated during marriage, then rents and bargaining can arise even
without uncertainty and frictions. As is well known from models of specific
human capital (see Becker (1993 ch. 3)) the accumulation of capital that
is useful only in a particular relation partly insulates from competition
the division of the gain from marriage. There is, therefore, a scope for
bargaining over such rents.
It has been recently pointed out by Lundberg and Pollak (2009) that
if the division resulting from bargaining in marriage is fully anticipated
prior to marriage and if, in addition, binding contracts cannot be made at
marriage, then the assignment into marriage must be based on the Gale
Shapley algorithm. Specifically, Lundberg and Pollak contrast their ‘BIM’
(Bargaining In Marriage) framework with the standard, ‘BAMM’ (Binding
Agreements on the Marriage Market) model, which is one of the possible
foundation of the Becker-Shapley-Shubik construct. In a BIM world, any
promise I may make before marriage can (and therefore will) be reneged
9 Specifically, m+f
the partial derivatives 2
are not log super modular because m > m0
and f > f 0 imply that
(m + f )(m0 + f 0 ) < (m + f 0 )(m0 + f ).

upon minutes after the ceremony; there is just no way spouses can com-
mit beforehand on their future behavior. Moreover, ‘upfront’ payments,
whereby an individual transfers some money, commodities or property
rights to the potential spouse conditional on marriage, are also excluded.
Then the intrahousehold allocation of welfare will be decided after mar-
riage, irrespective of the commitment made before. Marriage decision will
therefore take the outcome of this yet-to-come decision process as given,
and we are back in a non transferable utility setting in which each partner’s
share of the surplus is fixed and cannot be altered by transfers decided ex
ante.
This result is an outcome of the assumed inability to credibly bid a
person prior to marriage either by payments up-front or by short term
commitments. This argument raises some important modeling issues about
the working of the marriage market. A first remark is that it is not clear
why premarital contracting is assumed away. Historically, contracts speci-
fying what one brings into marriage and what the husband and wife take
away upon divorce were universal (see Anderson, 2007). In modern soci-
eties prenuptial contracts still exist, although they are less prevalent. One
possibility is that formal contracting and the associated enumeration of
contingencies would "crowd out" the emotional trust on which the part-
ners rely. This argument, however, has somewhat ambiguous implications,
because the mere existence of such emotional trust seems to imply the exis-
tence of at least some minimum level of ‘emotional commitment’ - an idea
that has been formalized by Browning (2009). Another important issue is
verification. Typically it is difficult for the courts to verify the division of
consumption or work within families. It must however be emphasized that
commitment on intrahousehold allocation is not needed to implement a
BAMM solution. Any transfer that (i) is decided ex ante, that is before
marriage, and (ii) can be used to alter the spouse’s respective bargaining
positions after marriage, can do the trick. For instance, if the husband can,
at (or just before) marriage, sign a legally enforceable contract specifying
the transfers that would occur in case of separation, then we are back to a
BAMM framework: I can now ‘bid’ my wife by offering her a very advanta-
geous contract, because even if we do not ultimately divorce, the additional
bargaining power provided to her by the ex ante contract will allow her to
get a larger share of household resources - and is therefore equivalent to
an ex post cash transfer. An even more striking example is the ‘payment
for marriage’ situation, in which the husband can transfer a predetermined
amount to his wife upon marriage (say, by offering her an expensive ring,
or putting the couple’s residence under her name, or even writing a check).
Again, the size of the transfer can be used in the bidding process, and the
relevant concept is again BAMM. Conversely, the BIM framework basically
requires that no ex ante contract can ever be signed, and no conditional
payment can ever be made.
A second concern is that even if we accept the total absence of com-
mitment, Gale-Shapley still need not be the relevant equilibrium concept.

To see why, consider the extreme situation in which marriage can be done
and undone at very low cost. Then at any moment of marital life, each
spouse has many close substitutes on the market, and the intrahousehold
allocation will typically reflect this fact. Although, technically, this is not
a BAMM situation (no binding agreement can be signed by assumption),
the relevant concept is still the TU model a la Becker-Shapley-Shubik, be-
cause each spouse receives exactly her/his reservation value and the latter
is fully determined by market equilibrium forces (at least when the number
of potential spouses is ‘large enough’). In other words, even in the extreme
no transfer/no commitment case, the BIM framework applies only insofar
as marriage decision can only be reversed at some cost, and only within
the limits defined by this cost.
It is clear, in practice, that entry into marriage is a major decision that
can be reversed only at some cost. However, as in any modeling choice,
"realism" of the assumptions is not the only concern. It is also important
to have a tractable model that allows one to predict the marriage market
outcomes under varying conditions. In this regard, the presence of trans-
action costs is quite problematic. To see this, consider again our example
7.3. Suppose that a new woman, 4, unexpectedly enters a marriage market
that has been in one of the two equilibria discussed in section 7.1. Let the
new payoffs matrix be as below:
Example 7.3a
Women
1 2 3 4
1 3, 2 2, 6 1, 1 2, 1
Men
2 4, 3 7, 2 2, 4 5, 4
3 1, 1 2, 1 0, 0 .5, .5
By assumption, woman 4 is preferred to woman 3 by all men and one would
expect that in the new assignment woman 3 will become single. Suppose,
however, that all existing couples bear a transaction cost of 0.75. Then it is
easy to see that if the original equilibrium was the one in which men moved
first, no man will marry woman 4 and she will remain single. In contrast, if
the original equilibrium was the one in which women moved first then man
2 will take woman 4 and his ex-wife (woman 1) will first propose to man 1
who will reject her and then to man 3 who will accept her, so that woman 3
will become single. Thus, in general, it is impossible to predict what would
happen when a new player enters the market, without knowing the bar-
gaining outcomes in all marriages, the potential bargaining outcome that
the entrant will have with all potential existing partners and the relational
capital accumulated in all existing marriages. Such information is never
available to the observer. In contrast, the Becker-Shapley-Shubik frame-
work can predict the outcome very easily, using only information about
the place of the new woman in the income distribution of women and the
form of the household production function that specifies the within couple
interaction between men and women of different attributes.
Given the different implications of alternative models of the marriage
market, it seems prudent to consider several alternatives, depending on the
application. In subsequent chapters we shall apply search models to analyze
marriage and divorce when match quality is uncertain, and we shall apply
the standard assignment model to discuss the determination of the division
of gains from marriage when men and women differ in their attributes.
Appendix: Supermodularity, Submodularity and As-

sortative Matching
This appendix proves that complementarity (substitution) in traits must
lead to a positive (negative) assortative mating; otherwise aggregate output
is not maximized. Assuming that h(x, y) is increasing in x and y we also
obtain that in the case of positive assortative mating, the best man marries
the best woman, and if there are more women than men the women with
low female quality remain single. If there is negative assortative mating,
the best man marries the worst woman among the married women but if
there are more women than men, it is the women with the lower female
attributes who remain single.
Super modularity
Let x̄ and ȳ denote the endowments of the "best" man and woman and
suppose that they are not married to each other and instead man x̄ marries
some woman whose female attribute is y 0 < ȳ and woman ȳ marries some
man whose attribute is x0 < x̄. Then stability of these matches requires the
existence of divisions such that
ux̄,y0 + vx0 , ȳ ≥ h(x̄, ȳ),

ux0 , ȳ + vx̄,y ≥ h(x0 , y 0 ),
ux̄,y0 + vx̄,y0 = h(x̄, y0),
ux0 , ȳ + vx0 , ȳ = h(x0 , ȳ),
which implies that
h(x̄, y0) + h(x0 , ȳ) ≥ h(x0 , y) + h(x̄, ȳ)
and contradicts (strict) super modularity. Thus complementarity implies

that the best man must marry the best woman. Eliminating this couple,
and restricting attention to the next best pair, we see that it must marry
too and so on. If there are more women than men, Then there must be
some woman ym such that all women with lesser quality are single.
Sub modularity
Suppose again that there are more women than men. Then, in this case
too, there must be some woman ym such that all women with lesser quality
are single. Otherwise, there must be a married woman with a lower quality
than some single woman, which under monotonicity implies that aggregate
output is not maximized. However, now man x̄ and woman ym must marry
each other. If they are not married to each other and, instead, man x̄
marries some woman whose female attribute is y 0 > ym and woman ym
marries some man whose attribute is x0 < x̄, then stability of these matches
requires the existence of surplus allocations such that
ux̄,y0 + vx0 ,ym ≥ h(x̄, ym ),

ux0 ,ym + vx̄,y ≥ h(x0 , y 0 ),
ux̄,y0 + vx̄,y0 = h(x̄, y0),
ux0 ym + vx0 ,ym = h(x0 , ym ),
which implies that
h(x̄, y0) + h(x0 , ym ) ≥ h(x0 , y 0 ) + h(x̄, ym )
and contradicts (strict) sub modularity. Eliminating this couple, and re-
stricting attention to the next best pair, that is the second best man among
men and the second worst woman among all married women must marry
too, and so on.
7.6 References
[1] Anderson, Siwan, ‘The Economics of Dowry and Bride Price’, Journal
of Economic Perspectives, 21 (2007), 151-174.
[2] Becker, Gary, ‘A Theory of Marriage: Part 1’, Journal of Political

Economy, 81 (1973), 813-846.

[4] Becker, Gary, and Kevin M. Murphy, Accounting for Tastes,

(Cambridge Mass: Harvard University Press, 1996)
[5] Becker, Gary, Human Capital, (Third edition, Chicago, Chicago Uni-
versity Press, 1993).
[6] Bergstrom, Ted and Mark Bagnoli, ‘Courtship as a Waiting Game’,

[7] Burdett, Kenneth and Melvyn G. Coles, ‘Long-Term Partnership For-

mation: Marriage and Employment’, The Economic Journal, 109
(1999), 307-334.
[8] Crawford, Vincent P., ‘Comparative statics in Matching Markets’,

Journal of Economic Theory, 54 (1991), 389-400.
[9] Chiappori, P.A., R. McCann and L. Nesheim, “Hedonic Price Equilib-

ria, Stable Matching, And Optimal Transport: Equivalence, Topology,
And Uniqueness”, Economic Theory, 42 2 (2010), 317-354
[10] Chiappori, P.A., and P. Reny, ‘Matching to share risk’, Mimeo, Uni-
versity of Chicago (2006).
[11] Diamond, Peter A., ‘Aggregate Demand Management in Search Equi-

librium’, Journal of Political Economy, 90 (1982), 881-894.
[12] Ekeland, I., "Notes on optimal transportation", Economic Theory,

42 2 (2010), 337-359.
[13] Gale, David and Lloyd Shapley, ‘College Admissions and the Stability
of Marriage’, American Mathematical Monthly, 69 (1962), 9-15.
[14] Gale, David, The Theory of Linear Economic Models, (New

York: McGraw-Hill, 1960).
[15] Galichon, A., and B. Salanié, ‘Matching with Trade-offs: Revealed

Preferences over Competing Characteristics’, mimeo, Columbia Univer-
sity, 2009
[16] Goldin, Claudia, ‘The Quiet Revolution That Transformed Women

Employment, Education and Family’, American Economic Review,
90, (papers and proceedings, 2006), 1-21.
[17] Lundberg, Shelly and Robert Pollak, ‘Marriage market equilibrium
and bargaining’, Delivered as the presidential address in the 2009 Society
of Labor Economics meeting in Boston.
[18] Mortensen, Dale T., ‘Matching: Finding a Partner for Life or Other-
wise’, American Journal of Sociology, 94 (1998, supplement), s215-
s240.
[19] Oppenheimer, Valerie K., ‘A Theory of Marriage Timing’, American
Journal of Sociology, 94 (1988), 563-591.
[20] Roth, Alvin E. and Marilda A.Oliveira Sotomayor, Two Sided
Matching: A Study in Game-Theoretic Modeling and Analy-
sis, (New York: Cambridge University Press, 1990).
[21] Shapley, Lloyd and Martin Shubik, ‘The Assignment Game 1: The
Core’, International Journal of Game Theory, 1 (1972), 111-130.
[22] Rubinstein, Ariel, ‘Perfect Equilibrium in a Bargaining Model’,
Econometrica, 50 (1982), 97-110.
[23] Shimer, Robert and Lones Smith, ‘Assortative Matching and Search’,
Econometrica, 68 (2000), 343-369.
[24] Smith, Lones, ‘The Marriage Model with Search Frictions’, University
of Michigan, (2006).
[25] Wolinsky, Asher, ‘Matching, Search and Bargaining’, Journal of Eco-
nomic Theory, 42 (1987), 311-33.
This is page 333
8
Sharing the gains from
marriage
In this chapter, we discuss in more detail the determination of the division
of the marital surplus and how it responds to market conditions. If each
couple is considered in isolation, then, in principle, any efficient outcome
is possible, and one has to use bargaining arguments to determine the
allocation. On the contrary, the stability of the assignments restricts the
possible divisions because of the ability to replace one spouse by another
one. The options for such substitution depend on the distributions of the
marital relevant attributes in the populations of the men and women to be
matched. In the present chapter, we precisely ask how the marriage market
influences the outcome in the ideal, frictionless case discussed previously.
Although the division within marriage is not always fully determined, some
qualitative properties of the division can be derived from information on
the joint distribution of male and female characteristics together with a
specification of the household production function.
As before, we discuss separately the cases of discrete and continuous
distributions. The general intuition goes as follows. In the discrete case,
competition puts bounds on individual shares but does not completely de-
termine them; this is because on the marriage market, each potential spouse
has only a finite number of ‘competitors’, none of which is a perfect substi-
tute - so some elements of ‘bilateral monopoly’ persist. In the continuous
case, however, competition between potential spouses tends to be perfect,
leading to an exact determination of the ‘prices’ - that is, in our case, bud-
get shares. In addition to the standard case of transferable utility, we also
consider the more general case in which the exchange rate of the spouses’
utilities varies along the Pareto frontier. We provide detailed examples that
illustrate how changes of the distributions of incomes or tastes of men and
women can affect the division of resources within couples. We conclude
with a discussion of recent developments in estimating equilibrium models
of the marriage market, including the gains from marriage and the division
of these gains.
The major insight obtained from the equilibrium analysis is that the
sharing of the gains from marriage depends not only on the incomes or
preferences of spouses in a given match but also and perhaps mainly on
the overall distributions of incomes and preferences in society as a whole.
Thus, a redistribution of income via a tax reform can influence the shares
of the gains from marriage even if the incomes in particular couple are un-
334 8. Sharing the gains from marriage
affected. Similarly a legal reform or a technological innovation that makes

it easier to prevent pregnancy can influence the division of resources within
married couples who chose to have children. In either case, the general equi-
librium effects arise from competition with potential spouses outside the
given marriage. Obviously, our assumptions regarding the agents’ ability
to transfer resources within marriage (and to a lesser extent the absence of
frictions) are crucial for such indirect effects. It is therefore a challenging
research agenda to find how important are these considerations in practice.
8.1 Determination of shares with a finite number

of agents
We start with matching between finite male and female populations.1 As
explained in the previous chapter, while the matching pattern (who mar-
ries whom) and the associated surplus is generally unique, the allocation of
surplus between spouses is not. Typically, there exists, within each couple,
a continuum of allocations of welfare that are compatible with the equilib-
rium conditions. That does not mean, however, that the allocation is fully
arbitrary. In fact, equilibrium imposes strict bounds on these allocations.
Depending on the context, these bounds may be quite large, allowing for
considerable leeway in the distribution of surplus, or quite tight, in which
case the allocation is practically pinned down, up to minor adjustments, by
the equilibrium conditions. We present in this section a general description
of these bounds.
8.1.1 The two men - two women case

As an introduction, let us consider a model with only two persons of each
gender discussed in Chapter 7. Assume for instance that z12 + z21 ≥ z11 +
z22 , implying that the stable match is ‘off-diagonal’ (man i marries woman
j 6= i, with i, j ∈ {1, 2}). Then all pairs (v1 , v2 ) satisfying the inequalities:
z12 − z11 ≥ v2 − v1 ≥ z22 − z21 , (8.1)

z21 ≥ v1 ≥ 0,
z12 ≥ v2 ≥ 0,
yield imputations v1 , v2 , u1 = z12 − v2, , u2 = z21 − v1 that support the

stable assignment along the opposite diagonal. The shaded area in Figure
8.1 describes all the pairs that satisfy the constraints required for stability
expressed in condition (8.1). The figure is drawn for the special case in
which woman 2 is more productive than woman 1 in all marriages (z22 >
1 Ellana Melnik participated in the derivation of the results of this section.

8. Sharing the gains from marriage 335
FIGURE 8.1. Imputations when z12 + z21 > z22 + z11
z21 , z12 > z11 ) and symmetry holds, z12 = z21 , implying that man 2 is also
more productive than man 1 in all marriages. The main feature here is
that the difference v2 − v1 is bounded between the marginal contributions
of replacing woman 1 by woman 2 as spouses of man 1 and man 2. Woman
2 who is matched with man 1 cannot receive in that marriage more than
z12 − z11 + v1 , because then her husband would gain from replacing her by
woman 1. She would not accept less than v1 + z22 − z21 , because then she
can replace her husband by man 2 offering him to replace his present wife.
The assumption that z12 − z11 > z22 − z21 implies that man 1 can afford
this demand of woman 2, and will therefore "win" her. In this fashion,
the marriage market "prices" the different attributes of the two women.
Symmetric analysis applies if we would replace (v1 , v2 ) with (u1 , u2 ).
Similarly, if z12 + z21 ≤ z11 + z22 , implying that the stable match is
‘diagonal’, then all pairs (v1 , v2 ) satisfying the inequalities
z22 − z21 ≥ v2 − v1 ≥ z12 − z11 , (8.2)

z11 ≥ v1 ≥ 0,
z22 ≥ v2 ≥ 0,
yield imputations v1 , v2 , u1 = z11 −v1 , u2 = z22 −v2 that support the stable
assignment along the diagonal. The shaded area in Figure 8.2 describes
all the pairs that satisfy the constraints required for stability expressed
FIGURE 8.2. Imputations when z12 + z21 < z22 + z11
in condition (8.2). Again the difference v2 − v1 is bounded between the

marginal contributions of replacing woman 1 by woman 2 as spouses of
man 1 and man 2. Because we assume that woman 2 is more attractive
than woman 1, she gets a larger part of the surplus in both cases and her
share in the surplus is always positive. Woman 1 who is less desirable may
get no surplus at all. If she is married to man 1 who is less attractive, she
may get the entire surplus. However, if she is married to man 2, he always
receives a positive share and she never receives the entire surplus.
The indeterminacy of prices in the marriage market reflects the fact

that the "objects traded" are indivisible and have no close substitutes.
Therefore, agents may obtain in the stable assignment utility levels that are
strictly higher than they would in alternative marriages. When this is true
for all alternative marriages it is possible to slightly shift utility between
the partners of each marriage, and still maintain all the inequalities of the
dual problem, without any effect on the allocation. An interesting feature,
noted by Shapley and Shubik (1972), is that in the core (that is, of the
set of imputations that support a stable assignment) "the fortunes of all
players of the same type rise and fall together". This is seen by the upward
tendency of the shaded areas in figures 8.1 and 8.2. In particular, there is
a polar division of the surplus that is best for all men, and also a polar
division that is best for all women.
As an illustration, let us come back to Example 7.4 in the previous Chap-
ter. Specifically, consider the table presenting three imputations, denoted
by a, b and c; for commodity, the Table is reproduced below. Note that these
imputations are arranged in such a manner that the reservation utility of
all men rise and those of all women decline.
Imputation a b c
W M W M W M
v1 2 u1 3 v1 2 u1 4 v1 1 u1 5
Individual shares
v2 5 u2 5 v2 4 u2 5.5 v2 3 u2 6
v3 1 u3 0 v3 0.5 u3 0 v3 0 u3 1
In each of these three imputations, individuals who are married to each
other receive their reservation utility, which together exhaust the marital
input, Thus v2 + u1 = z12 = 8, u2 + v3 = z23 = 6 and v1 + u3 = z31 = 2.
For marriages that do not form, the sum of the reservation utilities exceeds
or equals the potential marital output. For instance, man 2 and woman
2 are not married to each other, and therefore v2 + u2 ≥ z22 = 9. This
requirement is strict for imputations a and b and holds as equality for
imputation c. Similarly, because man 1 and woman 1 are not married to
each other, we must have v1 + u1 ≥ z11 = 5. This holds as strict inequality
for imputations b and c, and as equality for imputation a. The significance
of the equalities is that they indicate the bounds within which it is possible
to change prices without any affect on the assignment. Hence, imputation
c is the best for men and the worst for women and imputation a is the best
for women and the worst for men.
8.1.2 Bounds on levels

The previous insights can readily be extended to a more general setting.
We now consider N men and M women and assume that the assignment
variables aij are all either zeros or ones. Start with the dual problem:
N
X M
X
min ui + vj (8.3)
u,v
i=1 j=1
subject to
ui + vj ≥ zij for i = 1, 2...N, j = 1, 2...M

Denote the solution for individual utilities (or individual ‘prices’) by

(ûi , v̂j ). From the results on duality (Gale, 1960, chapter 5) we know that
the solution to this problem yields the same value as the solution to the
primal problem. That is
N
X M
X X
ûi + v̂j = âij zij ,
i=1 j=1 i,j
where â denotes the assignment that solves the primal.

Now compare (8.3) to the dual problem when man N is eliminated:
⎛ ⎞
N
X −1 M
X
min ⎝ ui + vj ⎠ (8.4)
u,,v
i=1 j=1
subject to
ui + vj ≥ zij for i = 1, 2...N − 1, j = 1, 2...M

Denote the solution for prices by (ūi , v̄j ). Again we know that the solution
to this problem yields the same value as the solution to the primal problem.
That is
N−1
X M
X X
ūi + v̄i = āij zij .
i=1 j=1 i,j
where ā denotes the assignment that solves the primal associated with (8.4).
Notice that the values (ûi , v̂j ) chosen in the dual problem (8.3) are feasi-
ble in the dual problem (8.4). It follows that the minimum attained satisfies
N−1
X M
X N−1
X M
X
ūi + v̄j ≤ ûi + v̂j ,
i=1 j=1 i=1 j=1
or X X
āij zij ≤ âij zij − ûN ,
i,j i,j
implying that
X X
ûN ≤ âij zij − āij zij .
i,j i,j
That is, the upper bound on the utility that man N can get is his marginal
contribution to the value of the primal program (that is, the difference
between the maximand with him and without him). Note that to calculate
this upper bound we must know the assignments in both cases, when N is
excluded and N is included. This is easily done if we assume positive or
negative assortative mating. For instance, with positive assortative mating

and M > N ,
X X N
X N
X −1
âij zij − āij zij = zi,M−N +i − zi,M−(N−1)+i . (8.5)
i,j i,j i=1 i=1
Similar arguments apply for any man and any woman. Using the bounds
for men and women who are married to each other at the stable assignment
we can put bounds on the possible divisions of the gains from marriage of
the husband and wife in each couple. Thus the husband’s share in the
couple M N is bounded by
XN N
X
zN,M − ( zi,M−N +i − zi,M−1−N +i )
i=1 i=1
≤ ûN ≤
XN N
X −1
zi,M−N +i − zi,M−(N−1)+i
i=1 i=1
or
N
X N
X −1 N
X N−1
X
zi,M−1−N +i − zi,M−N+i ≤ ûN ≤ zi,M −N +i − zi,M−(N −1)+i .
i=1 i=1 i=1 i=1
(8.6)
8.1.3 Bounds on Differences

With positive or negative assortative mating we can also put bounds on the
change in utilities as one moves along the assignment profile. Let there be
a positive assortative mating (that is, the matrix zij is super modular) and
suppose that M > N . Then man N is married to woman M , and woman
M − 1 is married to man N − 1. At a stable assignment
uN + vM = zN,M
uN −1 + vM −1 = zN −1,M−1
uN + vM −1 ≥ zN,M−1
uN −1 + vM ≥ zN −1,M
Eliminating uN and uN−1 and substituting into the inequalities we get
zN,M − vM + vM−1 ≥ zN,M−1

zN−1,M−1 − vM−1 + vM ≥ zN −1,M
Hence,
zN,M − zN,M−1 ≥ vM − vM−1 ≥ zN −1,M − zN−1,M−1
and we get the upper and lower bounds on vM − vM−1 . Now we also know
that woman M − 2 and man N − 2 marry each other. Using the fact that
M − 1 and N − 1 also marry each other we get by the same argument that
zN−1,M−1 − zN −1,M −2 ≥ vM −1 − vM−2

≥ zN −2,M−1 − zN−2,M−2 (8.7)
and so on all the way to the lowest married couple. Because we assume
more women than men, M > N , woman M − N + 1 will marry man 1. For
this particular couple, we have
u1 + vM−N+1 = z1,M−N+1
u1 + vM−N ≥ z1,M−N
vM−N = 0
The boundary condition is therefore
z1,M−N+1 − z1,M−N ≥ vM−N +1 .
We see that along the stable assignment the prices must form an increasing
sequence. This is a consequence of complementarity.
When we set the bounds on couple N, M in (8.7), we referred only to
couple N − 1, M − 1. However, there are M − 1 stability constraints, one
for each woman that man N is not married to:
uN ≥ zN,M−1 − vM−1
uN ≥ zN,M−2 − vM−2 ...
and also N − 1 stability constraints for woman M regarding for each man
that she is not married to. We now show that the most binding constraint
from all these constraints is the one expressing that man N (woman M )
does not marry woman M − 1 (man N − 1). That is,
zN,M−1 −vM−1 ≥ zN,M−2 −vM−2 ≥ ··· ≥ zN,M−N +1 −vM−N+1 ≥ zN,M−N .
Note, first, that if man N does not want to marry woman M − 1, then he
does not want to marry woman M −2 either; that is, the stability constraint
related to woman M −1 is more binding than that related to woman M −2.
Indeed, we want to show that:
zN,M −1 − vM−1 ≥ zN,M−2 − vM−2 . (8.8)

Equation (8.8) can be rewritten as :
zN,M−1 − (zN −1,M−1 − uN−1 ) ≥ zN,M−2 − vM−2 ,
or
uN−1 + vM−2 ≥ zN,M−2 − zN,M−1 + zN−1,M−1 .
In the stable assignment uN−1 +vM −2 ≥ zN −1,M −2 , so it is enough to show

that
zN −1,M−2 ≥ zN,M−2 − zN,M−1 + zN−1,M−1
. But, this follows directly from the assumption that zij is super modular.
Therefore, the lower bound woman M − 1 imposes is higher than the lower
bound woman M − 2 imposes on man N . By the same arguments, we now
generally show that woman M −k0 s constraint is more binding than woman
M − k − 10 s constraint. Now we have,
zN,M−k − vM−k ≥ zN,M−k−1 − vM−k−1 ,

zN,M−k − (zN −k,M−k − uN−k ) ≥ zN,M−k−1 − vM−k−1 ,
uN −k + vM−k−1 ≥ zN,M−k−1 − zN,M−k + zN −k,M −k .
Again, we know that in a stable assignment, uN −k +vM−k−1 ≥ zN −k,M −k−1 ,

so it is enough to show that
zN −k,M−k−1 ≥ zN,M−k−1 − zN,M−k + zN −k,M−k
which follows from the super modularity assumption that requires this
condition. Therefore we can say that the lower bound woman M −k imposes
is higher than the lower bound woman M − k − 1 imposes on man N, and
finally conclude that the highest lower bound on man N 0s share is imposed
by woman M − 1. In a very similar way it can be shown that the highest
lower bound on woman M 0s share is imposed by man N − 1.
8.2 The continuous case

We now consider a continuous distribution of agents, in which equilibrium
conditions typically pin down the intrahousehold allocation of welfare. The
difference between the continuous case and the discrete case analyzed in
the previous section is that, with a continuum of agents and continuous
distributions, each agent has a very close substitute. In this case, the upper
and lower bounds in 8.6 and 8.7, respectively, approach each other and in
the limit coincide.
8.2.1 Basic results

The setting, here, is a slight generalization of the one considered in subsec-
tion 7.2.3 of Chapter 7. There exists a continuum of men, whose incomes
x are distributed on [0, 1] according to some distribution F , and a contin-
uum of women, whose incomes y are distributed on [0, 1] according to some
distribution G.2 The measure of all men in the population is normalized
to 1, and the measure of women is denoted by r. Also, we still consider
a transferable utility (TU) framework. The innovation is that the ‘marital
output’ is now the sum of two components: an economic output, which is
a function h (x, y) of individual incomes, and a fixed non monetary gain
from marriage, denoted θ, which is perceived by the spouses in addition to
the economic benefits:
z (x, y) = h (x, y) + θ
As before, h is assumed to be supermodular.

An allocation rule specifies the shares of the wife and husband in every
marriage. If r > 1 and all men are married, we can index the marriage by the
husband’s income x (then his spouse’s income is ψ (x)). The marital output
is then h(x, ψ (x)) + θ and the marital shares are u(x) for the husband and
v(ψ (x)) for the wife. If r < 1 and all women are married, we can index the
marriage by the wife’s income y (then the husband’s income is φ (y)). The
marital output is then h(φ (y) , y) + θ and the marital shares are u(φ (y))
for the husband and v(y) for the wife.
As discussed before, the allocation rule that supports a stable assignment
must be such that the implied utilities of the partners satisfy
u (x) + v (y) ≥ h(x, y) + θ ∀x, y, (8.9)
with equality if the partners are married to each other and inequality if
they are not.3 The utility levels v(x) and u(y) that satisfy (8.9) can be
interpreted as the demand prices that men with income x and women with
income y require to participate in any marriage. Marriages that form are
consistent with the demands of both partners and exhaust family resources.
Marriages that do not form are those in which resources are insufficient to
satisfy the demands of both partners.
2 Obviously, the support could be changed to any intervals [a, A] and [b, B] - the only
cost being more tetious notations.

3 Note that by deducting h(y, 0) + h(0, z) from both sides of equation (8.9) it can be
written, equivalently, in terms of the surplus that the marriage generates, relative to
remaining single. Also, because the values of remaining single are independent of the
assignment, the condition for stable assignment can be formulated as maximization of
the aggregate surplus.
In particular, (8.9) implies that

u (x) = θ + max (h(x, y) − v (y)) ,
y
v (y) = θ + max (h(x, y) − u (x)) . (8.10)
x
That is, each partner gets the spouse that maximizes his/her “profit” from
the partnership, taking into account the reservation utility (the ‘price’) of
any potential spouse. The first order conditions for the maximizations in
(8.10) give:
v0 (y) = hy (φ (y) , y),
u0 (x) = hx (x, ψ (x)). (8.11)
These equations have an important implication - namely that, as we move
across matched couples, the welfare of each partner changes according to
the marginal contribution of his/her own income to the marital output,
irrespective of the potential impact on the partner whom one marries. The
reason for this result is that, with a continuum of agents, there are no rents
in the marriage market, because everyone receives roughly what he\she
would obtain in the best next alternative.4 Therefore, a change in mari-
tal status as a consequence of a marginal change in income has negligible
impact on welfare, and the only gain that one receives is the marginal
contribution of one’s own trait. Although the change of spouse provides
no additional utility, the spouse that one has influences the marginal gain
from an increase in own traits, reflecting the interactions between the traits
in the production of marital output.
Another important condition that needs to be satisfied in a stable assign-
ment is that, if there are unmarried men, the poorest married man (whose
income is denoted x0 ) cannot get any surplus from marriage. Similarly, if
there are unmarried women, the poorest married woman (whose income is
denoted y0 ) cannot get any surplus from marriage. Otherwise, the unmar-
ried men or women who are slightly less rich could bid away the marginal
match. This condition exploits the assumption that there is a continuum
of agents. Hence, if r < 1 then u (x0 ) = h (x0 , 0) and v(0) = θ. Conversely,
if r > 1 then v (y0 ) = h (0, y0 ) and u(0) = θ. If r = 1, then any allocation
of the gains in the least attractive match with x = y = 0 that satisfies
u(0) + v(0) = θ is possible.
This initial disparity between the two spouses is modified as they move
up the assignment profile. The main features that influence the evolution
of utility differences within couples are the local scarcity of males and fe-
males at different levels of incomes and the strength of the interaction
4 The absence of rents must be distinguished from the positive surplus that the mar-
riage creates. A positive surplus, h(y, z) + θ > h(y, 0) + h(0, z), simply means that there
are positive gains from marriage, relative to the situation in which both partners become
single, but this is rarely the best next alternative.
in traits. Assuming, for instance, that r > 1 and all men are married
then marriages can be indexed by the husband’s income. As one moves
across all married couples, the utility of the husbands rises at the rate
du(x)
dx = hx (x, ψ (x)) , while the utility of their assigned wives rises at the
rate dv(y) dy 0
dy dx = hy (x, ψ (x)) ψ (x). In this case, if men are everywhere lo-
0
cally scarce (that is, ψ (x) < 1), then the utility of the husband rises faster
than the utility of the wife. Conversely, if there are less women than men
(r < 1) and women are everywhere locally scarce (that is, φ0 (y) < 1), the
utility of the wife rises faster than the utility of the husband. Intuitively, an
overall scarcity of men benefits men at the top of the income distribution
to a larger extent because these men are desired by all women; by the same
token, an overall scarcity of women benefits the women at the top of the
income distribution to a larger extent, because these women are desired by
all men.
Integrating the expressions in (8.11) and using the boundary conditions
described above, one can obtain a unique allocation rule, provided that
r 6= 1. Basically, one first finds the allocation in the least attractive match,
in which the minority type has no income, using the no rent condition.
Then, the division in better marriages is determined sequentially, using
the condition that along the stable matching profile each partner receives
his\her marginal contribution to the marital output. The key remark is that
the allocation rule is fully determined by the sex ratio r and the respective
income distributions of the two sexes. The incomes of the partners in a
particular marriage have no direct impact on the shares of the two partners,
because the matching is endogenously determined by the requirements of
stable matching.
Technically, therefore, assuming for instance r > 1:
Z y
v (y) = h (0, y0 ) + hy (φ (t) , t) dt,
y0
Z x
u (x) = θ + hx (s, ψ (s)) ds,
0
y0 = Ψ (1 − 1/r) , (8.12)
(and analogous conditions can readily be derived for for r < 1). If r = 1,
Z y
v (y) = k + hy (t, φ (t))dt,
Z0 x
u (x) = k0 + hx (ψ (s) , s)ds,
0
k + k0 = θ. (8.13)
where k and k0 are arbitrary.
The first terms in the RHS of equations (8.12) and (8.13) are the utilities
of the partners in the match of the lowest quality and the integrals describe
the accumulated marginal changes, as we move up the stable assignment

profile to marriages with higher incomes. Because of the interaction in
traits, the change in the marital contribution depends on the income of
the spouse that one gets. Note that marginal increases in x0 or y0 have
no effect on u(x) or v (y) , respectively, because the marginal persons with
these incomes are just indifferent between marrying and remaining single.
In marriages that involve individuals from the bottom of the male and
female income distributions, members of the larger sex group typically have
higher income. Thus, if r > 1 and all men are married, the men in the lowest
quality matches have almost no income, while their wives have strictly
positive income. The wife receives her utility as single h (0, y0 ) and the
husband receives the remaining marital output. If r = 1, the allocation
in the lowest quality match is indeterminate and, consequently, there is a
whole set of possible sharing rules that differ by a constant of integration.
8.2.2 A tractable specification

Let us now slightly generalize our previous approach by assuming that male
incomes x are distributed on a support [a, A] according to some distribution
F and female incomes y are distributed on a support [b, B] according to
some distribution G; the assumption of different supports for men and
women is useful for empirical applications. We introduce now a simplifying
assumption, namely that the output function h depends only on total family
income. That is,
h(y, x) = H(y + x), (8.14)
with H (0) = 0. This assumption makes sense in our transferable utility

setting, since under TU a couple behaves as a single decision maker. Note
that basically all examples of intrahousehold allocation with TU given in
Chapter 3 satisfy this property.
Under this assumption, hy (y, x) = hx (y, x) = H 0 (y + x), and assortative
matching requires hxy (y, x) = H 00 (y + x) > 0, so that H is increasing and
convex. As above, we let ψ (x) (resp. φ (y)) denote the income of Mr. x’s
(Mrs. y’s) spouse. Finally, we maintain the convention that a single person
with income s (= x, y) achieves a utility level H (s).
We are interested in how changes in the sex ratio and the distributions
of income of the two sexes affect the allocation rule that is associated with
a stable matching. In this analysis, we shall distinguish between two issues:
(i) the shape of the allocation rule in a cross section of marriages - that is,
how do the shares vary as we move up the assignment profile to couples
with higher incomes, and (ii) changes of the allocation rule as parameters
of the marriage market, such as the sex ratio or the male and female income
distribution, change.
Allocation of marital output: general properties

We start by analyzing the properties of the allocation of marital output
between spouses, as described by equations (8.12) and (8.13). In the lowest
quality matches, the partner that belongs to the majority group has higher
income than the minimum of the corresponding income distribution, but,
because of competition with lower income singles, receives no rent, and has
the same income as a single. In contrast, the partner that belongs to the
minority group receives a rent that equals to the total surplus generated,
because there are no lower income singles to compete with. These properties
exactly define the allocation of welfare between the spouses.
Under assumption (8.14), equation (8.11) becomes:
du(x)
= H ‘ (x + ψ (x)) , (8.15)
dx
dv(y)
= H ‘ (φ (y) + y) (8.16)
dy
Therefore, for any married couple (x, ψ (x)) = (φ (y) , y),
du(x) dv(y)
= .
dx dy
In words: the return, in terms of intrahousehold allocation of marital out-

put, of an additional dollar of income is the same for males and females.
This symmetry between genders, however, is not maintained when moving
from one couple to another, because in general the change in husband’s
income between the couples does not equal to the change in the wife’s in-
come - reflecting the local scarcity of the respective genders, as discussed
in Chapter 7.
Integrating (8.15) and (8.16), and assuming for instance that r > 1, we
have:
Z x
u (x) = H (a + y0 ) − H (y0 ) + θ + H 0 (s + ψ (s)) ds (8.17)
a
Z y
0
v (y) = H (y0 ) + H (φ (t) + t) dt (8.18)
y0
Again, since women are assumed to be on the long side of the market,
the poorest married woman, with income y0 , must be indifferent between
marriage and singlehood; all the surplus generated by her marriage, namely
H (a + y0 ) − H (y0 ) − H (a) + θ, goes to the husband, generating a utility
H (a + y0 ) − H (y0 ) + θ. Moving up along the income distributions, the
allocation evolves as described by (8.15) and (8.16).
G F
FIGURE 8.3. A linear upward shift
The case r < 1 is similar and gives:

Z x
u (x) = H (x0 ) + H 0 (s + ψ (s)) ds (8.19)
x0
Z y
v (y) = H (x0 + b) − H (x0 ) + θ + H 0 (φ (t) + t) dt (8.20)
b
Linear shifts of distributions

We now introduce an additional assumption that considerably simplifies
the analysis. Specifically, we assume that (i) there are as many men as
women (r = 1), and (ii) that men’s income distribution is a linear upward
shift of the income distribution of women. That is,
F (t) = G (αt − β) for all t (8.21)
for some α < 1, β > 0. An illustration is provided in Figure 8.3.

This condition is satisfied, for instance, if the income distributions of both
men and women are lognormally distributed with parameters (μM , σ M ) for
males and (μF , σ F ) for females, under the condition that σ M = σ F - a form
that fits existing data pretty well.5
5 Alternatively, the property is also satisfied if the two income distributions are uni-
form and the support of the male distribution is [a, A] while the support of the female
distribution is [b, B]; then b = αa − β and B = αA − β.
The linear shift property implies that, under assortative matching and
with populations of equal size, a man with income x is paired with a woman
with income y = αx − β. With the previous notations, therefore, φ (y) =
(y + β) /α and ψ (x) = αx − β. Equations (8.15) and (8.16) then become
du(x)
= H 0 ((α + 1) x − β) , (8.22)
dx
and
dv(y)
= H 0 (((α + 1) y + β) /α) , (8.23)
dy
yielding upon integration :
α
v (y) = K + H (φ (y) + y) (8.24)
1+α
and
1
u (x) = K 0 + H (x + ψ (x)) (8.25)
1+α
where
K + K 0 = θ.
In words, the marriage between Mr. x and Mrs. y = ψ (x) generates
a marital output θ + H (x + ψ (x)), which is divided linearly between the
spouses. The non monetary part, θ, is distributed between them (he receives
K, she receives K 0 ) in a way that is not determined by the equilibrium
conditions (this is the standard indeterminacy when r = 1) but must be
the same for all couples (note that K or K 0 may be negative). Regarding
the economic output, however, the allocation rule is particularly simple; he
receives some constant share α/ (1 + α) of it, and she gets the remaining
1/ (1 + α).
8.2.3 Comparative Statics

We now turn to examine the impact of changes in the sex ratio and income
distribution.
Increasing the proportion of women

We begin by noting an important feature of the model, namely that if
all marriages yield a strictly positive surplus then the allocation rule has
a discontinuity at r = 1. Indeed, examining the expressions in (8.17) and
(8.18) we see that if r approaches 1 from above we get in the limit
Z x
u (x) = H (a + b) − H (b) + θ + H 0 (s + ψ (s)) ds (8.26)
a
Z y
v (y) = H (b) + H 0 (φ (t) + t) dt (8.27)
b
while if r approaches 1 from below we get in the limit,
Z x
u (x) = H (a) + H 0 (s + ψ (s)) ds (8.28)
a
Z y
v (y) = H (a + b) − H (a) + θ + H 0 (φ (t) + t) dt (8.29)
b
The marital surplus generated by the marriage of lowest income couple,

here (a, b), is equal to H (a + b) −H (a)− H (b) +θ. When the two sexes are
almost equal in number, a small change in the sex ratio shifts all the surplus
to one of the partners in the lowest quality match, the one whose sex is in
the minority, and this discontinuity is then transmitted up the matching
profile to all participants in the marriage market.6 This knife-edge property
is an undesirable property of the simple model without friction. One can
get rid of it either by assuming no rents for couples at the bottom of the
distribution, or by limiting our attention to marginal changes in the ranges
r > 1 or r < 1 which do not reverse the sign of these inequalities.
Consider, now, such a change - that is, a marginal increase in the pro-
portion of women r that maintains either r > 1 or r < 1 and assume that
the shape of of income distributions of both men and women remain un-
changed. From the matching rule 1 − F (x) = r (1 − G (y)) , we see that,
as a consequence of such change, any married man with a given income x
will now be matched with a woman with a higher y and each woman with
a given y, is now matched to a man with a lower x. That is, the match-
ing function ψ (x) shifts upwards and the matching function φ (y) moves
downwards. As we move along a stable assignment profile, the utility of
all married men grows with their own income at a higher rate, because
hx (x, ψ (x)) = H 0 (x + ψ (x)) is higher for all x and the utility of all mar-
ried women grows at slower rate because hy (φ (y) , y) = H 0 (φ (y) + y) is
lower for all y. It then follows from (8.15) and (8.16) that the utility of all
married men rises and the utility of all married women declines; those who
remain single are unaffected. Assuming, for instance, that r > 1, we have:
φ (y, r) = Φ [1 − r (1 − G (y))]
∙ ¸
1
ψ (x, r) = Ψ 1 − (1 − F (x))
r
y0 = Ψ (1 − 1/r) ,
6 The result that g is the only source of gain from marriage for couples at the bottom
of the income distribution reflects the assumptions that h(0, 0) = 0 and that there is a
positive density of the income distribution at zero. In general, participants at the bottom
of the income distribution have a positive income, so that the lowest quality match may
create a monetary surplus, because of the positive interaction of traits.
and
∂φ (y, r)
= − (1 − G(y) Φ0 [1 − r (1 − G (y))] < 0
∂r ∙ ¸
∂ψ (x, r) 1 0 1
= (1 − F (x)) Ψ 1 − (1 − F (x)) >0
∂r r2 r
∂y0 1 0
= Ψ (1 − 1/r) > 0.
∂r r2
Differentiating (8.17) and (8.18) with respect to r therefore gives:
∂u (x) ∂y0
= (H 0 (a + y0 ) − H 0 (y0 ))
Z x ∂r ∂r
00 ∂ψ (s, r)
+ H (s + ψ (s)) ds > 0
a ∂r
∂v (y) ¡ 0 ¢ ∂y0
= H (y0 ) − H ‘ (y0 + a)
Z y ∂r ∂r
00 ∂φ (t, r)
+ H (φ (t) + t) dt < 0
y0 ∂r
The case r < 1 is similar and left to the reader.

We conclude:
A marginal increase in the proportion of women to men in the marriage

market, improves (or leaves unchanged) the welfare of all men and reduces
(or leaves unchanged) the welfare of all women; the impact is stronger for
higher income households.
An important implication of this property is that for any couple, the sex
ratio can be used as a distribution factor : its variations affect the intra-
household allocation of resources without changing neither total income
nor the spouses’ preferences. The empirical relevance of this remark has
been established empirically by several authors. For instance, Chiappori,
Fortin and Lacroix (2002), using a collective model of labor supply, find
that, other things equal, a one percentage point increase in the sex ratio
(defined as the ratio of men to women in the relevant marriage market)
induces husbands to transfer some 2, 000 dollars (1988) of income to their
spouse (see Chapter 5).
Shifting female income upward

Recalling our assumption that men have the higher income in the sense
that their distribution dominates in the first degree the income distrib-
ution of women, that is, F (t) < G(t) for t ∈ (0, 1), we now consider a
first degree upward shift in the distribution of female income, holding the
male distribution constant. That is, the proportion of females with incomes
exceeding y rises for all y, so that women become more similar to men in
terms of their income, as we observe in practice. Such an upward first order
shift in the distribution of female income affects the matching functions in
exactly the same way as a marginal increase in the female/male sex ratio.
Thus, if all men maintain their income, they all become better off. Similarly,
any woman who would maintain her income would become worse off. This
remark should however be interpreted with care, because it is obviously
impossible for all women to maintain their income: when the distribution
of female incomes shifts to the right, some (and possibly all) females must
have higher income. In particular, those women who maintain their rela-
tive rank (quantile) in the distribution will maintain their position in the
competition for men, and will be matched with a husband with the same
income as before. Such women will be better off, as a consequence of the
increase in their own income.
As a special case, consider the linear shift case described above; to keep
things simple, assume moreover that β = 0. Suppose, now, that the income
of every woman is inflated by some common factor k > 1 and consider
a married couple with initial incomes (x, y). After the shift, the partners
remain married but the wife’s income is boosted to ky while the husband’s
income remains equal to x. If uk and vk denote the new individual utilities,
we have from (8.25) and (8.24):
kα
vk = K+ H (ky + x) and (8.30)
kα + 1
1
uk = K0 + H (ky + x)
kα + 1
Differentiating in k around k = 1 gives:
∂vk α αy
= H (y + x) + H 0 (y + x) and
∂k (α + 1)2 α+1
∂uk α y 0
= − 2 H (y + x) + α + 1 H (y + x) (8.31)
∂k (α + 1)
One can readily check that both changes are positive (for the second one,
it stems from the convexity of H). We conclude that the shift has two im-
pacts. First, the increase in total income generates some additional surplus
(the term in yH 0 (y + x)), which is shared between spouse in proportion of
their respective incomes (that is 1 and α). In addition, a redistribution is
triggered by the shift. Specifically, since the wife’s share of total income is
increased, so is her consumption; the husband therefore transfers to his wife
2
an amount equal to a fraction α/ (α + 1) of total surplus. One can readily
check that the transfer is proportionally larger for wealthier couples, since
the ratio H (y + x) / (y + x) increases with (y + x) due to the convexity of
H.
Empirical illustration
It is a priori not clear how important is income for matching and how to
measure it. Actual incomes are rarely available, and wages are measured
with a lot of noise and vary over the life cycle. For an empirical application,
we estimate the predicted hourly wage of white men and women aged 25-40
in the CPS data and use these predictions as measures of the male and
female incomes for this age group.7 We then obtain the following results
(see Figure 8.11):
1. The log normal distribution fits these predictions well.
2. The standard deviations for men’s and women’s predicted log wages
are similar and both grow over time,
3. The mean predicted log wages of men are higher than for women but
the discrepancy declines over time.
4 Male distributions of predicted log wage dominate in the first degree
the female distributions in all years but the gap declines over time.
5. Within couples, there is high positive correlation between the pre-
dicted log wages of husbands and wives and this correlation rises over time
indicating a high and increasing degree of positive assortative mating.
6. Finally, the ratio of men to women in the CPS sample of whites aged
25 − 40 has dropped from 1.045 in 1976 to 0.984 in 2005.
Figure 8.12 shows the male and female income distributions estimated
from the CPS data. As seen the cumulative distribution of male incomes
is below the cumulative distribution of female in both years but the gap is
lower in 2005, indicating a first degree dominance of the male distributions.
For both man and women, the cumulative distributions are less steep in
2005, representing the general rise in inequality between 1976 and 2005.
We use this information, together with the assumption that the mari-
2
tal output is given by h(x, y) = (x+y) 4 (so that the marital surplus from
marriage is yx2 ), to calculate the predicted response of the shares in marital
surplus to the observed changes in the male and female income distributions
and in the sex ratio between the years 1976-2005. The use of log normal dis-
tribution and the specification of h(x, y) allows us to use conditions (8.21)
and (8.14) and to calculate the shares using numerical integrations of (8.15)
and (8.16). Figures 8.13 and 8.14 show the estimated shares in the marital
surplus for men and women in 1976 and 2005. We see that men had a larger
estimated share in 1976, while women had the larger share in 2005. Part
7 These results are obtained by running regressions with every year for white men
and women aged 25 − 40 of wages on schooling experience and occupation, excluding

self employed. We use up to 53 occupation dummies, which allows for a large variance
given schooling and age and also captures a more permanent feature of wages because
an occupation tend to relatively stable over the life cycle. For men and women who
reported no occupation, we imputed the mode occupation in their schooling group. For
men who did not work, we imputed wages conditioned on working and for women we
also corrected for selection using the Heckman technique.
of this reversal is due to the narrowing wage gap between men and women
and part of it is due to the reduction in the female-male sex ratio over the
period.
8.2.4 Taxation
Changes in the income distribution can also arise from a government inter-
vention in the form of taxes and subsidies. For instance, we may consider a
linear transfer scheme, such that the after tax (subsidy) income of a person
with income s is κ + (1 − τ ) s, with κ > 0 and 0 < τ < 1. Let us assume
that the scheme is revenue neutral, so that its only impact is to redistribute
income between and within couples, and let us for the time being disregard
behavioral responses to the tax changes. We have that:
Z 1 Z 1
xF (x) dx + r yG (y) dy
0 0
Z 1 Z 1
= (κ + (1 − τ ) x) F (x) dx + r (κ + (1 − τ ) y) G (y) dy(8.32)
0 0
and
x̄ + rȳ
κ=τ (8.33)
1+r
R1 R1
when x̄ = 0 xF (x) dx, ȳ = 0 yG (y) dy denote average incomes of male
and females, respectively, so that x̄+r ȳ
1+r is average household income. Here,
τ is the taxation rate, and κ is the lump sum subsidy funded by income
taxation.
We can think of such an intervention as a change in the household pro-
duction function from h (x, y) to h̃ (x, y) = h (κ + (1 − τ ) x, κ + (1 − τ ) y).
Such a transformation preserves the sign of the cross derivative with re-
spect to the before tax incomes x and y. Therefore, the same pattern of a
positive assortative mating is maintained and the matching functions ψ (x)
and φ(y) remain the same. However, the introduction of tax and transfer
influences the gains from marriage, which depend on the after tax incomes
of the partners, and the division of these gains. By construction, a pro-
gressive transfer-tax system raises the income of the poor and reduces the
income of the rich. Due to positive assortative matching, the progressivity
of the program is magnified, because an individual whose after tax income
has increased (decreased) is typically assigned to a spouse whose after tax
income has increased (decreased). Put differently, the intervention affects
the surplus generated by marriage, holding the pre tax incomes fixed. For
low income matches, the surplus increases and for high income matches
it declines. In addition, the division of the surplus between husbands and
wives is affected in general.
When assumption (8.14) holds and only total family income matters,
the household production function is modified from h (x, y) = H (x + y) to
h̃ (x + y) ≡ H̃(2κ + (1 − τ ) (x + y)). Assume, in addition, that male and

female income distributions satisfy condition (8.21) so that ψ (x) = αx − β
and φ (y) = y+βα . Then, for a larger female population (r > 1), utilities
become:
ṽ (y) = H (κ + (1 − τ ) y0 ) +
Z y+φ(y)
αds
(1 − τ ) H 0 (κ + (1 − τ ) s)
0 α+1
Z ψ(x)+x
ds
ũ (x) = θ + (1 − τ ) H 0 (κ + (1 − τ ) s) (8.34)
0 α +1
with k = τ x̄+r ȳ
1+r . The impact of a change in the marginal income tax, t,
on the utilities of women and men, respectively, is
µ ¶
∂ṽ (y) x + ry
= − y0 H 0 (α + βy0 ) − αD (x + y)
∂t 1+r
∂ ũ (x)
= −D (x + y) , (8.35)
∂t
where Y denotes total family income and
Z Y
ds
D (Y ) = H 0 (κ + (1 − τ ) s)
0 α+1
Z Yµ ¶
x + ry ds
+ (1 − τ ) s− H 00 (κ + (1 − τ ) s) (8.36)
.
0 1 + r α +1
The term D (Y ) is typically positive for richer households (who there-
fore lose from the introduction of the tax/benefit system) and negative
for poorer ones. In this simple context, the corresponding gain or loss is
allocated between husband and wife in respective proportions 1 and α.
In addition, since single women are at the bottom of the female income
distribution, their utility is typically increased by the tax/benefit scheme
(this is the case whenever their income is below the mean). Equilibrium
then requires the gain of the marginal woman (that is, of the wealthiest
single or poorest married woman)³ to be´ forwarded to all women in the
x+ry
distribution; hence the term 1+r − y0 H 0 (α + βy0 ) in equation (8.35)
representing the gain of the marginal woman. Note that this boost in in-
come does not go to the poorer spouse, but to the spouse whose population
is in excess supply. Should males outnumber females (r < 1), they would
receive the corresponding benefit. The precise impact of these changes is
hard to evaluate in general and we therefore turn to a specific example.
8.2.5 An example
We now provide a simple example in which the shares can be easily cal-
culated. In addition to (8.21) we assume that incomes are uniformly dis-
tributed. We use again our example in Chapter 2 with public goods where
2
h(y, x) = (y+x)
4 , which satisfies (8.14). For this example, men and women
have the same marginal contribution to marriage, hx (y, x) = hy (y, x) =
y+x
2 . Assume that the incomes of men and women are uniformly distrib-
uted on [0, 1] and [0, Z] , respectively, where Z ≤ 1. If Z < 1, then the
income distribution of men dominates in a first degree the income distrib-
ution of women, because
½ t
Z if 0 ≤ t ≤ Z
G(t) = (8.37)
1 if Z < t ≤ 1
exceeds F (t) = t, for all t in the interval (0, 1). We are also in the ‘linear
upward shift’ case described above, with α = Z and β = 0.To simplify
further, we set θ = 0 so that the lowest quality matches generate no surplus.
Therefore, there is no indeterminacy of the allocation rule when r = 1 and
no discontinuity in the allocation rule.
Under the assumed uniform distributions, the assignment functions are
linear and given by
y
x = φ (y) = 1 − r(1 − ), (8.38)
Z
Z
y = ψ (x) = [(r − 1) + x]. (8.39)
r
and the local scarcity of men is constant and given by Zr . Under the sim-
plifying assumption that θ = 0, the shares of the husband and wife in the
marital output can then be rewritten in the form
Z
y2 1 y t
v(y) = + [1 − r(1 − )]dt,
4 2 y0 Z
Z
x2 1 x Z
u(x) = + [(r − 1) + s]ds,
4 2 x0 r
½ Z
y0 = r (r − 1) if r>1
,
0 if r ≤ 1
½
1 − r if r < 1
x0 = . (8.40)
0 if r ≥ 1
2 2
Notice that v(y) − y4 and u(x) − x4 are the shares of the husband and
wife in the marital surplus. Inspecting the integrals in (8.40), we see that
the gender in short supply always receives a larger share of the surplus. In
contrast, the shares of marital output of husbands and wives depend also
on the location of the couple in the income distribution.
If there are more women than men, r > 1, the match with the lowest
output is the one in which the husband has income x = 0, and the wife has
income y0 = Z (r−1)
r . His surplus and utility are at this point zero, while
y02
she receives the whole marital output 4 , which also equals her utility as
single. Because men are always locally scarce, Zr > 1, it follows from (8.40)
that their utility must grow along the stable assignment at a faster rate
than the utility of their assigned wives. It is readily seen that the husband’s
share is higher in matches with sufficiently high income. In particular, the
2
best match with x = 1 and y = Z, yields an output of (1+Z) 4 , of which
1 Z Z Z2 Z
the husband receives 4 + 2 − 4r and the wife receives 4 + 4r , which is a
smaller share.
If there are more men than women, r < 1, the match with the lowest
output is the one in which the wife has income y = 0, and the husband’s
income is x0 = 1 − r, and it is now the wife that has the lower utility. The
local scarcity parameter can now be higher or lower than 1. If Zr > 1, men
are always locally scarce, and it follows from (8.40) that the husband will
have a higher share in the output of all marriages. If, however, Zr < 1 and
women are always locally scarce, then the utility of women grows along the
stable assignment profile at a faster rate than the utility of their assigned
husbands, and they may eventually overtake them. Indeed, the wife’s share
2
in the best match is Z2 + Z4 − rZ 1 rZ
4 and the husband’s share is 4 + 4 , which
is smaller if r is sufficiently small.
This example illustrates clearly the impact of changes in the sex ratio
r and the distribution of female income as indexed by Z, on the welfare
of women and men. Recall that marginal increases in x0 or y0 have no
effect on u(x) or v (y) , respectively. Inspection of the integrands in (8.40),
shows that u(x) must increase in r and Z, while v (y) must decrease in r
and Z. As we noted above, the result that women are worse off when the
mean income of women rises sounds surprising. However, the reason that a
woman who maintains her income is worse off when Z rises is that there are
more women with income above her, which means that she cannot ”afford”
anymore a husband with the same income as before. However, any woman
who keeps her position in the income distribution, (that is, whose income
rose at the same proportion as Z) will obtain a husband with the same
x as before the change. Then it can be shown that if r > 1, her surplus
does not change, and if r < 1, her surplus rises.8 In either case, her welfare
must rise, reflecting the rise in her own income. This example can be easily
generalized for the case in which there are positive non monetary gains,
θ > 0.
The example allows us to examine numerically the impact of a progressive
transfer-tax system. Assume that male income is distributed uniformly on
[0, 1], while the female income is distributed uniformly on [0, 75]. Set θ =
0.025 and τ = .7. Now consider a balanced transfer scheme such that
κ(1 + r) = (1 − τ )(x + ry). We discuss here two separate cases, one in which
women are the majority and r = 1.1 and the other when women are the
minority and r = 0.9. In the numerical example, x = 0.5 and y = 0.375.
Thus, for a marginal tax of τ = 0.7, the balanced budget constraint implies
that κ = 0.13 when r = 1.1, and κ = 0.132. when r = 0.9.Figures 8.15,
8.16 and Table 8.1 summarize the results.
When women are in the majority, their share is usually less than half
but rising in the income of their assigned husband (see Figure 8.15). The
tax-subsidy intervention moderates this increase, because in low quality
matches, the wife’s share is determined by her income, and women with low
income gain from the progressive system. When women are in the minority,
their share in the marital output declines and the progressive tax system
moderates this decline (see Figure 8.16) because in low quality matches,
the husband’s share is determined by his income, and men with low income
gain from the progressive system. The difference in slopes between the two
figures reflects the role of the non monetary gains, θ, that are captured by
the men when r > 1 and by the women when r < 1. This effect weakens
as one moves to high income couples where the monetary gains become
increasingly important.
Table 8.1 provides the numerical values of the shares. In the benchmark:
the income of men is uniform on [0, 1], the income of women is uniform
on [0, .75], the gain from marriage is g = 0.025,the tax rate on income is
β = 0.7 and the implied value of α that balances the budget is α = 0.1322
at panel a and α = 0.1303 at panel b. We then examine the equilibrium
8 The surplus of the husband and the surplus of the wife are readily obtained by
calculating the integrals in (8.40). For r ≥ 1, we obtain
y2 Zy 2 1Z
sh (y) = u(y) − = + (r − 1)y,
4 4r 2 r
ψ (y) y Zy 2
sw (ψ (y)) = − sh (y) = .
2 4r
For r ≤ 1, we obtain
z2 rz 2 z(1 − r)
sw (z) = v(z) − = + ,
4 4Z 2
φ (z) z rz 2
sh (φ (z)) = − sw (z) = .
2 4Z
shares for some hypothetical couples. Panel a describes the case with more
women than men, r = 1.1. Then, all men marry and a proportion 0.9091 of
the women remains single. The man with the lowest income, 0, is matched
with a woman whose income is 0.0682, the man with the mean income,
0.5, is matched with a woman whose income is 0.4091, and the man with
the highest income, 1, is matched to the woman with highest income, 0.75.
Following the intervention; the after tax income of the man with lowest
income rises to 0.1304, and that of his matched wife rises to 0.1781, the
after tax income of the average man is reduced to 0.4804 and that of his
matched wife rises to 0.3947, while the after tax of the wealthiest man is
reduced to 0.8304 and that of his matched wife is reduced to 0.6554. Thus,
the tax and transfers scheme reduces inequality both between and within
couples.
Although the impact of the intervention on the couples with the average
man or average woman is relatively small, some noticeable changes occur
at the bottom and the top of the income distribution. At the bottom, the
intervention raises the utilities of both men and women but women obtain a
larger share of the total utility if r > 1 and a smaller share if r < 1. It seems
surprising that a progressive policy that transfers resources to poor women
reduces their share in the marital surplus. But when r < 1, poor women are
married to men who are wealthier than they are, and the intervention makes
these men less "useful" to their wives. At the top of the distribution, the
intervention lowers substantially the utilities of both men and women but
women gain relatively more than men if r < 1 and relatively less if r > 1. We
see that the impact of the tax-subsidy intervention on each spouse reflects
three different effects: an increase (decrease) in own income, an increase
(decrease) in the spouse’s income, and the increase in the incomes of the
individuals who are just indifferent between marriage and singlehood. The
first two effects influence the marital output that the matched partners can
generate together. The third effect reflects the changes in the sharing of this
output that are caused by the competition in the marriage market. In order
to separate these effects, we examine the impact of the tax for couples for
which the intervention does not affect total family income, and, therefore,
marital output does not change. This comparison is shown in panels c and
d of Table 8.1. We see that in both panels the wife gains income relative to
the husband. However, when women are in the majority, the wife in such
couples loses both in output and surplus terms. In contrast, the wife gains
if women are in the minority. This difference can be traced to the impact
of the intervention on the lowest quality matches, where the intervention
causes a larger gain to the wife than to the husband when women are in
the minority, r < 1, while the opposite is true when r > 1 (see panels a and
b). These effects are transmitted along the matching profile to all couples
in the marriage market.
The general conclusion that one can draw from these examples is that
in a frictionless market, where the shares are determined jointly with the
assignments, the simple intuition based on bargaining between two iso-

lated partners fails. For instance, we see in panel c that, although family
income remains fixed and the wife’s share in the total income rises, she
ends up with lower share of marital output. In other words, the allocation
rule that determines the wife’s and husband’s utility in a particular mar-
riage, reflects the traits of all participants in the marriage markets and,
therefore, a change in the income distribution in the economy (society) at
large can change the shares within specific marriages in a way that would
not be directly predictable from the change in the within-household income
distribution.
Table 8.1: Sharing of Marital Output and Surplus
Panel a: Women are the Majority, r = 1.1

U. Sh S. Sh
Hus. I. Wife I. Tax Hus. U. Wife U. Hus. S. Wife S.
W/(W+H) W/(W+H)
0 0.0682 no 0.025 0.0012 0.025 0 0.0444 0
0.1304 0.1781 yes 0.0409 0.0079 0.0366 0 0.1625 0
0.5 0.4091 no 0.1472 0.0845 0.0847 0.0426 0.3646 0.3348
0.4804 0.4167 yes 0.1463 0.0798 0.0887 0.0364 0.353 0.2913
1 0.75 no 0.4795 0.3111 0.2295 0.1705 0.3935 0.4261
0.8304 0.6554 yes 0.3548 0.222 0.1825 0.1146 0.3849 0.3858
Panel b: Men are the Majority, r = 0.9

U. Sh S. Sh
W/(W+H) W/(W+H)
0.1 0 no 0.0025 0.025 0 0.025 0.9091 1
0.2022 0.1322 yes 0.0102 0.0427 0 0.0384 0.807 1
0.55 0.375 no 0.1178 0.1211 0.0422 0.0859 0.5069 0.6707
0.5172 0.3947 yes 0.1084 0.1245 0.0415 0.0856 0.5347 0.6735
1 0.75 no 0.4188 0.3719 0.1688 0.2313 0.4704 0.5781
0.8322 0.6572 yes 0.2975 0.2821 0.1243 0.1741 0.4867 0.5834
Panel c: Women are the Majority, r = 1.1

U. Sh S. Sh
W/(W+H) W/(W+H)
0.4762 0.3929 no 0.1366 0.0772 0.0799 0.0387 0.3612 0.3261
0.4637 0.4054 yes 0.139 0.0748 0.0852 0.0338 0.35 0.2837
Panel d: Men are the Majority, r = 0.9

U. Sh S. Sh
W/(W+H) W/(W+H)
0.5264 0.3553 no 0.1072 0.1122 0.0379 0.0806 0.5115 0.6804
0.5007 0.381 yes 0.101 0.1184 0.0383 0.0821 0.5396 0.6819
8.2.6 Matching on preferences: Roe vs. Wade and female

empowerment
In the matching models presented so far, income is the trait on which
people match. But other determinants can also be considered. In a recent
paper, Chiappori and Oreffice (2007) use a matching model to analyze the
impact of the legalization of abortion on power allocation within couples.9
In their framework, people differ in their preferences towards children; the
corresponding matching patterns - and the resulting allocation of resources
- can be studied before and after legalization.
That the legalization of abortion should alter the balance of powers
within couples is not surprising;10 indeed, Oreffice (2007) has provided
an empirical study based on the collective approach to household behav-
ior, that confirms the ‘empowerment’ consequences of Roe vs.. Wade. Still,
the mechanism by which this empowerment occurs deserves some scrutiny.
While it is not hard to convince oneself that some women (for example,
career-oriented women with little taste for family life) will gain from le-
galization, whether all women will is another matter. A strong objection
is that women have heterogeneous preferences for fertility (or different at-
titudes toward abortion); some do not consider abortion as an option, ei-
ther for religious and ethical reasons or because they do want children.
Whether the legalization will benefit these women as well is not clear.
From an economist’s perspective, moreover, the new context will affect the
matching process on the market for marriage, and in particular the way
the surplus generated by marriage is shared between spouses. In principle,
such ‘general equilibrium’ effects could annihilate or even reverse the di-
rect impact of the reform, particularly for these women who are unlikely
to derive much direct benefit from it.
Preferences
To investigate these issues, Chiappori and Oreffice consider a model in
which a continuum of men and women derive utility from one private com-
posite good c (the price of which is normalized to 1) and from children; Let
the dummy variable k denote the presence (k = 1) or the absence (k = 0) of
children in the household. Men have identical, quasi-linear preferences over
consumption and children. The utility of single men only depends on their
consumption; that is, men cannot derive utility from (and do not share the
9 The version presented here is a slightly simplified version of the original paper;
in particular, we assume here that men have identical preferences, and concentrate on
preference heterogeneity among women.
1 0 See for instance Héritier (2002).
costs of) out-of-wedlock children, due to the fact that they do not live in
the same household. On the other hand, married men’s utility is of the form
UH (cH , k) = cH +uH k, where the parameter uH > 0 is identical for all men
in the economy. Women differ in their preferences toward children. Specif-
ically, female utility functions take the quasi-linear form U (c, k) = c + uk.
Here, each woman is characterized by the individual-specific taste para-
meter u, which is distributed according to the density f over the interval
[0, U ]. We assume that any woman (single or married) who wants a child
can have one. However, if she plans to have no children, unwanted births
may still occur with some probability p, which depends on the available
contraceptive technology and the legality of abortion.
The quasilinear structure of the male and female preferences implies that
utility is transferable within marriage. For each spouse, the utility depends
on the couple’s fertility decision and on the share of composite good that
he or she receives.
As before, we normalize the mass of men to be 1, and we denote by r
the total mass of women on the market; here, we assume that r > 1, that
is that women are on the long side of the market. Male income is denoted
by Y . Women without children have income, y; however, if a woman has
children, her income drops to y 0 , with y 0 < y, reflecting both the loss in her
earning capacity due to childbearing and the cost of raising the child. Hence
a single woman without children consumes her income y; if she decides to
have a child (or if an unwanted pregnancy occurs), she also consumes her
income (which has dropped to y 0 ) and receives a utility u from her child,
which is independent of her marital status.
Regarding couples, we assume that uH < y − y 0 , that is that the gain
received by the husband from having a child does not offset by itself the
loss in income experienced by the wife. This assumption implies, in our
framework, that the couple’s decision to have a child or not will also depend
on the wife’s preferences. Therefore married women must agree with their
husband on two issues. One is the fertility decision; that is, they must
decide whether to have kids or not, and the decision depends (in particular)
on the wife’s preferences towards children. The other decision relates to
the distribution of resources within the household (that is, the allocation
of total income between male and female consumption of the composite
good). Both decisions will be ultimately determined by the equilibrium on
the market for marriage. Finally, we model the legalization of abortion (and
generally the availability of some birth control technology) as an exogenous
decrease in the probability p of experiencing an unwanted pregnancy.
Fertility decisions
We first consider the fertility decisions of singles and couples, starting with
single individuals. Single men do not make decisions: they consume their
income, and get a utility which equals to Y . Single women, on the other
hand, will decide to have children if and only if the benefit compensates the
income loss, that is if u ≥ y − y 0 , leading to a utility which equals y 0 + u. In
the alternative case when u < y − y 0 , single the women chooses not to have
a child and any pregnancy will be involuntary. As pregnancy occurs with
probability p, the expected utility is y(1 − p) + p (y 0 + u). In what follows,
the threshold y − y 0 is denoted ū; women whose utility parameter is larger
than or equal to ū will be referred to as ‘high’ type.
In our transferable utility context, couples maximize their marital sur-
plus. The total benefit, for a couple, of having a child is uH + u, whereas
the cost is y −y 0 . It follows that a married couple will plan to have a child if
u ≥ y−y 0 −uH - then total utility is Y +y 0 +uH +u. The threshold y−y 0 −uH
is denoted u; note that u < ū. If u < y − y 0 − uH , only unwanted kids are
born, leading to an expected total utility Y + (1 − p) y + p (y 0 + uH + u).
Women with taste parameter u smaller than u will be said to be of ‘low’
type, while those between u and ū will be called ‘intermediate’. To sum-
marize:
• women of ‘high’ type (u ≥ ū) always choose to have a child
• women of ‘intermediate’ type (u < u < ū) choose to have a child only
when married
• women of ‘low’ type (u ≤ u) never choose to have a child
Stable match
We can now derive the properties of the stable match. The key element is
provided by Figure 8.4, which plots the maximum utility Φ (u) a man can
achieve when marrying a woman of taste u (in other words, Φ (u) denotes
his utility if he was to appropriate all the surplus generated by marriage).
The function Φ is increasing; that is, it is always better (for the husband)
to marry a wife with a larger taste coefficient u.
More precisely, women whose parameter u is greater than ū (the ‘high’
type), and who would plan to have a child even when single, are the most
‘attractive’ from the male’s perspective. While they differ in taste, this
difference is irrelevant from a husband’s viewpoint, since they require the
same compensation cH for getting married (namely, to be left with a private
consumption which equals their income with a child, y 0 ). Women between u
and ū (the ‘intermediate’ type) come next in males’ preferences. They plan
to have a child only when married, and the minimum compensation they
require is cI (u) = (y − u) (1 − p) + py 0 . This required compensation de-
creases with the individual utility u; hence men strictly prefer intermediate
women with a higher u. Finally, women with a u smaller than u (the ‘low’
type) never plan to have a child. Again, these women are equivalent from a
husband’s perspective, since they require the same compensation for getting
married, namely their consumption as single, that is cL = (1 − p) y + py 0 .
As often in matching models, the properties of the stable match crucially

depend on the identity of the marginal spouse (that is, the ‘last’ married
woman). We denote by u (r) the taste parameter of the marginal women
(that is, either the ‘last’ single woman or the ‘first’ married woman). Tech-
nically, u (r) is defined by the fact that the measure of the set of women
with a taste parameter larger than u (r) equals the measure of men, which
is 1; that is, the value u (r) solves the equation
Z U
r f (t) dt = 1. (8.41)
u(r)
Competition between women in the marriage market implies that women

who generate a larger surplus for their husband are a more desirable match.
Hence, whenever a women belonging to the intermediate type is married,
then all women with a larger taste parameter are married as well - this is
the case depicted in Figure 8.4. The intuition is that women with a larger
preference for children have a comparative advantage: the compensation
they need from their husband to accept marriage is smaller, because they
value highly the prospect of having a child. In general, the identity of this
marginal woman depends on the location of u (r) with respect to the two
thresholds u and ū.
An obvious property of stable matches in this context is that all males
receive the same utility; indeed, they are assumed identical, and the ab-
sence of friction implies that any difference of welfare between males would
be competed away. Since the marginal woman is indifferent between be-
ing married or single, her husband gets all the surplus generated by the
relationship, namely Φ (u (r)). Then all other men receive the same utility.
Graphically, this corresponds to the horizontal line going through Φ (u (r))
in Figure 8.4.
A crucial insight, at this point, is the following. Take any woman with a
taste parameter u larger than u (r). Then the difference Φ (u) − Φ (u (r))
represents the surplus received by this woman.11 In Figure 8.4, for instance,
the surplus received by any woman of ‘high’ type is depicted by a bold
arrow.
Using this geometric intuition, the characterization of the equilibrium is
straightforward. Three cases should be distinguished:
RU
• If 1/r ≤ W = ū f (t) dt, the excess supply of women is ‘large’, in the
sense that there are less men than high type women. Then u (r) ≥ ū,
and the marginal married women belongs to the high type. Only
(some of) these women are matched. Women of the same type who
1 1 If her husband’s utility was Φ (u) he would get all the surplus generated by the
marriage. Since his equilibrium utility is only Φ (u (r)), the difference Φ (u) − Φ (u (r))
represents the part of the surplus appropriated by the wife.
Singles Married
Y + uH
Female surplus
Y + puH
u u(r) u u
FIGURE 8.4. Maximum husband’s utility as a function of the wife’s taste - in-
termediate ESW
Singles Married
Y + uH
Female surplus
Y + pu H
u(r)u u u
FIGURE 8.5. Maximum husband’s utility as a function of the wife’s taste - small
ESW
remain single decide to have a kid; all other women remain single and
decide not to have children (although they may have one involuntar-
ily). Regarding welfare issues, note that, in that case, married women
receive no surplus from marriage; their consumption is the same as if
single.
RU
• If W < 1/r < W̄ = u f (t) dt, as depicted in Figure 8.4, the mar-
ginal wife belongs to the intermediate type. All married women have a
child, and consume the same amount, which is such that the marginal
wife is indifferent between getting married and remaining single. All
married women (but the marginal one) get a positive surplus from
marriage, and high type women receive the maximal surplus.
• Finally, when the excess supply of women is small enough (technically,

1/r ≥ W̄ ), the marginal wife belongs to the low type (that is u (r) ≤ u
- see Figure 8.5). Her fertility is the same with and without marriage
- namely, no planned child. Stability requires that her consumption
is also the same, and equals to (1 − p) y + py 0 . The same conclusion
applies to all married, low type women. Other married women belong
to the high or intermediate type, hence decide to have a child; their
consumption is defined by the fact that men, who are in short supply,
must be indifferent between the various potential spouses. Again,
this condition generates a positive surplus for all women of high and
intermediate types; high type women receive the largest surplus.
The variation in women’s utility across the three types of equilibria
exhibits interesting patterns. Not surprisingly, women are better off the
smaller their excess supply on the market. However, when women’s excess
supply is either large or small, their welfare does not depend on the size
of the imbalance. In the intermediate case, on the contrary, a marginal
increase in the number of men continuously reduces the taste parameter
u (r) of the marginal woman, which ameliorates the welfare of all married
women.
Changes in the birth control technology

We can now come to the main issue, namely the impact of a technological
change in birth control that reduces the probability of unwanted pregnan-
cies. A key assumption is that all women (including single) are given free
access to the technology; a natural example could be the legalization of
abortion that took place in the 1970’s.
The situation is depicted in Figure 8.6 (which, for expositional conve-
nience, considers the case in which the risk of unwanted pregnancies goes
to zero). The new technology decreases the maximum utility attainable
by husbands of low or intermediate type women, resulting in a downwards
shift of the graph of the function Φ. This leads to the following conclusions:
Not surprisingly, women who do not want to have a child (either be-
cause they belong to the low type or because they are single) benefit from
the technology, precisely because unwanted pregnancies become less likely.
In the extreme situation in which unwanted pregnancies are eliminated,
the monetary gain is thus p (y − y 0 − u) .More interesting is the fact that
women who decide to have a child also benefit from the technology, al-
though to a lesser extent than singles. The intuition is that the intrahouse-
hold distribution of resources is driven by the marginal women; for a small
or intermediate excess supply of women, the marginal women is indifferent
between getting married and remaining single without kid. Her reservation
utility is thus improved by the new technology. The nature of a match-
ing game, however, implies that any improvement of the marginal agent’s
situation must be transmitted to all agents ‘above’ the marginal one.
In the case of an intermediate excess supply depicted in Figure 8.6, the
benefit experienced by all married women, assuming the new technologies
drives the risk of unwanted pregnancies to zero, is p (y − y 0 − u (r)) (where,
again, u (r) denotes the taste parameter of the marginal married woman).
This benefit continuously increases with the number of men M . When the
excess supply is small, the gain is puH , still smaller than p (y − y 0 ) (the
Y + uH
Y + p.uH Increase in
female surplus
Y
u
u(M)
FIGURE 8.6. An Increase in the Risk of Pregnancy
gain for single women) but nevertheless positive. On the other hand, when
the excess supply of women is ‘large’, married women do not benefit from
the new technology, because the marginal woman does not use it. Hence
the consequences of the new technology for married women’s welfare are
intimately related to the situation that prevails on the marriage market.
Finally, men cannot gain from the introduction of the new technology.
When the excess supply of women is large, their utility is not affected.
When the excess supply of women is small, so that the marginal wife does
not want a child, the total welfare of the household is increased, but so
is the reservation utility of the wife; the husband is left with the same
consumption, but loses the benefit he would have received from an un-
wanted birth. The intermediate case is even more spectacular. Here, all
marriages result in a child being born, so the total surplus generated by
marriage is not affected by the innovation. What changes, however, is the
intrahousehold allocation of the surplus. The new technology improves the
reservation utility of the marginal woman, hence her share of resources in-
creases. Stability requires this shift to be reproduced in all couples. All in
all, the new technology results in a net transfer from the husband to the
wife, which equals the expected gain of the marginal single woman, that
is p (y − y 0 − u (r)), without any change on the fertility of married couple
(who actually do not use the new technology).
We thus conclude that in our model an improvement in the birth control
technology, such as the legalization of abortion, generally increases the wel-

fare of all women, including those who want a child and are not interested
in the new technology. Note, however, that the mechanism generating this
gain is largely indirect. The reason why even married women willing to have
a child benefit from the birth control technology is that the latter, by rais-
ing the reservation utility of single women, raises the ‘price’ of all women
on the matching market (although this logic fails to apply in situations of
‘large’ excess supply of women).
An interesting, although somewhat paradoxical implication is that re-
serving the new technology to married women (as was initially the case
for the pill, at least for younger women) would actually reverse the em-
powerment effect. The option of marriage to women with a low taste for
children, who are willing to accept a lower compensation from the husband
for getting married and gaining access to the new technology, toughens the
competition for husbands. Therefore, women of the high or intermediate
type are made worse off by the introduction of the new technology. Only
women with a very low taste parameter (that is, below the lower marginal
value) gain from the innovation. This comparison emphasizes the complex
and partly paradoxical welfare impact of a new technology. On the one
hand, its effects can go well beyond the individuals who actually use it,
or even consider using it. Our model suggests that a major effect of le-
galizing abortion may have been a shift in the intrahousehold balance of
powers and in the resulting allocation of resources, even (and perhaps espe-
cially) in couples who were not considering abortion as an option. On the
other hand, the new technology benefits all married women only because
it is available to singles. A technological improvement which is reserved
to married women will have an impact on their fertility, partly because it
changes the mechanisms governing selection into marriage. But its impact
on women’s welfare is largely negative, except for a small fraction of women
who choose marriage as an access to the new technology.
8.3 Matching with general utilities

We now switch to the general framework in which we relax the assump-
tion that utility is transferable. The tractability of the transferable utility
framework comes at a cost. The most obvious drawback is that under TU,
couples behave as singles; in particular, their demand function (that is,
the amount spent on each of the public or private commodities) does not
depend on the Pareto weights. In other words, changes in male and fe-
male income distributions may trigger a reallocation of resources (or more
precisely of one commodity) between spouses, but under TU, it cannot
have income effects, and cannot result in, say, more being spent on chil-
dren health or education. While this framework may be useful in many
contexts, it is clearly too restrictive in other situations.

In this section, we explore the more general framework introduced in
Chapter 7, in which utility is not (linearly) transferable. That is, although
compensations between spouses are still possible, they need not take place
at a constant ‘exchange rate’: there is no commodity the marginal utility of
which is always identical for the spouses. In particular, the matching model
is no longer equivalent to a linear optimization problem. The upside is that,
now, any change affecting the wife’s and husband competitive positions
(for example, a change in income distributions) will potentially affect all
consumptions, including on public goods - which allows for a much richer
set of conclusions. The downside is that the derivation of individual shares
from the equilibrium or stability conditions is more complex. It remains
feasible, however. We first present the general approach to the problem,
then we concentrate on a specific and tractable example.
8.3.1 Recovering individual utilities: the general strategy

We use the same framework as in Chapter 7. Male income is denoted by
x and female income is denoted by y; the Pareto frontier for a couple has
the general form
u = H(x, y, v) (8.42)
with H(0, 0, v) = 0 for all v. If a man with income x remains single, his
utility is given by H(x, 0, 0) and if a woman of income y remains single
her utility is the solution to the equation H(0, y, v) = 0. By definition,
H(x, y, v) is decreasing in v; we assume that it is increasing in x and y,
that is that a higher income, be it male’s or female’s, tends to expand
the Pareto frontier. Also, we still consider a continuum of men, whose
incomes x are distributed on [0, 1] according to some distribution F , and a
continuum of women, whose incomes y are distributed on [0, 1] according
to some distribution G; let r denote the measure of women. Finally, we
assume that an equilibrium matching exists and that it is assortative - that
is, that the conditions derived in Chapter 7 are satisfied; let ψ (x) (resp.
φ (y)) denote the spouse of Mr. x (of Mrs. y).
As previously, the basic remark is that stability requires:
u (x) = max H(x, y, v (y))
y
where the maximum is actually reached for y = ψ (x). First order conditions
imply that
∂H ∂H
(φ (y) , y, v (y)) + v 0 (y) (φ (y) , y, v (y)) = 0.
∂y ∂v
or:
∂H
∂y (φ (y) , y, v (y))
v 0 (y) = − ∂H . (8.43)
∂v (φ (y) , y, v (y))
Again, we have a differential equation in v. It is more complex than in the

TU case, because the right hand side depends on v (y) in a potentially non
linear way. Still, under mild regularity conditions, such an equation defines
v up to a constant, the value of which can be determined from the condition
that the last married person in the ‘abundant’ side of the market receives
no surplus from marriage.
Note, in particular, that from the assumptions made in Chapter 7, we
have that:
∂H
∂y (φ (y) , y, v (y))
v 0 (y) = − ∂H >0 (8.44)
∂v (φ (y) , y, v (y))
In words, richer people are always better off. Finally, once v has been
computed, the condition
u = H(x, ψ (x) , v (ψ (x))) (8.45)
exactly defines u.
This framework has been applied by Chiappori and Reny (2007), who
consider a population of heterogeneous agents with different risk aversions
matching to share risks arising from identically distributed random in-
comes. They show that (i) a stable match always exists, (ii) it is unique,
and (iii) it is negative assortative: among married couples, men with lower
risk aversion match with more risk averse women and conversely.
8.3.2 A specific example

We now present another application due to Chiappori (2009).
Preferences
There is a continuum of males, whose income y is distributed over [a, A] ac-
cording to some distribution F , and a continuum of females, whose income
y 0 is distributed over [b, B] according to some distribution G. To simplify,
we consider the linear shift case, where the matching functions are given by
φ (y 0 ) = (y 0 + β) /α and ψ (y) = αy − β; also, we assume that the number
of female is almost equal to, but slightly larger than that of men.12
Males have identical preferences, represented by the Cobb-Douglas util-
ity:
um = cm Q (8.46)
where cm denotes his consumption of some private, Hicksian composite
commodity and commodity Q is publicly consumed within the household;
1 2 This last assumption is simply used to pin down the constant in the allocation of
marital outcome; it can readily be modified as needed.

all prices are normalized to 1. Similarly, women all share the same pref-
erences, characterized by some minimum level of consumption c̄, beyond
which private and public consumptions are perfect substitutes:
uf (cf ) = −∞ if cf < c̄
= cf + Q if cf ≥ c̄
In particular, if a woman is single, her income must be at least c̄; then her
utility equals her income.
An important feature here is that men and women have different prefer-
ences: private and public consumption are complements for men and per-
fect substitutes for women. We shall further assume that household income
is always larger than c̄; then female utilities are of the quasilinear form
cf + Q. In particular, any efficient solution involves cf = c̄, because beyond
c̄, spending a dollar on private consumption for the wife is inefficient: spent
on the public good, the same dollar is as valuable for the wife and strictly
better for the husband.
Efficient allocations
We first characterize the set of efficient allocations. An efficient couple
solves the program:
max cm Q (8.47)
under the constraints
cm + cf + Q = y + y 0 (8.48)
uf = cf + Q ≥ U (8.49)
0
where y + y is household total income and U is some arbitrary utility level.
A first remark is that at any efficient allocation, the wife’s utility U cannot
fall below ((y + y 0 ) + c̄) /2. As the wife receives the same consumption c̄ in
any efficient allocation, her utility varies only with the amount of the public
good, Q. Once c̄ has been spent, the husband’s maximal utility is obtains
when he receives his optimal bundle of private and public consumption,
namely Q = cm = ((y + y 0 ) − c̄) /2; this choice generates a wife’s utility
of ((y + y 0 ) + c̄) /2. If U > ((y + y 0 ) + c̄) /2, however, providing her with
U requires more resources to be spent on the public good (and less on
his private consumption) than what he would choose by himself. Then the
constraint (8.49) is binding. Therefore, the Pareto frontier is given by
um = H ((y + y 0 ) , uf ) = (uf − c̄) ((y + y 0 ) − uf ) , (8.50)
(y+y0 )+c̄
where uf ≥ 2 . Moreover, one can readily compute the correspond-
ing consumptions; namely, Q = uf − c̄ and cm = (y + y 0 ) − uf . Fig-
ure 8.7 displays the Pareto frontier when total total income has been set
to (y + y 0 ) = 5 and the wife’s minimal consumption to c̄ = 1, so that
((y + y 0 ) + c̄) /2 = 3.
v4
0
3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8 5.0 5.2
u
FIGURE 8.7. Pareto frontier
Because of the public consumption, our simple model exhibits what

Lundberg and Pollack call ‘production dominance’; that is, any single man
and any single woman can do better by marrying. To see why, just note
that a single man with income y chooses Q = cm = y/2 for a utility of y 2 /4,
while a single woman with income y 0 > c̄ achieves a utility that equals y 0 .
Now, by marrying, they achieve an income (y + y 0 ). If y 0 ≤ y + c̄, he can
2
achieve (y + (y 0 − c̄)) /4 > y 2 /4 while she gets c̄ + (y + y 0 ) /2 > y 0 . If, on
the contrary, y > y + c̄ then he can achieve (y 0 − c̄) y > y 2 /4, while she
0
remains at y 0 . Therefore, in a frictionless model like this one (and without

non-monetary gains or costs), either all women or all men marry: singles
can only be on one side of the marriage market.
Assortativeness
The Pareto frontier just derived has a particularly tractable form. Indeed,
let us analyze the stability conditions along the lines previously described.
(y+y0 )+c̄
For v ≥ 2 , we get:
∂H (y + y 0 , v) ∂H (y + y 0 , v)
0
= v − c̄, = − (2v − (c̄ + (y + y 0 ))) (8.51)
∂ (y + y ) ∂v
implying that
∂ 2 H (y + y 0 , v) ∂ 2 H (y + y 0 , v)
2 = 0 and =1 (8.52)
∂ (y + y 0 ) ∂ (y + y 0 ) ∂v
As we have seen in Chapter 7, these conditions are sufficient for the

existence of a unique stable match involving assortative matching.
Intrahousehold allocation of welfare

We now turn to the allocation of welfare within the couple. Equation (8.43)
becomes:
∂H 0 0 0
∂y0 (φ (y ) + y , v (y ))
v0 (y 0 ) = − ∂H 0 0 0
∂v (φ (y ) , y , v (y ))
0
v (y ) − c̄
=
2v (y 0 ) − (c̄ + y 0 + φ (y 0 ))
αv (y 0 ) − αc̄
= . (8.53)
2αv (y 0 ) − (α + 1) y 0 − (αc̄ + β)
Recovering the wife’s utility requires solving this differential equation. For
that purpose, we may, since v is strictly increasing, define the inverse func-
tion ω by:
v (y 0 ) = v ⇔ y 0 = w (v)
Then equation 8.53 becomes:
1 αv − αc̄
= ,
ω0 (v) 2αv − (α + 1) ω (v) − (αc̄ + β)
or
(α + 1) 2αv − (αc̄ + β)
ω 0 (v) + ω (v) = ,
αv − αc̄ αv − αc̄
which is a standard first order, linear differential equation. The general
solution is:
− α+1 2α β + c̄α + 2αβ
ω (v) = K (v − c̄) α
+ v− ,
2α + 1 (α + 1) (2α + 1)
where K is an integration constant.
To find K, we consider the marginal couple in which the wife receives the
lowest female income b and the husband receives the lowest male income
a = (b + β) /α. Since we assumed that the number of women exceeds that
of men, the utility of the marginal woman must be at its minimum level,
namely ((a + b) + c̄) /2. Thus we have:
µ ¶
− α+1 2α β + c̄α + 2αβ 1 b+β
ω (b) = K (b − c̄) α
+ b− = + b + c̄
2α + 1 2α2 + 3α + 1 2 α
which yields:
µ µ ¶ ¶
1 b+β 2α β + c̄α + 2αβ α+1
K= + b + c̄ − b+ (b − c̄) α
2 α 2α + 1 (α + 1) (2α + 1)
Utilities
9
1
2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4
x
FIGURE 8.8. Husband’s and Wife’s Utilities, Public Consumption and the Hus-
band’s Private Consumption
To illustrate, suppose that β = 0, α = .8, a = 2, b = 1.6, c̄ = 1. Then

K = 0.65
and
.65
ω (v) = 0.615v + 2. 25 − 0.171,
(v − 1)
while the husband’s utility is:
u = H (y + y 0 , v)
= (v − c̄) (2.25ω (v) − v)
The resulting utilities are plotted in Figure 8.8. The horizontal line indicates
the husband’s income y. The wife’s utility is represented by the thick line,
while the husband’s is in dotted and thick. Also, the consumption of the
public good Q = v − c̄ is represented by dashed line, while the consumption
of the husband, cm , is represented by the thin line.
As one moves up the assignment profile, the total income of the couples
and utilities of both husband and wife rise. The consumption of the public
good also rises. The private consumption of the wife remains constant at
c̄ = 1, while the private consumption of the husband, cm , first declines and
than rises.
All the comparative statics exercises can be adapted to this general
framework. For instance, suppose that we keep c̄ = 1 and a = 2 but shift
Utilities
10
2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4
x
FIGURE 8.9. Impact of a Rightwards Shift in Female Income Distribution on

Husband’s and Wife’s Utilities
the income distribution of women to the right so that α = 1 and b = 2.

Then, K = 1.12 so that:
−2 2 1
ω (v) = 1.12 (v − 1) + v− ,
3 6
while his utility is still
u = (v − 1) (2ω (v) − v)
The husband’s and the wife’s utilities for these two cases are displayed in
Figure 8.9, where couples are indexed by male income (which remains in-
variant). For α = .8, we represent, as before, the wife’s utility by a thick
line and husband’s by a dotted and thick line. Thin lines (dashed for males
and solid for females) represent u and v when α = 1. We see that the shift
of the female distribution to the right benefits both men and women. More
interesting are the spending patterns. Figure 8.10 displays public (thick)
and husband’s private (thin) consumptions, both before (solid) and after
(dashed) the shift. We see that most of the additional income is spent on
the public good; increases in the husband’s private consumption are quan-
titatively small, and tend to shrink with income. In other words, while the
husband does benefit from the increase in the wife’s income, most of his
gain stems from a higher level of public consumption (which actually ben-
efits both partners). We conclude that in this model, unlike the TU case,
Utilities
5
2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4
x
FIGURE 8.10. Impact of a Rightwards Shift in Female Income Distribution on

Public Consumption and the Husband’s Private Consumption
changes affecting the wife’s situation do affect the structure of consump-

tion; moreover, improving the wife’s status boost public spending within
the couples - a fact that has been abundantly confirmed by empirical inves-
tigation, especially if we think of children as a primary example of public
consumption (see Chapter 5).
8.4 Matching by Categories

The matching model and the associated allocation rules that we have dis-
cussed so far assume some idealized conditions that are not likely to hold
in practice. The most common way to make the model more applicable
is to introduce frictions and some bargaining over the resulting surplus.
There is, however, an alternative modeling choice that goes part of the
way towards reality and is based on the recognition that the researcher
observes only part of the data that motivates and restricts choices. This
is particularly true in marriage markets where explicit market prices do
not exist and the division within families of consumption or time is rarely
observed. This path has been followed by Choo and Siow (2008) and Chi-
appori, Salanié and Weiss (2010); the presentation given here follows the
latter contribution.
To incorporate unobserved heterogeneity, we consider a case in which
the researcher observes marriage patterns within broad categories, such as

schooling level, race or occupation and observes only some of the individual
attributes that distinguish individuals within these classes. That is, in addi-
tion to their observed class, individuals are characterized by some observed
attributes such as income or age and by some idiosyncratic marriage related
attributes that are observed by the agents in the marriage market but not
by the researcher. We assume that the marital output that is generated by
the match of man i, and woman j can be written in the form
ζ ij = zI(i)J(j) + αiJ(j) + β jI(i) . (8.54)
The first component zI(i)J(j) depends on the class of the two partners, the
second component αiJ(j) depends on man i and the class of woman j and
the third component depends on woman j and the class of man i. This
specification embodies a strong simplifying assumption; the interaction be-
tween two married partners is always via their class identity. In particular
we do not have a term that depends on both i and j.13 We further assume
that
αiJ(j) = a0I(i)J(j) xi + εiJ(j)
β jI(i) = b0I(i)J(j) xj + εI(i)j (8.55)
where xi and xj are the observed attributes of man i and woman j, respec-
tively, aIJ and bIJ are vectors of coefficients that represent the marginal
contribution of each male (female) attribute to a marriage between a man
of class I and woman of class J. The error terms εiJ(j) represent the un-
observed contribution of man i to a marriage with any woman of class J.
Similarly, εI(i)j represents the contribution of woman j to a marriage with
any man of class I.
A basic property of the matching model with transferable utility that we
discussed in Chapter 7 is the existence of prices, one for each man, vi , and
one for each woman, uj , that support a stable outcome. At these prices,
the matching is individually optimal for both partners in each match. Thus,
equilibrium implies that i is matched with j iff
ui = ξ ij − vj ≥ ξ ik − vk for all k, and ui ≥ ξ i0 ,
vj = ξ ij − uj ≥ ξ kj − uk for all k, and vj ≥ ξ j0 . (8.56)
Under the special assumptions specified in (8.54) and (8.55), Chiappori,
Salanié and Weiss (2010) prove the following Lemma:
Lemma 8.1 For any stable matching, there exist numbers UIJ and VIJ , I =
1, ..., M, J = 1, ..., N , with the following property: for any matched couple
1 3 This simplifying assumption has been introduced in the context of transferable util-
ity by Choo and Siow (2006). Dagsvik(2002) considers a more general error structure in
the context of non transferable utility (e.g. an exogenous sharing rule).
(i, j) such that i ∈ I and j ∈ J ,
ui = UIJ + αiJ
vj = VIJ + β Ij (8.57)
where
UIJ + VIJ = zIJ
In words: the differences ui −αiJ and vj −β jI only depend on the spouses’

classes, not on who they are. Note, incidentally, that (8.56) is also valid for
singles if we set UI0 = ζ I0 and U0J = ζ 0J .
The economic interpretation of this result is as follows. The contribution
of women j and j0 who are in the same class J to a marriage with all men
in class I differ by β Ij 0 − β Ij . If vj 0 − vj > β Ij 0 − β Ij no men in I will marry
woman j0 because she is ‘too expensive’ relative to woman j. Conversely,
if vj 0 − vj < β Ij 0 − β Ij no man in I will marry woman j. Hence, in an
equilibrium in which both women j and j0 find a match with men of the
same class I it must be the case that vj 0 − vj = β Ij 0 − β Ij .
To empirically implement these ideas, a first step is to specify the distri-
bution of the unobserved heterogeneity components ε. Given the structure
of the model, it is natural to assume that these error terms are identi-
cally and independently distributed according to a type 1 extreme value
(Gumbel) distribution. We can now write the probability (as viewed by the
researcher) that man i marries a woman of a particular class (or remains
single) in the familiar multinomial- logit form (see McFadden, 1984)
Pr ( i ∈ I matched with j ∈ J)
¡ ¢
exp UI(i)J + aI(i)J xi
= P ¡ ¢ ¡ ¢,
K exp UI(i)K + aI(i)K xi + exp UI(i)0 + aI0 xi
Pr (i is single)
¡ ¢
exp UI(i)0 + aI(i)0 xi
= P ¡ ¢ ¡ ¢. (8.58)
K exp UI(i)K + aI(i)K xi + exp UI(i)0 + aI0 xi
Analogous expressions hold for women. The terms UI(i)K + aI(i)K xi rep-
resent the systematic part (excluding the unobserved εiK(j) ) of the share
that man i receives upon marriage with a woman in class K. The spouse’s
personal attributes xj and her idiosyncratic contribution εI(i)j have no di-
rect bearing on the probability of marriage, because in equilibrium they are
already captured by the unknown constants UIK . Similar remarks apply
to the probability of marriage of women. The unknown parameters con-
stants UIJ and VIJ adjust endogenously to satisfy the requirement that
the choices of men and women are consistent with each other in the sense
of market clearing.
In principle, it is possible to calculate these coefficients directly by solving

the market equilibrium (that is the linear programming problem) associ-
ated with a stable assignment. More interestingly, one can use data on
actual marriage patterns and the observed attributes of participants in a
"marriage market" to estimate the gains from marriage of these individu-
als (relative to remaining single).14 Basically, the preferences for different
types of spouses are "revealed" from the choice probabilities of individuals.
Taking the simplest case without covariates, we see
Pr (i ∈ I is matched with j ∈ J)
ln = UIJ − UI0
Pr (i ∈ I is single)
Pr (j ∈ J is matched with i ∈ I)
ln = UIJ − UJ0 . (8.59)
Pr (j ∈ J is single)
Estimating separate multinomial logit for men and women, one can esti-
mate the utilities for each gender in a marriage of each type. Summing the
estimated utilities one can recover, for each matching of types (I, J), the
systematic output of the marriage ξ IJ (which, under the normalization that
being single yields zero utility, equals the total surplus ZIJ ). The estimated
matrix ZIJ can then be analyzed in terms of the assortative matching that
it implies. Of particular interest is whether or not this matrix is super-
modular (implying positive assortative mating) or not. As noted by Choo
and Siow (2006) and Siow (2009), in the absence of covariates the super
modularity of ZIJ is equivalent to the supermodularity of
2
(μ(I, J))
ln
σ(I)σ(J)
where μ(I, J) is the total number of type (I, J) marriages and σ(I) and
σ(J) are the number of single men and single women, respectively. Such
supermodularity requires that for all I 0 > I and J 0 > J
μ(I 0 , J 0 )μ(I, J)
ln > 0.
μ(I, J 0 )μ(I 0 , J)
Siow (2009) uses census data on married couples in the US, where the
husband and wife are 32-36 and 31-35 respectively. In each couple, the wife
and the husband can belong to one of five possible schooling classes (less
than high school, high school, some college, college and college plus). He
compares the marriage patterns in the years 1970 and 2000 and finds that
in each of the two years strict supermodularity fails to hold as in some
of the off diagonal cells, the log odds ratio is negative. Looking at the
1 4 Because the probabilities in (8.58) are unaffected by a common proportionality fac-
tor, some normalization is required. A common practice is to set the utility from being
single to zero for all individuals.
whole matrix, one cannot conclude that there is more positive assortative
matching in 2000 than in 1970, although some specific local log odds have
increased over time.
Chiappori, Salanié and Weiss (2010) have extended Choo and Siow’s 2006
model by assuming that the same determinants of assortative matching
operate over a long period of time, during which the distribution of male
and female characteristics changes. In practice, their main classification
is by education level, and they exploit the remarkable increase in female
education over the last decades. In their model, while the surplus generated
by the matching of a man in class I with a woman in class J is allowed
to vary over time, the supermodular part of the surplus is not; therefore
the gains from assortative matching are assumed constant over the period.
This assumption generates strong testable predictions; interestingly, they
are not rejected by the data. In addition, one can then (over)estimate the
model; in particular, the parameters of the surplus function and their drifts
can be recovered. From these, it is possible to trace the time changes in the
common factors UIJ and VIJ driving the intrahousehold allocation of the
surplus, as well as the expected utility of each gender by education level.
Note that, as always, this utility is estimated in variations from singlehood;
it thus comes in addition to any direct benefit affecting all individuals
irrespective of their marital status.
This approach has important practical implications. Many theoretical
models suggest that education has two types of benefits. One (the so-called
‘college premium’) is collected on the labor market; it represents the wage
differential generated by a college degree, irrespective of a person’s marital
status. A second, and often omitted aspect is the impact of education on
marriage prospects (the ‘marital college premium’). An educated person
is more likely to marry an educated spouse, resulting in higher household
income and surplus; moreover, education typically boost the amount of
intra-marital surplus received by the person. This second phenomenon has
been recognized by the theoretical literature (see for instance Chiappori,
Iyigun and Weiss 2009, and also the next Chapter in this book), but its
empirical evaluation has often be perceived as elusive. The approach pro-
posed by Chiappori, Salanié and Weiss exactly addresses this issue. Using
CPS data, they show that, indeed, the marital college premium is strong,
and that it has significantly increased for women over the last decades -
which may help explaining the remarkable growth in female education over
the last decades.
Several extensions are currently being pursued. Perhaps the most promis-
ing is the explicit modeling of multidimensional matching - recognizing the
fact that, ultimately, several factors contribute to the formation of marital
surplus, hence to the matching process. The reader is referred to Galichon
and Salanié (2010) for a recent and path breaking contribution along these
lines.
8.5 Appendix: Extreme Value distributions

We collect here some useful properties of extreme value distributions (See
Ben-Akiva and Lerman, 1985, ch. 5 and Johnson et al. 1995, ch. 22). The
type 1 extreme value distribution for the maximal extreme is
− x−a
F (x) = e−e b
, b > 0.
− x−a
1 − x−a
b e−e
b
f (x) = be .
The moment generating function is E(etx ) = eat Γ(1 − bt); the mean is
E(x) = a + kb, where k = .57721 = −Γ0 (1) is Euler’s constant, the variance
2 2
is V (x) = b 6π , the mode is a and the median is a−b log(log 2) ∼
= a+.36611b.
Parameter a is thus seen to be a location parameter, while b controls the
variance. This distribution is sometimes named after E. J. Gumbel and we
shall say that x v G(a, b). The distribution is skewed to the right (mean >
median > mode). The distribution of the minimal extreme is obtained by
reversing the sign of x and is skewed to the left.
2
The standard form G(0, 1) has mean k and variance π6 . To get an ex-
treme value with zero mean we can set a = −kb and use G(−kb, b).
The basic properties are the following:
• If x v G(a, b) then αx + β v G(αa + β, αb).
• If x1 and x2 are independent Gumbel variates such that x1 v G(a1 , b)
and x2 v G(a2 , b) then x∗ = (x1 −x2 ) has a logistic distribution, that
is,
1
F (x∗ ) = 1 ∗
.
1 + e b 2 −a1 −x )
(a
• If x1 , x2.... xn are iid Gumbel variables with G(a, b) and v1 , v2, ..vn are
some constants then
v1
e b
P r{v1 + x1 ≥ max[v2 + x2 ....vn + xn ]} = P vi .
i e b
• If x1 , x2.... xn are independent Gumbel variables with distributions

G(ai , b) then
X ai
max(x1 , x2 , ....xn ) v G(b ln( e b ), b).
i
• In particular, if x1 , x2.... xn are iid Gumbel variables with G(a, b) and

v1 , v2, ..vn are some constants then
X vi +a
E{max[v1 + x1 , v2 + x2 ....vn + xn ]} = b ln( e b ) + kb
i
X vi
= b ln( e b ) + a + kb.
i
Thus, if x1 , x2.... xn are iid Gumbel variables with zero mean then
X vi
E{max[v1 + x1 , v2 + x2 ....vn + xn ]} = b ln( e b ).
i
If we choose one alternative as a benchmark, say alternative 1, and normal-

ize its value to zero, the expected utility relative to this alternative is fully
determined by, and inversely related to the probability that the benchmark
alternative is selected. The marginal impact of an increase in the value of
specific alternative, j, is
vj
∂E{max[v1 + x1 , v2 + x2 ....vn + xn ]} e
= P b vi ,
∂vj ie b
which is the probability that alternative j will be selected, pj .
8.6 References
[1] Akerlof, George A. and Rachel E. Kranton, ‘Economics and Identity’,
Quarterly Journal of Economics, 115 (2000), 715-753.
[2] Ben-Akiva, Moshe and Steven R. Lerman ,’Discrete Choice Analy-
sis’ (MIT Press, 1985).
[3] Bately Richard ‘On Ordinal Utility Cardinal Utility and Random Util-
ity’ Theory and Decision (2008) 64:37—63 .
[4] Botticini, Maristella and Aloysius Siow, ‘Why Dowries?’, The Amer-
ican Economic Review, 93 (2003), 1385-1398.
[5] Browning, Martin and M. Gørtz, ‘Spending Time and Money Within
the Household’, unpublished manuscript, Department of Economics, Uni-
versity of Oxford (2006).
[6] Choo, Eugene and Aloysius Siow, ‘Who Marries Whom and Why’,
[7] Chiappori, Pierre-Andre, Bernard Fortin and Guy Lacroix, ‘Marriage
Market, Divorce Legislation, and Household Labor Supply’, Journal of
[8] Chiappori, Pierre-Andre, Murat Iyigun and Yoram Weiss, ‘Investment
in Schooling and the Marriage Market’, The American Economic
Review, forthcoming (2009).
[9] Chiappori, Pierre-Andre, Bernard Salanié and Yoram Weiss ‘Assor-
tative Matching on the Marriage Market: A Structural Investigation’
(Columbia University, Department of Economics, in progress)
[10] Dagsvik, John K., ‘Aggregation in Matching Markets’, International

Economic Review, 41 (2000), 27-57.
[11] Edlund, Lena, ‘The Price of Marriage: Net vs. Gross Flows and the
South Asian Dowry Debate’, The Journal of the European Eco-
nomic Association, papers and proceedings, 4 (2006), 542-551.
[12] Fernandez, Cristina and Almudena Sevilla-Sanz, ‘Social Norms and

Household Time Allocation’, Institute for Social and Economic Research,
Working Paper 2006-38, (2006).
[13] Gale, David ‘The Theory of Linear Economic Models’ (

McGraw-Hill, 1960).
[14] Johnson, Norman L., Samuel Kotz and N. Balakrishnan ,’Continuous

Univariate Distributions’ (Wiley, 1995).
[15] Lise, Jeremy and Shannon Seitz, ‘Consumption Inequality and Intra-
Household Allocations’, Labor and Demography, EconWPA, No.
0504001, (2005).
[16] Mare, Robert D. ‘Educational Assortative Mating in Two Generations’

University of California Los Angeles, Department of Sociology, 2008.
[17] McFadden, Daniel L., ‘Econometric Analysis of Qualitative Response

Models’, in Griliches, Z. (eds.) and Intriligator, M. D. (eds.), ‘Handbook
of Econometrics’, (Elsevier, Vol 2, 1984).
[18] Siow, Aloysius ‘Testing Becker’s Theory of Positive Assortative Match-

ing’ University of Toronto (2009).
[19] Train, Kenneth E. Discrete Choice Methods with Simulation

(Cambridge University Press, posted internet version, 2009).
3.0
2.5
2.0
1.5
1.0
0.5
0.0
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
Mean of predicted log wages, men Mean of predicted log wages, women
Standard Deviation of predicted log wages, men Standard Deviation of predicted log wages, women
Correlation between husband's and wife's predicted log wages Female to male ratio
FIGURE 8.11. Parameters of the predicted log wage distribution of men and
women and sex ratio, U.S. 1976-2005. Source: Current Population Surveys.
FIGURE 8.12. Cumulative distributions of predicted hourly wages of men and

women
FIGURE 8.13. The surplus of married men and women in 1976, female to male
ratio = 1.045
FIGURE 8.14. The surplus of married men and women in 2005, female to male
ratio = 0.984
FIGURE 8.15. Wife’s relative share in the surplus, women are the majority
(r=1.1)
FIGURE 8.16. Wife’s relative share in the surplus, women are the majority
(r=0.9)
This is page 391
9
Investment in Schooling and
the Marriage Market
The purpose of this chapter is to provide a simple equilibrium framework
for the joint determination of pre-marital schooling and marriage patterns
of men and women.1 Couples sort according to education and, therefore,
changes in the aggregate supply of educated individuals affects who mar-
ries whom and the division of the gains from marriage. Unlike other at-
tributes such as race and ethnic background, schooling is an acquired trait
that is subject to choice. Acquiring education yields two different returns:
First, a higher earning capacity and better job opportunities in the labor
market. Second, an improvement in the intra-marital share of the surplus
one can extract in the marriage market. Educational attainment influences
intra-marital shares by raising the prospects of marriage with an educated
spouse and thus raising household income upon marriage, and by affect-
ing the competitive strength outside marriage and the spousal roles within
marriage.
The gains from schooling within marriage strongly depend on the de-
cisions of others to acquire schooling. However, since much of schooling
happens before marriage, partners cannot coordinate their investments.
Rather, men and women make their choices separately, based on the antic-
ipation of marrying a “suitable” spouse with whom schooling investments
are expected to generate higher returns. Therefore, an equilibrium frame-
work is required to discuss the interaction between marriage and schooling.
Such a framework can address some interesting empirical issues. For in-
stance, it is well documented that the market return to schooling has risen,
especially in the second half of the 20th century. Thus, it is not surprising
that women’s demand for education has risen. What is puzzling, however,
is the different response of men and women to the changes in the returns to
schooling. Women still receive lower wages in the labor market and spend
more time at home than men, although these gaps have narrowed over
time. Hence, one could think that women should invest in schooling less
than men, because education appears to be less useful for women both at
home and in the market. In fact, while women considerably increased their
investment in education in the last four decades, men hardly responded to
the higher returns to schooling since the 1970s, eventually enabling women
1 This chapter is based on Chiappori, Iyigun and Weiss (2009).

392 9. Investment in Schooling and the Marriage Market
to overtake them in educational attainment.2 It has been shown by Chi-

appori, Iyigun and Weiss (2009) that by introducing marriage market con-
siderations as an additional motivation for investment in schooling one can
explain the interrelated investment patterns of women and men.
The returns to pre-marital investments in schooling can be decomposed
into two parts: First, higher education raises one’s wage rate and increases
the payoff from time on the job (the labor-market return). Second, it
can improve the intra-marital share of the surplus one can extract from
marriage (the marriage-market return). Educational attainment influences
intra-marital spousal allocations directly (due to the fact that education
raises household income) and indirectly (by raising the prospects of mar-
riage with an educated spouse and also changing the spousal roles within
marriage). In this chapter, we take the labor market returns as given and
show how the marriage market returns are determined endogenously to-
gether with the proportions of men and women that marry and invest in
schooling.
9.1 Is pre-marital investment efficient?

An important issue that we shall address relates to the efficiency of premar-
ital investment. To see where the problem may arise, assume that, after
marriage, the spouses’ income is used to purchase private and public goods.
It follows that, because of the public consumption component, an invest-
ment made today will have an external, positive effect on the welfare of the
future spouse: if I invest more today, then after my marriage my household
will be wealthier and spend more on public consumptions, which will ben-
efit my wife as well. An old argument has it that this external effect will
not be taken into account when the investment is done - if only because,
at that date, agents probably don’t even know who their future spouse will
be.3 That is also true when ex-post bargaining within marriage determines
the division of the gains between the two partners. Because each person
bears the full cost of his\her investment prior to marriage and receives only
part of the gains, there is a potential for under investment. This is known
as the “hold-up problem.”
Convincing as it may sound, this argument is not robust. Once the match-
ing game is taken into account, it becomes invalid, because the equilibrium
conditions imply a full internalization of the externality. This important
2 Since the late1970s, the returns to schooling have risen steadily for men too. Still,
men’s college graduation rates have peaked for the cohort born in the mid-1940s (that
is, around the mid-1960s). And, after falling for the cohorts that followed, men’s college
graduation rates have reached a plateau for the most recent cohorts. See Goldin (1997)
and Goldin et al (2006).
3 See for instance Bergstrom et al (1986) and MacLeod and Malcomson (1993).
9. Investment in Schooling and the Marriage Market 393
result, due to Peters and Siow (2002) and Iyigun and Walsh (2007), can be
illustrated on a very simple example. Consider a woman, a, and a man, b,
who live for two periods. During the first period, they each receive some
income xs (s = a, b) that they can use for direct consumption or to invest
in human capital; therefore xs = cs + is , where cs denotes consumption
and is investment. The second period income depends on the investment:
y s = φ (is ), where φ is increasing and concave. Once married, the couple
can spend its total income y a + y b on private consumptions q a and q b and
public consumption Q. Individual utilities have the form:
U s = cs + q s Q
which satisfies the TU property.

In this very simple setting, one can readily compute the optimal level of
investment. Indeed, in our TU framework, efficient allocations solve:
max U a + U b = c + qQ
where c = ca + cb , q = q a + q b , under the constraint:

¡ ¢
q + Q = φ (xa − ca ) + φ xb − cb
In the second period, the optimal consumptions are given by:

¡ ¢
φ (xa − ca ) + φ xb − cb
q=Q=
2
so the program becomes:
¡ ¡ ¢¢2
a bφ (xa − ca ) + φ xb − cb
max c + c +
4
First order conditions give:
¡ ¢
φ (xa − ca ) + φ xb − cb 0 s
φ (x − cs ) = 1, s = a, b
2
which implies that ia = ib = i, where the common level of investment i
satisfies:
φ (i) φ0 (i) = 1
Let us now solve the dynamic game in which agents first non cooper-
atively determine their investments, then match on the marriage market
in a frictionless context. Note, first, that once second period incomes have
been generated, the output of a couple male with income y a - female with
income y b is:
¡ a ¢2
¡ a b¢ y + yb
h y ,y =
4
which is supermodular (hya yb = 1/2 > 0)

To keep things simple, let us further assume that the model is fully
symmetric in gender; that is, for each male there exists exactly one female
who has the same income in the initial situation. It is then natural to solve
for a symmetric equilibrium, in which a pair of identical individuals of
opposite sex invest the same amount and generate the same second period
income which put them at the same place in their respective distributions.
Supermodularity implies assortative matching, so the two individuals will
be matched together. Let us (y s ) denote the second period utility of person
s at the stable match; from Chapter 8, we know that:
¡ ¢
0s s ∂h y a , y b
u (y ) =
∂y s
¡ a ¢
y + yb
= = ys (9.1)
2
since y a = y b by symmetry.
Let us now consider the first period investment decision. Agent s chooses
is knowing that the second period income φ (is ) will, through the matching
game, result in a second period utility equal to us (φ (is )). The first period
investment therefore solves:
max
s
xs − is + us (φ (is ))
i
The first order condition gives:
u0s (φ (is )) .φ0 (is ) = 1
and from (9.1):

φ (is ) .φ0 (is ) = 1
which is exactly the condition for efficiency.
Our example clearly relies on a series of strong, simplifying assumptions.
Its message, however, is general. The equilibrium condition ((9.1) in our
case) precisely states that the marginal gain an individual will receive from
a small increase in his trait (here is income) is equal to the marginal impact
of the increase over the output generated at the household level. But this
is exactly the condition for efficiency. Although part of the consumption is
public (which explains the convexity of the output as a function of total in-
come and ultimately the assortative matching), this externality is internal-
ized by the competitive nature of the matching game. My initial investment
has actually three benefits: it increases my future income, which will result
in more consumption tomorrow; it ‘buys’ me a better spouse, since second
period matching is assortative in income; and it improves the fraction of
the marital surplus that I receive. The first effect, by itself, would not be
sufficient to induce the efficient level of investment - that is the essence of
the externality argument. But the logic of competitive matching requires

the three aspects to be considered - and the unambiguous conclusion is
that efficiency is restored.
Finally, what about the opposite line of argument, according to which
agents actually invest too much? The story goes as follows: since agents
compete for the best spouse, a ‘rat race’ situation follows, whereby all
males overinvest in human capital. Well, again, the argument is incorrect in
a matching setting in which transfers are feasible between spouses. Indeed,
one should take into account not only the ‘quality’ (here the wealth) of
the spouse who will be attracted by a higher second period wealth, but
also the ‘price’ that will have to be paid (in terms of surplus sharing). In
a matching game, wealthier spouses come with a higher reservation utility,
thus require giving up a larger fraction of the surplus; as illustrated by the
previous example, this is exactly sufficient to induce the right investment
level. An important remark, however, is that this conclusion would not hold
in a Gale-Shapley framework, in which transfers are not possible and the
spouses’ respective gains are exogenously determined (and do not respond
to competitive pressures on the marriage market). In such a setting, the
‘rat race’ effect is much more likely to occur!
The model developed in this chapter assumes a large competitive mar-
riage market without frictions and we shall demonstrate that premarital
investments in schooling are efficient.
9.2 The Basic Model

We begin with a benchmark model in which men and women are completely
symmetric in their preferences and opportunities. However, by investing in
schooling, agents can influence their marriage prospects and labor market
opportunities. Competition over mates determines the assignment (that
is, who marries whom) and the shares in the marital surplus of men and
women with different levels of schooling, depending on the aggregate num-
ber of women and men that acquire schooling. In turn, these shares together
with the known market wages guide the individual decisions to invest in
schooling and to marry. We investigate the rational-expectations equilib-
rium that arises under such circumstances.
9.2.1 Definitions
When man i and woman j form a union, they generate some aggregate
material output ζ ij that they can divide between them and the utility of
each partner is linear in the share he\she receives (transferable utility).
Man i alone can produce ζ i0 and woman j alone can produce ζ 0j . The
material surplus of the marriage is defined as
zij = ζ ij − ζ i0 − ζ 0j . (9.2)
In addition, there are emotional gains from marriage and the total marital
surplus generated by a marriage of man i and woman j is
sij = zij + θi + θj , (9.3)
where θi and θj represent the non-economic gains of man i and woman j

from their marriage.
9.2.2 Assumptions
There are two equally large populations of men and women to be matched.4
Individuals live for two periods. Each person can choose whether to acquire
schooling or not and whether and whom to marry. Investment takes place
in the first period of life and marriage in the second period. Investment
in schooling is lumpy and takes one period so that a person who invests
in schooling works only in the second period, while a person who does not
invest works in both periods. To simplify, we assume no credit markets.5 All
individuals with the same schooling and of the same gender earn the same
wage rate, but wages may differ by gender. We denote the wage of educated
men by w2m and the wage of uneducated men by w1m , where w2m > w1m . The
wage of educated women is denoted by w2w and that of uneducated women
by w1w , where w2w > w1w . Market wages are taken as exogenous and we do
not attempt to analyze here the feedbacks from the marriage market and
investments in schooling to the labor market. We shall discuss, however,
different wage structures.
We denote a particular man by i and a particular woman by j. We
represent the schooling level (class) of man i by I(i) where I(i) = 1 if i is
uneducated and I(i) = 2 if he is educated. Similarly, we denote the class
of woman j by J(j) where J(j) = 1 if j is uneducated and J(j) = 2 if
she is educated. An important simplifying assumption is that the material
surplus generated by a marriage of man i and woman j depends only on
the class to which they belong. That is,
sij = zI(i)J(j) + θi + θj . (9.4)
We assume that the schooling levels of married partners complement each

other so that
z11 + z22 > z12 + z21 . (9.5)
4 We address the impact of the sex ratio in a separate section below.

5 Allowing borrowing and lending raises issues such as whether or not one can borrow
based on the income of the future spouse and enter marriage in debt.
Except for special cases associated with the presence of children, we

assume that the surplus rises with the schooling of both partners. When
men and women are viewed symmetrically, we also have z12 = z21 .
The per-period material utilities of man i and woman j as singles also
depend on their class, that is ζ i0 = ζ I(i)0 and ζ 0j = ζ 0J(j) and are as-
sumed to increase in I(i) and J(j). Thus, a more educated person has a
higher utility as a single. Men and women who acquire no schooling and
never marry have life time utilities of 2ζ 10 and 2ζ 01 , respectively. A per-
son that invests in schooling must give up the first period utility and, if
he\she remains single, the life time utilities are ζ 20 for men and ζ 02 for
women. Thus, the (absolute) return from schooling for never married men
and women are Rm = ζ 20 − 2ζ 10 and Rw = ζ 02 − 2ζ 01 , respectively.6 The
return to schooling of never married individuals depends only on their own
market wages and we shall refer to it as the labor-market return. However,
investment in schooling raises the probability of marriage and those who
marry have an additional return from schooling investment in the form
of increased share in the material surplus, which we shall refer to as the
marriage-market return to schooling. In addition to the returns in the la-
bor market or marriage market, investment in schooling is associated with
idiosyncratic costs (benefits) denoted by μi for men and μj for women.
The idiosyncratic preference parameters are assumed to be independent
of each other and across individuals. We denote the distributions of θ and
μ by F (θ) and G(μ) and assume that these distributions are symmetric
around their zero means. This specification is rather restrictive because
one might expect some correlations between the taste parameters and the
observable attributes. For instance, individuals that have a low cost of
schooling may also have a high earning capacity and individuals may de-
rive different benefits from marriage depending on the observed quality of
their spouses. One may also expect a correlation between the emotional
valuations of the marriage by the two spouses. Thus, the model is very ba-
sic and intended mainly as an illustration of the possible feedbacks between
the marriage market and investment in schooling.
9.2.3 The Marriage Market

Any stable assignment of men to women must maximize the aggregate sur-
plus over all possible assignments (Shapley and Shubik, 1972).7 The dual
of this linear programming problem posits the existence of non-negative
shadow prices associated with the constraints of the primal that each per-
6 Because we assume away the credit market, the rate of return from schooling in-
vestment depends on consumption decisions and is in utility terms.

7 Note that the maximization of the aggregate surplus is equivalent to the maxi-
mization of aggregate output because the utilities as singles are independent of the
assignment.
son can be either single or married to one spouse. We denote the shadow
price of woman j by uj and the shadow price of man i by vi . The comple-
mentarity slackness conditions require that
zI(i)J(j) + θi + θj ≤ vi + uj , (9.6)
with equality if i and j are married and inequality otherwise.
The complementarity slackness conditions are equivalent to
vi = M ax{M ax[zI(i)J(j) + θi + θj − uj ], 0}
j
(9.7)
uj = M ax{M ax[zI(i)J(j) + θi + θj − vi ], 0},
i
which means that the assignment problem can be decentralized. That is,
given the shadow prices uj and vi , each agent marries a spouse that yields
him\her the highest share in the marital surplus. We can then define ūj =
uj + ζ 0j and v̄i = vi + ζ i0 as the reservation utility levels that woman j
and man i require to participate in any marriage. In equilibrium, a stable
assignment is attained and each married person receives his\her reservation
utility, while each single man receives ζ i0 and each single woman receives
ζ 0j .
Our specification imposes a restrictive but convenient structure in which
the interactions between agents depend on their group affiliation only, that
is, their levels of schooling. Assuming that, in equilibrium, at least one
person in each class marries, the endogenously-determined shadow prices
of man i in I(i) and woman j in J(j) can be written in the form,
vi = M ax(VI(i) + θi , 0) and uj = M ax(UJ(j) + θj , 0) (9.8)
where
VI = M ax[zIJ − UJ ] and UJ = M ax[zIJ − VI ] (9.9)
J I
are the shares that the partners receive from the material surplus of the
marriage (not accounting for the idiosyncratic effects θi and θj ). All agents
of a given type receive the same share of the material surplus zIJ no matter
whom they marry, because all the agents on the other side rank them in the
same manner. Any man (woman) of a given type who asks for a higher share
than the “going rate” cannot obtain it because he (she) can be replaced by
an equivalent alternative.
Although we assume equal numbers of men and women in total, it is
possible that the equilibrium numbers of educated men and women will
differ. We shall assume throughout that there are some uneducated men
who marry uneducated women and some educated men who marry edu-
cated women. This means that the equilibrium shares must satisfy
U2 + V2 = z22 (9.10)
U1 + V1 = z11 (9.11)
We can then classify the possible matching patterns as follows: under strict
positive assortative mating, educated men marry only educated women and
uneducated men marry only uneducated women. Then,
U1 + V2 ≥ z21 , (9.12)
U2 + V1 ≥ z12 . (9.13)
If there are more educated men than women among the married, some
educated men will marry uneducated women and condition 9.12 also will
hold as equality. If there are more educated women than men among the
married, equation 9.13 will hold as equality. It is impossible that all four
conditions will hold as equalities because this would imply
z22 + z11 = z12 + z21 , (9.14)
which violates assumption 9.5 that the education levels of the spouses are
complements. Thus, either educated men marry uneducated women or ed-
ucated women marry uneducated men but not both.
When types mix and there are more educated men than educated women
among the married, conditions 9.10, 9.11 and 9.12 imply
U2 − U1 = z22 − z21 , (9.15)

V2 − V1 = z21 − z11 .
If there are more educated women than men among the married, then
conditions 9.10, 9.11 and 9.13 imply
V2 − V1 = z22 − z12 , (9.16)

U2 − U1 = z12 − z11 .
One may interpret the differences U2 − U1 and V2 − V1 as the (additional)

return to schooling in marriage for women and men, respectively.8 The
quantity z22 − z21 , which reflects the contribution of an educated woman
to the material surplus of a marriage with an educated man, provides an
upper bound on the return that a woman can obtain through marriage,
while her contribution to a marriage with an uneducated man, z12 − z11 ,
provides a lower bound. When there are more educated women than men,
analogous bounds apply to men. When types mix in the marriage market
equilibrium, we see that the side that is in short supply receives the mar-
ginal contribution to a marriage with an educated spouse, while the side
8 The total return from schooling in terms of the output that men receive is Rm if
they remain single and Rm + V2 − V1 if they marry. Similarly, the total return from
schooling in terms of the output that women receive is Rw if they remain single and
Rw + U2 − U1 if they marry.
in excess supply receives the marginal contribution to a marriage with an

uneducated spouse.
We do not exclude the possibility of negative equilibrium values for some
VI or UJ . This would happen if the marginal person in a class is willing to
give up in marriage some of the material output that he\she has as single,
provided that the non-monetary benefit from marriage is sufficiently large.
Then, all men (women) in that class are also willing to do so and the
common factors, VI or UJ may become negative. However, stability implies
that the returns to schooling in marriage, V2 − V1 and U2 − U1 are positive
in equilibrium, provided that the marital surplus rises with the education
of both spouses.
9.2.4 Investment Decisions

We assume rational expectations so that, in equilibrium, individuals know
VI and UJ , which are sufficient statistics for investment decisions. Given
these shares and knowledge of their own idiosyncratic preferences for mar-
riage, θ, and costs of schooling, μ, agents know for sure whether or not they
will marry in the second period, conditional on their choice of schooling in
the first period.
Man i chooses to invest in schooling if
ζ 20 − μi + M ax(V2 + θi , 0) > 2ζ 10 + M ax(V1 + θi , 0). (9.17)
Similarly, woman j chooses to invest in schooling if
ξ 02 − μj + M ax(U2 + θj , 0) > 2ζ 01 + M ax(U1 + θj , 0). (9.18)
Figure 9.1 describes the choices made by different men. Men for whom θ <
−V2 do not marry and invest in schooling if and only if μ < Rm ≡ ζ 20 −2ζ 10 .
Men for whom θ > −V1 always marry and they invest in schooling if and
only if μ < Rm + V2 − V1 . Finally, men for whom −V2 < θ < −V1 marry if
they acquire education and do not marry if they do not invest in schooling.
These individuals will acquire education if μ < Rm + V2 + θ. In this range,
there are two motives for schooling: to raise future earning capacity and to
enhance marriage. We shall assume that the variability in θ and μ is large
enough to ensure that all these regions are non-empty in an equilibrium
with positive VI and UJ . In particular, we assume that, irrespective of
marital status, there are some men and women who prefer not to invest in
schooling and some men and women who prefer to invest in schooling. That
is, μmax > max[Rm +z22 −z12 , Rw +z22 −z21 ] and μmin < min[Rm , Rw ]. We
shall also assume that θmin < −z22 so that, irrespective of the education
decision, there are some individuals who wish not to marry. Note, finally,
that because the support of F (.) extends into the positive range, there are
always some educated men and women who marry and some uneducated
men and women who marry.
The proportion of men who invest in schooling is

Z 1
−V
m m
G(R )F (−V2 ) + [1 − F (−V1 )]G(R + V2 − V1 ) + G(Rm + V2 + θ)f (θ)dθ,
−V2
(9.19)
the proportion of men who marry is
Z 1
−V
[1 − F (−V1 )] + G(Rm + V2 + θ)f (θ)dθ, (9.20)

−V2
and the proportion of men who invest and marry is

Z 1
−V
[1 − F (−V1 )]G(Rm + V2 − V1 ) + G(Rm + V2 + θ)f (θ)dθ. (9.21)

−V2
The higher are the returns from schooling in the labor market, Rm , and in
marriage, V2 −V1 , the higher is the proportion of men who acquire schooling.
A common increase in the levels V2 and V1 also raises investment because
it makes marriage more attractive and schooling obtains an extra return
within marriage. For the same reason, an increase in the market return Rm
raises the proportion of men that marry. Analogous expressions hold for
women.
9.2.5 Equilibrium
In the marriage market equilibrium, the numbers of men and women who
marry must be the same. Using equation 9.20 and applying symmetry, we
can write this condition as
ZV2 ZU2
F (V1 )+ G(R +V2 −θ)f (θ)dθ = F (U1 )+ G(Rw +U2 −θ)f (θ)dθ. (9.22)
m
V1 U1
Under strictly positive assortative mating, the numbers of men and women
in each education group are equal. Given that we impose condition 9.22, it
is necessary and sufficient to require that the numbers of men and women
who marry but do not invest in schooling are the same. Using condition
9.21 and symmetry, we can derive this condition as
F (V1 )G(−Rm + V1 − V2 ) = F (U1 )G(−Rw + U1 − U2 ). (9.23)

Together with conditions 9.10 and 9.11, conditions 9.22 and 9.23 yield a
system of four equations in four unknowns that are, in principle, solvable.
If there is some mixing of types, equation 9.23 is replaced by an inequality

and the shares are determined by the boundary conditions on the returns
to schooling within marriage for either men or women, whichever is ap-
plicable. If there are more educated men than women among the married,
9.23 becomes
F (V1 )G(−Rm + V1 − V2 ) < F (U1 )G(−Rw + U1 − U2 )
and educated women receive their maximal return from marriage while
men receive their minimal return so that condition 9.15 holds. Conversely,
if there are more educated women than men among the married, we have
F (V1 )G(−Rm + V1 − V2 ) > F (U1 )G(−Rw + U1 − U2 )
and educated men receive their maximal return from marriage while ed-
ucated women receive their minimal return so that condition 9.16 holds.
Together with conditions 9.10 and 9.11, we have four equations in four un-
knowns that are again, in principle, solvable. For a proof of existence and
uniqueness see the Appendix.
The two types of solutions are described in Figures 9.2 and 9.3, where we
depict the equilibrium conditions in terms of V1 and V2 after we eliminate
U1 and U2 using 9.10 and 9.11. The two positively-sloped and parallel lines
in these figures describe the boundaries on the returns to schooling of men
within marriage. The negatively-sloped red line describes the combinations
of V1 and V2 that maintain equality in the numbers of men and women who
wish to marry. The positively-sloped blue line describes the combinations
of V1 and V2 that maintain equality in the numbers of men and women that
acquire no schooling and marry. The slopes of these lines are determined
by the following considerations: An increase in V1 (and a reduction in U1 ),
keeping V2 and U2 constant, induces more men and fewer women to prefer
marriage. An increase in V2 holding V1 constant has a similar effect. Thus,
V1 and V2 are substitutes in terms of their impact on the incentives of
men to marry and U1 and U2 are substitutes in terms of their impact on
the incentives of women to marry. Therefore, equality in the numbers of
men and women who wish to marry can be maintained only if V2 declines
when V1 rises. At the same time, an increase in V1 (and a reduction in U1 ),
keeping V2 and U2 constant, increases the number of men that would not
invest and marry and reduces the number of women who wish to acquire
no schooling and marry. Therefore, equality in the numbers of uneducated
men and women who wish to marry can be maintained only if V2 rises
when V1 rises so that the rates of return to education within marriage are
restored.
As long as the model is completely symmetric, that is Rm = Rw and
z12 = z21 , the equilibrium is characterized by equal sharing: V2 = U2 =
z22 /2 and U1 = V1 = z11 /2. With these shares, men and women have iden-
tical investment incentives. Hence, the number of educated (uneducated)
men equals the number of educated (uneducated) women, both among the
singles and the married. Such a solution is described by point e in Figure
9.2, where the lines satisfying conditions 9.22 and 9.23 intersect. There
is a unique symmetric equilibrium. However, with asymmetry, when ei-
ther Rm 6= Rw or z12 6= z21 , there may be a mixed equilibrium where
the line representing condition 9.22 intersects either the lower or upper
bound on V2 − V1 so that condition 9.23 holds as an inequality. Such a case
is illustrated by the point e0 in Figure 9.3. In this equilibrium, educated
men obtain the lower bound on their return to education within marriage,
z21 −z11 . The equilibrium point e0 is on the lower bound and above the blue
line satisfying condition 9.23, indicating excess supply of educated men.
The Impact of the Sex Ratio

Although we assume in this chapter an equal numbers of men and women
in the population, one can extend the analysis to examine the impact of an
uneven sex ratio on the marriage market equilibrium. Let r T 1 represent
the ratio of men to women in the population. Then we modify equations
9.22 and 9.23 as follows, respectively:
ZV2 ZU2
rF (V1 ) + r G(R + V2 − θ)f (θ)dθ = F (U1 ) + G(Rw + U2 − θ)f (θ)dθ.
m
V1 U1
(9.24)
rF (V1 )G(−Rm + V1 − V2 ) = F (U1 )G(−Rw + U1 − U2 ). (9.25)
Note that, even if Rm = Rw and z12 = z21 , the equilibrium with an

uneven sex ratio will not be characterized by equal sharing. For example,
if r > 1 and there are more men than women in the population, then 9.24
implies that V2 and U1 will need to decline and V1 and U2 will need to
rise to ensure that there are equal numbers of men and women who want
to marry. As a result, the marriage-market return for the sex in excess
supply (men) will fall and that of the sex in short supply (women) will rise,
regardless of whether the marriage market equilibrium is strict or mixed.
For r closer to unity, equation 9.25 may still hold, implying a strict sorting
equilibrium with equal numbers of educated men and educated women
among the married. However, with more uneven sex ratios, equation 9.25
may not hold even if Rm = Rw and z12 = z21 . Then, when r > 1 (r <
1) there will be a mixed equilibrium where the line representing condition
9.24 intersects the lower (upper) bound on V2 − V1 . In such cases, condition
9.25 will no longer hold as equality.
Efficiency
We can now demonstrate that in our model individuals’ pre-marital in-
vestments are efficient. Consider, first, a mixed equilibrium in which some
married men are more educated than their wives and consider a particular
couple (i, j) such that the husband is educated and the wife is not. The
question is whether by coordination this couple could have gained by, for
example, changing investments and allowing redistribution between them.
If woman j had gotten educated, the partners together would have gained
ζ 22 − ζ 21 in terms of marital output but the cost of schooling for woman
j would have been her forgone earnings in the first period ζ 01 plus her
idiosyncratic non-monetary cost, μj . The couple would gain from such a
shift only if μj + ζ 01 < ζ 22 − ζ 21 or, equivalently,
μj < z22 − z21 + Rw . (9.26)
But, in the assumed marriage market configuration, z22 − z21 = U2 − U1

and, by assumption, woman j chose not to invest and marry. Therefore, by
9.17,
μj > M ax(U2 +θj , 0)−U1 −θj +Rw ≥ U2 −U1 +Rw = z22 −z21 +Rw . (9.27)
We thus reach a contradiction, implying that there is no joint net gain

from such a rearrangement of investment choices. Nor is it profitable from
the point of view of the couple that the husband would have refrained from
schooling. The couple could gain from such a rearrangement only if the
reduction in the costs of the husband’s schooling exceeds the lost marital
output, μi + ζ 10 > ζ 21 − ζ 11 , or equivalently,
μi > z21 − z11 + Rm . (9.28)
But, in the assumed marriage market configuration, z21 − z11 = V2 − V1

and, by assumption, man i chose to invest and marry. Therefore, by 9.17
μi < Rm +V2 +θi −M ax(V1 +θi , 0) ≤ V2 −V1 +Rm = z21 −z11 +Rm . (9.29)
So, again, we have a contradiction, implying that there is no joint net gain
from such a rearrangement of investment choices. Similar arguments hold
if we consider a mixed equilibrium in which some educated women marry
uneducated men.
Next, consider a strictly assortative equilibrium and a married couple
(i, j) such that neither spouse is educated. Could this couple have been
better off had the partners coordinated their educational investments so
that they both had acquired education? This would be profitable if the
joint gain ζ 22 − ζ 11 in terms of marital output exceeds the total costs of
the two partners ζ 01 + ζ 10 + μj +μi . That is, if
μj + μi < z22 − z11 + Rm + Rw . (9.30)

But, by assumption, man i and woman j married and did not invest, im-
plying that
μj > U2 − U1 + Rw , (9.31)
μi > V2 − V1 + Rm .
By adding up these two inequalities, and using the equilibrium conditions

z22 = U2 +V2 and z11 = U1 +V1 , we see that it is impossible to satisfy (9.29).
Hence, there is no joint gain from such a rearrangement of investments. By
similar arguments, there is no joint gain for a couple in which both partners
are educated from a coordinated reduction in their investments.
We conclude that the equilibrium shares that individuals expect to re-
ceive within marriage induce them to fully internalize the social gains from
their premarital investments. An important piece of this argument is that
the marriage market is large in the sense that individual perturbations in
investment do not affect the equilibrium shares. In particular, a single agent
cannot tip the market from excess supply to excess demand of educated
men or women. This efficiency property of large and frictionless marriage
markets has been noted by Cole et al. (2001), Felli and Roberts (2002)
Peters and Siow (2002) and Iyigun and Walsh (2007). In contrast, markets
with frictions or small number of traders are usually characterized by in-
efficient premarital investments (Lommerud and Vagstad, 2000, Baker and
Jacobsen, 2007).9
9.3 Gender Differences in the Incentive to Invest

In this section, we discuss differences between women and men that can
cause them to invest at different levels. We discuss two possible sources of
asymmetry:
• In the labor market, women may receive lower wages than men; this
could lower the schooling return for working women.
• In marriage, women may be required to take care of the children; this
would lower the schooling return for married women.
Either of the above causes can induce women to invest less in schooling.
Therefore, the lower incentives of women to invest can create equilibria
9 Peters (2007) formulates premarital investments as a Nash game in which agents
take as given the actions of others rather than the expected shares (as in a market
game). In this case, inefficiency can persist even as the number of agents approaches
infinity. The reason is that agents play mixed strategies that impose on other agents
the risk of being matched with an uneducated spouse, leading to under-investment in
schooling.
with mixing, where educated men are in excess supply and some of them
marry less-educated women.
To illustrate these effects we shall perform several comparative statics
exercises, starting from a benchmark equilibrium with strictly positive as-
sortative matching, resulting from a complete equality between the sexes
in wages and household roles such that w1m = w1w = w1 , w2m = w2w = w2
and τ = 0.
9.3.1 The Household

We use a rudimentary structural model to trace the impact of different
wages and household roles of men and women on the marital output and
surplus. We assume that, irrespective of the differences in wages or house-
hold roles, men and women have the same preferences given by
u = cq + θ, (9.32)
where c is a private good, q is a public good that can be shared if two people
marry but is private if they remain single, and θ is the emotional gain from
being married (relative to remaining single). The household public good is
produced according to a household production function
q = e + γt, (9.33)
where e denotes purchased market goods, t is time spent working at home
and γ is an efficiency parameter that is assumed to be independent of
schooling.10
This specification implies transferable utility between spouses and allows
us to trace the impact of different market wages or household roles on
the decisions to invest and marry. Time worked at home is particularly
important for parents with children. To simplify, we assume that all married
couples have one child and that rearing it requires a specified amount of
time t = τ , where τ is a constant such that 0 ≤ τ < 1. Initially, we shall
assume that, due to social norms, all the time provided at home is supplied
by the mother. Also, individuals who never marry have no children and for
them we set τ = 0.11
m
If man i of class I with wage wI(i) marries woman j of class J with wage
w m w
wJ(j) , their joint income is wI(i) + (1 − τ )wJ(j) . Any efficient allocation
10 A plausible generalization is to allow the mother’s schooling level to affect positively

child quality. This would be consistent with the findings of Behrman (1997) and Glewwe
(1999), for example. However, the qualitative results will be unaffected as long as school-
ing has a larger effect on market wages than on productivity at home. The fact that
educated women participate more in the labor market than uneducated women supports
such an assumption.
1 1 We make no distinction here between cohabitation and marriage. So either no one
cohabits, or, if two individuals cohabit, they behave as a married couple.

of the family resources maximizes the partners’ sum of utilities given by

m w
[wI(i) + (1 − τ )wJ(j) − e](e + τ γ) + θi + θj . In an interior solution with
a positive money expenditure on the public good, the maximized material
output is
m w
[wI(i) + τ γ + (1 − τ )wJ(j) ]2
ζ ij = . (9.34)
4
Note that the wages of the husband and wife complement each other in
generating marital output, which is a consequence of sharing the public
good.12
An unmarried man i solves
M ax ci ei (9.35)
ei ,ci
subject to
m
ci + ei = wI(i) , (9.36)
m
and his optimal behavior generates a utility level of ζ i0 = (wI(i) /2)2 . A
single woman j solves an analogous problem and obtains ζ 0j = (wJ(j) /2)2 .
w
Therefore, the total marital surplus generated by the marriage in the sec-
ond period is
m w
[wI(i) + τ γ + (1 − τ )wJ(j) ]2 − (wI(i)
m 2 w
) − (wJ(j) )2
sij = + θi + θj
4
≡ zI(i)J(j) + θi + θj . (9.37)
The surplus of a married couple arises from the fact that married partners
jointly consume the public good. If the partners have no children and τ = 0,
the gains arise solely from the pecuniary expenditures on the public good.
In this case, the surplus function is symmetric in the wages of the two
spouses. If the couple has a child, however, and the mother takes care of it,
then the mother’s contribution to the household is a weighted average of her
market wage and productivity at home. We assume that w2w > γ > w1w so
that having children is costly for educated women but not for uneducated
women. The surplus function in (9.37) maintains complementarity between
1 2 The first-order condition for e is

m w
[wI(i) + (1 − τ )wJ(j) − e] − (e + τ γ) ≤ 0.
m +(1−τ )ww
Hence, e = [wI(i) −τ γ] / 2 in an interior solution. The maximized material
J(j)
m + τ γ + (1 − τ )ww ]2 / 4. If e = 0, the maximal material
output in this case is [wI(i) J(j)
m w ]τ γ, which would imply an additive surplus function,
output is [wI(i) + (1 − τ )wJ(j)
contradicting our assumption of complementarity. A sufficient condition for a positive e
is w1m + (1 − τ )w1w > τ γ if the wife works at home and w1w + (1 − τ )w1m > τ γ if the
husband works at home. We assume hereafter that these conditions hold.
the wages of the husband and wife, which is a consequence of sharing the
public good. However, the assumed asymmetry in household roles between
men and women implies that a higher husband’s wage always raises the
surplus but a higher mother’s wage can reduce the surplus. In other words,
it may be costly for a high-wage woman to marry and have a child because
she must spend time on child care, while if the mother does not marry, her
2
utility as a single remains wJ(j) /4. In addition, it is no longer true that
13
z21 = z12 .
Since we have assumed here that, due to social norms, all the time pro-
vided at home is supplied by the mother, all the gains from marriage arise
from sharing a public good and the wages of the partners complement each
other so that z11 + z22 > z12 + z21 . In later sections, we discuss endogenous
specialization whereby couples act efficiently and the partner with lower
wage works at home. For sufficiently low time requirements, that is, τ close
to 0, complementarity continues to hold. However, for τ close to 1, the
wages of the two partners become substitutes, that is, z11 + z22 < z12 + z21 ,
because wage differentials between spouses increase the gain from special-
ization (see Becker, 1991, ch. 2). Thus, whether couples act efficiently or
according to norms influences the equilibrium patterns of assortative mat-
ing.14
9.3.2 The Impact of the Wage Gap

We are now ready to examine the implications of gender wage differences.
The gender difference in wages can be an outcome of discrimination associ-
ated, for instance, with fewer opportunities for investment on the job. Such
discrimination can reduce or increase the incentives of women to invest,
depending on whether discrimination is stronger at the low or high levels
of schooling.
1 3 For instance, when the wages of men and women are equal but τ > 0, we have
τ (w2 − w1 ) w2 + w1
z21 − z12 = [(1 − τ ) + τ γ] > 0.
2 2
1 4 Forfixed household roles, the second cross derivative of the surplus function with
respect to wages is positive, implying complementarity. But with endogenous household
roles, the relevant measure of complementarity is embedded in the maximized marital
gains that can change discontinuously as household roles change. Suppose that w2m >
w2w > w1m . Let
f (τ ) ≡ 4(z11 + z22 − z12 − z21 )

= [w1m + τ γ + (1 − τ )w1w ]2 + [w2m + τ γ + (1 − τ )w2w ]2
−[w2w + τ γ + (1 − τ )w1m )]2 − [w2m + τ γ + (1 − τ )w1w ]2 .
Then, f (τ ) > 0 if τ = 0 and f (τ ) < 0 if τ = 1, where ∀ τ ∈ [0, 1], f 0 (τ ) < 0.

Define the (relative) wage gap among educated individuals as d2 =

w2w /w2m and let the gender wage gap between uneducated individuals be
d1 = w1w /w1m . Starting from the benchmark equilibrium with strictly posi-
tive assortative mating and equal shares (point e in Figure 9.4), we examine
the impact of a difference in the market returns from schooling of women
and men. Specifically, we consider an increase in the wage of educated men,
w2m , combined with a reduction in the wage of educated women, w2w , hold-
ing the wage of uneducated men at the benchmark value, w1 . To isolate the
role of market returns, we assume that the increase in the wage of educated
men exactly compensates the reduction in the wage of educated women so
that marital output is unaffected and symmetry is maintained.15 In other
words, the change in wages affect directly only the returns as singles, Rm
and Rw . For now, we assume that discrimination is uniform across schooling
levels so that d1 = d2 ≡ d < 1 and women have a lower market return from
schooling investment than men.16 Later, we shall discuss a case in which
discrimination against educated women is weaker so that d1 < d2 < 1.
With uniform discrimination, the returns to investment in schooling for
never married men and women, respectively, are
w2m 2 wm
Rm = z20
m m
− 2z10 =( ) − 2( 1 )2 , (9.38)
2 2
and
w2w 2 ww
Rw = z02
w w
− 2z01 =( ) − 2( 1 )2 = d2 Rm < Rm . (9.39)
2 2
The higher market return from schooling of men encourages their invest-
ment in schooling and also strengthens their incentives to marry, because
schooling obtains an additional return within marriage. In contrast, the
lower return to schooling for women reduces their incentives to invest and
marry. These changes create excess supply of men who wish to invest and
marry. Consequently, to restore equilibrium, the rates of returns that men
receive within marriage must decline implying that, for any V1 , the value
of V2 that satisfies conditions (9.22) and ( 9.22) must decline. These shifts
in the equilibrium lines are represented by the broken blue and red lines in
Figure 9.4.
1 5 When wages change z

I(i)J(j) usually changes. Also, when wages differ by gender, we
generally do not maintain symmetry in the contribution of men and women to marriage
m ww
so that z12 6= z21 . It is only in the special case in which the product wI(i) remains
J(j)
invariant under discrimination that the marital surplus generated by all marriages is
intact. The qualitative results for shares are not affected by this simplification.
1 6 In standard human capital models where the only cost of investment is forgone earn-
ings and the only return is higher future earnings, uniform discrimination has no impact
on investment. In this model, however, the absolute market returns are added to the
returns within marriage, which together determine investment decisions (see equations
(16) and (17)). Therefore, the absolute market returns to schooling matter in our model.
For moderate changes in wages, strictly positive assortative mating con-

tinues to hold. However, the equilibrium value of V2 declines and educated
men receive a lower share of the surplus than they do with equal wages
in any marriage. That is, as market returns of men rise and more men
wish to acquire education, the marriage market response is to reduce the
share of educated men in all marriages. When the gap between Rm and
Rw becomes large, the equilibrium shifts to a mixed equilibrium, where
some educated men marry uneducated women. That is, because of their
higher tendency to invest, some educated men must “marry down.” This
equilibrium is represented by the point e0 in Figure 9.4, where the bro-
ken red line representing equality in the numbers of men and women that
wish to marry (condition ( 9.22)) intersects the green line representing the
lower bound on the share that educated men obtain in the marital surplus,
z21 − z11 . As seen, both V1 and V2 are lower in the new equilibrium so that
all men (women), educated and uneducated, receive lower (higher) shares
of the material surplus when men have stronger market incentives to invest
in schooling than women.
These results regarding the shares of married men and women in the
material surplus must be distinguished from the impact of the shares in
the material output. If men get a higher return from schooling as singles
(due to the fact that their labor-market return from schooling is higher than
that of women), then their share of the material output can be higher even
though they receive a lower share of the surplus. The same remark applies
to our subsequent analysis as well; one can obtain sharper comparative
static results on shares of the material surplus than those on shares of the
material output.
9.3.3 The Impact of Household Roles

Recall that we assume that the wife alone spends time on child care. To
investigate the impact of this constraint, we start again at the benchmark
equilibrium and examine the impact of an increase in τ , holding the wages
of men and women at their benchmark values, that is w1m = w1w = w1 and
w2m = w2w = w2 . Such an increase reduces the contribution that educated
women make to marital output and raises the contribution of uneducated
women. That is, z11 and z21 rise because uneducated women are more pro-
ductive at home, γ > w1 , while z12 and z22 decline because educated women
are less productive at home, γ < w2 . Consequently, both equilibrium lines
corresponding to conditions (9.22) and (9.23) shift down so that V2 is lower
for any V1 . At the same time, the boundaries on the rate of return from
schooling that men can obtain within marriage shift as z21 − z11 rises and
z22 − z12 declines. These changes are depicted in Figure 9.5.
For moderate changes in τ , strictly positive assortative mating with equal
sharing continues to hold. As long as a symmetric equilibrium is main-
tained, the returns to schooling that men and women receive within mar-
riage, V2 −V1 and U2 −U1 , are equal. Hence, men and women have the same
incentives to invest. But because the material surplus (and consequently
utilities within marriage) of educated men and women, z22 /2, declines with
τ , while the material surplus of uneducated men and women, z11 /2, rises,
both men and women will reduce their investments in schooling by the
same degree.
As τ rises further, the difference in the contributions of men and women
to marriage can rise to the extent that an educated man contributes to
a marriage with uneducated woman more than an educated woman con-
tributes to a marriage with an educated man.17 That is,
z21 − z11 > z22 − z21 . (9.40)
Condition (9.40) implies that the lower bound on the return to schooling
that men receive within marriage exceeds the upper bound on the return to
schooling that women receive within marriage. In this event, the symmetric
equilibrium in Figure 9.5 is eliminated and instead there is a mixed equi-
librium with some educated men marrying uneducated women (point e0 in
Figure 9.5). This outcome reflects the lower incentive of educated women to
enter marriage and the stronger incentive of men to invest because their re-
turn from schooling within marriage, V2 −V1 = z21 −z11 , exceeds the return
to schooling that women can obtain within marriage. Consequently, some
educated men must “marry down” and match with uneducated women.
9.3.4 Division of Labor and Career Choice

We can further refine the family decision problem by letting the partners
decide who shall take care of the children. Reinterpreting τ as a temporal
choice, imagine that one of the partners must first spend τ units of time
on the child and later enter the labor market and work for the remainder
of the period (length 1 − τ ).
1 7 Consider the expression
h(w1 , w2 , τ ) ≡ 2z21 − z11 − z22 = 2[w2 + τ γ + (1 − τ )w1 ]2

−[w1 + τ γ + (1 − τ )w1 ]2 − [w2 + τ γ + (1 − τ )w2 ]2
as a function of w1 and w2 and τ . For w1 = w2 = γ, h(γ, γ, τ ) = 0 and
h1 (γ, γ, τ ) = −4γτ ,
h2 (γ, γ, τ ) = 4γτ .
Therefore, for a positive τ , w1 slightly below γ and w2 slightly above γ, h(w1 , w2 , τ ) > 0.
Also
h3 (w1 , w2 , τ ) = (w2 − w1 )[w2 (4 − 2τ ) + 2τ (2γ − w1 )] > 0
and for all w2 > γ > w1 , h(w1 , w2 , 0) < 0 and h(w1 , w2 , 1) > 0. Hence, the larger is τ
the broader will be the range in which h(w1 , w2 , 0) > 0.
An important idea of Becker (1991, ch. 2) is that wage differences among

identical spouses can be created endogenously and voluntarily because of
learning by doing and increasing returns. Thus, it may be optimal for the
household for one of the spouses to take care of the child and for the other
to enter the labor market immediately, thereby generating a higher wage in
the remainder of the period. Thus, by choosing schooling ahead of marriage
one can influence his\her household role within marriage.
Because we assume transferable utility between spouses, household roles
will be determined efficiently by each married couple, as long as there is
ability to commit to a transfer scheme, whereby the party that sacrifices
outside options when he\she acts in a manner that raises the total surplus
is compensated for his/her action. In particular, the partners will assign
the spouse with the lower wage to take care of the child. In the previous
analysis, there was no need for such a commitment because the division of
the surplus was fully determined by attributes that were determined prior
to marriage via competition over mates who could freely replace partners.
However, if time spent on child care affects one’s labor market wages subse-
quently, the cost of providing childcare can differ between the two spouses.
Thus, implementing the efficient outcome might require some form of com-
mitment even if (re)matching is frictionless. A simple, enforceable, prenup-
tial contract is one in which both partners agree to pay the equilibrium
shares VI to the husband and UJ to the wife in case of divorce. By making
those shares the relevant threat points of each spouse, this contract sus-
tains the equilibrium values VI and UJ in marriage, which is sufficient to
attain the efficient household division of labor.
If there is discrimination against women and they receive lower market
wages than men, then the wife will be typically assigned to stay at home,
which will erode her future market wage and reinforce the unequal division
of labor. Similarly, if there are predetermined household roles such that
women must take care of their child, then women will end up with lower
market wages. Thus, inequality at home and the market are interrelated.18
Models of statistical discrimination tie household roles and market wages
through employers’ beliefs about female participation. Typically, such mod-
els generate multiple equilibria and inefficiency (Hadfield, 1999, Lommerud
and Vagstad, 2000). Here, we do not require employers’ beliefs to be cor-
rect. Instead, we think of household roles and discrimination as processes
that evolve slowly and can be taken as exogenous in the medium run.
1 8 Related papers that emphasize the dual-feedback mechanism between the intensity
of home work and labor market wages are Albanesi and Olivetti ( 2009) and Chichilnisky
(2005).
9.3.5 Why Women May Acquire More Schooling than Men

We have examined two possible reasons why women may invest less than
men in schooling. The first is that women may receive lower return from
schooling investment in the market because of discrimination. The second
reason is that women may receive a lower return to schooling in marriage
because of the need to take care of children (due to social and cultural
norms or the biological time requirements of child care).
Over time, fertility has declined and women’s wages have risen in indus-
trialized countries, a pattern being replicated in many developing countries
too. This is consistent with increased investment in education by women.
The fact that women are now slightly more educated than men, on average,
appears surprising given the fact that women still earn substantially less
than men. However, in dealing with investments in education, the crucial
issue is whether the gender wage gap rises or declines with schooling, or
equivalently, whether women obtain a higher rate of return from schooling.
There is some evidence that this is indeed the case and that the gender
wage gap declines with schooling (see Chiappori et al, 2009 and Dougherty,
2005).
Now consider a comparison of the following two situations. An “old”
regime in which married women must spend a relatively large fraction of
their time at home and a “new” regime in which, because of reductions
in fertility and improved technology in home production, married women
spend less time at home and work more in the market (see chapter 1 tables
8a and 8b).19 Assume further that women suffer from statistical discrimina-
tion because employers still expect them to invest less on the job. However,
this discrimination is weaker against educated women because they are ex-
pected to stay longer in the labor market than uneducated women. Finally,
assume that in the old regime norms were relevant but in the new regime
the roles are determined efficiently (for some evidence, see Chiappori et
al, 2009). It is then possible that in the new regime women will invest in
schooling more than men. The presence of discrimination raises the re-
turn of women relative to men because schooling serves as an instrument
for women to escape discrimination. The fact that women are still tied up
in home work lowers their return from schooling relative to men because
women obtain lower returns from schooling within marriage. However, as
women raised their labor force participation due to technological changes or
break of norms, this second effect weakens and the impact of discrimination
can dominate.
1 9 Greenwood et al.(2005) and Fernandez (2007) discuss the impact of technological
advance and change in norms on the rise in female participation. Mulligan and Rubin-
stein (2008) emphasize the role of higher rewards for ability (reflected in the general
increase in wage inequality) in drawing married women of high ability into the labor
market.
In Figure 9.6, we display the transition between the two regimes. We

assume that d2 > d1 so that discrimination against women is lower at
the higher level of schooling. This feature generates stronger incentives for
women than men to invest in schooling. However, the fact that women must
spend time working at home has the opposite effect. We then reduce the
amount of time that the mother has to spend at home, τ , and raise the
wage that educated women receive (so that d2 rises), which strengthens
the incentives of women to invest in schooling and to marry. Therefore,
holding the marriage surplus zIJ constant, an increase in V2 relative to V1
is required to maintain equality between the number of men who wish to
invest and marry and the number of women who wish to invest and marry.
This effect is represented by the upwards shifts in the broken red and blue
lines in Figure 9.6.20 The impact is assumed to be large enough to generate
an equilibrium in which the two equilibrium requirements — equality of
the numbers of men and women who acquire no schooling and marry (the
broken blue line) and equality of the total numbers of men and women who
wish to marry (the broken red line) — yield an intersection above the upper
bound on the returns from schooling that men can receive within marriage.
Therefore, strictly positive assortative mating cannot be sustained as an
equilibrium and the outcome is a mixed equilibrium in which there are
more educated women than men among the married and some educated
women marry uneducated men. This new mixed equilibrium is indicated
by the point e00 in Figure 9.6.
9.4 A Numerical Example

Suppose that μ and θ are uniformly and independently distributed. Al-
though wages vary across the two regimes, we assume that in both regimes,
educated women are more productive in the market and uneducated women
are more productive at home. We further assume that in both regimes, men
earn more than women with the same schooling level but educated women
earn more than uneducated men. Finally, in both regimes, women have a
higher market return from schooling. The transition from the old regime
to the new regime is characterized by three features: (i) productivity at
home is higher and women are required to work less at home; (ii) men and
women obtain higher market returns from schooling; and (iii) couples move
from a traditional mode to an efficient one in which the high-wage spouse
works in the market.
All the above economic changes raise the gains from marriage and would
2 0 Because the marital surplus matrix, z , also changes, the equilibrium curves did
IJ
not shift up. In fact, for the parameters of Figure 9.6, there is a range over which the
equilibrium line representing market-clearing in the marriage market shifts down. This,
however, has no bearing on the equilibrium outcome.
cause higher marriage rates. To calibrate the model, we assume that the
variance in the preference for marriage rises over time which, other things
being the same, reduces the propensity to marry. We thus assume that in
both periods μ is distributed over the interval [−4, 4], while θ is distributed
over the intervals [−4, 4] and [−8, 8] in the old and the new regimes,
respectively. It is important to note that the shift in the distribution of
θ has no impact on the equilibrium surplus shares, which are our main
concern. However, it changes the proportion of individuals who invest and
marry given these shares. Table 9.1 reflects these assumptions.
Table 9.1: Parameters in the old and the new regimes
Parameter Old Regime New Regime

Wage of uneducated men w1m = 2 w1m = 2.375
Wage of uneducated women w1w = 1.2 w1w = 1.425
Wage of educated men w2m = 3 w2m = 4.0
Wage of educated women w2w = 2.4 w2w = 3.2
Wage difference among the uneducated d1 = .6 d1 = .6
Wage difference among the educated d2 = .8 d2 = .8
Market return to schooling, men Rm = .25 Rm = 1.18
Market return to schooling, women Rw = .72 Rw = 1.54
Work requirements τ = .8 τ = .3
Productivity at home γ=2 γ = 2.5
Distribution of tastes for schooling [−4, 4] [−4, 4]
Distribution of tastes for marriage [−4, 4] [−8, 8]
Norms Wife at home Efficient
The marriage market implications of these changes are summarized in

Tables 9.2-9.5 below.
Table 9.2: Impact of parameter changes on marital surplus
Old regime
Uned. wife Educ. wife

Uned. husband z11 = 2.33 z12 = 1.72
Educ. husband z21 = 3.25 z22 = 2.76
New Regime

Uned. husband z11 = 2.33 z12 = 3.90
Educ. husband z21 = 3.75 z22 = 5.66
A decrease in the amount of time worked at home, raises the contribution

of an educated woman to the material surplus and lowers the contribution

of an uneducated woman. Therefore, in the old regime with τ = .8, the
material surplus declines with the education of the wife when the husband
is uneducated, while in the new regime with τ = .3, it rises. This happens
because educated women are more productive in the market than unedu-
cated women but, by assumption, equally productive at home. In the old
regime, if an educated wife would marry an uneducated man (which does
not happen in equilibrium) she would be assigned to household work even
though she has a higher wage than her husband. In the new regime, cou-
ples act efficiently, household roles are reversed and educated women do
marry uneducated men. Note that for couples among whom both husband
and wife are uneducated, the wife continues to work at home in the new
regime, because she has the lower wage. The parameters are chosen in such
a way that technology has no impact on the marital surplus of such couples.
In the new regime, uneducated women work less time at home but their
productivity at home is higher as well as the wage that they obtain from
work.
Table 9.3: Impact of parameter changes on the equilibrium shares
Old regime
Uneducated Educated
Men V1 = .76 V2 = 1.68
Women U1 = 1.57 U2 = 1.09
New Regime
Uneducated Educated
Men V1 = 1.13 V2 = 2.88
Women U1 = 1.20 U2 = 2.78
Compared with the old regime, educated women receive a higher share
of the marital surplus in the new regime, while uneducated women receive
a lower share. These changes reflect the higher (lower) contributions to
marriage of educated (uneducated) women. The marital surplus shares of
both educated and uneducated men rise as a consequence of the rising
productivity of their wives.
The implied returns from schooling within marriage in the old regime
are
U2 − U1 = 1.09 − 1.57 = z22 − z21 = 2.76 − 3.25 = −.49 ,

V2 − V1 = 1.68 − .76 = z21 − z11 = 3.25 − 2.33 = .92 .
That is, men receive the lower bound on their return from schooling within
marriage while women receive the upper bound on their return from school-
ing. This pattern is reversed in the new regime:
U2 − U1 = 2.78 − 1.20 = z12 − z11 = 3.90 − 2.33 = 1.58,

V2 − V1 = 2.88 − 1.13 = z22 − z12 = 5.66 − 3.90 = 1.75,
where women receive their lower bound and men receive their upper bound.
Both men and women receive a higher return from schooling within mar-
riage in the new regime, reflecting the increased efficiency although the rise
for women is much sharper.
Table 9.4: Impact of parameter changes on the investment and marriage rates∗
Old Regime
Married Unmarried All
Educ. .452, .335 .153, .215 .606, .550

Uned. .211, .323 .183, .122 .394, .450
All .662, .666 .334, .334 1
New Regime
Educ. .577, .590 .207, .226 .784, .816

Uned. .077, .063 .139, .121 .216, .184
All .653, .653 .347, .347 1
∗ First and second entries in each cell refer to men and women resp.
In the old regime, more men invest in schooling than women and some
educated men marry down to match with uneducated women. This pat-
tern is reversed in the new regime and women invest in schooling more
than men and some educated women marry down to join uneducated men.
That is, women increase their investment in schooling more than men. Al-
though market returns have risen for both men and women, the returns for
schooling within marriage have risen substantially more for women. The
basic reason for that is the release of married women from the obligation
to spend most of their time at home, due to the reduction in the time re-
quirement of child care and the change in norms that allow educated women
who are married to uneducated men to enter the labor market. Uneducated
men gain a higher share in the surplus in all marriages because of their new
opportunity to marry educated women, while uneducated women lose part
of their share in the marital surplus in all marriages because they no longer
marry educated men. Notice that the proportion of educated women who
remain single declines from .215/.550 = 0.39 to 226/.816 = 0.28 in the new
regime. In contrast, the proportion of educated men who marry remains
roughly the same, 0.153/0.606 = 0.28 and 0.207/0.784 = 0.26 in the old
and new regimes, respectively. This gender difference arises because, under
the old regime, women were penalized in marriage by being forced to work
at home.
We can use these examples to discuss the impact of norms. To begin
with, suppose that in the old regime couples acted efficiently and, if the
wife was more educated than her husband, she went to work full time and
the husband engaged in child care. Comparing Tables 9.2 and 9.5, we see
that the impact of such a change on the surplus matrix is only through the
rise in z12 . Because women receive lower wages than men at all levels of
schooling, the household division of labor is not affected by the norms for
couples with identically educated spouses; for all such couples, the husband
works in the market and the wife takes care of the child. However, the norm
does affect the division of labor for couples among whom the wife has a
higher education level than her husband. This is due to our assumptions
that educated women have a higher wage than uneducated men in the
labor market and their market wage exceeds their productivity at home.
In contrast to the case in which the mother always works at home, we
see in Table 9.5 that the education levels now become substitutes, namely
z11 + z22 < z12 + z21 , implying that we can no longer assume that there will
be some educated men married to educated women and some uneducated
men married to uneducated women. More specifically, an educated woman
contributes more to an uneducated man than she does to an educated man
(that is z12 − z11 > z22 − z21 ) so that uneducated men can bid away the
educated women from educated men. Thus changes in norms can influence
the patterns of assortative mating.
Table 9.5: Impact of norms on material surplus
Old regime, efficient

Uned. husband z11 = 2.33 z12 = 2.40
Educ. husband z21 = 3.25 z22 = 2.76
New Regime with norms

Uned. husband z11 = 2.33 z12 = 3.23
Educ. husband z21 = 3.75 z22 = 5.66
Consider, next, the possibility that the norms persist also in the new
regime and the mother must work at home even if she is more educated
than her husband. Again, the norm bites only in those marriages in which
the wife is more educated than the husband. In the new regime, positive
assortative mating persists independently of the norms. However, the mix-
ing equilibrium in which some educated women marry uneducated men is
replaced by strict assortative mating in which educated men marry only ed-
ucated women and uneducated men marry only uneducated women. Thus,
again, norms can have a qualitative impact on the type of equilibrium that
emerges.
The new marriage and investment patterns are presented in the lower
panel of Table 9.6. The main difference is that educated women are less
likely to marry when the norms require them to work at home, where they
are relatively less efficient.
Table 9.6: Impact of norms on investment and marriage rates (new regime)∗
Efficient work pattern
Educ. .577, .589 .207, .126 .784, .816

Uned. .077, .063 .139, .121 .216, .184
All .653, .653 .347, .347 1
Wife work pattern

Educ. .583, .583 .207, .227 .790, .810

Uned. .070, .070 .140, .120 .210, .190
All .653, .653 .347, .347 1
∗ The first and second entry in each cell refer to men and women resp.
Consider, finally, the impact on the shares in the material surplus when
norms are replaced by an efficient allocation in the new regime (see Table
9.7). The removal of social norms that the wife must work at home benefits
uneducated men and harms uneducated women. This example illustrates
the differences between the predictions of general equilibrium models with
frictionless matching, like the one we present here, and partial equilibrium
models that rely on bargaining. The latter would predict that no woman
would lose from the removal of norms that forces women in general to stay
at home and take care of the child, but as this example demonstrates, the
market equilibrium can change and uneducated women are hurt because
they can no longer marry with educated men.
Table 9.7: Impact of norms on the equilibrium shares in the new regime
Efficient pattern of work
Uneducated Educated
Men V1 = 1.13 V2 = 2.89
Women U1 = 1.20 U2 = 2.78
Wife always works at home

Uneducated Educated
Men V1 = 1.06 V2 = 2.89
Women U1 = 1.28 U2 = 2.77
9.5 Conclusions
In standard models of human capital, individuals invest in schooling with
the anticipation of being employed at a higher future wage that would com-
pensate them for the current foregone earnings. This chapter added another
consideration: the anticipation of being married to a spouse with whom
one can share consumption and coordinate work activities. Schooling has
an added value in this context because of complementarity between agents,
whereby the contribution of the agents’ schooling to marital output rises
with the schooling of his\her spouse. In the frictionless marriage market
considered here, the matching pattern is fully predictable and supported
by a unique distribution of marital gains between partners. Distribution
is governed by competition because for each agent, there exists a perfect
substitute that can replace him\her in marriage. There is thus no scope for
bargaining and, therefore, premarital investments are efficient.
We mentioned two interrelated causes that may have diminish the incen-
tives of women to invest in schooling in the past: lower market wages and
larger amount of household work. With time, the requirement for wives
to stay at home have relaxed and discrimination nay have decreased too
but probably not to the same extent21 . Although we did not fully specify
the sources of discrimination against women in the market, we noted that
such discrimination tends to decline with schooling, which strengthens the
incentive of women to invest in schooling. This is a possible explanation for
the slightly higher investment in schooling by women that we observe to-
day. We do not view this outcome as a permanent phenomenon but rather
as a part of an adjustment process, whereby women who now enter the la-
bor market in increasing numbers, following technological changes at home
2 1 Whether discrimination has declined is debated; see Mulligan and Rubinstein (2008).
and in the market that favor women, must be “armed” with additional
schooling to overcome norms and beliefs that originate in the past.
We should add that there are other possible reasons for why women
may invest in schooling more than men. One reason is that there are more
women than men in the marriage market at the relatively young ages at
which schooling is chosen, because women marry younger. Iyigun and Walsh
(2007) have shown, using a similar model to the one discussed here, that in
such a case women will be induced to invest more than men in competition
for the scarce males. Another reason is that divorce is more harmful to
women, because men are more likely to initiate divorce when the quality of
match is revealed to be low. This asymmetry is due to the higher income
of men and the usual custody arrangements (see Chiappori and Weiss,
2007). In such a case, women may use schooling as an insurance device
that mitigates their costs from unwanted divorce.
9.6 References
[2] Baker, Matthew J. and Joyce P. Jacobsen, ‘Marriage, Specialization,

and the Gender Division of Labor’, Journal of Labor Economics, 25
(2007), 763-793.

[4] Behrman, Jere R., ‘Mother’s Schooling and Child Education: A Survey’,
PIER Working Paper No. 97-025, (1997).
[5] Chiappori Pierre-Andre and Yoram Weiss, ‘Divorce, Remarriage, and

[6] Chiappori, Pierre-Andre, Iyigun, Murat, and Yoram Weiss, ‘Investment

in Schooling and the Marriage Market’, American Economic Review,
99 (2009), 1689-1713.

Economics, 12 (2008), 828-844.
[8] Cole, Harold Linh, Mailath, George J., and Andrew Postlewaite, ‘Ef-
ficient Non-Contractible Investments in Finite Economies’, Journal of
Economic Theory, 101 (2001), 333-373.
[9] Dougherty, Christopher, ‘Why Are the Returns to Schooling Higher for
Women than for Men?’, Journal of Human Resources, 40 (2005),
969-988.
[10] Felli, Leonardo and Kevin W.S. Roberts, ‘Does Competition Solve the
Hold-Up Problem?’, CEPR Discussion Paper No. 3535, (2002).
[11] Fernandez, Raquel, ‘Culture as Learning: The Evolution of Female

Labor Force Participation over a Century’, NBER Working Paper No.
13373, (2007).
[12] Glewwe, Paul, ‘Why Does Mother’s Schooling Raise Child Health in
Developing Countries? Evidence from Morocco’, The Journal of Hu-
man Resources, 34 (1999), 124-159.
[13] Goldin, Claudia, ‘Career and Family: College Women Look at the
Past’, in Blau, Francine D., (eds.) and Ehrenberg, Ronald G., (eds.),
Gender and Family Issues in the Workplace, (New York, Russell
Sage Foundation, 1997).
[14] Goldin, Claudia, Katz, Lawrence F. and Ilyana Kuziemko, ‘The Home-
coming of American College Women: The Reversal of the College Gender
Gap’, Journal of Economic Perspectives, 20 (2006), 133-156.
[15] Greenwood, Jeremy, Seshadri, Ananth and Mehmet Yorukoglu, ‘En-

gines of Liberation’, The Review of Economic Studies, 72 (2005),
109-133.
[16] Hadfield, Gillian K., ‘A Coordination Model of the Sexual Division of

Labor’, Journal of Economic Behavior & Organization, 40 (1999), 125-
153.
[17] Iyigun, Murat and Randall P. Walsh, ‘Building the Family Nest: Pre-
marital Investments, Marriage Markets, and Spousal Allocations’, Re-
view of Economic Studies, 74 (2007), 507-535.
[18] Lommerud, Kjell Erik and Steinar Vagstad, ‘Mommy Tracks and Pub-
lic Policy: On Self-Fulfilling Prophecies and Gender Gaps in Promotion’,
CEPR Discussion Paper No. 2378, (2000).

Women’s Relative Wages Over Time’, The Quarterly Journal of Eco-
nomics, 123 (2008), 1061-1110.
[20] Peters, Michael, ‘The pre-marital investment game’, Journal of Eco-

nomic Theory, 137, (2007), 186-213.
[21] Peters, Michael and Aloysius Siow, ‘Competing Premarital Invest-

ments’, Journal of Political Economy, 110 (2002), 592—608.
[22] Shapley, Lloyd and Martin Shubik, ‘The Assignment Game 1: The
Core’, International Journal of Game Theory, 1 (1972), 111-130.
9.7 Appendix: Existence and Uniqueness of

Equilibrium
Substitute z11 − V1 for U1 and z22 − V2 for U2 in equation (21), and define
Ψ(V1 , V2 ) as
ZV2
Ψ(V1 , V2 ) ≡ F (V1 ) + G(Rm + V2 − θ)f (θ)dθ (A1)
V1
Z−V2
z22
−F (z11 − V1 ) − G(Rw + z22 − V2 − θ)f (θ)dθ .

z11 −V1
Note, first, that
Zz22
Ψ(0, 0) = F (0) − F (z11 ) − G(Rw + z22 − θ)f (θ)dθ < 0 (A2)
z11
and that
Zz22
Ψ (z11 , z22 ) ≡ F (z11 ) − F (0) + G(Rm + z22 − θ)f (θ)dθ > 0 , (A3)
z11
since z11 > 0 implies that F (z11 ) − F (0) > 0. By continuity, we conclude
that there exists a set of couples (V1 , V2 ) for which Ψ (V1 , V2 ) = 0.
In addition, we have
∂Ψ(V1 , V2 )
= f (V1 )[1 − G(Rm + V2 − V1 )]
∂V1
(A4)
w
+ f (z11 − V1 )[1 − G(R + z22 − z11 − (V2 − V1 )] > 0
and
∂Ψ(V1 , V2 )
= G(Rm )f (V2 ) + G(Rw )f (z22 − V2 )]
∂V2
(A5)
ZV2 ZU2
+ g(Rm + V2 − θ)f (θ)dθ + g(Rw + U2 − θ)f (θ)dθ > 0 .
V1 U1
By the implicit function theorem, Ψ (V1 , V2 ) = 0 defines V2 as a differ-

entiable, decreasing function of V1 over some open set in R. Equivalently,
the locus Ψ (V1 , V2 ) = 0 defines a smooth, decreasing curve in the (V1 , V2 )
plane.
Using (22), define Ω(V1 , V2 ) as
Ω (V1 , V2 ) ≡ F (V1 ) [1 − G (Rm + V2 − V1 )] (A6)

w
−F (z11 − V1 ) [1 − G (R − z11 + V1 + z22 − V2 )].
Note that Ω is continuously differentiable, increasing in V1 and decreasing
in V2 . Moreover,
lim Ω (V1 , V2 ) = 1, (A7)
V1 →∞
lim Ω (V1 , V2 ) = −F (z11 − V1 ) < 0.
V2 →∞
By continuity, there exists a locus on which Ω (V1 , V2 ) = 0; by the implicit

function theorem, it is a smooth, increasing curve in the (V1 , V2 ) plane. In
addition,
Ω (V1 , V2 ) = A (V1 , V2 − V1 ) , (A8)
where
A (V, X) = F (V ) [1−G (Rm + X)]−F (z11 − V ) [1−G (Rw − z11 + z22 − X)].
(A9)
Since
∂A (V, X)
= f (V ) [1−G (Rm + X)]+f (z11 − V ) [1−G (Rw − z11 + z22 − X)] > 0
∂V
(A10)
and
∂A (V, X)
= −F (V ) g (Rm + X)−F (z11 − V ) g (Rw − z11 + z22 − X) < 0,
∂X
(A11)
the equation A (V, X) = 0 defines X as some increasing function φ of V .
Therefore,
Ω (V1 , V2 ) = A (V1 , V2 − V1 ) = 0 (A12)
gives
V2 = V1 + φ (V1 ) , (A13)
0
where φ (V ) > 0. Thus in the (V1 , V2 ) plane, the slope of the Ω (V1 , V2 ) = 0
curve is always more than 1. In particular, the curve must intersect the
decreasing curve Ψ (V1 , V2 ) = 0, and this intersection (V1∗ , V2∗ ) is unique.
Finally, stability requires that
U1 + V2 ≥ z21 and U2 + V1 ≥ z12 (A14)
which implies that, at any stable match, we have
z21 − z11 ≤ V2 − V1 ≤ z22 − z12 , (A15)
and
z12 − z11 ≤ U2 − U1 ≤ z22 − z21 . (A16)
Three cases are thus possible:
1. If z21 − z11 ≤ V2∗ − V1∗ ≤ z22 − z12 , then (V1∗ , V2∗ ) is the unique equi-
librium (see figure A.1).
Indeed, it is the only equilibrium with perfectly assortative matching.

Moreover, a point such that
Ψ (V1 , V2 ) = 0 and V2 − V1 = z21 − z11 (A17)
cannot be an equilibrium, because at that point Ω (V1 , V2 ) > 0, which

contradicts the fact that the number of educated men should exceed
that of educated women for such an equilibrium to exist. Similarly, a
point such that
Ψ (V1 , V2 ) = 0 and V2 − V1 = z22 − z12 (A18)
cannot be an equilibrium, because at that point Ω (V1 , V2 ) < 0, which

contradicts the fact that the number of educated women should ex-
ceed that of educated men for such an equilibrium to exist.
2. If z21 − z11 > V2∗ − V1∗ , then the unique equilibrium (see figure A.2)
is such that
Ψ (V1 , V2 ) = 0 and V2 − V1 = z21 − z11 . (A19)
Indeed, a perfectly assortative matching equilibrium is not possible

because the only possible candidate, (V1∗ , V2∗ ), violates the condition
z21 − z11 ≤ V2∗ − V1∗ ≤ z22 − z12 . A point such that
Ψ (V1 , V2 ) = 0 and V2 − V1 = z22 − z12 (A20)
cannot be an equilibrium, because at that point Ω (V1 , V2 ) < 0 which

contradicts the fact that the number of educated women should ex-
ceed that of educated men for such an equilibrium to exist.
3. Finally, if V2∗ −V1∗ > z22 −z12 , then the unique equilibrium (see figure
A.3) is such that
Ψ (V1 , V2 ) = 0 and V2 − V1 = z22 − z12 . (A21)

Indeed, a perfectly assortative matching equilibrium is not possible

because the only possible candidate, (V1∗ , V2∗ ), violates the condition
z21 − z11 ≤ V2∗ − V1∗ ≤ z22 − z12 . A point such that
Ψ (V1 , V2 ) = 0 and V2 − V1 = z21 − z11 (A22)
cannot be an equilibrium, because at that point Ω (V1 , V2 ) > 0 which

contradicts the fact that the number of educated men should exceed
that of educated women for such an equilibrium to exist.
b Marriage, no
No marriage, investment
no investment
Rm + V2 −V1
Rm
−V2 −V1 θ
Marriage and
investment
Investment,
no marriage
FIGURE 9.1. Regions for Marriage and Investment

V2
z22
Uneducated married men=

Uneducated married women
e
z22
2
z22 − z12 Upper bound
Married men=Married women
Lower bound
z21 − z11
z11 z11 V1
2
FIGURE 9.2. Equilibrium with Strictly Positive Assortative Matching

V2
z22
Upper bound
z22
2 Uneducated married men=
e' Uneducated married women
z22 − z12
Lower bound
z21 − z11
z11 z11 V1
2
FIGURE 9.3. Mixed Equilibrium with More Educated Men than Educated
Women
V2
z22
Upper bound

Lower bound
e
z22
2
z22 − z12
e'
z21 − z11
z11 z11 V1
2
FIGURE 9.4. The Impact of an Increase in the Wage of Educated Men Combined
with a Reduction in the Wage of Educated Women
V2
z22
z '22
Upper bound

Lower bound
e
z22
2
z22 − z12 e'

z '22 − z '12
z '21 − z '11
z21 − z11
z11 z11 z '11 V1

2
FIGURE 9.5. The Impact of an Increase in the Wife’s Work at Home

V2
z '22
z22
Upper bound
e ''
e'
z '22 − z '12
z '21 − z '11
z22 − z12 Lower bound
z21 − z11 Uneducated married men=

z '11 z11 V1
FIGURE 9.6. The Impact of a Decrease in the Wife’s Work at Home Combined
with an Increase in the Wage of Educated Women
This is page 435
10
An equilibrium model of
marriage, fertility and divorce
This chapter provides a simple model of the marriage market that includes
fertility, divorce and remarriage and addresses some of the basic issues
associated with the higher turnover in the marriage market. For this pur-
pose, we introduce search frictions, heterogeneity and unexpected shocks
to match quality. The model is simple enough to identify the welfare im-
plication of increasing turnover. The main result is that the prospects of
remarriage generate multiple equilibria due to a positive feedback whereby
a higher aggregate divorce rate facilitates remarriage, which, in turn, raises
the incentives of each couple to divorce. Moreover, when multiple equilib-
ria exist, an equilibrium with higher divorce and remarriage rates generates
higher expected welfare for all participants in the marriage market. This
is a direct outcome of the positive search externalities that are embedded
in the model. The main lesson is that a high aggregate divorce rate can be
beneficial because it facilitates the recovery from negative shocks to match
quality, allowing couples to replace bad marriages by better ones. Related
papers are Aiyagari et al (2000), Brien et al (2006) and Chiappori and
Weiss (2006).
10.1 A model of the Marriage Market

Consider a society in which there is an equal number of men and women
and all individuals are ex ante identical and live for two periods. Alone,
each person consumes their own income Y . If married, the partners share
consumption and each consumes 2Y . In addition, marriage entails a non
monetary return θ that both partners enjoy. This ‘quality of match’ is
randomly distributed and different couples draw different values of θ at the
time of marriage. However, the future quality of match is uncertain.
Meetings are random. At the beginning of each period, each person ran-
domly meets a person of the opposite sex of his/her age group in a given
cohort. We assume that marriage binds for at least one period. At the end
of the first period divorce can occur but remarriage is possible only with
unattached individuals who never married before or have divorced. In the
first period, one meets an eligible partner with certainty. The probability of
each individual to meet a single person of the opposite sex in their second
period of life equals the proportion in the population of unattached indi-
436 10. An equilibrium model of marriage, fertility and divorce
viduals of the opposite sex, divorced or never married. This assumption is

crucial for our analysis and implies an ‘increasing returns meeting technol-
ogy’ whereby, the more singles are around, the easier it is for each single
person to find a match. The logic behind this assumption is that meetings
often occur at work or school and are ‘wasted’ if the person you meet is
already married.
Marriage also provides the partners with the option to produce (exactly)
two children (there is no out of wedlock birth). The production of children
entails a cost to the parents in the first period, c, and a benefit which both
parents enjoy in the subsequent period. The utility of a child is independent
of household income but depends on the proximity to their natural parents.
It equals q ∗ if the children live with both natural parents and to q 0 if they
live with only one of the parents or in a step family; we assume q ∗ > c > q 0 .
Both parents treat the utility of the child as a public good and it enters
additively into their preferences. Partners with children find divorce more
costly, because the welfare of the children is higher if children are raised
with their natural parents.
Upon meeting, the quality of match θ is revealed and the matched part-
ners decide whether to marry or not. If they choose to marry, they can
further decide whether they wish to have children. Because of the delayed
benefits, the production of children is a relevant option only for partners
in the first period of their life. During each period, there is a shock ε to
the quality of match, which is revealed at the end of the period. Hav-
ing observed the shock at the end of the first period, the partners decide
whether to divorce or not. The random variables θ and ε are assumed to
be independent across couples. In particular, for each remarried person the
values of θ in the first and second marriage are independent. We denote the
distributions of θ and ε by G (θ) and F (ε) with densities g (θ) and f (ε)
respectively. We assume that these distributions have zero mean and are
symmetric around their mean.
We assume that all goods in the household, consumption, match quality
and children are public and both partners enjoy them equally. Hence, by
assumption, men and women benefit equally from marriage or divorce.
The assumptions of public goods and equal numbers of men and women
generate perfect symmetry between genders that allows us to set aside, in
this chapter, conflict and bargaining between the partners.
10.1.1 Individual Choices

The last stage: the remarriage decision
We first analyze the marriage, fertility and divorce decisions of individu-
als who take the conditions in the marriage market as given. We proceed
from the last available choice, marriage at the second period and work
backwards. Two unattached individuals who meet at the beginning of the
10. An equilibrium model of marriage, fertility and divorce 437
second period will marry if and only if their drawn θ satisfies
θ ≥ −Y. (10.1)
That is, conditioned on meeting, marriage occurs whenever the sum of

monetary and non monetary gains from marriage is positive. This simple
marriage rule holds because each partner gains Y + θ from the marriage
and, if one of the partners has a child then, by assumption, the benefits
from that child are the same whether the child lives with a single parent
or in a step family. There are thus no costs associated with remarriage.
We denote the probability of remarriage conditioned on a meeting in the
second period by
γ = 1 − G(−Y ), (10.2)
and the expected quality of match conditioned on marriage in the second
period by
β = E(θ/θ ≥ −Y ). (10.3)
Note that although the expected value of θ is zero, the expectation condi-
tioned on remarriage, β, is positive, reflecting the option not to marry if
the drawn θ is low.
The probability of meeting an unattached person of the opposite sex
at the beginning of the second period is denoted by u. The probability
that an unattached person will meet an eligible single person whom he
or she will choose to marry is p = uγ. Note that men and women face
the same remarriage probability p, because we assume perfect symmetry
between men and women. The expected utility of an unattached person,
conditioned on having children is, therefore,
V2,n = p(2Y + β) + (1 − p)Y + nq 0 , (10.4)
where n = 1 if children are present and n = 0, otherwise.
The intermediate stage: the divorce decision

A married person will choose to divorce if and only if the θ drawn at the
beginning of the first period and the ε drawn at the end of the first period
are such that
2Y + θ + ε + nq ∗ < V2,n . (10.5)
This can be rewritten as ε + θ < hn , where
hn ≡ −Y + p(Y + β) − n(q ∗ − q 0 ) (10.6)
is the expected net gain from divorce.

The probability of divorce for a married couple with initial quality of
match θ is given by F (hn − θ). This probability depends on both indi-
vidual circumstances, represented by θ and n, and on market conditions,
represented by p = uγ. Specifically, the probability of divorce rises with the

number of singles who are eligible for remarriage, u, and is lower among
couples who have children or are well matched. That is, surprises such as
shocks to the quality of the match, represented here by ε, are less disruptive
if the current marriage is good, the cost of separation is high or remarriage
is unlikely. The influence of remarriage prospects on the decision to divorce
creates a link between the aggregate divorce rate and the individual deci-
sion to divorce. If many choose to divorce then the number of singles, u, is
high, which would raise the probability of remarriage, p, and the net gain
from divorce, hn , and thus the probability of divorce.
The first stage: the marriage and fertility decisions

Two unmarried individuals who meet at the beginning of the first period
and observe their drawn quality of match, θ, must decide whether to marry
and whether to have children upon marriage. Their expected life time utility
upon marriage, conditioned on n, is given by
Z∞
W1,n (θ) = 2Y +θ−nc+ (2Y +nq ∗ +θ+ε)f (ε)dε+F (hn −θ)V2,n . (10.7)
hn −θ
Differentiating W1,n (θ) with respects to θ yields (details are given in the
appendix):
∂W1,n
= 2 − F (hn − θ). (10.8)
∂θ
That expected utility is increasing in the quality of match is intuitively
clear, because a couple with high θ can always replicate the divorce and
remarriage decisions of a couple with low θ. The value of marrying without
children, W1,0 (θ), and the value of marrying with children, W1,1 (θ), are
continuous, increasing and convex functions of θ. A person who chooses
not to marry at the beginning of the first period has expected lifetime
utility given by:
V1 = Y + V2,0 . (10.9)
Thus, a first marriage will occur if and only if:
max (W1,0 (θ), W1,1 (θ)) ≥ V1 . (10.10)
This maximum function inherits the properties of the individual W1,n func-
tions; that is, it is continuous, increasing and convex in θ. Because the val-
ues of marriage with and without children both rise with θ, the decision
whether to marry has the form of a stopping rule. That is, couples will
marry if and only if θ ≥ θm , where θm is determined by the condition
that (10.10) holds as an equality.1 Because the maximum is an increasing

function of θ, θm is unique; see Figure 10.1.
The decision whether to have children can also be represented as a stop-
∂W ∂W
ping rule, because (10.8) implies that ∂θ1,1 > ∂θ1,0 > 0 for all θ. That
is, the quality of the first match is more important if the couple has chil-
dren and are thus less likely to divorce (recall that children impede divorce,
h1 < h0 ). Therefore, there is a unique value of θ, θc , that solves W1,1 (θ) =
W1,0 (θ); see Figure 10.1. Thus a very simple rule arises: those couples for
whom θ < θm will not marry. Those couples for whom θ ≥ θm will marry
but they may or may not have children, depending on the costs and ben-
efits from having children. If the cost of having children is relatively high
then θc > θm and only those married couples for whom θ > θc will have
children while couples for whom θc > θ ≥ θm will choose to marry but have
no children. This is the case illustrated in figure 10.1. If the cost of having
children is relatively low then θc < θm and all people that marry will have
children. In terms of Figure 10.1 this is equivalent to moving W1,1 (θ) up
until the two curves intersect at a value of θ below θm .
An interesting testable implication of this model is that individuals are
less selective in their first marriage decision than in their remarriage deci-
sion. That is, θm ≤ −Y (see the appendix). Conditional on θ, marriage in
the first period is always more attractive because of the option to sample
ε. There is no downside risk because one can divorce if ε is low. Such an
option is not available in the second period. The option to have children
makes this preference for early marriage even stronger.
Another testable result is that individuals become more selective in their
first marriage decisions if more eligible singles are available for remarriage in
the second period. That is, θm is increasing in the remarriage probability, p.
This follows directly from the observation that the probability of remarriage
has a stronger effect on someone who chose not to marry and is thus sure
to be single in the second period than on someone who married and will
be single next period with probability less than one. That is,
∂W1,n ∂V1
= (Y + β)F (hn − θ) < Y + β = . (10.11)
∂p ∂p
It is also the case that the critical value for having children, θc , rises with
the probability of remarriage, p, implying that a couple will be less inclined
to have children when p is higher. This follows because childless couples
are more likely to divorce and therefore the positive impact of p on couples
∂W ∂W
without children is stronger, ∂p1,0 > ∂p1,1 ; see the appendix.
1 We are here implicitly assuming that the support of θ is wide enough so that some
people do not marry.

Summary
We have identified two basic forces that guide marriage, divorce, and fer-
tility choices: individual circumstances, represented here by θ and ε and
market forces represented here by p. Couples who drew a good match qual-
ity upon meeting are more willing to marry and to invest in children because
they expect the marriage to be more stable. High turnover in the marriage
market has the opposite effect; it discourages marriage and investment in
children, because of the higher risk of divorce. These two forces interact and
reinforce each other. If individuals expect high turnover, they invest less
in children and are therefore more likely to divorce, which raises turnover.
High turnover can raise the probability of divorce even in the absence of
children because partners are more willing to break a marriage when the
prospects for remarriage are good.
10.1.2 Aggregation
We can now aggregate over couples with different realizations of θ and de-
fine the aggregate rate of divorce (per number of individuals in the cohort)
assuming that the cost of children is large enough so that θc > θm .
Zθc Z∞
d= F (h0 − θ)g(θ)dθ + F (h1 − θ)g(θ)dθ. (10.12)
θm θc
Given the value of p that individuals expect, the implied proportion of

singles at the beginning of period 2 is:
u = U (θm (p), θc (p)) ≡ G(θm ) + d (10.13)
and the aggregate number of remarriages (per number of individuals in the
cohort) is p = γu.
Our results on individual behavior imply that U (., .) is increasing in its
two arguments. Specifically, from equations (10.12) and (10.13) and the
fact that children raise the cost of divorce, h0 > h1 , we obtain:
∂U
= (1 − F (h0 − θm ))g(θm ) > 0,
∂θm
∂U
= [F (h0 − θc ) − F (h1 − θc )]g(θc ) > 0. (10.14)
∂θc
Having shown that both θm (p) and θc (p) are increasing in the remarriage
probability, p, we conclude that U (θm (p), θc (p)) is also increasing in p.
10.1.3 Equilibrium
Equilibrium is defined by the condition that the value of p that individu-
als expect is the same as the aggregate number of singles implied by the
expectation. That is,

p = U (θm (p), θc (p)). (10.15)
The function U (., .), viewed as a function of p, is a non decreasing function
from [0, 1] to [0, 1]. Therefore, by the Tarski fixed point theorem (see Mas-
Colell et al (1995), section MI), there is at least one equilibrium point in
the interval [0, 1] at which expectations are realized.
We may narrow down the range of possible equilibria, based on some a
priori information. Because of the advantages of joint consumption and the
zero mean and symmetry assumptions on G(θ) and F (ε), more than half of
the population will choose to marry, and those who subsequently received
a sufficiently favorable shock to the quality of match will remain married
even if the probability of finding a new mate is 1, implying that p < 1
in equilibrium. If there is not much heterogeneity in θ and the support
of the shock ε is small, everyone will marry and no one will divorce so
that p = 0 in equilibrium. However, with sufficiently large variability in
θ and ε, an equilibrium p will be positive, because even in the absence of
remarriage prospects, couples who draw a sufficiently low quality of match
will not marry, and married couples who suffered a large negative shock
will divorce, so that U (θm (0), θc (0)) > 0.
Because of the positive feedback, whereby an increase in the expected
number of singles induces more people to become single, there may be mul-
tiple equilibria. Having assumed that all individuals are ex ante identical,
we can rank the different equilibria based on their common expected value
of life time utility:
W1 = E max(W1,1 (θ), W1,0 (θ), V1 ). (10.16)
The expectation is taken at the beginning of the first period prior to any
meeting, when the quality of prospective matches is yet unknown. An equi-
librium with a higher number of unattached individuals at the beginning
of the second period will generally have less marriages, more divorces and
fewer couples with children. Despite these apparently negative features,
equilibria with higher p are in fact Pareto superior, because of the better
option for couples who suffered a bad shock to their first marriage to re-
∂W
cover by forming a new marriage. To see this, note that by (10.11), ∂p1,n
and ∂V∂p are positive, implying that an increase in p causes an increase in
1
the expected welfare of all members of society, irrespective of the value of

θ that they draw. In other words, the search frictions, represented here by
random meetings with members of the opposite sex, irrespective of whether
or not they are already attached, imply that those who choose to divorce
or remain single exert a positive externality on other members of society
who find it easier to find a mate for remarriage. This externality dominates
the welfare comparisons because all other factors, such as the damage to
children, are internalized by the partners. The presence of children implies
that married couples are more reluctant to divorce, which yields a lower
equilibrium value for p. However, it is still true that all couples, including
couples with children, will be better off in an equilibrium with a higher p
if multiple equilibria exist.
10.2 An Example
We now introduce a simple example with multiple equilibria and discuss
their properties. Assume that ε takes only two values, −a and +a with equal
probability, while θ is¡distributed
¢ uniformly on [−b, b]. For this example,
we assume that 2a > q ∗ − q 0 ; that is, the variance of the match quality
shock is large relative to the loss for children from divorce, so that even
couples with children may divorce if the revised quality of their match is
low enough.
The expected utility profile if marriage takes place, conditional on having
children or not (n = 0, 1), is:
⎧
⎪
⎪ 3Y + θ + p(Y + β) + n(q 0 − c)
⎪
⎪ if − b ≤ θ < hn − a
⎪
⎪
⎨ 0 ∗
W1,n (θ) =
7Y
2 + 2 + p(Y2+β) + a2 + n( q2 + q2 − c)
3θ
(10.17)
⎪
⎪ if hn − a ≤ θ ≤ hn + a
⎪
⎪
⎪
⎪ 4Y + 2θ + n(q ∗ − c)
⎩
if hn + a < θ ≤ b
For a given n and conditional on marriage, couples who draw θ such that
θ + a < hn will divorce for sure at the end of the first period.2 Couples
who draw θ such that θ − a > hn will stay married for sure (if they marry).
Couples who draw θ in the intermediate range hn − a ≤ θ ≤ hn + a
will divorce if the shock is negative and remain married otherwise. Using
equations (10.4) and (10.9), the value of not marrying in the first period is
given by:
V1 = 2Y + p(Y + β) (10.18)
which is independent of θ.
We now wish to identify the points θm and θc that trigger marriage
and having children, respectively. For this purpose, it is useful to inspect
Figure 10.2, in which we plot W1,0 (θ), W1,1 (θ) and V1 .3 Note that the
kinks in W1,0 (θ) always appear at higher values of θ than the kinks in
W1,1 (θ). This happens because the expected gains from divorce are higher
for couples without children, h0 − h1 = q ∗ − q 0 > 0. It can be seen that an
2 Marriage followed by certain divorce can occur if the gains from joint consumption
are sufficiently large to offset the low quality of the current match (Y + θ > 0).
3 In this figure we have h + a > h − a. This follows from the assumption that
1 0
2a > q ∗ − q 0 .
intersection of the two curves can occur only in the intervals [h1 − a, h0 − a]
or [h1 + a, h0 + a]. Moreover, it can be verified that if the costs from having
∗ 0
children are relatively high, that is, q ∗ > c > q +q
2 , then the only possible
intersection is in the region [h1 + a, h0 + a]; see the appendix for a proof.4
We obtain θc by equating W1,0 (θ) for the intermediate region (h0 − a <
θ < h0 + a) with W1,1 (θ) for θ > h1 + a. This gives:
θc = p(Y + β) − Y + a − 2(q ∗ − c), (10.19)

Using this expression that determines θc , we can now determine θm . Refer-
ring again to Figure 10.2, we see that max (W1,0 (θ), W1,1 (θ)) is represented
by the upper envelope of the W1,0 (θ) and W1,1 (θ) profiles. We thus have
to consider three segments of this envelope. In the first case (with low V1 ),
V1 intersects the envelope at a value of θ below h0 − a, where couples
would be indifferent between singlehood and a marriage without children
followed by a certain divorce. In the second segment the intersection occurs
at θ ∈ [h0 − a, θc ] ,where couples would be indifferent between singlehood
and a marriage without children followed by divorce if a negative shock
occurs (this is the case illustrated in Figure 10.2). In the third case, (high
V1 ) the intersection is above θc ,where couples would be indifferent between
singlehood and a marriage with children that remains intact with certainty.
In the appendix we show that:
⎧
⎨ −Y if p(Y + β) > a
p(Y +β)−a
θm = −Y if a ≥ p(Y + β) ≥ 3(q ∗ − c) − 2a .
⎩ Y +β 3 (q∗ −c)
p 2 − 2 − Y if p(Y + β) < 3(q ∗ − c) − 2a
(10.20)
¡ ∗ 0
¢ q ∗ +q 0
Note that the assumptions 2a > q − q and c > 2 ensure that
interval [3(q ∗ − c) − 2a, a] is non-empty.
From equations (10.19) and (10.20) we see that both θm and θc rise with
the expected remarriage rate, p. That is, the likelihood of marrying and
having children decline with p. This happens because matched partners
anticipate that they are more likely to divorce if the prospect of remarriage
rises. Both θm and θc decline with income, implying that the likelihood of
marrying and having children rise with income. This happens in our model
because of the complementarity between the incomes of the spouses that
is induced by joint consumption of public goods. A dollar increase in Y
raises the consumption of each married person by 2 dollars, while their
consumption as a single will rise by only one dollar.
The proportion of singles at the beginning of the second period that
is associated with a given p consists of those who did not marry in the
beginning of the first period, G (θm (p)) and the divorcees at the end of the
first period among the married. These divorcees constitute of all the married
4 The interested readers may try the case with low costs of children, see appendix.
for whom θm < h0 − a, half of the married for whom h0 − a ≤ θm ≤ θc and

none of the married for whom θm > θc . Therefore, equation (10.13) for the
proportion of singles at the beginning of period 2 can be written as
⎧ G(h0 (p)−a)+G(θc (p))

⎪
⎪ 2
⎪
⎪
⎪
⎪ if p(Y + β) > a
⎨ G(θm (p))+G(θc (p))
U (θm (p), θc (p)) = 2 (10.21)
⎪
⎪
∗
if a ≥ p(Y + β) ≥ 3(q − c) − 2a
⎪
⎪
⎪
⎪ G (θm (p))
⎩
if p(Y + β) < 3(q ∗ − c) − 2a
Because in this particular example, the reservation rules for marriage and
for having children are linear functions of p we obtain under the assumption
that G(.) is uniform that U (θm (p), θc (p)) is also a piecewise linear function
of p. Consequently multiple equilibria can arise. Within the confines of our
example, multiple equilibria occur only if there is not too much heterogene-
ity in the quality of match. We therefore choose a relatively small b and
obtain Figure 10.3. As seen in this figure, there are three equilibria at p = 0,
at p = 0.25 and at p = 0.5. Details of these three equilibria are presented
in Table 10.1. In all three equilibria, everyone marries whomever they meet
(this holds in both periods5 ), but the higher is the equilibrium level of p,
the lower is the proportion of families that choose to have children and
the higher is the proportion that divorces. At the low equilibrium, where
everyone expects a remarriage rate of p = 0, all couples have children and
no one divorces. This implies that there will be no singles in the second
period, which justifies the expectations. At the equilibrium in which every-
one expects a remarriage rate of p = 0.25, half of the couples have children
and, of those who do not have children, half divorce upon the occurrence
of a bad shock. This implies that at the beginning of the second period, a
quarter of the population will be single, which justifies the expected remar-
riage rate. At the equilibrium with p = 0.5, no couple has children and half
of them divorce upon the realization of a bad shock so, in this case too, ex-
pectations are realized. Thus, all three equilibria share the basic property
that expectations are fulfilled. However, the intermediate equilibrium at
p = 0.25 is not stable with respect to an arbitrary change in expectations.
That is, if the expected remarriage rate, p, rises (declines) slightly then the
aggregate number of singles U (θm (p), θc (p)) rises (declines) too.6
For these examples, one can easily calculate the equilibrium value of ex
ante welfare, W1 (see equation (10.16)). If p = 0.5, W1,0 (θ) is the highest
for all θ, implying that all couples marry, have no children and divorce with
5 Inthe second period, this implies that γ = 1 and β = 0.

6 If
b goes to zero and all matches are ex ante identical, the middle section disappears
and the equilibrium function becomes a step function yielding only two stable equilibria.
probability 0.5, so that

7 1 1 5
W1 = EW1,0 (θ) = Y + Y + a= (10.22)
2 4 2 6
If p = 0, W1,1 (θ) is the highest for all θ, implying that all couples marry,
have children and do not divorce, so that
3
W1 = EW1,1 (θ) = 4Y + (q ∗ − c) = (10.23)
6
The calculation of welfare is a bit more complex if p = 0.25. In this case,
the maximum is given by W0 (θ) if θ ≤ 0 and by W1 (θ) if θ ≥ 0. Thus,
7
2Y + 14 Y + 12 a + 32 E(θ/θ ≤ 0) 4Y + (q ∗ − c) + 2E(θ/θ ≥ 0) 4
W1 = + =
2 2 6
(10.24)
These calculations illustrate that ex-ante welfare rises as we move to equi-
librium points with higher p, reflecting the positive externality associated
with an increase in the aggregate number of singles.
10.3 Income uncertainty and ex-post heterogeneity

The simple model assumed perfect symmetry among spouses and that all
individuals have the same incomes which remain fixed over time. We now
allow income to change over time, which creates income heterogeneity ex-
post. As before, all men and women have the same income, Y , in the first
period of their life. However, with probability λ income in the second period
rises to Y h and with probability 1−λ it declines to Y l . To maintain ex-ante
symmetry, we assume that the incomes of men and women follow this same
process. To simplify, we shall assume now that the quality of the match, θ,
is revealed only at the end of each period. The realized value of θ at the end
of the first period can trigger divorce, while the realized value of θ at the
end of the second period has no behavioral consequences in our two period
model. Since there are gains from marriage, and the commitment is only
for one period, everyone marries in the first period. However, in this case,
changes in incomes as well as changes in the quality of match can trigger
divorce. We continue to assume risk neutrality and joint consumption.
The main difference from the previous model is that at the beginning
of the second period there will be two types of potential mates, rich and
poor. Let α be the expected remarriage rate and π the proportion of high
income individuals among the divorcees, and let y = πY h + (1 − π)Y l be
the average income of the divorcees. Then the expected values of being
unattached in the beginning of the second period for each type are
V j (α, π) = Y j + αy, j = l, h. (10.25)

This expression is obtained because type j consumes Y j alone and expects

to consume Y j + y when married and the expected value of the quality
of a new match θ in the second (and last) period is zero. Clearly, a richer
person has a higher expected value from being unattached.
At the end of the first period, the quality of the current match and
the new income values (Y h or Y l ) for each spouse are revealed, and each
partner can choose whether to stay in the current match or divorce and
seek an alternative mate. An hh couple divorces if:
2Y h + θ < Y h + αy ⇒ θ < αy − Y h (10.26)
An ll couple divorces if:
2Y l + θ < Y l + αy ⇒ θ < αy − Y l (10.27)
Note that, despite the lower value of being unattached for the two spouses,
a poor couple is more likely to divorce, because the current marriage is less
attractive.
In a mixed couple, type h will wish to divorce if
Y h + Y l + θ < Y h + αy,
which is the same as condition (10.27), while type l will wish to divorce if
Y h + Y l + θ < Y l + αy,
which is the same as condition (10.26). But inequality (10.26) implies in-
equality (10.27), which holds for a wider range of θ. Thus, the condition for
marital dissolution for mixed couples is (10.27). For mixed couples there
will be disagreement on the divorce decision if
αy − Y h ≤ θ < αy − Y l .
In this case, divorce is always triggered by the high income spouse who can
do better outside the marriage.
In equilibrium, the expected remarriage rate, α, equals the divorce rate,
that is,
α = λ2 G(αy − Y h ) + (1 − λ2 )G(αy − Y l ). (10.28)

Equation 10.28 involves two endogenous variables, the expected remarriage
rate α and the expected income of a divorcee, y. However, these two vari-
ables are interrelated and the equilibrium condition (10.28) can be reduced
to one equation in one unknown, αy, which is the variable part of the ex-
pected gains from divorce. Then, we can deduce the separate equilibrium
values of both α and y.
As a first step, note that the proportion in the population of high income
divorcees of each gender is
απ = λ[λG(αy − Y h ) + (1 − λ)G(αy − Y l )] (10.29)

Taking the difference between (10.28) and (10.29), we have
α(1 − π) = (1 − λ)G(αy − Y l ). (10.30)
Using the definition of y, we have
Yh−y
1−π = . (10.31)
Yh−Yl
Then, substituting from (10.31) into (10.30) we get
h l
αy l (Y − Y )
α= + (1 − λ)G(αy − Y ) . (10.32)
Yh Yh
Finally, eliminating α in (10.28), we can then rewrite the equilibrium con-
dition as an equation in αy
λY h Yl
αy = λ2 G(αy − Y h )Y h + (1 − λ2 )G(αy − Y l )[ + ]. (10.33)
1+λ 1+λ
To analyze this equation, we note that the expected income of a divorcee,
y, is bounded between Y l (which occurs if only low income individuals
divorce, π = 0) and Y h (which occurs if only high income individuals
divorce, π = 1) and that the divorce rate α is bounded between 0 and 1.
Therefore, αy is bounded between 0 and Y h . Assuming that G(−Y h ) > 0,
equation (10.33) has a positive solution for αy because the right hand side
of (10.33) is positive at αy = 0 and smaller than Y h at αy = Y h and G (.)
is continuous. However, because both sides of (10.33) are increasing in αy,
this equation may have multiple solutions. Given an equilibrium value for
αy, we can find the equilibrium divorce rate, α, from equation (10.28) and
the equilibrium share of the rich among the divorcees, π, from the ratio of
(10.29) to (10.28).
The comparative statics of this system are somewhat complicated, but
the basic principles are quite clear. An increase in the proportion of the
rich in the second period, λ, has two opposing effects on the equilibrium
divorce rate. First, it raises the monetary gain from maintaining the current
marriage. Second, it raises the average quality of divorcees and thus the
prospects of finding a good match, which encourages divorce. The relative
importance of these considerations depends on the initial proportions of the
two types, the values of low and high income and the distribution of match
quality. We cannot provide general results but simulations suggest that
the divorce rate tends to increase with the proportion of the rich when the
proportion of the rich is low in the second period. An increase in the income
of the poor or the rich tends to reduce divorce. The positive income effects
reflect the increasing gains from remaining married when consumption is

a public good. There is no simple mapping from income risk or income
inequality to the rate of divorce, but starting from equality an increase in
the difference Y h − Y l raises the divorce rate. An increase in the variability
of the quality of match generally leads to a rise in the divorce rate.
The simple model outlined above generates positive assortative mating in
the second period. This happens here because the good matches hh are less
likely to break, and all types have the same remarriage probability α. So
that there is a larger proportion of h among those who stay married than
in the population. This can be immediately seen by noting that the term
in square brackets in (10.29) is smaller than 1, so that απ < λ. Conversely,
there is a larger proportion of l among the singles than in the population,
because they are more likely to divorce and are equally likely to remain
single. This process of selective remarriage, via differential incentives to
divorce, is quite different from the usual models (see Burdett and Coles
1999) that are built on the idea that the high type is more selective in the
first marriage. In the search model, rejection of unsatisfactory mates is done
when one is single, reflecting the assumption that a match is “for ever”.
In our model, rejection happens when married, after θ is revealed. This
reflects our assumption that marriage is an “experience good”. It seems
that the two approaches lead to the same outcome.
10.4 Conclusion
The simple models discussed in this chapter make several important points
that carry a general message for the empirical and theoretical analysis of
the family. First, the marriage, fertility and divorce decisions are closely
interrelated. Couples decide to marry and to have children based on the
risk of divorce and the prospect of remarriage. Conversely, the fact that
couples chose to marry, or have children, has implications for their subse-
quent divorce decisions. Second, in a marriage market, as in other search
markets, individual decisions can be quite sensitive to the choices of oth-
ers. In particular, if many choose to remain single, not to have children, or
to divorce, this will strengthen the incentive of each couple separately to
behave in a similar manner. Such markets are susceptible to sudden and
large structural changes as may have happened following the introduction
of the contraceptive pill in the 1970’s. As we have seen, search external-
ities may have important policy and welfare implications. In particular,
societies with high marital turnover may in fact yield better outcomes for
the typical adult, because such an equilibrium allows easier recovery from
bad shocks. In this chapter, we assumed that children are always worse off
as a consequence of divorce. In the subsequent chapter, we shall discuss
child support transfers and show that children are not necessarily harmed
by divorce and, conditional on the divorce of their parents, may in fact be

better off in a high divorce environment.
10.5 Appendix
10.5.1 Properties of the expected utility, with and without
children
Using (10.7), (10.4) and (10.6):
Z∞
W1,n (θ) = 2Y + θ − nc + (2Y + nq ∗ + θ + ε)f (ε)dε + F (hn − θ)V2,n ,
hn −θ
V2,n = p(2Y + β) + (1 − p)Y + nq 0 ,

hn = −Y + p(Y + β) − n(q ∗ − q 0 ).
Hence,
V2,n = hn + 2Y + nq ∗ (10.34)
Differentiating W1,n (θ) with respects to θ yields
Z∞
1+ f (ε)dε + (2Y + nq ∗ + θ + hn − θ)f (hn − θ) − f (hn − θ)V2,n , (10.35)
hn −θ
where we use the fact that the derivative of an integral with respect to the
lower bound equals the value of the integrand at that point. Cancelling and
collecting terms, we obtain
∂W1,n
= 2 − F (hn − θ), (10.36)
∂θ
∂W ∂W
as stated in (10.8). Note that 1 ≤ ∂θ1,n ≤ 2 and that ∂θ1,n is increasing
in θ. Hence, the expected values with and without children, W1,1 (θ) and
W1,0 (θ) respectively, are increasing and convex functions of of θ, with slopes
∂W ∂W
bounded between 1 and 2. Also, because h1 < h0 , ∂θ1,1 > ∂θ1,0 . Finally,
examining the partial impact of p, holding θ fixed we see that
Z∞
∂W1,n (θ) ∂
= [ (2Y + nq ∗ + θ + ε)f (ε)dε
∂p ∂p
hn −θ
+F (hn − θ)(hn + 2Y + nq ∗ )]
= F (hn − θ)(Y + β), (10.37)
∂W1,0 ∂W1,1
implying that ∂p > ∂p .
10.5.2 Properties of the trigger for having children, θc

The trigger for θc is determined by the condition W1,1 (θc ) = W1,0 (θc ). If
∂W ∂W
there is a solution for θc , it must be unique because ∂θ1,1 > ∂θ1,0 . Using
(10.7) and (10.34), the requirement that W1,1 (θc ) = W1,0 (θc ) implies
Z∞
−c + (2Y + q ∗ + θc + ε)f (ε)dε + F (h1 − θc )(h1 + 2Y + q ∗ )
h1 −θc
Z∞
= (2Y + θc + ε)f (ε)dε + F (h0 − θc )(h0 + 2Y ). (10.38)
h0 −θc
or
Z∞
∗
−c + q + (θc + ε)f (ε)dε + F (h1 − θc )h1 (10.39)
h1 −θc
Z∞
= (θc + ε)f (ε)dε + F (h0 − θc )h0 .
h0 −θc
By (10.6),
dh0 dh1
= = Y + β. (10.40)
dp dp
Differentiating both sides of (10.39) with respect to p and θc , we obtain
(1 − F (h1 − θc ))dθc + F (h1 − θc )(Y + β)dp

= (1 − F (h0 − θc ))dθc + F (h0 − θc )(Y + β)dp, (10.41)
implying that
dθc
= Y + β.
dp
10.5.3 Properties of the trigger for marriage, θm

By definition,
max(W1,1 (θm ), W1,0 (θm ) = V1 (10.42)
Because W1,1 (θm ) and W1,0 (θm ) both increase in θ, while V1 is indepen-
dent of θ, the solution for θm must be unique if it exists. The solution must
also satisfy θm ≤ −Y, because
Z∞
W1,0 (−Y ) = V1 + (−p(Y + β) + ε)f (ε)dε ≥ V1 . (10.43)
p(Y +β)
There are two cases to consider.

Case 1, W1,0 (θm ) = V1 > W1,1 (θm ), which implies θc > θm . In this case
Z∞
2Y + θm + (2Y + θm + ε)f (ε)dε + F (h0 − θm )V2,0 = Y + V2,0 , (10.44)
h0 −θm
or
Z∞
Y + θm + (2Y + θm + ε)f (ε)dε = (1 − F (h0 − θm ))(h0 + 2Y ). (10.45)
h0 −θm
Differentiating totally both sides of (10.45) yields
[1 + (1 − F (h0 − θm ) + f (h0 − θm )(2Y + h0 )]dθm

−f (h0 − θm )(h0 + 2Y )(Y + β)dp
= f (h0 − θm )((h0 + 2Y )dθm
+[(1 − F (h0 − θm )) − (h0 + 2Y )f (h0 − θm )](Y + β)dp.(10.46)
Cancelling equal terms and rearranging, we obtain
∂θm 1 − F (h0 − θ)
= (Y + β) > 0 if θc > θm . (10.47)
∂p 2 − F (h0 − θ)
Case 2, W1,1 (θm ) = V1 > W1,0 (θm ), which implies θc < θm . In this case
Z∞
2Y + θm − c + (2Y + q ∗ + θm + ε)f (ε)dε + F (h1 − θm )V2,1 = Y + V2,1 ,
h1 −θm
(10.48)
or
Z∞
Y +θm −c+ (2Y +q ∗ +θm +ε)f (ε)dε = (1−F (h1 −θm ))(h1 +2Y +q ∗ ).
h1 −θm
(10.49)
Using the same calculations as in the previous case, we obtain
∂θm 1 − F (h1 − θ)
= (Y + β) > 0 if θc < θm . (10.50)
∂p 2 − F (h1 − θ)
We conclude that
∂θc ∂θm
> . (10.51)
∂p ∂p
10.5.4 Calculations for the example

Properties of θc in the example
We first prove that if the costs of having children are relatively high, that
∗ 0
is if q ∗ > c > q +q
2 , then an intersection of W1,0 (θ) with W1,1 (θ) cannot
occur in the region [h1 − a, h0 − a]. The proof is by contradiction. Assume
for some θ ∈ [h1 − a, h0 − a], W1,0 (θ) = W1,1 (θ). Then this θ must satisfy
7 3 1 1 1 1
Y + θ + p(Y + β) + a + ( q 0 + q ∗ − c) (10.52)
2 2 2 2 2 2
= 3Y + θ + p(Y + β).
Solving for θ and denoting the solution by θc , we have
θc = p(Y + β) − Y − a + 2c − (q ∗ + q 0 ). (10.53)
Recalling equation (10.6) for n = 0:
h0 = −Y + p(Y + β),
q ∗ +q 0
we obtain, using c > 2 ,
θc = h0 − a + 2c − (q ∗ + q 0 ) > h0 − a. (10.54)
Properties of θm in the example

Proof of (10.20). Consulting Figure 10.2 and allowing V1 to move up or
down, we see that we have to consider three cases for equation 10.42
First, low values of V1 give an intersection with W1,0 (θ) below θ = h0 −a.
Equating V1 with W1,0 (θ) this gives:
θm = −Y (10.55)
This requires that:
−Y = θm ≤ h0 − a = −Y + p(Y + β) − a (10.56)
⇒ p(Y + β) ≥ a
For intermediate values of θ ∈ [h0 − a, θc ] we equate V1 with W1,0 (θ) eval-

uated in the intermediate region of equation (10.17). This gives:
1
θm = (p(Y + β) − a) − Y. (10.57)
3
Since we have θm ≤ θc this value and (10.19) requires that:
p(Y + β) ≥ 3(q ∗ − c) − 2a (10.58)

Finally we can consider high values of θ, such that θ ≥ θc . Equality for equa-
tion (10.10) requires equating V1 with W1,1 (θ) evaluated for θ ≥ θc (that
is, the third region of equation (10.17)). This gives
1 1
θm = p(Y + β) − (q ∗ − c) − Y (10.59)
2 2
This case requires θm > θc which gives:
p(Y + β) < 3(q ∗ − c) − 2a (10.60)
Properties of the proportion of singles, U (θm (p), θc (p)), in the example

(proof of 10.21)
The proportion of singles at the beginning of the second period consists of
those who did not marry in the beginning of the first period, G (θm (p)),
and of the divorcees at the end of the first period among the married.
The proportion of divorcees depends on the location of θm . If V1 is low
and intersects W1,0 (θ) below h0 − a, then all of the married for whom
θm < θ < h0 − a divorce for sure, and all of the married for whom h0 − a <
θ < θc divorce upon a bad shock, that is with a probability of 12 , while
those married with children for whom θ > θc do not divorce. Therefore,
U (θm (p), θc (p)) = G (θm (p)) + (G (h0 (p) − a) − G (θm (p)))

1
+ (G (θc (p)) − G(h0 (p) − a))
2
1
= [G (θc (p)) + G(h0 (p) − a)]. (10.61)
2
For intermediate values of V1 , the intersection with W1,0 (θ) is in the
range [h0 − a, θc ], where the married with children for whom θm < θ < θc
divorce upon the occurrence of a bad shock. In this case,
1
U (θm (p), θc (p)) = G (θm (p)) + (G(θc (p)) − G(θm (p)))
2
1
= (G(θc (p)) + G(θm (p))). (10.62)
2
Finally, for high values of V1 , the intersection is with W1,1 (θ) above θc ,
where all married people have children, and no one divorces. In this case,
U (θm (p), θc (p)) = G (θm (p)) . (10.63)
10.5.5 Low costs of raising children

For completeness, we discuss briefly the case with low costs of raising chil-
∗ 0
dren, q +q
2 > c > q 0 . In this case, the intersection is at θ ∈ [h1 − a, h0 − a].
Therefore, we equate W1,0 (θ) evaluated in the first region of equation

(10.17) with W1,1 (θ) evaluated in the intermediate region of equation
(10.17), implying
θc = p(Y + β) − Y − a + 2c − q ∗ − q 0 , (10.64)
and
⎧
⎪ −Y
⎪
⎪
⎪
⎪ if p(Y + β) > a + q 0 + q ∗ − 2c
⎪
⎨ p(Y +β)−a+2c−q ∗ −q 0
3 −Y
θm = 0 ∗ . (10.65)
⎪
⎪ if a + q + q − 2c ≥ p(Y + β) ≥ c + q ∗ − 2q 0 − 2a
⎪
⎪ p(Y +β) q ∗
−c
⎪
⎪ − 2 −Y
⎩ 2
if p(Y + β) < c + q ∗ − 2q 0 − 2a
¡ ¢ ∗ 0
Note that the assumptions 2a > q ∗ − q 0 and c < q +q 2 ensure that
interval
£ ¤
c + q ∗ − 2q 0 − 2a, a + q 0 + q ∗ − 2c is non-empty.
The aggregate number of singles associated with a given p is
⎧ G(h1 (p)+a)+G(θc (p))
⎪
⎪ 2
⎪
⎪ 0 ∗
⎪
⎪ if p(Y + β) > a + q + q − 2c
⎨ (G(h1 (p)+a)+G(θm (p)))
U (θm (p), θc (p)) = 2 .
⎪
⎪ if a + q 0 + q ∗ − 2c ≥ p(Y + β) ≥ c + q ∗ − 2q 0 − 2a
⎪
⎪
⎪
⎪ G(θm (p))
⎩
if p(Y + β) < c + q ∗ − 2q 0 − 2a
(10.66)
10.6 References
[1] Aiyagari, S. Rao, Jeremy Greenwood and Nezih Guner, ‘On the State
of the Union’, Journal of Political Economy, 108 (2000), 213-244.
[2] Brien, Michael J., Lee A. Lillard and Steven Stern, ‘Cohabitation, Mar-
riage, and Divorce in a Model of Match Quality’, International Eco-
nomic Review, 47 (2006), 451-494.
[3] Burdett, Kenneth and Melvyn G. Coles, ‘Long-Term Partnership For-

mation: Marriage and Employment’, The Economic Journal, 109
(1999), 307-334.
[4] Chiappori, Pierre-Andre and Yoram Weiss, ‘Divorce, Remarriage and

Welfare: A general Equilibrium Approach’, Journal of the European
Economic Association, 4 (2006), 415-426.
FIGURE 10.1. Expected utility profiles.

FIGURE 10.2. Expected utility profiles for example

FIGURE 10.3. Equilibrium Divorce Rates
Table 10.1: Example with Multiple Equilibria

p = 0.0 p = 0.25 p = 0.5
Critical value for marriage, θm −1.31 −1.22 −1.14

Critical value for children, θc −.25 .00 .25
Percent married 100 100 100
Percent with children 100 50 0
Percent divorced with children 0 0 0
Percent divorced without children − 50 50
Percent single 0 25 50
Lifetime utility 4.083 4.094 4.208
Parameter values:
Income, Y = 1
Range for the match quality, θ ∈ [− 16 , 16 ]
Size of shock to match quality, a = ± 11 12
Utility of children in intact family, q ∗ = 1
Utility of children following divorce, q 0 = 0
Cost of raising children, c = 11
12
Probability of remarriage, γ = 1
Expected quality of match conditioned on remarriage,
β = E(θ/θ ≥ −Y ) = 0
This is page 459
11
Marriage, Divorce, Children
11.1 Introduction
The purpose of this chapter is to examine in more detail the role of children
in marriage and divorce.1 In particular, we wish to discuss the determi-
nation of expenditures on children and their welfare under various living
arrangements, with and without the intervention of the courts. There is a
growing concern that the higher turnover in the marriage market causes
more children to live with single mothers or step parents. In the US, year
2005, 68 percent of children less than 18 years old lived with two parents
(including step parents), 23 percent lived only with their mother, 5 percent
lived only with their father and the rest lived in households with neither
parent present. This may be harmful to the children.2 Part of the problem
is that, following separation, fathers are less willing to transfer resources
to the custodial mothers (that is, their ex-wives). A major objective of our
analysis is to explain how transfers between separated parents are deter-
mined and how they vary with marriage market conditions.
Separation may entail an inefficient level of expenditures on children for
several reasons: 1) If the parents remarry, the presence of a new spouse who
cares less about step children reduces the incentives to spend on children
from previous marriages. 2) If the parents remain single then, in addition to
the loss of the gains from joint consumption, the custodial parent may de-
termine child expenditures without regard to the interest of their ex-spouse.
3) Parents that live apart from their children can contribute less time and
goods to their children and may derive less satisfaction from them. These
1 This chapter extends the results reported in Chiappori and Weiss (2007) to include
both time and money as inputs to the child welfare. See also Weiss and Willis (1985,
1993), Del-Boca (2003), and Case et al (2003).
2 There is substantial evidence that children of divorced parents do not perform as
well as comparable children in intact families. See Argys et al.(1998), Lamb et al.(1999),
Hetherington and Stanley-Hagan (1999), Gruber (2004) and Stafford and Yeung (2005).
Such empirical evidence should be interpreted with some care, for two reasons. First,
dysfunctional families are more likely to generate both divorce and poor child perfor-
mance. B j́orklund and Sundstrom (2006) argue that inferior performances of divorced
children can largely be attributed to selection effects. Second, even if divorce causes
poor performance at the individual level, the impact of the aggregate divorce rate on
the welfare of children is a different issue. As shown by Piketty (2003) the increase in
the divorce rate in France has reduced the gap in school performance between children
of divorced parents and children from intact families.
460 11. Marriage, Divorce, Children
problems are amplified if the partners differ in income and cannot share
custody to overcome the indivisibility of children. The custodial parent is
usually the mother who has some comparative advantage in caring for chil-
dren but has lower income. The father has often limited access to the child
and low incentive to provide for him. The outcome is that the level of child
expenditures following separation is generally below the level that would
be attained in an intact family, reducing the welfare of the children and
possibly the welfare of their parents.
An important consequence of having children is that they create ex-post
wage differences between men and women. The basic reason for such dif-
ferences is biological in nature. The mother is the one who gives birth as
she is more capable of taking care of the child at least initially. As noted by
Becker (1993) this initial difference may have large economic consequences.
When the mother takes care of the child, her future earning capacity erodes.
Then, because of the reduced earning capacity of the mother and her inher-
ent advantage in child care, a pattern of specialization arises, whereby the
father works more in the market and the mother works more at home; see,
also, Chichilnisky (2005) and Albanesi and Olivetti (2009). This pattern
is most pronounced if the couple remains married and can coordinate ac-
tivities. Following separation, however, the allocation of time may change,
and a custodial mother may spend less time on her child if she remarries,
because a foster father cares less about the child than a natural father.
The ex-post asymmetry between parents can have strong implications for
the divorce decision and the incentive to produce children. Because men
maintain or increase their earning capacity during marriage, they have
higher expected gains from divorce. Under divorce at will, they will initiate
the divorce, at some situations in which the mother would like to maintain
the marriage. If transfers within marriage are limited due to a large com-
ponent of public consumption, separations will be inefficient, implying that
the gains from having children are smaller to the mother than to the father.
Because the production of children requires both parents, the mother may
avoid birth in some situations in which the husband would like to have a
child. The consequence is then an inefficient production of children.
To overcome these problems, the partners have an incentive to sign bind-
ing contracts that will determine some transfers between the spouses. The
purpose of the transfers is to induce an efficient level of child expenditures
following divorce and to guarantee efficient separation and child produc-
tion by restoring the symmetry between the parents. It is generally not
possible to obtain such a first best outcome, because of some important
limitations on transfers. First, transfers within marriage can only partially
compensate for common factors that affect both partners, such as the fail-
ure of the marriage. If the partners separate then transfers can compensate
for differences in the gains and costs from divorce, but these transfers are
limited too. In particular, it is not possible to condition the transfer on
the allocation within a household which is usually not observed by a third
11. Marriage, Divorce, Children 461
party.
Legal intervention is required to enforce binding contracts. In practice,
enforcement of alimony and child support contracts is imperfect. This is
not simply a matter of lack of resources or determination on the part of
the legal authorities. There is a basic conflict between private needs and
social needs that results from the externalities that prevail in the marriage
market. One issue is that parents and child interests may conflict, even if
parents care about their children. For instance, a mother may choose to
remarry even if the child under her custody is harmed, because she gains
more than the child from the presence of a new spouse. Another issue
is the impact of the divorce and fertility decisions of a given couple on
the prospects for remarriage and the gains from remarriage of others. In
marriage markets with frictions, competition does not force a couple to
internalize the impact on potential mates, because meetings are to a large
extent random and rents prevail. Therefore, a contract that a couple is
willing to sign is not necessarily optimal from a social point of view. A
related issue is that contracts that couples are willing to sign may at the
time of marriage, before the quality of match is observed, may be inefficient
ex-post after divorce has occurred and the impact of the contract on the
divorce and fertility decisions is not relevant any more. In this case, the
partners may wish to renegotiate, thereby creating a lower level of welfare
for both of them from an ex-ante point of view.
The benefits from having children depend on the contracts that the par-
ents employ to regulate these decisions and on the prospects of remarriage
that are determined in the marriage market. Consequently, the incentives
to produce children depend not only on the risk of divorce, triggered by
changing circumstances in a specific household, such as falling out of love,
but also on the general situation in the marriage market. The larger is the
proportion of couples that divorce, the better are the remarriage prospects.
In the absence of children, or with children but adequate transfers, this
would increase the probability of divorce. However, with children, remar-
riage may have a negative effect on the child because the new husband of
the custodial mother may be less interested in its welfare. We may refer
to this problem as the "Cinderella effect" (see Case et al., 1999). This ef-
fect reduces the incentive of the non custodial father to support the child,
because part of the transfer is "eaten" by the new husband. In addition,
non custodial parents who are committed to their custodial ex-spouse are
less attractive as potential mates for remarriage. Thus, the larger is the
proportion of such individuals among the divorcees, the less likely it is that
a particular couple will divorce, and the more likely it is that each couple
will have children. In this chapter, we use a simple model to illustrate the
interactions among these considerations in a general equilibrium framework
and highlight the potential consequences for parents and children.
11.2 The Model

We consider here a given cohort with equal number of men and women.
Individuals live for two periods and can be married or single in each of
these periods. A household consists of one or two adults and possibly one
child. We treat fertility as a choice variable and each couple decides on
whether or not it should have a child in the first period. We assume that
childless men and women are identical and both earn the same wage wf .
However, if a couple has a child then, because the mother is the one who
gives birth, her second period wage drops to a lower level, wm .
11.2.1 The technology and preferences

The household pools the incomes of its members and allocates it to buy an
adult good a and a child good c. Each parent has one unit of time which can
be allocated between market work and child care. Let hm and hf denote
the time spent by mother and father in market work, respectively. Then,
the amount of time they spend at home is tj = 1 − hj , where 0 ≤ hj ≤ 1
for j = m, f .
The household production function is
q = αa + t + g(c) (11.1)
where
t = βtf + γtm . (11.2)
The output q is interpreted as the child’s utility or ‘quality’. The para-
meter α describes the marginal effect of the adult good, a, on the child’s
quality, the parameters β and γ represent the productivities of the father
and mother, respectively, in household work and t is total time spent on
the child, measured in efficiency units. The function g(c) is assumed to be
increasing and concave, with g(0) = 0. The linearity in t is assumed to
allow corner solutions whereby family members specialize either in house-
hold work or market work. To determine the pattern of specialization under
different household structures, we assume
γ > wm (1 + α)
β < wf (1 + α) (11.3)
where wm is the wage of the mother and wf is the wage of the father.
That is, the mother is more productive at home, while the father is more
productive in the market. This may hold either because the mother has an
absolute advantage in home production γ > β or that she has an absolute
disadvantage in market work, wm < wf , because of the erosion in her wage
due to her withdrawal from the labor force during child birth.
The adult good a is shared by all members of a household. The marginal

utility of each adult from the adult good is set to 1, while the marginal
utility of the child is set to the constant α that is smaller than 1. In contrast,
the child good, c, is consumed only by the child. However, indirectly, child
consumption matters to the parents of the child, who care about its welfare.
The utility of a child is defined to be identical to its quality, q, and the utility
of each parent is defined as the sum a + q. Thus, both parents care about
their joint child, wherever the child lives. In this sense, child quality is a
collective good for the natural parents.3
Taken together, these assumptions impose a quasi-linear structure that
implies that the time and specific goods spent on the child do not depend
on the household’s total income if a positive amount of the adult good is
consumed. In that case, any additional income is spent only on the adult
good. Income effects on the child are present, however, if the adult good is
not consumed.
Married couples also enjoy a match specific ”love” factor which we denote
by θ. This factor is random and not known at the time of marriage. The
quality of match θ is revealed at the end of each period.4 We assume that θ is
independent across couples and is distributed with some known distribution
F (θ) that is symmetric and has a mean of zero.5 Individuals, who marry
at the beginning of the first period, observe θ at the end of the first period
and can then decide whether or not to break the marriage and look for a
new match. If marriage continues it will have the same θ. If a new marriage
is formed its θ will be a random draw from F (θ).
A negative shock to θ can cause dissolution of the marriage. Following
divorce, the parents may remain single or remarry, so that the child may
live in a household that consists of one or two adults. Household structure
affects both the technology and the household decision making. We assume
that a parent can spend time on a child only if they live in the same
household, but may spend money on the child even if they live apart. If
both parents live together with their child in an intact family, all household
goods are public and there is no conflict as to how much should be spent
on the child. However, if the family breaks the parents may have conflicting
interests, because the costs of caring for the child good will not be the same
when they do not share the adult good. In addition, if the custodial parent
3 A parent that lives apart from the child may enjoy it to a lesser degree, and we may
set the parent’s utility to a + δq if the parent and child live in a separate households.
The parameter δ may be interpreted as a discount factor that captures the idea that
"far from sight is far from heart".
4 In contrast to chapter 10, we simplify here by eliminating the premarital signal of
the quality of the match.

5 The zero mean assumption implies that in the second period agents marry only "for
money". In the first period, however, the average married couple enjoys a positive non
monetary gain, because the option of divorce eliminates some of the downward risk. The
model can be easily generalized to the case with θ̄ > 0.
remarries, then child quality is influenced by the foster parent who may
care less about the child than his natural parents.
11.2.2 The legal framework

We consider a modern society in which individuals can marry or divorce
at will. However, the partners can sign binding contracts, enforced by law,
that specify the custody arrangement and child support payments follow-
ing divorce or remarriage. Such contracts may be signed at the time of
divorce or at the time of marriage. An interim contract signed at the time
of divorce takes the presence of children and the separation as given, and
it’s main objective is to influence the expenditures on children under the
different household structures that may arise if each parent remarries or
remains single. An ex-ante contract, signed at the time of marriage, aims
to influence the fertility and separation decisions as well. We discuss here
simple and familiar contracts in which the mother obtains custody and the
father commits, at the time of divorce, to pay the mother a fixed amount
that is not contingent on whether one or both of the parents remarry. Such
binding contracts are in fact enforced by law.
Except for the enforcement of arrangement that the partners may reach,
the law may also intervene by setting standards within which the partners
can operate. Custody is most often given to the mother on the ground that
she can take better care of the child, while the father obtains visitation
rights. Unless stated otherwise, we shall assume that, following divorce the
mother is the sole custodian.6 Often, custody assignment is associated with
some amount of child support that is mandated by law. The guiding prin-
ciple is that the custody assignment and the mandated payments should
minimize the harm to the child. Such legal constraints may affect the agree-
ments that partners would reach when bargaining in the "shadow of the
Law" (see Mnookin and Kornhauser, 1977).
11.2.3 The meeting’s technology

As in chapter 10, we assume that, each period, a person meets a random
draw from the population of the opposite sex in the same age group. If
this person is already married then such a meeting is ”wasted” and no new
marriage is formed. This feature creates "increasing returns" in meetings
(see Diamond and Maskin, 1979), whereby it is more likely to meet a single
6 See Cancian and Meyer (1998). However, the share of joint physical custody has
increased over time. Halla (2009) examines the impact of state differences in this trend
and concludes that the option of joint custody has raised the incentives of men to marry,
with little impact on divorce.
person if there are more singles around.7

Only if two singles meet, they can form a new household. We denote
by p the proportion of singles (divorcees) that one meets in the second
period of life. Because the expected quality of match, θ, which is revealed
with a lag, equals zero and the material gains from marriage are positive,
every meeting of two divorcees results in a marriage, so that p is also the
probability of remarriage. A new element in this chapter is that, in the
second period, one can meet individuals that differ in their attractiveness
as partners for remarriage, because of the presence of children, lack of
income or commitments to previous spouses. However, the consequences of
having a child may differ for men and women. For women it implies a lower
wage. For men it may imply commitments to the ex-wife. In either case, a
person with a child is a less attractive match.
11.3 Household structure and child care

We begin our analysis with the allocation of resources by couples with one
child. This allocation depends on whether the partners live with their child
or are separated.
11.3.1 Intact family

If both parents live with their child in an intact family, the utility of hus-
band and wife is the same for all allocations and the two spouses will
agree to maximize their common utility subject to the household budget
constraint
a + c = wm hm + wf hf . (11.4)
Because the mother is assumed to have the comparative advantage in home
production, the father spends all his discretionary time in the market, hf =
1, while the mother will spend all her discretionary time at home. To verify
the optimality of this outcome, note that, due to assumption (11.3), an
increase the father’s work in the market raises the utility of both parents
by wf (1 + α) − β, while an increase in the mother’s work at home raises
their common utility by γ − wm (1 + α).
7 Lauman et al. (1994, Table 6.1 ) report that about half of the marriages arise from
meeting in school, work, and private party and only 12 percent originate in specialized
channels such as social clubs or bars. The establishment of more focused channels, where
singles meet only singles, is costly and they will be created only if the ”size of the
market” is large enough. Also, as noted by Mortensen (1988), the search intensity of
the unattached decrease with the proportion of attached people in the population. The
reason is that attached individuals are less likely to respond to an offer, which lowers
the return for search (see Chapter 7).
Given this specialization pattern, the amount spent on the child is de-
termined by equalizing the marginal utilities from a and c, that is,
1 + α = g 0 (c). (11.5)
We denote the unique solution to (11.5) by c∗ and assume that c∗ < wf so
that a positive amount is spent on the adult good a.
The utilities of the three family members in an intact family are
uc = g(c∗ ) + γ + α(wf − c∗ )
um = uf = g(c∗ ) + γ + (1 + α)(wf − c∗ ). (11.6)
We shall denote the above common utility of father and mother in an intact
family (with a child) by u∗ .
11.3.2 Separation, custody and voluntary transfers

If the parents separate, one of the parents receives custody over the child.
Only the custodial parent can contribute household time to the child, but
both parents can participate in the child expenditures. The non custodial
parent continues to care about the child and may wish to transfer resources
to it voluntarily. Transfers can be earmarked in the form of tuition and
health care, for instance, or fungible in which case the custodian treats it
as regular income that can be allocated according to the custodian’s prefer-
ences. Generally, an ear-marked transfer is preferred by the non custodial
parent, because part of the fungible transfer is ‘taxed’ by the custodial
parent (and also the new spouse if remarriage occurs) and does not reach
the child. Realistically, the father can rarely transfer money directly to the
child, especially when he is young. We shall, therefore, discuss here only
fungible transfers. The transfers are determined at the time of separation,
prior to meeting a new partner and are binding when the new marital sta-
tus of the parents is realized. Such voluntary commitments are enforced by
law and typical examples are child support and alimony agreements. We
shall discuss separately non contingent contracts, such as child support in
which the payment usually does not depend on the marital status of the
parents, and contingent contracts, such as alimony in which the payment
may stop if the ex-spouse remarries.
In practice, custody is often given to the mother, based on the idea that
she can or willing to take better care of the child. In terms of our model,
this is rationalized by the assumed comparative advantage that the mother
has in housework. However, it is possible that the child is better of with
the father because of the higher adult consumption that he provides. The
custody choice is, therefore, related to the transfers that occur between the
divorced parents. If transfers are sufficiently high, it is possible to restore
at least partially the efficient division of labor that is attained under mar-
riage, so that the child and consequently the parents are less harmed by
the divorce. For the time being, we shall assume that the mother receives
custody and will address this issue again after we derive the equilibrium
level of transfers.
Single custodial mother

If the custodial mother remains single, she will choose the amount of time
that she spends at work hm , her adult consumption am and the amount
of child goods c so as to maximize her own utility subject to her budget
constraint, taking as given the amount that the ex-husband transfers to
her, s. Her utility is then defined as the solution to the program
um (s) = max {(1 + α)am + γ(1 − hm ) + g(c)} (11.7)
am ,c,hm
subject to
am + c = wm hm + s,
0 ≤ hm ≤ 1.
The mother’s choices as a function of s are summarized in Figure 11.1
below. Because of the quasi-linear structure of the problem, the solution has
three distinctly different regions. For low levels of s, the mother withdraws
some time from the child and works in the market part time. She then
spends all her disposable income on child goods. The optimum conditions
in this region are
wm g 0 (c) = γ, (11.8)
c = wm hm + s.
Thus, the mother spends a fixed amount of money, ĉ on child goods and
works in the market the minimal amount of time required to achieve this
target. As s rises, the mother reduces her market work until it reaches zero,
spending more time on child care.
For high levels of s, the mother does not work in the market and allocates
her disposable income between the child and adult goods. The optimum
conditions in this region are
g 0 (c) = 1 + α, (11.9)
c + am = s.
That is, the mother will spend a fixed amount of money, c∗ , on the child
and adjust her adult consumption according to the level of s.
For intermediate values of s, satisfying
γ
> g 0 (s) > 1 + α, (11.10)
wm
the mother will not work and will not consume adult goods, so that all her
income and free time are devoted to the child.
This pattern of behavior reflects our assumption that the mother has
comparative advantage in child care γ > wm (1 + α) which is seen to imply
that, for the mother, the child comes first and she spends resources on
herself only when she is sufficiently wealthy. The utility of the child is then
⎧
⎨ g(ĉ) + γ(1 − ĉ−swm ) if s ≤ ĉ,
q(s) = g(s) + γ if ĉ < s < c∗ , (11.11)
⎩
g(c∗ ) + γ + α(s − c∗ ) if s ≥ c∗ ,
and the utility of the mother is
⎧
⎨ g(ĉ) + γ(1 − ĉ−s
wm ) if s ≤ ĉ,
um (s) = g(s) + γ if ĉ < s < c∗ , (11.12)
⎩
g(c∗ ) + γ + (1 + α)(s − c∗ ) if s ≥ c∗ .
Remarried custodial mother

If the custodial mother remarries, she may spend less time and money on
the child, because her new husband receives little or no benefits from such
spending. This is in contrast to the case of an intact family where, by
assumption, both parents benefit equally from the time and money spent
on the child. To sharpen our results, we assume that the new husband
derives no utility at all from the step child and depends only on the adult
good, a, that the remarried couple purchases.8 The mother, however, cares
about both the adult good, a, and the child good, c. All this means that
the child good is a private good for the wife in the new household. Because
of the potential conflict of interests, bargaining is required to determine the
amount spent on the adult good in the new household.9 Assuming that the
bargaining outcome is efficient, it must be on the Pareto frontier within the
new household. Let yh = wf − s0 denote the income that the new husband
brings into marriage, net of his obligations to his ex-wife, s0.10 Then, the
Pareto utility frontier is given by
⎧
⎪
⎪ (1 + α)a + γ + g(yh + s − a)
⎪
⎪ if yh < a < yh + s − ĉ
⎪
⎪
⎨ (1 + α)a + γ (w + y + s − a − ĉ) + g(ĉ)
wm m h
um = (11.13)
⎪
⎪ if yh + s − ĉ ≤ a ≤ yh + wm + s − ĉ
⎪
⎪
⎪
⎪ (1 + α)a + g(yh + wm + s − a)
⎩
if yh + wm + s − ĉ < a < yh + wm + s
8 The new husband’s utility also depends on the utility of his child from the previous
marriage, which is taken as given in the bargaining of the remarried couple.

9 Akashi-Ronquest (2009) reports lower child investments following remarriage (com-
pared with intact families) and that an increase in the hourly wage of a biological mother
significantly improves her child investment when her husband is a stepfather of the child,
while there is no such effect for mothers living with the biological father of the child.
The author interprets these findings as bargaining on child quality in step families.
1 0 Since we shall examine only symmetrical equilibria, there is no loss of generality in
assuming that all other fathers make the same payments.

and described in Figure 11.2, which is drawn for the case in which the
transfer from the father, s, exceeds the efficient level of expenditure on the
child good, c∗ . For levels of a close to yh , the mother spends all her time
on the child and the implied expenditure on the child good c exceeds c∗ .
In this case, both spouses want to reduce c and increase a, although the
child may be hurt from such a substitution. As a is raised sufficiently so
that c reaches c∗ the newly formed couple enters the region of conflict.
Any further increase in a and the associated decrease in c, benefits the new
husband but reduces the mother’s (and the child’s) utility. Initially, the
mother continues to spends all her time on the child but when a reaches
yh +s−ĉ, she starts to work part time and continues to do so until a reaches
yh + wm + s − ĉ. In this segment, the Pareto frontier is linear because the
child good is held at a fixed level, c = ĉ, and any increase in a is achieved by
an increase in hm which raises the father’s utility by wm dhm and reduces
the mother’s utility by ((1 + α)wm − γ)dhm . At high levels of a, exceeding
yh +wm +s− ĉ, the mother works full time in the market and as a rises, the
amount of child good is reduced until it reaches zero and the new husband
obtains all the household resources, yh + wm + s.
To proceed with the analysis, one must determine how the conflict be-
tween the spouses is resolved and which particular point on the Pareto
frontier is selected. For simplicity, we assume that the new husband obtains
all the surplus from remarriage, so that the point on the Pareto frontier
is selected so as to make the mother indifferent between remarriage and
remaining single. This allows us to illustrate the general equilibrium issues
in a relatively simple manner. The reader may interpret the model as a
worst case scenario from the point of view of the mother and child.11 The
efficient level of adult consumption is then defined as the solution of the
following maximization program
a(s, yh ) = max a (11.14)

a,hm ,c
subject to
a + c = wm hm + yh + s,
(1 + α)a + γ(1 − hm ) + g(c) ≥ um (s),
0 ≤ hm ≤ 1.
For remarriage to take place, it must be the case that the solution of this
1 1 Wecould use instead a symmetric Nash-Bargaining solution to determine the bar-

gaining outcome (see Chiappori and Weiss, 2007). The Nash axioms imply that the
bargaining outcome must maximize the product of the gains from remarriage, relative
to remaining single, of the two partners. This model yields similar qualitative results,
because the mother is assumed to have lower income and therefore her options outside
marriage are worse than those of men. The magnitudes of the welfare loss of the child
and mother would, of course, be smaller if the mother gets a larger share of the gains
from marriage.
program satisfies a(s, yh ) ≥ yh , otherwise the new husband would be better

off as single.
The first order conditions for the efficient allocation within the new
household are
−1 − λ(g 0 (c) − (1 + α)) = 0, (11.15)

wm − λ((1 + α)wm − γ) = 0 if 0 < hm < 1,
wm − λ((1 + α)wm − γ) ≤ 0 if hm = 0,
wm − λ((1 + α)wm − γ) ≥ 0 if hm = 1,
where λ is a Lagrange multiplier such that λ1 equals the slope of the Pareto
frontier in the new household (see equation (11.13) and Figure 11.2)). In
an interior solution with 0 < hm < 1 we have
wm
λ= < 0, (11.16)
(1 + α)wm − γ
because a marginal increase raises the utility of the husband by wm and

reduces the utility of the wife by (1 + α)wm − γ. As in the case of a single
mother household, wm g 0 (c) = γ and c = ĉ as long as the mother works
part time in the market. If hm is at the boundaries of 0 or 1, the level of
child expenditures c is determined by the requirement that the utility of
the mother is equated to her reservation utility um (s) and
1
λ= . (11.17)
(1 + α) − g 0 (c)
In this section, we shall consider only equilibria with moderate levels of

transfers such that the mother spends some time in the labor market when
she is remarried. This allows us to exploit the linear Pareto frontier that
arises in this case, which substantially simplifies the calculations. Such inte-
rior solution requires that the net income of the new husband is sufficiently
high to motivate the mother to work part time but not so large as to cause
her to work full time.
If the remarried mother has no surplus then
a(s, yh ) + q(s, yh ) = um (s). (11.18)
When the mother remarries, she obtains more of the adult good that she
shares with her new husband but the child’s utility q(s, yh ) is lower. This
implies that the mother’s remarriage has a negative impact on the father,
but he can mitigate this effect by transferring money to the mother, that
is by increasing s. Specifically,
∂a(s, yh ) γ
= λ(u0m (s) − ) ≥ 0, (11.19)
∂s wm
∂a(s, yh )
= 1 − λ(1 + α) > 1,
∂yh
∂q(s, yh ) ∂a(s, yh )
= u0m (s) − > 0,
∂s ∂s
∂q(s, yh ) ∂a(s, yh )
= − ≤ 0.
∂yh ∂yh
An increase in the transfer s raises the utility that the mother would re-
ceive as single and improves her bargaining position in the newly formed
household. Consequently, the remarried mother works less and spends more
time with the child, which raises the utility of the child. 12 However, an
increase in s also has the unintended effect of raising the new husband’s
utility, who "eats" part of the transfer. An increase in the net income of the
new husband raises his gain from marriage a(s, yh )−yh because the mother
spends less time with the child and more time in the market. The mother
is willing to do such a sacrifice of child quality because she is compensated
by a higher level of adult consumption, jointly with the new husband.
The result that remarried mothers work more in the market may seem
counterfactual.13 We emphasize that market work is just one way of trans-
ferring resources from the child to the new husband and the crucial as-
sumption is the availability of a linear transfer in some non-trivial range.
For instance, the mother may spend less time with the child and more time
with the new husband in joint leisure activities. As long as such substitu-
tions are available at a fixed rate of exchange, the results are the same as
if the remarried mother would spend time working in the market.
11.4 Transfers, the interim perspective

Following separation, the parents can be in four different states, depending
on the new marital status of their ex spouses:
1. Both parents are single, which happens with probability (1 − p)2 .
1 2 Note that ∂a(s,yh ) = 0 if s = ĉ, ∂a(s,yh ) = 1 if s = c∗ and 1 > ∂a(s,yh ) > 0 for
∂s ∂s ∂s
ĉ < s < c∗ .
1 3 As seen in Chapter 1, Figure 1.13 , the raw data suggests that divorced women work
more than married women. However Seitz (1999) shows that correcting for selection and
unobserved attributes, there is no significant difference in labor supply of divorced and
married women, while remarried women work significantly more than married women,
as our model suggests.
2. The father remains single while the mother is remarried, which hap-
pens with probability p(1 − p).
3. The mother remains single but the father is remarried, which happens
with probability (1 − p)p.
4. Both parents remarry, which happens with probability p2 .
Note that, by assumption, the probability of remarriage is the same for
the husband and wife, and that meetings and subsequent remarriages are
independent across parents.
Anticipating these contingencies, the father may be willing to commit to
transfer money to the custodial mother with the intention to influence the
welfare of the child, of whom he continues to care.14 Each father makes
his choice of s separately, taking the choice of others, s0 as given. These
payments are made at the time of divorce, before the marital status of the
ex-spouses is known. We, therefore, must use expectations in determining
the optimal level of the transfer. The expected utility of the father is,
therefore,
Vf = (1 − p)2 [wf − s + q(s)] + (1 − p)p[wf − s + q(s, wf − s0 )]
+p(1 − p)[a(s0 , wf − s) + q(s)]
+p2 [a(s0 , wf − s) + q(s, y − s0 )] (11.20)
and
∂Vf ∂q ∂a
= (1 − p)[q 0 (s) − 1] + p[ − ] (11.21)
∂s ∂s ∂yh
We first note that the father will never choose voluntarily transfer s that
exceeds c∗ because, in this case, the single mother would spend the marginal
dollar on the adult good. The father then receives a marginal benefit of α
γ
from the transfer if the mother remarries single and 1 + a − γ−(1+α)w m
if
she remarries. But his expected cost in terms of the adult good is higher,
because a transfer of a dollar costs the father 1 dollar if he remains single
∂a γ
and ∂y h
= γ−(1+α)w m
if he remarries (see equations (11.11) and (11.19)).
Under our maintained assumption that the remarried mother works part
time, equation (11.21) can be rewritten as
(
∂Vf (1 − p)( wγm − 1) + p( wγm − ∂y ∂a
) if 0 ≤ s < ĉ,
= h
∂a (11.22)
∂s (1 − p)(g (s) − 1) + p(g (s) − 2 ∂yh ) if ĉ ≤ s ≤ c∗ ,
0 0
where
∂a γ
= 1 − λ(1 + α) =
∂yh γ − (1 + α)wm
1 4 Another possible motive is that the father maintains an altruistic motive towards
his ex-wife. In this chapter, however, we ignore this added altruistic link and confine our
attention only to the case in which parents care about their children.
The two branches in (11.22) reflect changes in the mother’s behavior as a

function of the transfer s that she would receive if she would have remained
single. In the region 0 < s < ĉ, the single mother would work part time,
∂V
so that c = ĉ and q 0 (s) = wγm . The independence ∂sf from s in this region
implies that either the father will contribute nothing or he will voluntarily
commit on a transfer of at least ĉ. Which of these two possibilities applies
depends on the basic parameters of the model and the probability of re-
marriage. The father is certainly willing to transfer resources to his ex-wife
if he and the mother remain single with high probability, because then the
marginal benefit in terms of child quality, wγm , exceeds the marginal costs
in terms of the forgone consumption of the father, which is 1. The father
will be more reluctant to contribute if p is large, because then the cost for
∂a
him is larger, ∂y h
. The father will contribute at all p if γ > (2 + α)wm
which means that the mother is highly effective in caring for the child. If
this requirement is not satisfied, there will be some critical p below which
the father will give at least the amount ĉ, but above which he will give
nothing. In the region ĉ < s < c∗ , the mother uses the transfer to increase
child consumption and because of the concavity of g(s), the marginal value
∂V
of the transfer ∂sf declines with s. Hence, an interior solution can exist in
this region.
We can now characterize the father’s incentives to support the mother.
Proposition 11.1 Let s∗ (p) be the optimal level of voluntary commitment
that the father is willing to make at the time of divorce. Then, s∗ (p) ≤ c∗ ,
declines in the probability of remarriage, p, and is independent of the income
of the new husband whom the mother may remarry. For a sufficiently high
comparative advantage of the mother in child care, that is, γ > (2 + α)wm ,
the optimal transfer, s∗ (p), exceeds ĉ. The transfer is then set to s∗ (p) =
c∗ if p < p1 , where p1 satisfies
∂a
1 + α = (1 − p1 ) + 2p1 . (11.23)
∂yh
Otherwise, if p ≥ p1 , s∗ (p) is determined by the unique solution to
∂a
g 0 (s) = (1 − p) + 2p . (11.24)
∂yh
The optimal transfer declines with the remarriage probability for two
reasons: the marginal impact of a transfer on child quality is larger (or
equal) when the mother is single, and the cost of giving are higher if the
father remarries.
The independence of the transfer from the new husband’s income implies
that although the expected utility of each husband depends on the transfers
by others, the marginal impact of s is not and, therefore, s is independent
of s0. This feature is reflected in the fact that
∂a γ
=
∂yh γ − (1 + α)wm
is a constant. However, if the mother does not work when she is remarried,
or works full time, the utility frontier for a remarried couple is no longer
linear and the marginal impact of the transfer to the mother will depend
on the net income of her new husband yh .
11.4.1 Partial equilibrium

Suppose that all couples have children. Then, at given probability of re-
marriage, p, the equilibrium outcome is that all fathers will transfer the
same amount s∗ (p) to the mother if the marriage dissolves. That is, given
that other fathers choose s0 = s∗ (p), each father independently chooses
s = s∗ (p). This equilibrium requirement is trivially satisfied here, because
the optimal choice of each father is (locally) independent of the choices
of others. We refer to the equilibrium as partial because, as we shall see
shortly, the remarriage and fertility rates must also be set at equilibrium
levels.
In this partial equilibrium, the mother works part time when she is re-
married but not as single. The reason for this difference is that she must
compensate her new husband for the option of sharing the adult good.
The amount of time that the mother spends in market work is
(1 + α)(wf − ĉ) + g(ĉ) − g(s∗ (p))

hm (p) = (11.25)
γ − (1 + α)wm
and a sufficient condition for an interior solution 0 < hm (p) < 1 for all p is
that:
γ − (1 + α)wm > (1 + α)(wf − ĉ) > g(c∗ ) − g(ĉ). (11.26)
Basically, the mother should have a sufficiently high comparative advantage

in child care to motivate her to spend some time at home and the net
income of the new husband should not be so high that the mother is driven
completely into the market, contrary to her comparative advantage.15
We see that when p rises and all husbands reduce their contribution,
the remarried mothers increase their hours of work and thus reduce the
amount of time spent with the child. The implied adult consumption in
the remarried household
a(p) = wm hm (p) + wf − ĉ (11.27)
rises in p but the child’s utility if the mother remarries
q(p) = αa(p) + γ(1 − hm (p)) + g(ĉ) (11.28)
1 5 The sufficient condition (11.26) is much stronger than we need because, as we shall
show shortly, the equilibrium remarriage rate is bounded by 12 .

declines in p, because the mother’s time is more important for the child
than the added adult good.
The expected utilities of the three family members, evaluated at the time
of divorce, are
Vm (p) = g(s∗ (p)) + γ,

Vc (p) = g(s∗ (p)) + γ − pa(p),
Vf (p) = g(s∗ (p)) + γ + (1 − p)(wf − s∗ (p)). (11.29)
Compared with an intact family with θ = 0, all three family members are
worse off if the marriage breaks. The child received less child goods because
the transfer from the father s∗ (p) is lower than c∗ (except at low probability
of remarriage p < p1 ) and also less time if the mother remarries. Both the
mother and the father suffer from the reduction in child quality. In addition,
there is a loss of resources resulting from the inability to share consumption
goods when the parents remain single. This cost is born mainly by the
mother. The assumption that the mother receives no surplus implies that
she pays for the adult good in terms of the child’s quality, so that her utility
is unaffected by remarriage but that of the child is reduced by a(p). The
father, on the other hand, gets the benefits from sharing a(p) with the new
wife and, in addition, he consumes the adult good when he is single. In
fact, he consumes as single more of the adult good than he would under
marriage. The outcome of this asymmetry is that the father’s expected
utility following separation is higher than the mother’s.
The expected utility of all family members in the aftermath of divorce
declines with the probability of remarriage, p. This is a surprising result,
given that remarriage is voluntary. It can be traced to the fact that a
higher remarriage rate does not only make it easier to remarry, which is
individually welfare enhancing, but also affects behavior in a way that may
be harmful to others. Thus, although the mother fully internalizes that the
child is worse off upon remarriage, this does not stop her from remarrying
if she is compensated by higher adult consumption. Nor does she take into
account the negative impact of her remarriage on her ex-husband. The
father’s incentives to transfer money to the custodial mother decline as the
probability of remarriage rises, because he anticipates that part of it will
be spent on adult goods that are not as useful to the child, mainly because
of a presence of a third party in the form of the new husband. As a result
of this reluctance to contribute, mothers are worse off even if they remain
single. Finally, each father is worse off mainly because the child is worse
off when the mother remarries and he cannot fully remedy that by the use
of transfers, due to the principal-agent issues that we described. This loss
of control is sufficiently costly to offset the gains that the father receives
when he remarries and obtains all the surplus.
11.5 Divorce
Having observed the realized quality of the current match, each spouse
may consider whether or not to continue the marriage. A parent will agree
to continue the marriage, if given the observed θ the utility in marriage
exceeds his/her expected gains from divorce. Under divorce at will, the
marriage breaks if
u∗ + θ < max (Vm , Vf ) − b (11.30)
where u∗ is the common utility of the husband and wife if the marriage
would continue (not incorporating the quality of the match) and b is a fixed
cost associated with divorce. The fixed costs reflect the emotional, legal and
relocation costs associated with the change in marital status that affects
the child and parents. We assume that these costs are higher for couples
with children and are shared equally by the two spouses.
The particular value of θ that triggers divorce is given by
θ∗ (p) = max (Vm (p), Vf (p)) − u∗ − b. (11.31)
The critical value θ∗ is seen to equal the expected gains from divorce,
relative to remaining married, evaluated at θ = 0, which is the mean value
of θ in the population. In other words, the couples that divorce are those
with a realized quality of match that is below the unconditional expectation
of the gains from divorce, before θ is observed. These expected gains are
negative, because an intact marriage with θ = 0 is better for all parties.
The probability that a couple will divorce is then
Pr{θ ≤ θ∗ } = F (θ∗ ) (11.32)
where F (.) is the cumulative distribution of θ.

Our previous analysis implies that
Proposition 11.2 If all couples have a child, then in a partial equilibrium

where all fathers choose the optimal transfer s∗ (p), the divorce decision at
any expected remarriage rate, p, is determined by the father. The critical
value of θ that triggers divorce θ∗ (p) is negative and declines in the proba-
bility of remarriage, p.
Because of the ex-post asymmetry between the partners, separation may

be inefficient. The father, who has strictly higher expected gain from divorce
than the mother, will initiate the divorce at some θ such that the mother
wants the marriage to continue and inflict on her a loss of match quality.16
1 6 We note, however, that if δ < 1 so that the non custodial father suffers from the
distance from his child, the father’s gains from divorce decline and may be lower than
the mother’s.
Couples without children

If the parents do not have a child, the "material" utility (not including
the love component θ) of each parent in an intact family is u∗ = 2wf ,
reflecting the assumptions of income pooling and joint consumption of the
adult public good. However, a parent that remains single consumes only
wf . If all couples do not have a child, the symmetry between the parents
is reestablished and both expect upon separation to receive
Vm (p) = Vf (p) = (1 − p)wf + p2wf . (11.33)
An important difference from the case with children is that the expected
utility of the two parents, as evaluated at the time of divorce, rises with the
probability of remarriage. This is simply an outcome of the option to share
consumption upon remarriage, without any negative impact of divorce on
child quality.
The critical value of θ that triggers divorce is now given by
θ∗ (p) = −(1 − p)wf (11.34)
which rises with p. That is, the higher is the probability of remarriage the
more likely it is that a particular couple will divorce. This result is in sharp
contrast to that in Proposition 11.2, illustrating the marked difference that
children might have on the divorce decisions.
11.6 Fertility
So far, we took the number of children as given and assumed that all couples
have children. We now examine the decision to have children.
We view children as an investment good that the parents produce at
some cost during the first period of marriage, before the quality of match
is revealed. To simplify, we assume that only one child can be produced. The
costs of having a child are the forgone earnings of the mother associated
with child birth and child rearing. We assume that the mother cannot work
in the first period if she gives birth, so that wf is lost in the first period.
Also, because of the mother’s withdrawal from the labor force, her second
period wage erodes from wf to wm . The benefits from the child that accrue
in the second period depend on the probabilities of divorce and remarriage
and on the parents’ ability to care for the child in the aftermath of divorce.
To avoid trivial solutions, we assume that children may be a bad or good
investment, depending on the circumstances. In particular, a couple that
obtains the average draw θ = 0 and chooses not to divorce gains from
having had children. This is equivalent to saying that children are desired
if divorce is not an option. However, when divorce is an option, children
may be a liability if they lock the parents into bad matches.
An important feature of the analysis is that the decisions of each couple

whether to divorce or to have a child depend not only on the circumstance
of the couple, for example, if it suffered a negative shock, but also on the
decisions of other couples to have children and to divorce, as well as on
the contracts that they sign. These decisions by other couples influence
the prospects of remarriage and the quality of potential mates. To simplify
our analysis, we focus here on the case in which, in equilibrium, all couples
have children or all couples do not have children. Therefore, we only need to
consider the benefits of a particular couple from having a child, conditioned
on whether or not all other couples have children.
Given the choices of others, the expected life time utility of a parent j
in a particular couple is
Z∞
Wj,n (p) = u0n + (u∗n + θ)f (θ)dθ + F (θ∗n (p))(Vj,n (p) − bn ) (11.35)
θ∗
n (p)
where, j = f for the (potential) father and, j = m for the (potential)

mother, and n is a choice variable that equals to 1 if the couple has children
and 0 otherwise. The term u0n in equation (11.35) represents the utility of
the two partners in the first period, which is 2wf if the couple has no
children and only wf if a child is born, because of the mother’s withdrawal
from the labor force during child birth. The term u∗n represents the parents
utility if marriage continues, which is 2wf if the couple has no children and
u∗ if there is a child. The fixed costs of separation bn are assumed to be
larger when the couple has children.17
The expected life time utility is higher for the partner with the higher
gains from divorce who determines the divorce decision. In fact, the ex-
pected life time utility can be rewritten as
⎧ θ∗
⎪
⎪
nR(p) ∗
⎪ un + un +
⎪
0 ∗
(θn (p) − θ)f (θ)dθ if Vj,n (p) ≥ Vi,n (p)
⎪
⎨ −∞
∗
Wj,n (p) = θnR(p)
⎪
⎪ u 0
+ u ∗
+ (θ∗n (p) − θ)f (θ)dθ
⎪
⎪ n n if Vj,n (p) < Vi,n (p)
⎪
⎩ −∞
∗
−F (θn (p))(Vi,n (p) − Vj,n (p))
(11.36)
where the term u0n + u∗n is the value of the marriage if it never breaks
θ∗
R(p) ∗
n
and the term (θn (p) − θ)f (θ)dθ is the option value of breaking the
−∞
marriage if it turns sour because of a bad draw of θ. The option to sample
from the distribution of θ is a motivation for marriage that exists even
1 7 It would be more realistic to allow discounting of future utilities in (11.35). However,
this does not add any new conceptual issues and to economize on notation we set the
discount factor to unity.
if marriage provides no other benefits. However, this option is available

only to the person with the higher gains from divorce, who determines the
divorce. When the marriage breaks, an event that happens with probability
F (θ∗n (p)), the spouse who does not initiate the divorce and is left behind
suffers a capital loss given by Vi,n (p) − Vj,n (p). The value of the option for
the spouse who determines the divorce, increases in the gains from divorce,
F (θ∗n (p), and also with the variability in the quality of match, because then
the ability to avoid negative shocks becomes more valuable.
We define the benefit of spouse j from having a child as
Bj (p) = Wj,1 (p) − Wj,0 (p). (11.37)
Because the production of children must involve both partners, a couple

will have a child if and only if both partners agree to have a child. That is
if,
B(p) = M in{Bm (p), Bf (p)} ≥ 0. (11.38)
Because the father determines the divorce decision, his life time utility
must also be higher, Wf,1 (p) ≥ Wm,1 (p). Now imagine that a particular
couple departs from the general pattern and chooses not to have a child. If
the husband remains single, he will get his income wf and if he remarries he
will get wf + wm hw , because his new wife is a custodial mother who spends
some of her time on the child. The wife also gets wf if she remains single,
because she has no child, and has the same wage as her husband. However
if she remarries she will get wf + wf − s∗ (p). For a sufficiently large gap
between wf and wm , a non custodial father brings more income into the
marriage than a custodial mother who generally works only part time.18
Therefore, if all other couples have a child, the wife in a couple that chooses
not to have a child expects to gain from divorce more than her husband,
Vf,0 (p) < Vm,0 (p). In this case, she will determine the divorce decision, and
consequently, her life time utility is higher, Wf,1 (p) < Wm,1 (p). It follows
that the mother has lower benefits from having the child, Bm (p) < Bf (p).
Proposition 11.3 The wife determines whether or not the couple have
children.
When the value of having a child is strictly lower for the wife, there may
be some p such that she will prefer not to have a child when the husband
would like to have a child. In such a case, the father may be willing to sign
a binding ex-ante contract which would transfer money to the mother upon
separation if this would induce her to have a child. We shall return to this
issue in the concluding section.19
that s∗ (p) ≤ c∗ , a sufficient condition is that wf > wm + c∗ .

1 8 Recalling
1 9 Wenote again that if δ < 1, so that the non custodial father suffers from the
distance from his child, the father’s benefit from having a child could be smaller than
the mother’s.
11.7 Equilibrium
Equilibrium requires consistency among the choices of the participants in
the marriage market and realization of their expectations. The first con-
sistency requirement is that the aggregate divorce rate coincides with the
expected remarriage rate. Assuming independence of the marital shocks
across couples and a large population, the proportion of couples that will
choose to divorce is the same as the probability that a particular couple
divorces. The decision of each couple to divorce depends on the expected
remarriage rate, p. Assuming that a person can remarry only with a di-
vorcee and that meetings are random, we require that, in equilibrium, the
realized aggregate divorce rate must equal the expected remarriage rate of
all agents. That is,
p = F (θ∗ (p)). (11.39)
Because the gains from divorce for a couple with θ = 0 are negative, the
threshold θ∗ (p) is negative and it then follows from our assumptions on
F (p) that any solution of (11.39) must be such that p < 12 .
When fertility is endogenous, we have the additional requirement that
the expected gain from divorce must reflect the optimal fertility choices of
the participants in the marriage market. Thus, in an equilibrium without
children we must have that
p = F (θ∗0 (p)), (11.40)
and B(p) < 0. That is, the expected gains from divorce are calculated based
on the assumption that all singles are childless, and given these expectations
no couple wishes to have a child. Similarly, in an equilibrium in which every
couple has a child we must have
p = F (θ∗1 (p)), (11.41)
and B(p) > 0.

The third requirement from equilibrium is that the participants contract-
ing choices must be optimal, given by s∗ (p). These equilibrium requirements
implicitly assume symmetric equilibria in which all agents behave in the
same manner. Such equilibria are a natural choice given that all agents
are initially identical, but other equilibrium may exist. In a more general
analysis, one can incorporate also mixed equilibria such that some couples
choose to have a child, some choose to remain childless and all couples are
indifferent between having and not having a child. However, because such
equilibria tend to be unstable, we are less interested in them and will not
introduce the additional notation that is required to characterize them.
For any given legal environment, it is convenient to rewrite the equilib-
rium condition in the form
F −1 (p) = θ∗ (p). (11.42)

This formulation separates the properties of the distribution of the unan-

ticipated shocks from the properties of the trigger θ∗ (p) that summarized
the impact of the expected remarriage rate on the expected gains from di-
vorce. Because F −1 (p) rises in p, while θ∗ (p) declines (rises) in p when a
child is (not) present, there may be two equilibrium points: a high divorce
(remarriage) without children and a low divorce (remarriage) with children.
Numerical Example 11.1

We now present a numerical example that illustrates some of the results.
We adopt here a slightly more flexible formulation, allowing the father to
suffer a utility loss when he lives separately from the child, δ < 1, and
allowing discounting of future utilities, ∆ < 1. With these modifications
(11.20) and (11.35) become:
Vf = (1 − p)2 [wf − s + δq(s)] + (1 − p)p[wf − s + δq(s, wf − s0 )]

+p(1 − p)[a(s0 , wf − s) + δq(s)] + p2 [a(s0 , wf − s) + δq(s, y − s0 )]
and
Z∞
Wj,ch (p) = u0ch + ∆{ (u∗ch + θ)f (θ)dθ + F (θ∗ch (p))(Vj,ch (p) − bch )}
θ∗
ch (p)
respectively, where 0 < δ, ∆ ≤ 1. In the figures, we set δ = 0.75 and

∆ = 0.625.
Figure 11.3 shows the optimal transfers that the father promises to his ex-
wife at the time of divorce as a function of the prospective remarriage rate,
p, and the implied consequences for the child when the mother remarries
and remains single.20 The optimal transfer from the father to his ex-wife,
s∗ (p), declines with p because the marginal impact of the transfer on the
child is lower when the mother remarries. As a consequence, the utility of
the child when the mother remains single, γ+ g(s∗ (p)), declines too. The
child’s utility when the mother remarries, αa(p) + γ(1 − hm (p)) + g(ĉ),
declines because a lower transfer implies that the mother works more and
spends less time on the child; but the reduction in the mother’s caring time
has a stronger effect than the child’s gains from the higher consumption of
the adult good. The child’s utility when the mother is single exceeds the
child’s utility when the mother remarries, because the mother is "paying"
for her gain of adult good by reducing the utility of the child. Therefore, the
2 0 The graphs are drawn for the case in which the mother’s productivity at home
is γ = 1.75, the mother’s wage is wm = 0.5 and the father’s wage is wf = 1. We

specify g(c) = g1 (1 − (1 + c)−g2 ) where g1 = g2 = 94 and α = 14 . Thus c∗ satisfies
g1 g2 (1 + c)−g2 −1 = 1 + α, implying c∗ = 0.5378 and g(c∗ ) = 1.3956. ĉ satisfies g2 (1 +
c)−g2 −1 = wγ implying ĉ = 0.1203.
f
child’ loss from remarriage equals a(p), which rises with p, as the bargaining
position of the mother worsens when the father transfers less.
As a consequence of the decline in the optimal transfer, the expected
utilities at the time of divorce of the child, mother and father all decline (see
Figure 11.4). Assuming a moderate loss for the father when he lives apart
from the child, δ = 0.75, the expected utility of the father is higher than
that of the mother through most of the relevant range of p. Consequently,
the father determines the divorce decision if 0.05 < p < 0.5, while the
mother determines the divorce decision if p < 0.05.21
In Figure 11.5, we plot the maximum of the husband’s and wife’s ex-
pected gains (losses) from divorce, including the fixed cost of divorce, for
couples with and without children. These gains rise for couples without chil-
dren because remarriage enhances joint consumption and decline for cou-
ples with children, because remarriage implies lower spending on the child
that dominate the gains from joint consumption. The intersections of these
curves with the inverse probability function at p = 0.214 and p = 0.334
represent potential equilibria, where the realized divorce rate equals the
expected remarriage rate. 22 A higher potential equilibrium point arises
when all couples do not have children because, by assumption, such cou-
ples have lower fixed cost of separation and, in addition, they do not suffer
from the reduced welfare of the child when the marriage breaks. To make
sure that the two intersections in Figure 11.5 satisfy all the requirements
for equilibrium, we must further verify that, at the higher intersection with
p = 0.334, no couple without a child wants to deviate and have a child
when all the others do not have a child, while in the low intersection with
p = 0.214, no couple with a child wants to deviate and have no child when
all others have a child.
Figure 11.6 shows the incentives of the husband and wife to deviate and
have no child when all other couples have a child and their child support
is set at the optimal level s∗ (p). The expected life time utilities of the
husband and wife when all couples have children decrease with the proba-
bility of remarriage, with the mother’s life time utility being slightly lower
than the father’s (except for very low p, p < 0.05), reflecting the father’s
higher expected gain from divorce. In contrast, the life time utilities that
the parents obtain upon deviating to not having a child rise with the prob-
ability of remarriage because of the gain from joint consumption. With this
structure, a deviation would occur only at a sufficiently high probability
2 1 The difference between the father’s and mother’s expected gains from divorce is
(g(s∗ (p)) + γ)(δ − 1) + (1 − p)(wf − s∗ (p)),
which is always positive if δ = 1, but can be negative for δ < 1.

2 2 The inverse probability is drawn for the case in which the match quality, θ, is
uniformly distributed over [−d, d] so that p = prob{θ ≤ x} = d+x 2d

and x = d(2p − 1).
In the figures, we set d = 2.
of remarriage. Because both partners are required to produce a child, it is

sufficient for a deviation to occur that one of the two parents refuses to
have a child. We see that the wife wants to deviate only if the remarriage
rate exceeds 0.28, while the husband wants to deviate only if the remarriage
rate exceeds 0.36. Thus, the intersection in Figure 11.5 at p = 0.214 is an
equilibrium with children. By a similar argument, it can be seen that the
intersection in Figure 11.5 at p = 0.334 is an equilibrium without children,
because neither the husband nor the wife wish to have a child if all others
do not have a child (see Figure 11.7).23
In Table 11.1, we provide some comparative static results. The first panel
shows the benchmark parameters. The second panel shows the impact of
changes in the variance of the quality of match, holding the mean constant.
The inverse probability is drawn for the case in which the match quality, θ,
is uniformly distributed over [−d, d], so that such an increase is represented
by an increase in d. The higher is d, the more likely it is that the realized
match will be sufficiently low to trigger divorce. Therefore, the equilibrium
divorce rate rises with d. At a low d, d = 1.5, the only equilibrium is the
one with children and for a high d, d = 2.5, the only equilibrium is without
children. For intermediate values of d (d = 2.0 and d = 2.2) there are two
equilibria for each value of d. It is then possible that a small exogenous
change that is, a rise in d from 2 to 2.5 will cause a large change in the
divorce rate, shifting the equilibrium from a divorce (remarriage) rate of p =
0.214, (with children) to 0.375 (without children), with a noticeable rise in
the utility of both parents. This change illustrates a social multiplier effect
where the higher willingness of each couple to divorce, as a consequence of
the exogenous shock (that is, the rise in d), increases the aggregate divorce
rate, which further increases the incentives to divorce. The rise in the life
time utility of the parents with d illustrates our observation that marriage
has an option value, because bad outcomes to the quality of the match can
be avoided through divorce. However, the child, who is a passive agent that
cannot directly influence the divorce decision, suffers from the dissolution
of the marriage.
The third panel of Table 11.1 illustrates the impact of an increase in the
fixed costs of divorce in the presence of children, b1 .24 An increase in b1
reduces the divorce rate of couples with children and thereby reduces the
expected life time utility of the parents who cannot so easily recover from
bad matches. The child, of course, gains from such a change, because he
is better off in an intact family and, by assumption, does not suffer from
a bad quality of match. Although such a change does not directly affect
2 3 If all other couples do not have a child, a husband would like to deviate and have
a child only if p < 0.19, while a wife would like to have a child only if p < 0.24. In
calculating the deviation, we take into account that when the couple will have the child
the father will commit to pay child support according to s∗ (p).
2 4 We assume no fixed cost of separation in the absence of children; that is, b = 0.
0
the outcomes if all couples do not have children, it still may influence the
equilibrium outcome through the impact on the incentives to have children.
Thus if b1 is reduced from 0.25 to 0.05 then the equilibrium without children
disappears and the only equilibrium is with children.
The last panel of Table 11.1 illustrates the impact of changes in the
utility of the father, as δ rises and he suffers less from living apart from the
child. Such an increase in proximity raises the utility of the father directly,
but it also raises his willingness to transfer money to the custodial mother
and, consequently, the child and mother gain too. Notice that for δ = 1, the
father would like to have a child but the mother prefers not to have a child
if all other couples do not have a child. This conflict could, in principle, be
resolved by ex ante contracting at the time of marriage.
11.8 Further issues

In this concluding section we discuss some departures from the standard
contract that we analyzed and examine their implications.
11.8.1 The custody assignment

If the father has the custody, he will spend only the minimal amount of
time on child care because, under our assumption that wf (1 + α) > β, any
hour spent on child care could be better used in the market. He will spend
on the child good c∗ and spend the remainder of his income on the adult
good. This is true whether or not he receives transfers from the mother.
Therefore, the mother has no incentive to transfer to the custodial father,
as any additional dollar is spent on the adult good and α < 1. If all couples
choose father custody then in a remarried couple, the new wife will work
in the market, because she cannot spend time on her child who lives in a
different household. The expected utilities of the family members are then
Vm (p) = α(wf − c∗ ) + g(c∗ ) + pαwm + wm + p(wf − c∗ ),

Vc (p) = α(wf − c∗ ) + g(c∗ ) + pαwm ,
Vf (p) = (1 + α)(wf − c∗ ) + g(c∗ ) + p(1 + α)wm . (11.43)
We see that under father’s custody, the child receives less time but con-
sumes more of the adult and child goods. Thus, the justification for the
prevalence of mother custody must rest on the assumption that, in the
case of children, time is more important then money, that is γ is large
relative to α(wf − c∗ ). For a small remarriage probability, the condition
γ > α(wf − c∗ ) is sufficient to ensure that mother custody is better for
the child. In this case, the father also prefers that the child will be with
the mother, because for a small p his expense on child support, s∗ (p), is
about the same as he would spend himself on the child, c∗ , and the po-
tential gain from the mother contribution of time exceeds the gains that
the father has from sharing adult consumption with the child. However, if
γ < α(wf − c∗ ) + wm then, for a small p, the mother would prefer that
the father will have the custody, because this would free her to earn some
extra money in the labor market. In this case, the child is a "hot potato"
that each parent prefers that the other will take care of it. This reflects,
of course, the potential for free riding that exists in the provision of public
goods. Thus, γ must exceed α(wf − c∗ ) + wm for the two parties to agree
on mother custody.25
An increase in the probability of remarriage p decreases the welfare of the
child under the mother’s custody but raises it under the father’s custody.
This difference is caused by the shift of the custodial mother towards market
work when she remarries. The custodial father works at the same intensity
whether he is married or not and the child gains from the added adult
consumption when the father remarries. Therefore, for a large probability
of remarriage father’s custody becomes more attractive and a larger gap
between γ and α(wf − c∗ ) is required to justify mother’s custody under
the voluntary commitments discussed so far. A possible resolution is to
mandate (and enforce) some minimal child support transfer from the non
custodial father to the custodial mother.
11.8.2 Mandated and contingent contracts

The courts often consider the "accustomed standard of living" of the par-
ties as a standard for divorce settlements. Because living alone is more
costly then living together and there is always a risk of remaining single,
it is impossible to restore the same standard of living for all parties. The
problem is exacerbated by the principal-agent issues emphasized here that
imply a level of transfers that is insufficient to restore efficiency. We now
discuss some alternative contracting options that may restore efficiency.
The law also singles out children as worthy of special consideration in
divorce settlements. This concern is justified because, as we have seen,
even if parents are altruistic and internalize the welfare of the child, the
child as a passive party can be hurt by the divorce. It is then natural to
apply the accustomed standard of living principle only to the child, as a
constraint on the parents’ contracting choices. For instance, the law may
mandate a level of child support s = c∗ . As we have shown, such a transfer
would indeed induce a single mother to choose the efficient level of child
care, spending all her time on child care. This, however, is not true for
2 5 For alternative models of custody assignment see Atteneder and Halla (2007) and
Rasul (2006).
a remarried mother, who still may be forced to work part time to comply
with the interest of her new husband. As long as the courts cannot interfere
with within-household allocations (that are hard to verify) and the father
cannot transfer directly to the child, because his money transfer is fungible
and can be consumed by the mother and her new husband, it is hard to
expect that the child interests will be maintained simply by mandating a
money transfer. There is, however, one notable exception. If the mother has
sufficient bargaining power to take all the gains from marriage, then she
would solve the mirror image of problem (11.13) and maximize her utility
subject to the constraint that the new husband is just indifferent between
remaining single and remarriage. Formally, this problem is the same as
problem (11.7) that the mother solves as single and setting s = c∗ would
indeed induce her to maintain the efficient outcome when she remarries.26
This brief discussion illustrates that in search markets with rents that
are subject to bargaining, it is important to specify the relative bargaining
power of the parties that determines the share of the surplus that each party
gets. In the Nash bargaining model this is determined by considerations
such as impatience and risk aversion that, of course, need not be equal
across genders. More broadly, social norms such as egalitarianism and sex
roles may also affect the bargaining outcome.
Another, and potentially more fruitful, direction is to enlarge the set of
contracts that the courts are willing to enforce. In principle, child support
payments should depend on the marital status of both parents, because
the costs and benefits of post divorce transfers depend on these states.27
In practice, child support is not contingent on marital status but there are
other payments such as alimony that are often contingent on the marital
status of the mother. Because we assume that all transfers are fungible,
the name attached to these payments does not really matter, but it does
matter how flexible they are and to what contingences they respond.
Now imagine that a father can pay different amounts to the custodial
mother depending upon whether or not she is remarried. Suppose further
that the father is forced by law to pay a fixed amount of child support
s = c∗ but can augment it by an additional payment σ that he pays the
custodial mother only if she is single. Then, the efficient allocation within
the remarried household is determined by
max E(a|σ, σ 0 ) = wm hm + wf − c − (1 − p)σ 0 (11.44)

hm ,c
2 6 Aiyagari et al. (2000) also discuss mandatory child support payments in a general
equilibrium framework. They show that an increase in such payments raises welfare of
parents and children.
2 7 In theory, the transfers should depend on the marital status of all agents that
participate in the marriage market. But this, of course, is highly impractical.

subject to the constraints:
(1 + α)[wm hm + wf − c − (1 − p)σ 0 ] + γ(1 − hm ) + g(c) ≥ um (c∗ + σ)
and 0 ≤ hm ≤ 1. The chosen values of hm and c depend on the transfer that

the father promises the mother if she remains single, σ, and the expected
value of the new husband’s gross income wf −(1−p)σ0 . Only the expectation
matters because the remarried partners are risk neutral with respect to a
and because the mother’s work time, hm , and the expenditures on the child
good, c, are determined before the marital status of the ex-wife of the new
husband is known.
In contrast to a non contingent transfer, a transfer given to the mother
only when she is single does not change the utility frontier of the remarried
couple and therefore must reduce the expected utility of the new husband.
This implies that with contingent payments, the father is able to attain a
larger impact on the child’s utility and is willing to contribute more to the
custodial mother. In fact, by setting σ = wf − c∗ − (1 − p)σ 0 the father
can eliminate all the gains from marriage of the new husband and restore
efficiency. We then obtain the following characterization (see Appendix).
Proposition 11.4 If all couples have children, then the commitment equi-
librium for a given remarriage probability p, is such that: For p < p0 , the
only symmetric equilibrium is one in which all fathers pay only the manda-
tory payment s = c∗ and σ = 0. For p > p1 , the only symmetric equilib-
rium is one in which all fathers voluntarily commits to pay their ex-wife
wf −c∗
σ = 2−p if she remains single. For p1 ≥ p ≥ p0 , both types of equilibrium
wf −c∗
can arise. The equilibrium σ = 2−p is efficient and
wf − c∗
Vc (p) = γ + g(c∗ ) + α ,
2−p
wf − c∗
Vm (p) = Vf (p) = γ + g(c∗ ) + (1 + α) . (11.45)
2−p
The pattern described in the proposition suggests reinforcement; one is

willing to commit to his wife if others do, but not if they do not. As is
well known, such positive feed backs can yield multiple equilibria. We also
see that higher probability of remarriage is conducive to equilibria in which
fathers are willing to commit on a payment that is conditioned on the event
that the mother remains single, because such promises are carried out less
often and are more likely to yield benefits.
When efficiency is restored, the child suffers only from the reduced adult
consumption that is caused by the risk of remaining single. If the mother
is sure to remarry, that is, p = 1, then the child is as well off as in an intact
family. That is, the father was practically replaced by the new husband with
no harm to the child. This favorable outcome was achieved by eliminating
the marital surplus completely and effectively eliminating the power of the
new husband from extracting any rents which may harm the child upon
remarriage. Importantly, the contingent transfer restores ex-post symme-
try between the parents, which implies that divorce will also be efficient.
Finally, because of the efficiency in child care, all family members benefit
from a higher probability of remarriage.
These results are in sharp contrast to the case of non contingent transfers
and raise the question why contingent contracts are not more prevalent.
The basic problem with such contracts is that they are not attractive when
the probability of remarriage is low, because then the father is very likely
to bear the costs, when the mother remains single, and correspondingly
unlikely to reap the benefits when she remarries. As we shall show in a
subsequent section, this problem can be mitigated if the courts would also
enforce contracts that are signed at the time of marriage. Before we turn
to that case, however, let us illustrate the impact of contingent contracts
with a numerical example.
Numerical Example 11.2
We now present an example with contingent contracts. The parameters
are the same as in numerical example 11.1, except that we now set δ = 1.
This change is made to increase the motivation of the father to support
the child and the motivation of the couple to have children, so that an
equilibrium with voluntary transfers and children can be supported.
Figure 11.8 presents three potential equilibrium points for the divorce
and remarriage rates associated with the following alternatives:
1. All couples have children and the father pays the mother a fixed
wf −c∗
payment c∗ and, in addition, a contingent payment σ = 2−p that
the mother receives only if she remains single.
2. All couples have children and the father pays the mother only the
fixed payment c∗ .
3. All couples have no children and no transfers are made upon divorce.
Case 3 is identical to the one discussed in the previous section. Case 2 is a

modification of the case discussed in the previous section; the child support
payment is still unconditional but fathers are forced to pay more than they
would pay voluntarily. However, because part of this payment is eaten by
the new husband of the remarried mother, the child and consequently the
father are still harmed by remarriage, which is the same qualitative result
that we had before. The main departure is case 1, where the contingent
payment given to the mother raises her bargaining power to the extent
that the child is not harmed from remarriage and, consequently, the gains
from divorce of both parents rise with the prospects of remarriage. For the
chosen parameters, this case yields an equilibrium divorce rate p = 0.35628

Figures 11.9 and 11.10 describe the impacts of deviations from the trans-
fer patterns when all parents have children. Figure 11.9 shows that if all
wf −c∗
parents commit on the transfer σ = 2−p that restores efficiency then for
p < 0.142, each father, taken separately, would be better off by unilater-
ally deviating to σ = 0. Thus, a symmetric equilibrium in which everyone
wf −c∗
commits to σ = 2−p cannot exist in this range. Figure 11.10 shows that
if all parents pay only the compulsory payment c∗ and σ = 0 then, for
p > 0.302, each father taken separately would be better off if he unilater-
ally commits to the mother to pay her all his disposable income, wf − c∗ ,
if she remains single. Thus an equilibrium where no one wishes to commit
cannot exist in this range. Proposition 11.4 states that if p1 > p > p0
there may be two partial equilibria, one in which every father commits on
a positive σ and one in which no father commits. However, for the chosen
parameters, the equilibrium divorce rate associated with having children
is above p1 = 0.302, implying that the equilibrium in which every father
commits at p = 0.356 is the only potential equilibrium when all parents
have children. Table 11.2 shows that the equilibrium at p = 0.356 is in-
deed a full equilibrium in the sense that, with the implied child support
transfers, all couples prefer to have children. For the assumed parameters,
this is the only equilibrium because, if all couples have no children, there
is an incentive to deviate to a situation with a child and full commitment
σ = wf − c∗ . Thus, the full equilibrium is unique.
Comparing the results in Table 11.2 to the last row in Table 11.1, we can
see the impact of different legal regimes when all parameters of the model
are the same. Suppose that all couples have children. Then, if transfers
are not contingent and determined optimally, the child’s expected utility
is 2.613. If fathers are forced to pay the mother a transfer of s = c∗ , the
child’s expected utility rises to 2.811 and if contingent transfers are also en-
forced, the child’s expected utility is 3.216, which is only slightly less than
the child’s utility in an intact family, 3.260. As we move across these alter-
natives, the transfer from each father to his ex-wife rises when the marriage
breaks and, consequently, the expected life time utility of each mother rises.
The surprising result, however, is that the father is also better off and his
expected utility levels are 3.456, 3.475 and 3.485, respectively. The result
that a compulsory increase in child support above the individually optimal
level, s∗ (p) raises the father’s expected utility reflects a positive contract
externality, whereby the commitment made by each father to his ex-wife
benefits other fathers when they remarry. The second increase, associated
2 8 Aninteresting point is that if the efficiency of child expenditures is restored by

appropriate transfers then the gains from divorce with children can exceed the gains
from divorce without children, despite the higher fixed costs of divorce associated with
children. The reason is that couples without children have higher joint consumption (in
terms of the adult good), which can make divorce more costly for them.
w −c∗
f
with raising the contingent payment, σ, from 0 to 2−p benefits each fa-
ther separately because of the rise in the expected utility of the child. The
rise in the remarriage prospects as p rises from 0.344 to 0.356 raises the
incentives of all fathers to contribute to their ex-wives. In this respect, a
higher aggregate divorce rate can serve as a coordination device that can
benefit children and raises the incentives to have children.
11.8.3 Transfers, the ex-ante perspective

At the interim stage, when the fertility has already been determined, the
purpose of the contract is to induce the custodial mother to spend all her
time on the child if she remarries. A contract that is signed at the time of
marriage can also influence the divorce and the father would be willing to
commit for a broader range of p. Thus, in contrast to the ex-post contract,
where the husband gains from the commitment to pay the single mother
only if she remarries, the ex ante contract can benefit the father even if the
mother remains single. Of course, the husband is willing to pay only if the
mother would not have the child in the absence of contract. It is easy to
find parameters of the model such that the mother would prefer to have a
child even without a contract, in which case there is no role for voluntary
ex-ante commitments by the husband. However, we shall focus here on the
case in which, in the absence of binding contracts, the mother does not wish
to have the child but if binding contracts are enforced then the mother may
prefer to have a child, depending on the expected remarriage rate and the
decision of others to have children.
Suppose that all couples have children and sign an ex-ante contract, at
wf −c∗
the time of marriage, that promises the mother σ = 2−p if the mother
remains single. Then, the gains from divorce are the same for the two
partners and separations are efficient for all p and, therefore, the expected
utility at the time of marriage is the same for the husband and wife. Since
under such contract both partners want the child at the same values of p,
the production of children is efficient too. Both partners would agree to
sign such a contract at the time of marriage, if their expected utility is
higher than it would be in the absence of contract and no children. Finally,
the ex-ante contract is renegotiation proof, because it coincides with the
interim contract.
It is puzzling why ex-ante contracts that are signed at the time of mar-
riage are not prevalent among all couples with children. The implemen-
tation of such contracts in earlier times suggests that the enforcement of
ex-ante contracts is not the issue. Rather, in modern societies with free
marriage, based in part on mutual attraction, the general sense is that
emotional commitments are more important than legal agreements and
thinking of contingencies and writing them down may "kill love". However,
prenuptial contracts are often signed, at the time of marriage, by couples
in which one (or both) of the partners bring into the marriage substantial
property, which is more common on second marriages.
11.9 Conclusion
As the last two chapters illustrate, marriage markets with search frictions,
in which the meeting technology displays increasing returns, may have mul-
tiple equilibria, because of the various search and contracting externalities.
In chapter 10, we did not allow any contracting and, as a consequence,
obtained the result that equilibria with higher turnover, that is, higher
divorce and remarriage rates, provides all participants with a higher wel-
fare. The reason is that an increase in the aggregate divorce rate, raises the
prospects of remarriage, which makes it easier to replace bad marriages
by better ones. In chapter 11, we allowed parents to transfer resources
in the aftermath of divorce, based on the insight that, in the presence
of children, marriage dissolution does not eliminate all ties between the
partners because both parents continue to care about their child, which
motivates post divorce transfers.29 However, the impact of transfers on the
marriage market and the welfare of children is quite complex, because the
willingness of each parent to transfer to his\her ex-spouse depends on the
transfers that potential mates for remarriages expect from their ex-spouses.
This contract externality can operate in different ways, depending on the
type of contracts that are enforced by law. If only unconditional transfers
are enforced, higher divorce and remarriage rates reduce the incentive to
transfer money to the custodial mother, because a dollar transferred to her
is less likely to reach the child than if she remarries. The consequence is
that children may be worse off in high divorce equilibria. The outcome is
completely reversed if the contracts environment is enriched and contin-
gent contracts are also enforced. If the non custodial father promises the
mother a payment that is contingent on her remaining single, then her bar-
gaining position vis-a-vis her new husband is improved and the welfare of
the child can be protected. Fathers have a stronger incentive to make such
commitments when the remarriage rate is high, because then the payments
to the custodial mothers are made relatively rarely, while the non custodial
fathers are rewarded for their commitments more often. The outcome, in
this case, is that equilibria with higher aggregate divorce (and remarriage)
can be welfare enhancing. In particular, children who would suffer from the
break of the marriage of their parents if it would happen in isolation, can
gain from being in environment in which a higher proportion of marriages
dissolve.
2 9 This is in sharp contrast to employment relationships that end in separation, in
which case the ex-partners are no longer tied with each other.
11.10 Appendix: Contingent contracts

The purpose of this appendix is to prove Proposition 11.4. We assume
throughout a mandatory child support payment of c∗ so that s = c∗ + σ
and σ ≥ 0. Hence, by (11.12), u0m (s) = 1 + α and q0(s) = α.
11.10.1 The choice of contract

The expected utility of the father is now
Vf = (1 − p)2 [wf − (c∗ + σ) + q(c∗ + σ)] + (1 − p)p[wf − c∗ + E(q|σ, σ 0 )],

+p(1 − p)[E(a|σ 0 , σ) + q(c∗ + σ)],
+p2 [E(a|σ0 , σ) + E(q|σ, σ 0 )]. (11.46)
From (11.44) we have, that in any household
∂E(a|σ, σ 0 )
= λu0m (c∗ + σ) = λ(1 + α) < 0,
∂σ
∂E(q|σ, σ 0 ) ∂E(a|σ, σ 0 )
= u0m (c∗ + σ) − = (1 + α)(1 − λ) > 0,
∂σ ∂σ
∂E(a|σ, σ 0 )
= −(1 − p)(1 − λ(1 + α)) < 0,
∂σ 0
∂E(q|σ, σ 0 ) ∂E(a|σ, σ 0 )
= − > 0, (11.47)
∂σ 0 ∂σ
where λ is the Lagrange multiplier for the participation constraint of the
wife. Therefore,
∂Vf
= (1 − p)2 [−1 + α] + (1 − p)p[(1 + α)(1 − λm )] +
∂σ
p(1 − p)[−(1 − p)(1 − λf (1 + α)) + α)] + p2 [−(1 − p)(1 − λf (1 + α))
+(1 + α)(1 − λm )], (11.48)
where λm and λf denote the Lagrange multipliers if the mother or father

remarry, respectively.
∂V
An interior solution for σ exists if ∂σf = 0 and
∂ 2 Vf ∂λm ∂λf
2
= (1 + α)p[− + (1 − p) ] < 0. (11.49)
∂σ ∂σ ∂σ
From the first order conditions to (11.13), if both the mother and the new
wm
wife of the father work part time then λm = λf = (1+α)w m −γ
. Otherwise,
1
λj = 1+α−g0 (c) for j = m, f . When the mother does not work, any increase
in σ increases the consumption of the child and decreases λm . Similarly,
when the father remarries and his wife does not work any increase in his
commitment to his ex-wife raises the consumption of the step child and
∂λf
decreases λf . Hence, ∂λ
∂σ ≤ 0 and ∂σ ≤ 0. But in a symmetric equilibrium
m
∂λf
all couples make the same choices and ∂λ
∂σ = ∂σ , which would imply that
m
∂ 2 Vf
∂σ 2 ≥ 0. Therefore, there is no interior symmetric equilibrium, and the
only symmetric equilibria are such that all couples must be at one of the
boundaries, σ = 0 or σ = wf −c∗ −(1−p)σ 0 . Notice that the upper boundary
is not determined by the budget constraint but by the requirement that
the mother is just indifferent between remarriage and remaining single.
wf −c∗
Suppose that all other couples set σ 0 = 2−p . Then by setting
wf − c∗
σ = wf − c∗ − (1 − p)σ 0 = (11.50)
2−p
the father can guarantee that the mother is just indifferent between mar-
riage and remaining single. This is seen by noting that the mother partici-
pation constraint becomes
wf − c∗ ∗ wf − c∗
(1+α)[wm hm + +c −c]+γ(1−hm )+g(c) ≥ γ+g(c∗ )+(1+α)
2−p 2−p
(11.51)
But
M ax{(1 + α)[wm hm + c∗ − c] + γ(1 − hm ) + g(c)}

c,hm
≤ M ax{γhm + (1 + α)(c∗ − c) + γ(1 − hm ) + g(c)}
c,hm
= γ + g(c∗ ). (11.52)
Therefore, (11.51) must hold as equality.

wf −c∗
It remains to show that σ = 2−p is indeed an optimal choice, given that
w −c∗ ∂V
others maintain σ0 = 2−p
f
. For a marginally lower σ we have that ∂σf
1 ∗
approaches ∞ because λm = 1+α−g 0 (c) approaches −∞ as c approaches c .
∂V
For marginally higher σ we have that ∂σf approaches −∞, because when
the mother chooses not to marry the father suffers a discrete loss since he
wf −c∗
pays the mother σ with certainty. Thus, σ = 2−p is a local maximum.
However, it need not be a global maximum. In particular, for a small p, it
∂V
is always the case that ∂σf < 0, because the father bears the costs with
high probability and the benefits with a low probability, so that σ = 0 is
wf −c∗
also a local maximum. However, in contrast to the selection of σ = 2−p ,
the selection of σ = 0 is a local maximum only if p is small. The difference
arises because at low levels of σ, c < c∗ so that λm and λf are finite and
∂Vf
∂σ must change sign from negative to positive as p rises from zero to one.
We have noted in the text that if remarried mothers work part time
∂V
then ∂σf is independent of σ 0 . However, if the mother does not work or
1 0
works full time so that λm = 1+α−g 0 (c) , then an increase in σ will raise c,
∂V
λm becomes more negative and ∂σf rises. That is, σ and σ 0 are strategic
complements.
The characterization in the text follows from the following observations:
For any fixed σ 0 , the global maximum is at σ = 0 if p is sufficiently small,
wf −c∗
say less than p0 , and at σ = 2−p if p is sufficiently large, say larger than
p1 . Because of complementarity, one is more inclined to give if others do,
and therefore p1 must exceed p0 .
11.11 References
[1] Aiyagari, Rau S., Greenwood Jeremy and Nezih Guner, ’On the State
of the Union’, Journal of Political Economy, 108 (2000), 213-244.
[2] Akashi-Ronquest, N., ‘The Impact of Biological Preferences on Parental
Investments in Children and Step-Children’, Review of Economics of the
Household, 7 (2009), 59—81.
[4] Argys, Laura M., Peters, H. Elizabeth, Brooks-Gunn, Jeanne and Ju-
dith R. Smith, ‘The Impact of Child Support on Cognitive Outcomes of
Young Children’, Demography, 35 (1998), 159-173.
[5] Atteneder, Christine and Martin Halla, ‘Bargaining at Divorce: The
Allocation of Custody’, IZA Discussion Paper No. 2544, (2007).
[6] Becker, Gary Stanley, A Treatise on the Family, enlarged edition,
(Harvard University Press, 1993).
[7] Bjorklund, Andres and Marianne Sundstrom, ‘Parental Separation and
Children’s Educational Attainment: A Siblings Analysis on Swedish Reg-
ister Data’, Economica, 73 (2006), 605-624.
[8] Cancian, Maria and Daniel R. Meyer, ‘Who Gets Custody?’, Demog-
raphy, 35 (1998), 147-157.
[9] Case, Anne C., Lin, I-Fen and Sarah S. McLanahan, ‘Household Re-
source Allocation in Stepfamilies: Darwin Reflects on the Plight of Cin-
derella’, American Economic Review, 89 (1999), 234-238.
[10] Case, Anne C., Lin, I-Fen and Sarah S. McLanahan, ‘Explaining
Trends in Child Support: Economic, Demographic, and Policy Effects’,
Demography, 40 (2003), 171-189.
[11] Chiappori, Pierre-André and Yoram Weiss, ‘Divorce, Remarriage and
Economics, 12 (2008), 828-844.
[13] Del-Boca, Daniela, ‘Mothers, Fathers and Children after Divorce: The
Role of Institutions’, Journal of Population Economics, 16 (2003),
399-422.
[14] Diamond, Peter A. and Eric Maskin, ‘An Equilibrium Analysis of
Search and Breach of Contract, I: Steady States’, The Bell Journal
of Economics, 10 (1979), 282-316.
[15] Gruber, Jonathan, ‘Is Making Divorce Easier Bad for Children? The
Long-Run Implications of Unilateral Divorce’, Journal of Labor Eco-
nomics, 22 (2004), 799-833.
[16] Halla, Martin, ‘The Effect of Joint Custody on Marriage and Divorce’,
IZA Discussion Paper No. 4314, (2009).
[17] Hetherington, Mavis E. and Margaret M. Stanley-Hagan, ‘Stepfami-
lies’, in Lamb, Michael E. (eds.), Parenting and children develop-
ment in non-traditional families, (Mahwah, New Jersey: Lawrence
Erlbaum Associates, 1999).
[18] Lamb, Michael E., Kathleen J. Sternberg and Ross A. Thompson, ‘The
Effects of Divorce and Custody Arrangements on Children’s Behavior,
Development and Adjustment’, in Lamb, Michael E. (eds.), Parenting
and children development in non-traditional families, (Mahwah,
New Jersey: Lawrence Erlbaum Associates, 1999).
[19] Laumann, Edward O., Gagnon, John H., Michael, Robert T. and Stu-
art Michaels, The Social Organization of Sexuality: Sexual Prac-
tices in the US, (Chicago: The University of Chicago Press, 1994).
[20] Mnookin, Robert H. and Lewis Kornhauser, ‘Bargaining in the Shadow
of the Law: The Case of Divorce’, The Yale Law Journal, 88 (1979),
950-997.
[21] Mortensen, Dale T., ‘Matching: Finding a Partner for Life or Other-
wise’, American Journal of Sociology, 94 (1988), supplement, s215-
s240.
[22] Picketty, Thomas, ‘The Impact of Divorce on School Performance:
Evidence from France, 1968-2002’, Centre of Economic Policy Research,
Discussion Paper No. 4146 (2003).
[23] Rasul, Imran, ‘The Economics of Child Custody’, Economica, 73
(2006), 1—25.
[24] Seitz, Shannon N., ‘Labor Supply, Divorce and Remarriage’, University
of Western Ontario, Department of Economics, Working Paper No. 9902,
(1999).
[25] Stafford, Frank and Wei-Jun Jean Yeung, ‘The Distribution of Chil-
dren’s Developmental Resources’, in Hamermesh, Daniel. S. (eds.) and
Pfann, Gerard Antonie (eds.), The Economics of Time Use, first
edition (Elsevier Science, 2005).
[26] Weiss, Yoram and Robert J. Willis, ‘Children as Collective Goods and
Divorce Settlements’, Journal of Labor Economics, 3 (1985), 268-
292.
ples: Evidence and Interpretation’, Journal of Labor Economics, 11
(1993), 629-679.
Table 11.1: Impact of Change in Parameter on the Equilibrium

Divorce (Remarriage) Rate and Life Time Utilities of Family
Members
Part 1: Benchmark Parameters
Husband’s wage wf = 1
Wife’s wage wm = .5
Mother’s productivity in child care γ = 1.75
Father’s productivity in child care β=0
Child’s marginal utility from the adult good α = .25
Distribution of shocks θ˜U [−d, d], d = 2
1
Utility from child expenditures g(c) = 2.25 ∗ (1 − (1+c)2.25 )
Child expenditure levels c = 0.5378, ĉ = 0.1203, g(c∗ ) = 1.4

∗
Proximity factor of father’s utility from child δ = .75

Discount factor ∆ = .625
Fixed cost of separation without a child b0 = 0
Fixed cost of separation with a child b1 = .35
Father’s second period utility γ + g(c∗ ) + (1 + α)(wf − c∗ ) = 3.72
Mother’s second period utility γ + g(c∗ ) + (1 + α)(wf − c∗ ) = 3.72
Child’s second period utility γ + g(c∗ ) + α(wf − c∗ ) = 3.26
Part 2: Change in the Variability of Shock, d

Part 3: Change in the Fixed Cost of Separation with a Child,

b1
Part 4: Change in the Proximity Factor, δ

Table 11.2: The Incentives to Deviate at Alternative Potential

Equilibria
wf −c∗
Equilibrium with children, σ = 2−p
Equilibrium with children, σ = 0
Equilibrium without children
Parameters are the same as in the benchmark of Table 11.1, except that
δ = 1, instead of δ = .75.
FIGURE 11.1. Effect of Transfer on Mother’s Work, Child’s Consumption and

Child’s Utility
FIGURE 11.2. The Pareto Utility Frontier

FIGURE 11.3. Optimal Transfers and Child’s Utility if the Mother Remmaries
or Remains Single
FIGURE 11.4. Expected Utilities of Father, Mother and Child at the Time of
Divorce
FIGURE 11.5. Potential Marriage Market Equilibria, with and without Children
FIGURE 11.6. The Impact of Deviation to not having a Child on the Expected
Life Time Utility when all other Couples have a Child
FIGURE 11.7. The Impact of Deviation to Having a Child on the Expected Life
Time Utility when all other Couples Have no Child
FIGURE 11.8. with Children under Different Payment Schemes and without
Children
FIGURE 11.9. The Impact of Deviation to a Fixed Payment when all other
Fathers give a Contingent Payment
FIGURE 11.10. The Impact of Deviation to a Contingent Payment when all other
Fathers give no Contingent Payment

Economics of Family

Uploaded by

Copyright:

Available Formats

Economics of Family

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

Economics of Family

Uploaded by

Copyright:

Available Formats

This is page i

Printer: Opaque this

Martin Browning Pierre-André Chiappori

I Models of Household Behavior 11

2 The gains from marriage 81

3 Preferences and decision making 103

3.2 Household production . . . . . . . . . . . . . . . . . . . . . 107

4 The collective model: a formal analysis 157

4.5.1 Lindahl prices . . . . . . . . . . . . . . . . . . . . . . 177

5 Empirical issues for the collective model 199

6 Uncertainty and Dynamics in the Collective model 243

6.4 Intertemporal Behavior . . . . . . . . . . . . . . . . . . . . 266

II Equilibrium Models of the Marriage Market 291

8 Sharing the gains from marriage 333

8.2.6 Matching on preferences: Roe vs. Wade and female

9 Investment in Schooling and the Marriage Market 391

10 An equilibrium model of marriage, fertility and divorce 435

11 Marriage, Divorce, Children 459

2.1 Bounds for the relative cost of equivalent bundles . . . . . . 89

5.1 Distribution factors . . . . . . . . . . . . . . . . . . . . . . . 224

1.23 Age Pyramid , US 2005. Source: US Census. . . . . . . . . . 64

2.1 Preferences over two public goods . . . . . . . . . . . . . . . 84

3.1 Demand for public good . . . . . . . . . . . . . . . . . . . . 150

4.1 Collective price responses . . . . . . . . . . . . . . . . . . . 196

4.2 A potentially compensating variation. . . . . . . . . . . . . 196

5.1 A Shift in the Sharing Rule . . . . . . . . . . . . . . . . . . 221

6.1 Game tree for investment in education . . . . . . . . . . . . 250

7.1 Positive Assortative Mating . . . . . . . . . . . . . . . . . . 310

8.1 Imputations when z12 + z21 > z22 + z11 . . . . . . . . . . . 335

9.1 Regions for Marriage and Investment . . . . . . . . . . . . 429

9.3 Mixed Equilibrium with More Educated Men than Educated

10.1 Expected utility profiles. . . . . . . . . . . . . . . . . . . . . 455

11.1 Eﬀect of Transfer on Mother’s Work, Child’s Consumption

two distinct solution concepts; one is a non-cooperative self-enforcing out-

motivate assortative matching by inherent preferences to marry someone

Becker, Gary, ‘A Theory of Marriage: Part 1’, Journal of Political

Lewis, Susan K. and Valerie K. Oppenheimer, ‘Educational Assorta-

1.1 Marriage, divorce and remarriage

1.1.2 Marital histories

1 Unfortunately it is not possible to extend the series beyond 1988.

sumption goods within households. One possible reason is that technolog-

1.2 Marriage, work and wages

by the number of all individuals, including those who do not work.

1.2.3 The marriage premium

1.3 Who Marries Whom?

than 10 hours a week.

the correlation due to age and schooling.

7 Completed fertility is defined as the mean number of children born to women of

1.4.2 Children under diﬀerent household arrangements

1.5 Saving and life stages

riage using UK expenditure data.

consistent coding throughout the period.

ferent (plausible) weights or scale factors gives similar results.

1.7 Tables and Figures

Year Never married Married Divorced Widowed

Age group USA Denmark

Single head Married couples Cohabiting couples

Single head Married couples Cohabiting couples

Sex,Age NumberofMarriages Divorced Widowed

Birth cohort 1931-36