Price Discrimination∗

Simon P. Anderson†and Régis Renault‡

This version August 2008.
keywords arbitrage, nonlinear pricing, discriminatory
pricing, Ramsey-Boiteux pricing, marginal cost of
public funds

1 Introduction

Whenever we take a trip by train or by plane, we are often well aware that the

price we paid was quite different from that paid by our fellow passengers with

whom we are sharing the carriage or cabin. We can bemoan this situation, if

we booked late and do not qualify for an age discount and our ticket was one of

the more expensive ones or perhaps be pleased at having gotten a good price.

The different prices are illustrations of what economists call discriminatory pric-

ing. This seems to be a textbook case where a service which is identical (same

journey, same date, same time, same comfort class) is sold at different prices1 .
∗ We gratefully acknowledge travel funding from the CNRS and NSF under grants INT-

9815703 and GA10273, and research funding under grant SES 0452864 . We thank Anita
Anderson, Alain Béraud, Catherine de Fontenay, Robin Lindsey, André de Palma, Emile
Quinet, and Sarah Tulman for their comments and suggestions. We would also like to thank
Melbourne Business School and the Autoridade da Concorrencia in Lisbon for their hospitality.

† Department of Economics, University of Virginia, PO Box 400182, Charlottesville VA

22904-4128, USA. sa9w@virginia.edu

‡ ThEMA, Université de Cergy-Pontoise, 33 Bd. du Port, 95011, Cergy Cedex, FRANCE

and Institut Universitaire de France. regis.renault@u-cergy.fr

1 Price discrimination is also common in other transportation services in addition to rail

and air travel. Odlyzko (2004) provides examples from maritime transport, inland waterways,

1
 Looking closer, it is a little oversimpliÞed to claim that all travelers have

actually received the same level of service. Less expensive tickets are often

associated with numerous restrictions which clearly indicate a lower level of

service. It is often necessary to buy the ticket a long time in advance, with

restrictive conditions on cancellation and reimbursement. The traveler can then

enjoy the same service as she would have had if she had paid full price — unless,

of course, her plans change at the last minute. However, to get the cheap fare

she has to accept some risk, had she been obliged to change or cancel her ticket,

or indeed she might have had to put up with some inconvenience due to having

not changed her ticket in order to not lose money. There are also reduced fares

for those satisfying certain “demographic” considerations. Senior citizens and

children often pay lower fares.

On May 5, for May 18, 2007 the prices for one-way second class travel be-

tween Charleroi Sud (Belgium) and Paris Nord (France), proposed on the inter-

net site www.thalys.com were as follows: Librys 59 E; Mezzo 44.5 E; Mezzo+

32.5 E; and Smilys 20.5 E. The rates for Mezzo+, Mezzo, and Smilys can

be reserved only if sufficient seats are still available, and only if one buys a

round trip ticket. It is less expensive to buy a round-trip Smilys ticket than

to buy a one-way Librys ticket. However, Smilys can only be bought when

reserving two weeks before the departure date and it is the only fare which is

non-exchangeable and completely non-reimbursable. Mezzo and Mezzo+ are

reimbursable up to Þfty percent of the price of the ticket, up to the departure

and turnpikes, as well as railroads, to argue that price discrimination has been (and is still)
prevalent in the development of these sectors.

2
date. The only difference between the two tickets is that the quota for Mezzo+

is sold out quicker than for Mezzo. It is also possible to get a Lys fare for 26

E if one has a membership in the program. This is an example of a two-part

tariff.

For these examples, the menu of prices that are proposed must take into

account the fact that travelers can personally arbitrage (or choose between) the

different options. There are also fare categories which are not subject to such

arbitrage, such as: Kid (15 E), for children under 12 years old; Kid & Co. (29.5

E), for adults accompanying a child under 12; Youth (29.5 E) for travelers under

26; and Senior (41.5 E) for travelers over 59. All of these classes, except for

Smilys, are also available in Þrst class — Librys, for example, costs 100 E.

Price discrimination arises when a Þrm sells different units of the same good

at different prices. This applies perfectly to cases in which certain groups

of customers beneÞt from special tariffs (i.e. students or senior citizens), or

nonlinear pricing where the price per unit depends on the number of units

brought (such as the price of a French Metro ticket when bought singly or in a

pack of ten).2

Nevertheless, several practices that involve selling different services can be

viewed as discriminatory. In such cases, price differences might also be explained

by cost differences without necessarily invoking a discriminatory motive (see,

for example, Lott and Roberts, 1991). To address this concern, some writers
2 Price discrimination is often also based on time of travel, such as evening or summer tariffs.

The simplest analysis assumes that the demands in each period are independent (see also the
analysis of peak-load pricing in the Backhaul section below). Gerstner (1986) formulates a
peak-load pricing model in which Þrms account for intertemporal demand shifting.

3
have proposed deÞnitions based on the comparison of price differences relative

to cost differences. Stigler (1987) proposed comparing the ratio of the prices

of two services with the ratio of their marginal production costs. By this

criterion, a situation is discriminatory if the two ratios are unequal. Phlips

(1983) on the other hand proposes comparing absolute differences. Then prices

are discriminatory if the difference in marginal costs is not equal to the difference

in prices.

It is difficult to Þnd a decisive argument for one deÞnition over the other.3

Both deÞnitions indicate that prices can be discriminatory even if price differ-

ences are small, just as it can be discriminatory if price differences are large.

Suppose an airline brings passengers to a Parisian airport from which its in-

ternational ßights leave, and has everyone pay the same price for a ßight to

New York. This pricing discriminates against travelers living near Paris (see

Tirole, 1988, 1993, for a similar example).4 The deÞnitions do not say whether

such discrimination harms economic efficiency: the airline’s pricing scheme al-

lows it to more effectively exploit its market power by bringing its transatlantic

travelers to Paris.

A Þrm with some market power and proposing different services can set its

prices to get the greatest proÞt, and its prices will not bear any simple relation

(absolute or relative) to marginal costs. Exercising market power is channelled

through the ability to price above marginal cost. Offering diverse services

can be seen as a way of discriminating insofar as it allows the Þrm to adjust

3 See Clerides (2004) for a deÞnition in which there is no discrimination if the pricing

structure is not subject to arbitrage by buyers.

4 For further discussion, see Section 3.5.

4
the service proposed and its pricing to a demand that differs from customer to

customer.5

Just as it is difficult to deÞne discriminatory pricing, it is not easy to classify

different discriminatory practices. The classic reference is Pigou (1938), who

distinguishes between three possible degrees of discrimination, depending on the

ability of the Þrm to distinguish between buyers who are prepared to pay a higher

price and those inclined to pay less. Pigou deÞnes Þrst-degree discrimination

as when consumers pay their maximal willingness to pay for each unit. This is

also called perfect price discrimination.

Pigou recognized that this Þrst form of price discrimination might not have

great practical relevance. He notes that the Þrm is better able to segment the

market between different groups of buyers who have different demands. Ideally,

the Þrm would like to segment the market into groups with similar willingness

to pay; such segments could be ranked from highest to lowest willingness to pay.

Such idealized segmentation constitutes second-degree price discrimination.

As Pigou notes, however, in practice a Þrm can only imperfectly compartmen-

talize consumers according to their willingness to pay. The Þrm must use

characteristics which it can directly monitor, such as the type of good that is

being transported (for example, livestock or pig iron6 ) for a railroad that trans-

ports freight, or the location of the buyer. This latter practice is third-degree

price discrimination.

Pigou’s classiÞcation underlines the fact that discriminatory pricing is meant

5 In this paper, we emphasize demand drivers and take marginal cost as constant. The
proper attribution of costs is a complex problem in itself.
6 See Leadbelly’s “Rock Island Line”.

5
for the Þrm to exercise its market power as well as possible. Rather than

proposing a reference case corresponding to no discrimination, as did Stigler

and Phlips, Pigou deÞned the benchmark where the ultimate objective of dis-

crimination is attained, that is, where each unit is sold at the highest possible

price.

From this viewpoint, Þrst-degree price discrimination is the theoretical bench-

mark. We will see, nonetheless, that third-degree discrimination is very relevant

to the understanding of the discriminatory practices which are based on directly

observable characteristics such as age, departure location, and time and date of

the journey.

On the other hand, second-degree discrimination only seems to be a par-

ticular case of third-degree discrimination which, as we will see, corresponds

to a situation which is particularly favorable for the Þrm, where the veriÞable

information that allows the Þrm to discriminate is perfectly correlated with

willingness to pay (for example, if older people are willing to pay more).

Nevertheless, numerous types of discriminatory pricing such as nonlinear

pricing or offering different comfort classes within a train or plane do not rely

on a veriÞable criterion and allow the user to choose her preferred option. Such

practices therefore cannot be viewed as third-degree price discrimination, and

economists have gotten into the habit of calling such practices second-degree

price discrimination (see Phlips, 1983, Tirole, 1988, 1993, chapter 3, Varian,

1989, Mougeot and Naegelen, 1994). This term is therefore currently used to

cover practices that are quite different from those originally envisaged by Pigou.

6
 The discussion above suggests that discriminatory pricing is tightly tied to

the exercise of market power.7 Under perfect competition, Þrms are constrained

to sell their output at the price that is imposed by the market. It is then

obviously impossible to sell different units of a good at different prices, or to

try to affect prices by proposing a range of different services (at least as long

as such services are sold in a competitive market). Even if the Þrm has some

market power, its ability to discriminate between different buyers can be undone

or mitigated by the buyers’ ability to arbitrage between the different options

proposed.8

This arbitrage can take two forms, depending on whether one or several

buyers are involved. If it is possible to transfer the good, buyers can exchange

the good or service between themselves and the Þrm cannot charge different

prices, because those buyers who beneÞt from the lowest price will be able to

buy in order to resell to those who would otherwise have to pay more. For

example, if someone has a membership card which allows her to obtain her

tickets at a cheaper price, she could buy a large number of them and resell them

to those without a membership card. Similarly, if a low fare is offered under

the condition that the ticket should be bought sufficiently far in advance, then

entrepreneurial individuals could buy a large quantity of these cheap tickets

in order to sell them just before the date when the tickets are valid. Even

though this type of arbitrage can be limited by transactions costs, it nonetheless

7 Nonetheless, McAfee, Mialon, and Mialon (2006) propose a simple model in which they

show that the extent of price discrimination has no theoretical connection to the extent of
market power.
8 One other limit on the ability to price discriminate is that Þrms cannot effectively propose

a price menu that is too complex. See Levinson and Odlyzko (2007) for a recent treatment.

7
represents a signiÞcant constraint on Þrms’ pricing strategies, so much so that

Þrms often put in place several techniques to stop it. They often require one

to present the membership card during the trip, or, for airlines, present a piece

of identiÞcation which has on it the name matching that of the ticket holder.

This type of rule also allows the Þrm to circumvent the second type of arbitrage,

in which one customer with several options does not choose the one which was

designed for her. For example, if a Þrm wants to discriminate on the basis

of age, by presenting a driver’s license the customer veriÞes that she is paying

the price that she should. When buyers can practice such arbitrage between

the different pricing options that are offered, Pigou (1938) says that demand is

transferrable.

In practice, Þrms can often discriminate without explicitly forbidding arbi-

trage. The Þrm then has to explicitly worry about potential arbitrage when it

is setting up a discriminatory tariff structure. While the Þrm cannot force its

customers to not arbitrage, it has to set up the right incentives in its pricing

plan.

The analysis of arbitrage behavior across consumers is relatively compli-

cated, and most of the literature on discriminatory pricing simply supposes

that transactions costs are high enough to render it impossible. We, too, will

implicitly invoke this assumption throughout the paper.9 On the other hand,

so-called personal arbitrage— by which a user can choose an option which is not

intended for her purchase— has been the subject of numerous studies, especially
9 See Alger (1999) for an analysis of the constraints that are imposed by the potential of

arbitrage involving several buyers.

8
over the last three decades.

Section 2 presents a certain number of base concepts and general princi-

ples which are needed to understand the rest of the paper.10 In Section 3

we will show how a Þrm can exploit information about customer demand in

order to discriminate, by supposing that buyers do not indulge in personal arbi-

trage. We will distinguish between perfect discrimination (3.1), discrimination

between several categories of buyers who are purchasing the same good (3.2),

and market segmentation where the Þrm offers different services for which it can

perfectly determine the buyers who are prone to buying each service. Section

4 looks at strategies which allow the Þrm to motivate buyers to not engage in

personal arbitrage when the Þrm cannot stop it by using veriÞable information.

We Þrst consider using nonlinear tariffs (4.1 and 4.2), and then the possibility

of discriminating by offering different qualities (4.3). In Section 5, we con-

clude and also discuss the link between discriminatory pricing and competition.

This discussion allows us to evaluate the robustness of the results obtained for

monopolies.

2 Preliminary results and basic concepts

2.1 Uniform monopoly pricing

We start with the simplest pricing structure, uniform pricing, where all units of

the same good are sold at the same price, such as a given journey where there

is only one comfort class and all travelers pay the same price.

Let D (p) be the quantity demanded at the price p. The simplest inter-
1 0 For a simple and broader treatment of this subject, see Varian (2000)

9
pretation is to suppose that D describes the distribution of willingness to pay

over different travelers. Each traveler only wants a single trip, and D (p) then

indicates the number of travelers willing to pay at least p.

It is also useful to deÞne the inverse demand for each quantity q. This is

the maximum price at which this number of trips can be sold, P (q). When

each traveler only wants a single trip, P (q) is the willingness to pay of the

marginal consumer: the individual who would not travel if the price were slightly

higher. In order to simplify the analysis, suppose that the marginal cost of

production is constant at rate c per unit, which is also therefore the average

variable production cost (and is the cost generated by each traveler). Under

perfect competition, the price would be given by this marginal cost, and the

number of travelers would be D (c). We assume that services are provided by

a private monopolistic Þrm, whose objective is to maximize its proÞt. We start

with the case where demand is linear,

a
q = D (p) = a − bp, a > 0, b > 0, and > c.
b

The latter condition ensures that it is optimal to produce a strictly positive

quantity. This situation is illustrated in Figure 1. The Þrm is constrained

under uniform pricing to choose a single point on the demand curve. If it

wants to carry q travelers, the highest uniform price that it could charge is:

a−q
p = P (q) = .
b

10
 Figure 1: Uniform Monopoly Pricing

The producer surplus, P S, which is proÞt gross of Þxed costs, is represented

by the rectangle above marginal cost (i.e. the mark-up per traveler), p − c,

multiplied by the number of travelers q.11 The optimal quantity must therefore

maximize:

∙ ¸
a−q
− c q,
b

and the Þrst order conditions for an optimal quantity can be written as:

a − 2q
= c.
b

The left hand side of this expression is marginal revenue, M R. This is a straight

line with the same price intercept as inverse demand D, but its slope is twice as

steep. It is easy to see from Figure 1 that the condition for equality between

marginal cost and marginal revenue is satisÞed for a quantity:

D (c) a − bc
qm = = ,
2 2

which is therefore half the quantity produced under perfect competition. The

uniform price chosen is therefore:

a + bc
pm = P (q m ) = ,
2b
1 1 See Anderson and Engers (2007b) for further discussion of the concept of producer surplus.

11
and this generates a producer surplus, P S, equal to:

2
(a − bc)
(pm − c) q m = ,
4b

which corresponds to the rectangle P S in Figure 2.

In the general case, the equality of marginal revenue and marginal cost al-

lows us to determine the optimal quantity.12 When we describe discriminatory

pricing, we will often use Þrst order conditions for prices (rather than quanti-

ties). Under uniform pricing, the Þrm chooses a uniform price pm in order to

maximize D (pm ) (pm − c) . The Þrst order necessary condition for a maximum

can be written in the form:

pm − c 1 D0 (pm )
m
= m
, where η (pm ) = pm < 0,
p |η (p )| D (pm )

where η (pm ) < 0 is the elasticity of demand. The left hand side is the mark-

up rate, also known as the Lerner Index. it is therefore higher when demand

is inelastic (meaning an elasticity closer to zero). Note that the elasticity of

demand at the uniform price chosen by the monopolist is always less than -1.

2.2 Welfare analysis and public policy

We have just seen that a private Þrm which maximizes its proÞt will choose a

price that is higher than the perfectly competitive price, and can thus increase

its proÞt. Obviously, this is done at the expense of the consumers (travelers),

who face higher prices. We now show how this loss of consumer well-being can
1 2 This quantity is strictly positive and unique if marginal revenue at zero is superior to

marginal cost, and if the slope of marginal revenue is always less than that of marginal cost.

12
be measured in monetary terms in a way that can be compared with the extra

proÞt extracted by the Þrm.

An individual’s consumer surplus is the difference between the maximum

price that the consumer is willing to pay for a trip and the price that she

actually pays. The inverse demand curve is constructed by ranking willingness

to pay in decreasing order, so that the q trips are sold to the q travelers willing to

pay most. Aggregate consumer surplus, CS, generated by the sale of quantity

D (p) at price p, therefore corresponds to the area between inverse demand and

the quantity from 0 to D (p).13 For the solution described in Figure 1, this

is represented by the triangle CS in Figure 2. The loss in consumer surplus

resulting from a change from pricing at marginal cost to pricing at the monopoly

level is given by the monopoly producer surplus plus the area DW L in Figure 2.

This loss of consumer surplus can be interpreted as the sum of what travelers

would have been willing to pay, collectively, in order to be able to access tickets

priced at marginal cost as opposed to the monopoly price. Insofar as this total

is larger than the monopoly producer surplus, there is a beneÞcial exchange

possibility for all market participants which has not been realized (the Þrm

would be ready to cut its price down to marginal cost if, in exchange, it could

receive compensation that is at least as large as its producer surplus).

Monopoly pricing therefore introduces an inefficiency which can be measured

by that part of the loss of consumer surplus which is not offset by an increase

in producer surplus. This deadweight loss is the area DW L in Figure 2. To

1 3 See Anderson and Engers (2007a) for further discussion of the concept of consumer surplus.

13
understand this inefficiency more concisely, it is convenient to introduce the

concept of social surplus. This is the sum of producer surplus and consumer

surplus. Pricing at marginal cost enables the maximal social surplus to be

attained. Deadweight loss is the reduction of social surplus caused by a higher

price.14

One reasonable objective for public policy could be to minimize deadweight

loss. This objective could be obtained by a public Þrm, or one subject to

regulation, pricing at marginal cost. Such a solution is not generally very

satisfactory. When there are increasing returns to scale, such pricing will not

cover production costs. For example, when marginal cost is constant, marginal

cost pricing will generate zero producer surplus, so the Þrm will not cover its

Þxed costs.

It follows that the Þrm must be partly Þnanced by taxpayers. It may then be

desirable to suffer some deadweight loss in order to avoid an overly large deÞcit.

Pricing at marginal cost might then be replaced by pricing at average cost, so

that the Þrm just covers its costs. Nevertheless, it seems rather arbitrary to

impose the condition that the Þrm should not make losses. An alternative

argument against pricing at marginal cost is that taxation induces inefficiency

in the allocation of resources (see Meade, 1944). Optimal pricing must therefore

strike a balance between maximizing social surplus in the markets served by the

Þrm and the efficiency cost of raising tax revenue in the rest of the economy.

The latter cost can be measured by the deadweight losses caused by the taxes in
1 4 Any price below marginal cost causes a deadweight loss because it leads to the sale of

some units for which consumers are willing to pay less than the extra social cost that their
production would engender.

14
the markets where they are levied.15 Under such circumstances, the Þrm will

only be fully Þnanced by its users if the cost of raising public funds is too large.

This reasoning also implies that if the cost of raising public funds is too large,

and if the government can appropriate the Þrm’s proÞts (which is effectively

the case if the Þrm is public), it may be desirable to earn a return exceeding

production costs. This extra revenue will enable the government to reduce Þscal

pressure on the rest of the economy.

We can now address the pricing problem of a public or regulated Þrm. Fol-

lowing Laffont and Tirole (1993), we introduce a parameter λ which is the

marginal deadweight loss of raising public funds: an extra euro raised means a

cost of 1 + λ euros.16 The Þrm’s objective function can then be written as:

CS + (1 + λ) P S.

When λ becomes large, the Þrm pays no heed to consumer surplus. This case

correspond to a private Þrm maximizing its proÞts. Otherwise, the Þrm chooses

p to maximize:

CS (p) + (1 + λ) D (p) (p − c) .

Using the relationship CS 0 (p) = −D (p)17 gives the Þrst order necessary

condition of :
1 5 For example, income taxes cause deadweight losses in the labor market.
1 6 The size of λ has been the topic of several studies. One reasonable estimate for the US
is 0.3 (see Ballard, Shoven, and Whalley, 1985, and Hausman and Poterba, 1987).
1 7 Consumer surplus at price p is the integral of the demand for prices larger than p.

15
 p−c λ 1
= .
p 1 + λ |η (p)|

The price thus obtained is a special case of what are called Ramsey-Boîteux

prices, which ensure maximization of social surplus under the constraint that

the Þrm returns a particular level of proÞts, for example to cover its Þxed costs.

Under this interpretation, λ therefore indicates the severity of the budget con-

straint and is the marginal social surplus gain that could be obtained by reducing

the proÞt level to be earned by one euro (see Ramsey, 1927, and Boîteux, 1956).

In the interpretations above, λ is necessarily positive, so that more weight

is placed on producer surplus than on consumer surplus. A higher λ leads

to a higher mark-up over marginal cost (with monopoly pricing resulting as λ

goes to inÞnity).18 This outcome is rather unsatisfying insofar as we might

wish for public policy to respond not only to economic efficiency but also to

redistribution. The current analysis does not need to take a stand on these

issues.

Thus, if a higher price (of a train ticket) allows the government to efficiently

collect revenues, these could be used to give lump sum transfers to consumers.

Nevertheless, a system of monetary transfers creates difficulties in itself because

of the perverse incentives it may induce, as well as perhaps for reasons of polit-

ical viability. If indeed transfers are not to be made directly from Government

revenues because it is intrinsically costly to do so, then the pricing scheme may

be used directly for transfers, and it may therefore be reasonable to put a larger
1 8 This follows from applying the Implicit Function Theorem to the Þrst order condition to

the Þrm’s problem above.

16
weight on consumer surplus than on producer surplus. This translates in our

formal analysis to −1 < λ < 0.19 This allows us to understand why we might

want to heavily subsidize certain services such as public transportation, even

pricing below marginal cost. Finally, there may be other reasons for pricing

below marginal cost, especially for transportation services. Up until now, we

have not taken into account the possibility of positive or negative externalities

such as pollution or congestion. This omission can easily be rectiÞed by re-

placing, in our analysis, marginal private cost with marginal social cost, which

would be greater or smaller depending on whether the externality were negative

or positive.

3 Discrimination and veriÞable consumer char-

acteristics
3.1 Perfect discrimination

We Þrst consider the case which is most advantageous to the seller. This arises

when the seller has perfect information about the demand from each possible

buyer. Perfect discrimination arises when the Þrm can use this information

fully (Þrst-degree discrimination, in Pigou’s terminology). In order to use the

information, the seller must be able to control the price and the characteristics

of each unit sold to each buyer. For example, if an airline perfectly knew the

needs and desires of all of its customers it could choose the price at which each
1 9 A negative lambda might also be applied in a welfare analysis used to evaluate different

market outcomes, and where the Þrms’ losses do not contribute to (or need to be Þnanced
from) the public purse. Private Þrms’ surplus would usually be weighted (much) less than
consumer surplus. For example, a consumer surplus standard, as is arguably used in some
antitrust circumstances, would put a weight on producer surplus of zero (λ = −1).

17
client would take the price, and dictate the date, time, and comfort class.

This principle can be illustrated very simply by supposing that the good sold

is perfectly homogeneous, and that each buyer only wants a single unit. In this

case, the tastes of the buyer are completely described by her willingness to pay.

A proÞt-maximizing monopolist will then have each buyer pay her maximum

willingness to pay, leaving the consumer no surplus. The Þrm will therefore

want to sell to all those whose valuations exceed marginal cost. This situation

is shown in Figure 1 for constant marginal cost and linear demand. We see here

that this pricing policy leads to maximal social surplus, all of which is captured

by the Þrm.

The above approach can easily be extended to deal with elastic individual

demand (so the quantity demanded decreases as the price rises). For illustra-

tion, suppose that the demand curve in Figure 1 represented a single consumer.

The inverse demand would then indicate the highest price that she would pay

to consume one extra unit. If the quantity q were sold at the uniform price

P (q), the net consumer surplus would be CS (P (q)); we can then deÞne the

gross consumer surplus (for an individual20 ) as V (q) = CS (P (q)) + P (q) q.

Optimal discriminatory pricing would then mean paying the willingness to pay

for each unit, and selling units as long as this willingness to pay exceeds mar-

ginal cost. This then induces the quantity q ∗ = D (c) sold at a tariff equal to

the corresponding gross consumer surplus, V (q ∗ ) which is the surplus that the

consumer would enjoy if she could consume the quantity q ∗ for free.
2 0 The corresponding aggregate concept simply sums the surpluses over individuals.

18
 Another method for getting to the same result would be to use a two-part

tariff (see also Oi, 1971). Such a tariff would specify an entry, or membership,

fee A that the consumer must pay in order to consume the good at all. If

she joins, then she can buy as much as she wants at a price p. Setting p = c

ensures that the consumer will therefore choose q ∗ , and she will therefore enjoy

a surplus of V (q ∗ ) − cq ∗ . She will join as long as the entry fee is not larger

than this, and so the seller will set a fee as large as possible subject to this

individual constraint; namely it will set an entry fee of A = V (q ∗ ) − cq ∗ . The

total price paid by the consumer is then A + cq ∗ = V (q ∗ ), which means that

her full surplus is extracted by the Þrm. The outcome is thus just the same as

for the preceding pricing system. Even though a two-part tariff seems simpler,

it still needs just as much information: while it may be easy to Þx price at

marginal cost, calculating the entry fee means knowing the consumer’s surplus,

and hence her full demand curve.21

The pricing solution for a public Þrm under the similar assumption of per-

fect knowledge and ability to discriminate is straightforward: it should choose

exactly the same tariff structure. This is because all surplus is extracted from

the consumer and earned by the Þrm for the public purse.
2 1 Nofurther complication is introduced for the preceding analysis when marginal costs are
not constant. The optimal quantity, q∗ , equates the demand price with marginal cost, and all
consumer surplus is extracted. A single (all-or-nothing) tariff equal to gross consumer surplus
at this quantity is optimal, and is equivalent to a two-part tariff with a per-unit price equal
to the demand price P (q∗ ) and an entry fee of A = V (q ∗ ) − P (q∗ ) q ∗ (= CS (q ∗ )).

19
3.2 One good and several groups of buyers

We next consider a situation in which a Þrm can observe a characteristic or

characteristics of buyers— such as age, job, or residential address— and observing

these characteristics allows the Þrm to infer something about demand. The Þrm

can exploit some correlation between the observed variable and the individual

demand in order to discriminate (perhaps such discrimination is not legal or

socially acceptable, for example if it is based on gender or race).

The Þrm can then choose a price that depends on the observed character-

istic, and the individual who does not have this characteristic can be excluded

from prices not meant for her. However, discrimination is imperfect insofar as

consumers are bundled together onto the same characteristics (e.g. the same age

group), and individuals still may differ by willingness to pay within the group

(within a group, selling different quantities with nonlinear pricing or introduc-

ing differentiation can be used to get potential buyers to reveal their tastes: see

Section 4). The Þrm is therefore obliged to set a uniform price for each category

(or group). Suppose, for example, there are two classes of buyers: youths under

26 and the rest of the population. Asking for identiÞcation (in the absence of

fake IDs), the Þrm can know which category the buyer is in. It can therefore

exclude arbitrage by which people in one group buy the good or service in or-

der to sell it to people in the other group (for example, airline tickets with the

traveler’s name on them). In the absence of such arbitrage, demand from each

group depends only on the price charged to members of the group.

Consider the case of a public Þrm: the specialization to a private Þrm will

20
simply be given by letting λ become arbitrarily large. Let pi be the price applied

to group i, and let Di (pi ) be the resulting demand curve, and CSi (pi ) be the

corresponding consumer surplus. We assume that marginal cost is constant

at rate c, so that the Þrm can choose prices for groups independently of each

other.22 The Þrm therefore chooses pi to maximize:

CSi (pi ) + (1 + λ) Di (pi ) (pi − c) .

This leads to a Þrst order condition which simply generalizes that, for a single

market:

pi − c λ 1
= .
pi 1 + λ ηi (pi )

This type of discrimination beneÞts the Þrm because it faces groups of con-

sumers whose elasticities differ. The less elastic demand is (the closer η is to

zero), the higher the price charged to the group. Thus, youths enjoy lower train

fares and airfares because they are more likely to reduce their demand if prices

were higher. Clearly the Þrm beneÞts from discrimination because it can always

opt to charge the same price across groups. However, only consumers for whom

demand is quite elastic beneÞt, to the detriment of the other consumers.23

For a proÞt maximizing Þrm, we can determine whether discrimination is

socially beneÞcial. Suppose for simplicity that λ = 0, so that social surplus

2 2 When marginal costs are not constant, the optimal solution is obtained by equalizing

marginal revenues across groups for any given total production level, and then choosing the
total quantity which equalizes marginal cost with this common revenue.
2 3 This discussion assumes that two demands are comparable in the sense that one is more

elastic than the other for all prices. It is possible to construct examples for which prices under
discrimination are higher or lower than under uniform pricing (see Nahata, Ostaszewski, and
Sahoo, 1990)

21
is simply the sum of consumer surplus and producer surplus. Note Þrst that,

for a Þxed total output, discrimination induces inefficiency in the allocation of

this quantity between two groups. The price paid by the consumers in a group

reßects what they are willing to pay in order to buy one extra unit. Social

surplus can be increased by shifting consumption from those willing to pay less

at the margin to those willing to pay more. Furthermore, since the output of

a monopolist under uniform pricing is too low, discrimination can only improve

welfare if it increases production. This necessary “output condition” was Þrst

proposed by Schmalensee (1981).

In general, the impact of discriminatory pricing on the total quantity pro-

duced is ambiguous. For example, suppose that the demand in each group i is

linear and given by:

Di (p) = ai − bi p.

Applying the earlier analysis of linear demand, if both groups are served under

uniform pricing the total output is:

a1 + a2 − (b1 + b2 ) c
,
2

while the quantities allocated to each group of buyers under discrimination are,

respectively:

(a1 − b1 c) (a2 − b2 c)
and .
2 2

22
 The total output is therefore identical24 , and discrimination is necessarily

harmful to social welfare. This conclusion is invalid if only one group is served

by the non-discriminating monopolist. Then, if marginal cost is constant,

discriminatory pricing is clearly preferable because it allows more markets to be

served. In this case, the group which is served under both pricing schemes is

not worse off because it pays the same price under both, while the other group

would not have been served at all under uniform pricing.

The above reasoning also gives us an unambiguous result when capacity is

Þxed. Discriminatory pricing cannot increase output, and is therefore clearly

detrimental. If an airplane is full, it is therefore better that all passengers

pay the same price. Nevertheless, this reasoning is only valid in the short

run, because in the longer run the possibility of discrimination can motivate the

airline to increase the number of ßights. In this context, too, discrimination

can engender a greater diversity amongst the traveling population.

As we argued in the introduction, second-degree discrimination as originally

envisaged by Pigou (1938) can be seen as a special case of the above analysis of

discrimination between different groups of buyers. If the observable character-

istic used by the Þrm to discriminate can allow it to perfectly separate buyers

into groups that can be ordered according to willingness to pay, then it can

practice second-degree price discrimination à la Pigou.

To illustrate, consider the following example given in Anderson and Renault

(2003b). Suppose that a traveler wishes to take a train trip and her willingness
2 4 This is a special property of linear demand, that marginal revenue corresponding to the

sum of demands is equal to the sum of the marginal revenues to the demands for each group.

23
to pay is perfectly positively correlated with age. Even though this would

seem to allow the Þrm to practice perfect discrimination (because the client’s

age perfectly reveals her demand), it might in practice be costly to specify too

many different prices. For example, it may be only possible to offer two prices.

The railroad company must then set a threshold age above which people cannot

get the low price.

Suppose, for example, that the inverse demand is given by P (q) = 1 − q.

With zero marginal cost, the monopoly price and output are both one-half. It

is then easy to see that the Þrm’s optimal strategy is to charge full fare of 2/3

for the oldest third of the population, and to set a reduced fare of 1/3 for the

younger ones. In this context, we can also determine the second best policy of

choosing the critical age with the objective of maximizing social surplus, subject

to the Þrm choosing its prices given this critical age. The second best solution is

that the Þrm only offers the reduced fare to the younger half of the population.

The Þrm would then choose a full fare of 1/2 for the older half of the population

and a half-priced fare of 1/4 for the younger travelers. In this sense, the critical

age chosen by the proÞt maximizing Þrm is too high.

3.3 Backhauling

We have assumed so far that the product is produced at constant marginal

cost. In the case of transportation, services are provided and these depend

upon the capacity offered. While it is out of our scope to cover the full range of

issues associated with proper cost attribution, scheduling of service, and route

network choice of passenger services, we nevertheless broach this issue with a

24
simple example of capacity that is provided on an outbound trip: the train has

to get back to the origin to make the next trip to the destination.

The analysis can be framed in terms of a monopoly passenger railway service:

see Rietveld and Roson (2002) for a recent application in this vein. Many

commuters might wish to make the trip to the Central Business District in the

morning rush-hour commute, but few people want to go in the reverse direction

soon afterwards.25

The analogue in the context of freight transportation is when a product

is shipped to a destination and there is a relatively weak demand for shipping

goods in the opposite direction. However, the trucks, ships, or freight trips must

still make the return trip to pick up another load. This is termed the “backhaul

problem”.26 We therefore frame the application to passenger transportation in

the classic fronthaul and backhaul context (e.g., Mohring, 1976). This is well

known in economic theory as a joint production problem, much as the textbook

mutton and wool joint production in raising a sheep. Analogously, once an

outbound trip to the Þnal market is created (fronthaul), then a return trip is

also created (backhaul).

The economic theory of pricing for competitive markets with backhaul is

well understood. Suppose, as above, that the round trip costs c, and that the

demand for such trips is given by a well-behaved downward-sloping demand,

2 5 Of course, in the evening, the “backhaul” is stronger than the “fronthaul” in the sense
that relatively empty trains come into the city and full ones go back out. For the present
purposes we shall identify the stronger demand as the “fronthaul” even if it happens after the
weaker demand (backhaul).
2 6 See Anderson and Wilson (2008) for a treatment of the backhaul problem when a dominant

Þrm faces a competitive fringe.

25
D (p). Suppose too that the demand for trips back from the Þnal market

is Db (p). Denote the inverses of these demand curves as P (Q) and Pb (Q)

respectively. For simplicity we assume that the incremental cost when carrying

passengers for the trip back is zero. The relevant demand price for round

trips is, therefore, the sum of the demand prices for the outbound and inbound

trips, censored to be non-negative (because the transporter can always come

back empty). That is, if Q is quantity, the demand price for the round trip is

P (Q) + Pb (Q) (where it is understood the demand prices are non-negative) and

this sum is equal to p in equilibrium. Denote the solution as Q̂: transport prices
³ ´ ³ ´
for each leg are then P Q̂ and Pb Q̂ . Clearly, if the backhaul demand is
³ ´
weak then Pb Q̂ can very well be zero. Some passengers can be carried, but

backhaul demand is not contributing anything to reducing the price on the front

haul, and effectively Q̂b < Q̂ where Q̂b is the number of travelers transported

on the backhaul.

It is now simple enough to see how to introduce incremental costs for back-

haul: they can be netted off the demand price for the backhaul. The same

principle applies in what follows: it suffices to net the incremental costs off the

demand price.

For a monopolist, say the commuter train where it has a large cost advantage

and is unconstrained, the appropriate principle for determining the quantity to

carry (and the corresponding prices) is that the sum of the marginal revenues

equal the marginal cost. Here again, the monopolist is not obligated to carry

as much back as it carries out, and so we truncate the marginal revenues at

26
zero. Then the solution for the fronthaul quantity, Q∗ , is given as the solu-

tion to max {M R (Q) , 0} + max {M Rb (Q) , 0} = c, where M R (.) denotes the

marginal revenue to the outbound demand curve (and M Rb (Q) for the back-

haul demand). In this case, with a weak backhaul demand the price charged

will not be zero but rather the revenue maximizing point on the (net) backhaul

demand curve. This “zero-cost monopoly price” (which is equivalently the rev-

enue maximizing price) will prevail when some backhauls carry no passengers.

Equivalently, any solution with Q∗b < Q∗ involves Q∗b = M Rb−1 (0). For any

solution to the monopoly problem, the prices on the two legs are given from the

inverse demand curves as P (Q∗ ) and Pb (Q∗b ).

In the preceding analysis, the discriminating Þrm behaves just like a Þrm

that offers different products with perfectly independent demands. In prac-

tice, discrimination founded on observable characteristics is often associated

with multi-product production. The next section addresses the application of

discriminatory pricing in such a context.

3.4 Yield Management

To set the stage, assume that a transportation vehicle (such as a train) has

a single seat to Þll.27 The operator knows that one traveler will purchase

the ticket, and that the traveler’s reservation price for the trip is uniformly

distributed on [0, 1] (think linear demand curve, as in Section 2.1 and Figure

1). There is no cost associated with selling the ticket or Þlling the seat. The
2 7 For references in this area, see Anderson and Schneider (2008), which also contains more

analysis of competitive cases. See also Sinsou (1999) for a detailed presentation of the algo-
rithmic methods Þrms can use for Yield and Revenue Management.

27
operator is to set a price to maximize its expected proÞt. Given that the

associated demand curve is linear, the price the operator should set solves the

classic monopoly problem, so p = q = 12 and the associated proÞt is π = 14 .

Now consider the case where two travelers will arrive, one after the other, and

their reservation prices are independently distributed on [0, 1]. The operator

sets a single price for both. What price should it set? To Þnd the answer, we

Þrst determine its proÞt. It is more transparent to write out the more general

problem with F (p) the probability an individual traveler’s reservation price

is below p (so F (.) is the cumulative distribution). Then the proÞt is π =

p [1 − F (p) + F (p) (1 − F (p))], where the term in brackets is the probability

that the Þrst traveler buys, plus the probability that she does not (F (p)) times

the probability that the second one does (1 − F (p)). In the case of linear

demand, F (p) = p for p [0, 1] and the Þrst-order condition implies that:

1 1
p= √ >
3 2

The second order condition is readily seen to be satisÞed; and proÞt is ap-

proximately 0.385. This is more proÞt because the operator can take a shot at

a higher price given that the second customer provides some insurance in case

the Þrst one refuses. This logic is Þne-tuned in the next example. Note that

the price charged (in the current case of a single price for all) is increasing in

the number of consumers, and the proÞt is too. As the number of travelers

gets large, retaining a single seat, the price goes to the highest possible in the

population (1) and the proÞt goes to 1 too since the probability of acceptance

also tends to 1.

28
 ProÞt is even higher if the operator can choose a separate price for each

arriving guest. Suppose there are again two travelers, and they arrive sequen-

tially. We want to Þnd what price will the operator will set for the Þrst one to

arrive, and what it will charge the second one if the Þrst traveler declines the

seat. To Þnd these prices, we work backwards. If the Þrst traveler does not

buy, we can simply use the answer from the monopoly case with one seat.

This gives the expected proÞt of 1/4 if the room is not Þlled on the Þrst

shot. We can now determine the proÞt for the choice of a price p1 to the

Þrst traveler. Noting that the probability that the Þrst traveler does not buy

is F (p1 ), this is π = p1 (1 − F (p1 )) + F (p1 ) 14 . The uniform distribution en-

5
tails an optimal price of 8. Plugging back into proÞt, this means a proÞt of
53 51 25
π = 88 + 84 = 64 = 0.39. This discriminatory strategy is larger than when

a single price must be set for both travelers, since the discriminatory problem

subsumes the non-discriminatory one.

We next consider the case of three travelers and two places. Recall Þrst that
1
with simple linear demand, the monopoly price is 2 and proÞt is 14 . These are

indeed the price and expected proÞt that are earned on the last traveler if one

place has already been sold. Likewise, if the very Þrst sale is a failure, there
1
are two places left and two travelers, so then the price is again 2 in each period

and the expected proÞt coming out of the failed Þrst sale is 12 . If the Þrst sale

was a success, there is one place left and there are two remaining consumers.
25
The continuation proÞt is then that derived just above, i.e., π = 64 .

We are now in a position to analyze the Þrst period’s price. This is given

29
by the solution to
∙ ¸
25 1
maxπ = (1 − F (p)) p + + F (p) .
p 64 2
£ 25
¤
With a uniform distribution, the proÞt expression is π = (1 − p) p + 64 + p 12 ,
7
which generated a Þrst order condition (1 − 2p) + 64 = 0, and hence the solution
71 71 5
p= 128 . The corresponding sequence of prices is then 128 , followed by 8 (an

1
increase) if the sale was a success or 2 (a decrease) if it was a failure. The last

price is 12 , if at least one previous sale foundered.

The type of exercise above can be applied to various other different pricing

practices observed in transportation. For example, the SNCF sells a Þrst batch

of tickets at a lower price than a second tranche, which in turn is lower than the

third tranche. The simplest set-up to analyze this practice is to assume that

there are three travelers, and two seats. In contrast to what was just described,

the pricing is now per seat, instead of the operator being able to condition per

traveler. This implies that this is a special (constrained) case of the earlier

analysis, and so leads to lower proÞts.28

The analysis can also be expanded to deal with uncertainty in the number

of buyers, possibility of “strategic” buyers (or coming back after observing a

Þrst price), correlation in traveler values, etc. We pick up on this last topic

because it is germane to transportation when Þrms may use the strength of

early demand response to gauge later demand strength (for example, airlines
2 8 The proÞt associated to the pricing© per¡ seat ¢ª busi-
ness model is π = (1 − F (p1 )) p1 + p2 1 − F 2 (p2 ) +
F (p1 ) {(1 − F (p1 )) [p1 + p2 (1 − F (p2 ))] + F (p1 ) (1 − F (p1 )) p1 } .
The Þrst term corresponds to the Þrst traveler buying the Þrst (lower-priced) seat, and then
one of the other two buys the second seat. The second term corresponds to the Þrst traveler
declining the Þrst seat, and then either the second buys it (and the third might buy the second
seat), or else only the third might buy the Þrst seat.

30
or passenger trains judging demand for special events like Olympic Games or

tennis matches), and because it illustrates nicely the experimentation motive

for varying prices.

Suppose then an operator has a single seat to Þll, and there are two potential

travellers. Both have the same valuation for traveling (perfect correlation), and

this valuation is uniformly distributed on [0, 1]. If the operator charges p to the

Þrst traveler, it expects to sell the seat with probability 1−p. If the Þrst traveler

does not buy, the operator knows the seat is worth less than p to the second.

The updated valuation for the second traveler is uniformly distributed on [0, p],

and so the optimal price to set, conditional on a Þrst refusal, is p/2. The second

traveler accepts with probability 1/2, so the expected proÞt, conditional on a

Þrst refusal, is p/4. This means the operator’s proÞt problem as a function of

the Þrst price set, p, is

n po
max (1 − p) p + p .
p∈[0,1] 4

This proÞt is maximized at p = 2/3. If the Þrst traveler refuses, the price

is dropped to 1/3. The case of uncorrelated demand, given earlier, has less

price dispersion, starting with a lower price (5/8) and ending with a higher one

(1/2). With correlated demand, the operator starts with a higher price that

includes an experimentation motive to Þnd out more about the second traveler’s

valuation.

31
3.5 Discriminating with several products

When a Þrm sells several goods with perfectly independent demands, the sit-

uation is formally very similar to the one we have just discussed. The main

difference is that marginal costs generally differ across products. With several

products, there is a possibility of discrimination when different goods are sold

at the same price or at prices that are close to each other. As we pointed out in

the introduction, it can be quite difficult to determine whether a price difference

is discriminatory.

The objective of this subsection is to brießy present some of the prominent

examples that are frequently considered to represent price discrimination. In

particular, we will discuss spatial discrimination and then look at discrimination

over time and tied sales. In all cases that we consider, the client group which

might buy one of the goods is clearly identiÞable and there is no possibility of

personal arbitrage. Using a multi-product offer to discriminate when arbitrage

is possible will be studied in the next section. We Þrst develop the optimal

pricing strategy for a Þrm which sells to consumers located at different distances

from its production point. We then show how this theory is relevant to a Þrm

which must transport travelers over different distances, and we suggest other

applications in transportation.

Suppose as a benchmark that consumers can access a transport service sold

at a competitive price, and they can use this service to transport the good from

its production point. Using this service, a consumer then pays for each unit

of the good, an f.o.b. price set by the Þrm at its production point plus the

32
transport cost.29

Without speciÞc information on individual demand, the Þrm must charge a

uniform tariff. This is the benchmark case of non-discrimination. However,

if the Þrm can take care of delivery itself and if it can circumvent consumers’

access to competing delivery services, it can extract more proÞt even though

the delivery point does not reveal information about the demand of the buyers

in question.

For illustration, suppose that demand is the same at every point in space. If

the Þrm charges a price px to buyers located at a distance x away, the demand

is D (px ) and the consumer surplus is CS (px ). Transport costs are linear and

it costs tx to transport one unit a distance x, with t > 0. The Þrm chooses px

to maximize

CS (px ) + (1 + λ) D (px ) (px − (c + tx)) .

The price chosen will then be similar to that derived earlier, where we now

simply use the full marginal cost c + tx. It can be shown that as long as the

demand is more “concave”30 than an exponential function, the rate of increase

in the Þrm’s price is less than t per unit distance (see Anderson, 1989). This

means that the Þrm does not pass on all of the transportation costs into the

price (“freight absorption”). In other words, the implicit f.o.b. price— the

price net of transportation costs— is smaller the farther that buyers are from the
2 9 The term f.o.b. stands for “free on board” or “freight on board” and means that the

buyer must pay transportation costs for delivery. In spatial economics, the term f.o.b. price
is synonymous with the term “mill price” to indicate the factory price (or the store price).
3 0 See Anderson and Renault (2003a) for a precise deÞnition of this concept. The condition

is equivalent to Seade’s (1987) condition on the elasticity of inverse demand.

33
production point.

Freight absorption is often used in practice31 . Tirole (1988) notes that there

are good reasons why there should be freight absorption even if demand curves

do not satisfy the condition above. First, any pricing policy which over-bills

transport costs might provoke arbitrage between consumers, with those who are

closer to the factory perhaps buying the good and transporting it themselves,

and reselling it to those farther away. Second, buyers far from the production

site might also be more likely to be closer to a rival’s production site, and

this might render their demand more elastic. The Þrm may consequently be

motivated to set a somewhat lower price.32

The social beneÞts from forcing a private Þrm to set a uniform price are

ambiguous. The discriminating Þrm serves a larger geographic area (which,

per se, improves social welfare) but it does so while imposing higher prices

on nearby consumers. For linear demand, the social welfare is highest under

spatial discrimination when λ = 0 (see Holahan, 1975), and therefore a fortiori

is highest when λ > 0 because the producer surplus is always higher when the

Þrm can discriminate.33

Now consider some applications in transportation. Consider Þrst an airline

selling long-distance ßights between Paris and New York. Customers’ points

of origin differ: they start out from different regional French airports. Our
3 1 Greenhut (1981) presents survey results on spatial pricing by Þrms.
3 2 See Lederer and Hurter (1986) for a model of competitive spatial discrimination and
Anderson, de Palma, and Thisse (1992, Chapters 8 and 9) for a review of the literature on
spatial economics.
3 3 The case of uniform pricing is a special case of discriminatory pricing. See Greenhut,
Norman, and Hung (1987) for other developments with monopoly, and Anderson, de Palma,
and Thisse (1989 and 1992b) for competition. Quinet (1998) describes in detail competition
between two airports.

34
benchmark is when ßights between different regional airports and Paris are sold

on a competitive market and are therefore sold at marginal cost. Suppose that

the airline can provide exclusive service to Paris, and that the demand for a

trip to New York is the same irrespective of the starting point of the travelers

(Paris, Bordeaux, Nice...), and that this demand is not “too convex”. The

freight absorption principle is manifested by more expensive tickets for tickets

who originate farther from Paris (because of the higher cost of service), but

price differences are less than the extra cost of serving these customers. In this

sense, the more distant customers are subsidized by those closer to Paris.

The analysis also applies to railway pricing, or indeed an urban transporta-

tion network. Consider two trips of different length, which therefore involve

different costs. Following Phlips’ (1983) deÞnition, pricing is discriminatory if

the price difference between two trips is not the same as the cost difference. If

the demand for the two trips is the same (so that we eliminate any price differ-

ences due to demand differences), and if the demand curve is not “too convex”,

then optimal prices will be closer than costs. In other words, the price should

rise more slowly than the serving costs as a function of distance.

Of course, the analysis of urban transportation prices is much more complex

than the sketch presented above. Demand typically depends on the distance

traveled and the other transport modes available. Pricing should take conges-

tion into account, etc., but this simpliÞed framework at least serves to indicate

the central role of discriminatory pricing.

There are numerous dimensions other than space over which buyers can be

35
sorted. Time is one obvious example. Airlines often offer lower prices for

return trips that include a Saturday night stay. This practice allows them to

discriminate between leisure and business trips. Tourists’ demand is relatively

elastic, whereas business travelers face important timing restrictions. Similarly,

prices which differ according to the time and date of departure can allow account

to be taken of congestion (which is the reasoning behind peak hour pricing),

but it can also be used to discriminate between clients with differing demand

elasticities.34

Another practice which is often cited as an example of third-degree price

discrimination is that of bundling or tying. Sales which combine two goods were

recognized by Stigler (1963) to be an efficient means of discriminating. Such

sales allow the Þrm to distinguish buyers interested in both goods from those

interested in only one or the other. Package sales, combining transportation

plus hotel, allow tourists to pay less for both services than if they were bought

individually.

These examples indicate that segmenting the market can be substantially

more subtle than just verifying IDs. However, strategies that do not rely on

strictly veriÞable information might be undone both by arbitrage across indi-

viduals or “within” the individual. The limits that personal arbitrage impose

on discriminatory pricing were laid out in the 19th century by the French Ponts

et Chaussées engineer Jules Dupuit. In his famous article on tolls (1849), he

developed an example of price discrimination for a footbridge. He supposed

3 4 There is a large literature on pricing under congestion, following Vickrey’s (1969) seminal

paper. De Palma and Lindsey (2000) treat the problem of two competing toll roads, and de
Palma and Lindsey (1998) study the role of information acquisition in this context.

36
that many workers would like to use the bridge, but that a universal price of 1

centime per user would not suffice to cover construction costs.

In an attempt to impose a special workers’ price that would be less than that

paid by the rest of the population, Dupuit proposed a discriminatory scheme

based on clothing: “for a crosser with a cap or a smock or jacket, the toll is

reduced to 1 c.” (p. 220) instead of 5 c. for the other travelers. However, he

also notes that “it is quite likely that revenues will be reduced because some 5

c. crossers will beneÞt, by dint of their tenue, from the price reduction that was

not meant for them” (p. 220). To staunch this potential arbitrage, he proposed

only applying the price reduction at certain times of day (when workers were

more likely to be present) or to require that worker present their pay stubs. In

the next section we give a systematic analysis of strategies that allow the Þrm

to prevent personal arbitrage.

4 Personal arbitrage

Travelers cannot be classed like merchandise by their appearance:

they have to be allowed to sort themselves. (Jules Dupuit, 1849)

Discrimination when there is personal arbitrage is effectively an informa-

tion revelation problem. The Þrm knows that there is heterogeneity in the

willingness to pay among buyers, but it doesn’t have a means of knowing this

information directly. In the example of Dupuit’s (1849) footbridge, those who

are willing to pay only one centime to cross may be relatively poor people who

are not workers with a pay stub, or indeed they may be people who do not truly

37
need to cross the bridge and who will not bother if it costs too much.

This problem of information revelation can only be resolved if individual

demand varies with price, or if the Þrm can vary some characteristics of its

product, which are differently evaluated by different consumers.

In the example of the footbridge, Dupuit noted that the second possibility

could be to price differently according to the time of day. To see how the sale

of different quantities can be exploited in the same example, suppose that the

workers were sold a coupon, valid for six return trips per week, at a price of 12

c. If the other users only want, at most, one return trip, they will only pay 5

c. per trip rather than buying the ticket tailored for the workers. This type

of “nonlinear pricing” is described in Subsections 1 and 2 below. Subsection 3

treats the use of multi-product offering in the presence of personal arbitrage.

4.1 Two-part pricing with heterogeneous customers

As we saw above, a two-part tariff speciÞes an entry fee A and a marginal price

p, at which those who paid the entry fee can get the good. When the Þrm

perfectly knows consumer tastes it can use this type of pricing to appropriate

the entire consumer surplus, and thus perfectly discriminate. Although, in

practice, transportation Þrms do not have such perfect information, they often

use this type of pricing. The SNCF uses different passes for various periods of

time. The French Metro (RATP) has the unlimited-ride Orange Card, which

has a marginal price of zero. As we shall now see, such a two-part tariff can be

useful when the Þrm does not know the consumers’ tastes. (see also Oi, 1971)

Suppose there’s a fraction α of consumers, corresponding to Type 1 con-

38
sumers, whose demand at price p is D1 (p) while the rest of the consumers

demand D2 (p). Suppose that D1 (p) < D2 (p) for all p ≥ 0 (the demand

curves are shown in Figure 3), so that Type 1 consumers are low-demand and

Type 2 consumers are high-demand. It follows that consumer surpluses satisfy

CS1 (p) < CS2 (p) for all p ≥ 0.

When the Þrm has full information it would choose a marginal price p = c

and have Type 1 consumers pay an access fee A1 = CS1 (c), and A2 = CS2 (c)

for the others. The number of trips would be q1∗ = D1 (c) and q2∗ = D2 (c)

for consumers of Types 1 and 2, respectively. These are the Þrst-best opti-

mal quantities.35 Under incomplete information, if the Þrm offered these two

choices, clearly Type 2 consumers would pay the lower entry fee A1 , which would

give them access to the same price per trip, and there would thus be personal

arbitrage.

One simple solution to circumvent this type of arbitrage is to offer a single

two-part tariff. If the Þrm offered the full-information prices tailored for Type

2 buyers, the consumers with the lower willingness to pay would not buy.

If there are enough Type 1 consumers (if α is large enough), this will not

be the best solution. Instead, offering the Þrst-best optimal price tailored for

the Type 1 consumers, the Þrm can put in place an allocation which maximizes

total surplus (because marginal price equals marginal cost and both types of

buyers are active).

This will be an optimal solution for a public Þrm when there is no marginal
3 5 Theyare the socially optimal quantities if the Þscal system allows lump-sum redistribution
with a zero marginal cost of public funds.

39
cost of public funds (λ = 0). If λ > 0, the Þrm is also concerned with its proÞts.

Even though the Þrm can extract the full surplus of Type 1 consumers, it must

leave a strictly positive surplus to the others, as we shall now see.

The Þrm’s objective can be split into two parts, the social surplus associated

with Type 1 buyers:

CS1 (p) + (1 + λ) (p − c) D1 (p) + λA (with weight α),

and the social surplus associated with Type 2 buyers:

CS2 (p) + (1 + λ) (p − c) D2 (p) + λA (with weight 1 − α).

Each of these components includes consumer surplus and producer surplus

plus the entry fee, which has a value of λ per euro because it is a transfer from

consumers to the Þrm.

First, whatever the marginal price p, it is always desirable to extract the

full consumer surplus from Type 1 consumers when λ > 0. If A < CS1 (p) <

CS2 (p), a small increase in the entry fee will have no impact on the quantities

consumed and will raise revenues, so the Þrm is better off. It will therefore

choose, as with full information, to set A = CS1 (p), so that consumers with

a low willingness to pay are just indifferent between joining and not joining.

Nevertheless, in contrast to the solution with full information, the Þrm now

wants to price above marginal cost. Even though the Þrm would maximize its

objective over the Type 1 consumers by choosing p = c, this does not maximize

social surplus with regard to the Type 2 consumers. The objective can then be

40
written as:

CS2 (p) + (1 + λ) (p − c) D2 (p) + λCS1 (p) ,

where the last term takes into account that an increase in marginal price must

be accompanied by a decrease in the entry fee to ensure that Type 1 consumers

buy. Evaluating the derivative with respect to p when p = c gives:

λ (D2 (c) − D1 (c)) > 0.

The Þrm can thus improve surplus by increasing the marginal price. As

the effect of such an increase on social surplus for Type 1 buyers is negligible

(because the derivative at p = c is zero), this price hike is desirable. The

optimal marginal price pmar is strictly between marginal cost and the price

which maximizes social surplus for Type 2 consumers. This latter price is

below the Ramsey-Boiteux price corresponding to demand D2 , because of the

negative effect of a higher marginal price on the membership fee.

It is straightforward to show that a fully nonlinear tariff, without the re-

striction that the second price be linear in quantity, performs better. Figure

3 shows that consumers with a high willingness to pay get a surplus (a rent)

measured by area KN LM J.

It is possible to reduce this rent without affecting quantities consumed.

Suppose that instead of offering a two-part tariff, the Þrm gives each con-

sumer the option of either the quantity q1 = D1 (pmar ) at entry fee T1 =

pmar q1 + CS1 (pmar ), or else the quantity q2 = D2 (pmar ) at entry fee T2 =

41
pmar q2 + CS2 (pmar ), plus the area KN J. Type 1 consumers will then prefer

the combination (q1 , T1 ) to (q2 , T2 ). Type 2 consumers would enjoy a surplus

of JN LM if they chose (q1 , T1 ), and they would get at least as much paying T2

for quantity q2 . The Þrm then wants to leave the Type 2 consumers with as

little rent as possible, that is exactly JN LM .

As we shall see, the pricing scheme that we have just described dominates

the simple two-part tariff, but it is not generally the optimal tariff. It has a

certain number of characteristics of the optimal tariff: Type 1 consumers have

zero surplus, and consume less than their Þrst-best optimal quantity (which

would be consumed at marginal cost pricing); Type 2 consumers have a strictly

positive surplus and are indifferent between the two choices offered. This latter

indifference condition is at the heart of the incentive problem which must be

solved to Þnd the optimal nonlinear tariff.36

4.2 Optimal nonlinear pricing

Continuing the theme from above, suppose that there is a choice between two

quantity-fee combinations (q1 , T1 ) and (q2 , T2 ). Up until now, we have not

considered the problems with personal arbitrage from buyers with weak demand.

Taking into account the possibility of such arbitrage can greatly complicate the

general analysis, and we shall therefore side-step the issue.37

In what follows, because quantities are no longer determined by marginal

prices, it will be more useful to write gross surplus as a function of the quantity
3 6 Forfurther advances on this topic, see Brown and Sibley (1986) and Wilson (1992).
3 7 It
is straightforward to show that the solution we shall describe is not subject to arbi-
trage by Type 1 buyers, insofar as it remains the optimal solution when we include an extra
constraint to take such arbitrage into account.

42
consumed. To this end, let Vi (q) denote the gross surplus of a buyer of type i

consuming quantity q. If personal arbitrage only arises from buyers with high

demand, it is clear that there should be no surplus left for consumers with low

demand. If their surplus was positive, raising the tariff T1 would raise revenues

without affecting quantities because this would render Package 1 ((q1 , T1 )) less

attractive to Type 2 consumers. This condition for zero surplus for Type 1

consumers can be written as:

T1 = V1 (q1 ) .

On the other hand, a strictly positive surplus must be left to Type 2 buyers,

because if the combination (q1 , T1 ) is acceptable for Type 1 buyers then the

Type 2 buyers would get a strictly positive surplus from this combination. To

get them to buy Package 2 ((q2 , T2 )), they must be enticed with at least as

much surplus. However, it is clear that it is not necessary to give them a

strictly greater surplus because T2 can be increased without changing anything

else. This indifference condition can be written as:

V2 (q2 ) − T2 = V2 (q1 ) − T1 ,

or indeed, using the equality T1 = V1 (q1 ) :

T2 = V2 (q2 ) − [V2 (q1 ) − V1 (q1 )] .

The term in brackets is the “informational rent” which the high demand buy-

ers must be guaranteed under personal arbitrage. If λ > 0, this rent represents

43
a cost for the Þrm.

Figure 4 illustrates these constraints for arbitrary quantities q1 and q2 . The

informational rent R1 is greater the higher the quantity consumed by Type 1

buyers. For the choice of q2 , it can be seen from Figure 4 that if q2 < q2∗ , an

increase of this quantity will raise social surplus for a Type 2 buyer as well as

raising the Þrm’s proÞts (with the informational rent unchanged) and it is not

necessary to change the allocation destined for Type 1 buyers. By a symmetric

argument, if q2 > q2∗ , the Þrm can improve surplus by reducing the quantity

addressed to Type 2 buyers.38 The Þrm thus chooses to produce the Þrst-

best socially optimal quantity (obtained by ignoring the marginal cost of public

funds) for buyers whose demand is high. This is also true for λ > 0. ProÞts

that can be earned by selling to Type 2 buyers depend on social surplus and

informational rent. Because the latter only depends on the quantities sold

to Type 1 buyers, it is optimal to choose for Type 2 buyers the quantity that

maximizes the corresponding social surplus.

If the Þrm could extract the full surplus of a Type 1 consumer, it would

like this consumer to buy q1∗ = D1 (c), because this would extract the maximal

surplus. This would also yield the greatest Þscal revenue, which coincides with

the social surplus. Because the Þrm is uncertain about the buyer type, it must

take into account the impact the quantity q1 has on the rent that would accrue

were the demand high (as opposed to low).

Taking into account this possibility leads the Þrm to choose a quantity below
3 8 Recall that the area between marginal cost and demand entails a negative surplus when
the latter is below the former.

44
the Þrst-best optimal one, q1∗ . At q1∗ the effect of a quantity reduction on social

surplus for a Type 1 consumer is negligible, while it allows the surplus to a Type

2 consumer to be reduced. Such a reduction is therefore desirable.

More generally, a decrease of q1 leads to a lower social surplus associated

with Type 1 consumers and a decrease in the informational rents to Type 2 con-

sumers. Figure 5 shows that the reduction in social surplus would be P1 (q1 )−c,

which has a social value of (1 + λ) per euro, because this amount is entirely ap-

propriated by the Þrm. The reduction in informational rent, which from the

Þgure is P2 (q1 ) − P1 (q1 ), is a transfer from Type 2 consumers to the Þrm and

whose social value per euro is therefore λ. The optimal quantity can thus be

seen from the Þgure: it is the value of q1 for which the ratio of the vertical

distance between the lower demand and marginal cost to the vertical distance

between the two demand curves is equal to:

λ 1−α
.
1+λ α

Formally, the Þrm chooses q1 to maximize:

α [CS1 + (1 + λ) P S1 ] − (1 − α) λ [V2 (q1 ) − V1 (q1 )]

= α [(1 + λ) (V1 (q1 ) − cq1 )] − (1 − α) λ [V2 (q1 ) − V1 (q1 )] ,

where the last term measures the impact of informational rent on the Þrm’s

objective when the buyer is of Type 2 (and a constant term involving V2 (q2 ) is

omitted). The Þrst order condition can be written as:

45
 λ 1−α 0
V10 (q1 ) = c + [V2 (q1 ) − V10 (q1 )] .
1+λ α

To interpret this result and compare it to the earlier pricing formulae, it is

useful to rewrite it using the fact that the gross surplus derivative of type i is

the price Pi (q), given by the inverse demand curve, which is the demand price

at which a consumer of type i would choose to consume q units. We can thus

write the formula in terms of mark-ups as:

P1 (q1 ) − c λ 1 − α [P2 (q1 ) − P1 (q1 )]

= .
P1 (q1 ) 1+λ α P1 (q1 )

This formula allows a comparison with uniform Ramsey-Boiteux pricing.

This latter pricing induces consumers to buy less than the Þrst-best optimal

quantity in order to generate a producer surplus and pricing above marginal

cost. The possibility of generating such a surplus depends inversely on the price

elasticity of demand: the more elastic demand is, the less feasible it is to raise

price without causing too large of a drop in quantity. With nonlinear pricing,

decreasing the quantity bought by low-demand consumers has a completely

different role. It allows informational rent to high demand consumers to be

reduced, while still motivating them to reveal their demand. The distortion is

larger when the impact on rent is higher, as measured by P2 (q1 ) − P1 (q1 ), and

1−α
the relative share of high-demanders is higher, as measured by α . Because

the informational rent is larger as q1 increases, the high-demanders would prefer

the distortion in q1 to be as small as possible; their welfare is highest when they

only constitute a small fraction of total buyers.

46
 Although it is clear that the price per unit will differ between groups of

buyers, nothing guarantees that it is decreasing in quantity. In practice, we

often see that those who buy more beneÞt from larger discounts. For example,

membership programs or frequent traveler programs allow those who travel more

to pay less.

The current analysis does not immediately apply to this type of rebate, in

part because we have not explicitly considered the possibility that large buyers

could buy up several packages targeted to small buyers (see Alger, 1999, for more

on this topic). More generally, the empirical implications of nonlinear pricing

merit a deeper study (recent work includes Cohen, 2002, Ivaldi and Martimort,

1994, Leslie, 2003, and McManus, 2007).

4.3 Multiple qualities and discrimination

While there are certain types of nonlinear pricing in transportation, another

frequent practice is to offer several classes of service on the same voyage. This

practice goes back a long way, and Jules Dupuit gives an extremely perceptive

analysis in his 1849 article.

It is clear that by increasing the number of classes indeÞnitely,

consumers could be made to pay all of the utility that they get from

using the railway. However, to do that requires distinguishing among

consumers who get different utility from their transportation, and

make them voluntarily sort themselves into one or another price cat-

egory. However, this is a great difficulty, which gives rise to a whole

47
host of measures which are generally quite poorly understood by the

public.

Hence, a good many people, on seeing travelers in third class,

traveling without a roof over the carriage, on poorly upholstered seats,

they denounce the barbarity of the railways. It would cost very little,

they say, to put some meters of leather and kilos of horse-hair [on

the seats], and it is beyond greed to withhold them.

It is not because of the several thousand francs which they would

have to spend to cover the third class wagons or to upholster the

benches that a particular railway has uncovered carriages and wooden

benches; it would happily sacriÞce this for the sake of its popularity.

Its goal is to stop the traveler who can pay for the second class

trip from going third class. It hurts the poor not because it wants

them to personally suffer, but to scare the rich. The proof is that

if today the State were to say to this railroad: here are one hundred

thousand francs to improve your carriages, this subsidy would be

certainly refused... improving the third class carriages could reduce

revenues by two million francs and ruin the company.

Thus, it is for the same reason that companies, after being cruel

to travelers in third class and miserly for those in second, become

prodigious for those in Þrst class. After having refused the poor

some necessary comforts, they give the rich what is superßuous.

48
 Walras (1875/1897) had a similar view of the logic which guided the pricing

of French railways in the middle of the nineteenth century.

French railroads ask, respectively, for 10 c. in Þrst class, 7.5 c.

in second class, and 5.5 c. in third class; but they put 24 travelers

in a Þrst class carriage, 30 in a second class one, and 40 in third

class. They also use less comfortable seats, etc. (... )

As it happens, the railroads consider, rightly or wrongly, that the

average price of 7.66 c., which is close enough to the 7.5 c. which is

the second class fare, as being the price of maximal proÞt; but they

do not want to miss the opportunity of getting more from travelers

who are prepared to pay more, nor to refuse to get less from travelers

who decided not to pay too much.

When people earlier rejoiced at the rule of 1857-1858 that required

the companies to put windows in third class, and when today they

want heating in winter, and they complain about the harshness of

the railroads, they are not understanding the key motivation.

If the third class carriages were comfortable enough that many

second class travelers and some of the Þrst class ones would go there,

net total product, as we understand it from the theory of monopoly,

would fall. That is all there is to it.

The railroads only have third class carriages to avoid missing

out on a large number of travelers who, rather than pay the Þrst or

second class prices, would have continued to travel in stage coaches.

49
 Following Dupuit and Walras, the choice of the level of comfort in the differ-

ent seating classes is effectively driven by the desire to make people pay a price

corresponding to their willingness to pay, and to avoid personal arbitrage from

those from whom the company wants to extract a high fare. Such arbitrage is

discouraged by introducing sufficiently high comfort differences between classes.

It is interesting to draw a parallel between the arguments of these authors and

our results for nonlinear pricing. We have shown that potential personal ar-

bitrage by high-demanders leads the Þrm to introduce a difference between the

quantities it proposes, a difference which is larger than it would be under per-

fect discrimination. Dupuit and Walras suggest that railways use a similar

logic when they choose the level of comfort. We now show that there is a

formal equivalence between these two forms of price discrimination. To show

the analogy between discrimination based on different qualities and nonlinear

pricing of a homogeneous good, consider the following model due to Mussa and

Rosen (1978). We also use an a graphical approach that is an alternative to

the previous one that used demand curves.

Suppose that there are two types of users, l and h, differing in their willing-

ness to pay for quality. Assume that their willingness to pay for an extra unit of

quality is θh and θl , respectively, where θh > θl . Now let q be the quality level

of the service proposed rather than the quantity sold, as it was under nonlinear

pricing. The gross surplus of a buyer of type θi consuming services of quality q

is then given by Vi (q) = θi q. Each customer only wants one unit of the good,

and her willingness to pay is her gross surplus. Her net surplus for a price T is

50
θi q − T . The marginal cost of service c (q) is increasing and strictly convex.

The combinations of quality q and tariff T that give an equal level of utility

to consumer i are illustrated in Figure 6. These indifference curves are straight

lines with slope θi (with quality on the horizontal axis and euro price on the

vertical). The indifference curve through the origin corresponds to zero surplus

and is indicated in the sequel by ICi,0 . Higher surplus levels correspond to

indifference curves farther to the right. The producer surplus for one unit of

service (one trip) from quality q sold at tariff T is given by T − c (q). If the Þrm

knew the value of θi for each traveler, it could perfectly discriminate and leave

no surplus to the buyer. The corresponding producer surplus is θi q − c (q), and

the Þrm will therefore choose the quality qi∗ that maximizes this (with social

surplus being simply the producer surplus weighted by 1 + λ).

Figure 6 shows the indifference curve for zero surplus and the marginal cost

curve as a function of quality. The optimal quality is that which maximizes the

distance between the indifference curve and marginal cost, which gives a value

qi∗ where the curves have the same slope: formally, θi = c0 (qi∗ ) (the vertical

distance between the two curves is always less than that between the indifference

curve and the tangent to marginal cost at qi∗ ). With perfect discrimination,

the optimal quality will then optimize the willingness to pay for quality with

marginal cost of increasing quality.

If the Þrm does not know the consumers’ willingness to pay for quality then,

if it were to offer the two Þrst-best optimal qualities, it would necessarily leave

some surplus to the high consumer type. Figure 7 shows that the combination

51
(qh∗ , Th∗ ) is to the left of the high-type user’s indifference curve, denoted by ICh,l ,

which goes through (ql∗ , Tl∗ ), and she therefore prefers this latter combination.

In order to motivate her not to choose the low quality, the Þrm must quote

her a price such that she is on the indifference curve ICh,l . Just as under

nonlinear pricing, it is potential personal arbitrage by the high-demanders which

constitutes the constraint for the Þrm. Again, because the Þrm is not concerned

with arbitrage by the low-demanders, it can extract their full surplus so that the

optimal combination will be on the indifference curve ICl,0 . High-demanders

enjoy an informational rent of R, which is measured as the vertical difference

between the indifference curve ICh,0 and the indifference curve ICh,l (which is

the difference between the perfectly discriminatory tariff and what they actually

have to pay).

The social surplus associated with high-demanders is then the Þrst-best so-

cial surplus. Although this rent does not depend on the high quality, the

optimal high quality is qh∗ , as with perfect discrimination. On the other hand,

as is seen from Figure 7, the Þrm wants to decrease the low quality below ql∗ .

This permits the Þrm to reduce the informational rent because the indifference

curve ICh,l shifts left while, for a small change, the loss of social surplus as-

sociated with low-demanders is negligible because ql∗ constitutes its maximum.

This model thus conÞrms Dupuit’s (1849) intuition. The comfort in third class

is deliberately reduced to dissuade travelers who are ready to pay for higher

levels of comfort from traveling at the cheaper fares. Today’s economy class

air travelers might sympathize. As Tirole (1988, 1993) points out, the comfort

52
offered to the Þrst class travelers actually is not “superßuous” because it is the

level chosen under perfect discrimination; contrary to what Dupuit and Walras

thought, it is only by adjusting the lower quality that the Þrm discourages per-

sonal arbitrage. There is therefore a perfect analogy between this model and

that of nonlinear pricing, and the results are the same (under reinterpretation).

Only passengers whose willingness to pay is high can retain some surplus, and

this surplus increases as the fraction of high demanders decreases.

One variant of this model explored by Chander and Leruth (1989), and

particularly relevant for transportation, supposes that the quality of comfort

class decreases in the number of users as it becomes more congested. The Þrm

then chooses two different prices, with the cheaper class having a lower quality

just because it attracts more travelers. The two classes in the Parisian Metro

until the 1980s give a striking illustration of this type of strategy.39 Second class

carriages only differed from Þrst class by their color, but fares were lower.40

The modern day counterpart to the insights of Dupuit and Walras can be

found without much difficulty in air travel. What was true in nineteenth century

train carriages sometimes seems to be not very far off from what is found on

modern economy class ßights. Airlines could scarcely charge such premiums

for Þrst and business class travel if economy class were more comfortable.41
3 9 Another illustration is that of a toll road with a parallel freeway to the same destination
(see de Palma and Lindsey, 2000).
4 0 This system was abolished by Charles Fiterman, Communist Transportation Minister in

the Þrst government of the Union of the Left.

4 1 In practice, the problem faced by the airlines is very complicated because the information

that it has about passengers is very sparse and evolves over time. See the sub-section on
Yield Revenue Management above.

53
5 Conclusion

We have focused on discriminatory pricing for a single Þrm (without rivals) and

suggested how this analysis is relevant for transportation. We analyzed a public

Þrm under the assumption that public funds are valued more than consumer

surplus, in order to account for inefficiencies in raising tax revenue. ProÞt

maximization arises as a limit case when λ goes to inÞnity.

This theoretical framework enables us to highlight the similarity between the

pricing problem of a public Þrm and that of a private Þrm. However, restricting

this analysis to monopolies is more restrictive for the private sector. Although

several transport modes are effectively monopolies, such as the railways and

urban transportation systems, other sectors have several competing carriers,

most notably the airlines. We now give a brief review of oligopoly competition42 .

For the discussion that follows, we consider private proÞt maximizing Þrms

and we let λ = 0, so that consumer surplus and producer surplus are equally

weighted.43

There is relatively little work on price discrimination under competition.

As regards discrimination without personal arbitrage, standard oligopoly the-

ory fairly quickly delivers some main conclusions. For one good sold to several

groups the model most directly comparable with monopoly is Cournot’s frame-

work, where Þrms choose outputs and price equates aggregate output with the

quantity demanded. If each Þrm has the same marginal costs ci for serving
4 2 Fora review of the state of the art
on price discrimination under oligopoly, see Stole (2007), and Armstrong and Vickers (2001).
4 3 The Competition Authority of the European Union seems to put more weight on consumer

surplus.

54
market i, Lerner’s formula becomes:

pi − ci 1 1
= ,
pi ni |η i (pi )|

where ni is the number of active Þrms in market i. If Þrms have different costs,

Lerner’s formula becomes:

pi − cm 1 1
= ,
pi ni |ηi (pi )|

where

ni
1 X
cm = cj
ni j=1

is the average cost for serving market i. This formula is directly comparable to

the monopoly one. If two groups differ by elasticity of demand, we again Þnd

that the price is higher for inelastic demand. However, all prices go to marginal

cost when the number of Þrms becomes large, and competition eliminates price

differences. These results can also be applied when Þrms discriminate with

different products, in which case the marginal cost differs from one market to

another. A richer framework can be analyzed by supposing that there is also

differentiation within each market. One common formulation for demand for

differentiated products is founded on discrete choice models, which are described

in greater detail in other chapters of this book. One interesting feature of these

models is that it is easy to introduce new features, such as quality differences or

congestion, into the consumer preferences which are at the heart of the model.44
4 4 Anderson, de Palma, and Thisse (1992a) provides background for this model.

55
 The multinomial logit model is a particularly useful formulation. If prices

are the strategic variable, the equilibrium price of Þrm j in market i is

µi
pij = cij + ,
1 − Dij

where Dij is the equilibrium demand addressed to Þrm j in market i and depends

on prices set by all active Þrms in market i. Anderson and de Palma (2001)

show that this price equilibrium has several intuitive properties. For example, if

consumers have a higher tendency to buy the product, then prices will be high

and the differences will be large across variants. Furthermore, the more Þrms

there are, or the more similar they are (from the point of view of a group of

consumers), the lower prices will be and the more similar they will be across

Þrms.

It can also be understood within this framework why introducing compe-

tition in a market can actually lead discriminating Þrms to exacerbate price

differences across different services. Borenstein and Rose (1994) note that fol-

lowing airline deregulation in the United States, price differences increased. The

theoretical explanation that they propose is that those travelers who are willing

to pay more for a trip (hence, those for whom the reservation value is high) are

also those who are the most loyal to a particular airline, which translates to a

greater effective heterogeneity between products. When competition is intro-

duced into the marketplace, the difference between the high business class fare

and the low economy class fare is ampliÞed by the differential intensity of com-

petition at the two service levels. Competition is more intense for the economy

56
segment.45 However, more recent work by Gerardi and Shapiro (2007) Þnds

that competition decreases the power of an airline to price discriminate. Price

dispersion within the airline industry falls as competition increases, especially

on routes where consumers have relatively heterogeneous elasticities of demand.

Looking at patterns within different groups of air travelers, Bilotkach (2005)

looks at across-airlines differences in economy class fares for the London-New

York market, aimed at different consumer types. He Þnds that fares targeted to

business ßiers differ across airlines, while fares targeted to leisure ßiers do not.

One issue in the economics of third-degree price discrimination under com-

petition concerns the level of producer surplus when discrimination is not always

possible (it could, for example, be illegal). Several authors came to the same

conclusion: in contrast to monopoly, it may be that proÞts are lower when Þrms

can discriminate. Hoover (1948, p. 57) anticipated this result in the context of

spatial discrimination.

The difference between market competition under f.o.b. pricing (with

strictly delineated market areas) and under discriminatory delivered

pricing is something like the difference between trench warfare and

guerrilla warfare. In the former case all the Þghting takes place

along a deÞnite battle line; in the second case the opposing forces are

intermingled over a broad area.

This indicates that the implications of discrimination for proÞts can be very
4 5 Borenstein and Rose base their theoretical arguments on Borenstein (1985) and Holmes
(1989).

57
different, depending on the degree of competition. However, it remains true

that welfare falls following the introduction of discrimination if total output

does not rise.

In contrast to the analysis of third-degree discrimination, the study of dis-

crimination under personal arbitrage in oligopoly is complicated and there are

relatively few general contributions. Notable exceptions include Champsaur

and Rochet (1989), Ivaldi and Martimort (1994), Stole (1995), Armstrong and

Vickers (2001), and Rochet and Stole (2001). The latter two articles show

that if duopolists offer relatively close services, nonlinear pricing can lead to

two-part tariff with zero proÞts for each Þrm (an extension of the Bertrand

Paradox). Rochet and Stole (2001) also show that if the services offered are

more differentiated, equilibrium pricing resembles the monopoly case.

Theoretical progress on these issues is quite tricky, but it is important to

continue to develop the theoretical framework which will allow empirical studies

to examine discrimination with personal arbitrage under oligopoly.

References

[1] Alger, Ingela (1999). Consumer Strategies Limiting the Monopolist’s Power:

Multiple and Joint Purchases. RAND Journal of Economics, 30, 736-757.

[2] Anderson, Simon P. (1989). Socially Optimal Spatial Pricing. Regional Sci-

ence and Urban Economics, 19, 60-86.

58
[3] Anderson, Simon P. and André de Palma (1989). Spatial Price Discrimina-

tion with Heterogeneous Products. Review of Economic Studies, 55, 573-

592.

[4] Anderson, Simon P. and André de Palma (2001). Product Diversity in

Asymmetric Oligopoly: Is the Quality of Consumer Goods Too Low? Jour-

nal of Industrial Economics, 49, 113-135.

[5] Anderson, Simon P., André de Palma and Brent Kreider (2001). Tax Inci-

dence In Differentiated Products Oligopoly. Journal of Public Economics,

81, 173-192.

[6] Anderson, Simon P., André de Palma and Brent Kreider (2001). Efficiency

of Indirect Taxes Under Imperfect Competition. Journal of Public Eco-

nomics, 81, 231-251.

[7] Anderson, Simon P., André de Palma and Jacques-François Thisse (1989).

Spatial Price Policies Reconsidered. Journal of Industrial Economics, 38,

1-18.

[8] Anderson, Simon P., André de Palma and Jacques-François Thisse (1992a).

Discrete Choice Theory of Product Differentiation, MIT Press, Cambridge,

Mass.

[9] Anderson, Simon P., André de Palma and Jacques-François Thisse (1992b).

Social Surplus and ProÞtability under Different Spatial Pricing Policies.

Southern Economic Journal, 58, 934-949.

59
[10] Anderson, Simon P. and Maxim Engers (2007a). Consumer Surplus, Article

for International Encyclopedia of the Social Sciences, 2nd Ed.

[11] Anderson, Simon P. and Maxim Engers (2007b). Producer Surplus, Article

for International Encyclopedia of the Social Sciences, 2nd Ed.

[12] Anderson, Simon P. and Victor Ginsburgh (1999). International Pricing

with Costly Consumer Arbitrage. Review of International Economics, 7,

126-139.

[13] Anderson, Simon P. and Luc Leruth (1993). Why Firms May Prefer not to

Price Discriminate via Mixed Bundling. International Journal of Industrial

Organization, 11, 49-61.

[14] Anderson, Simon P. and Régis Renault (2003a). Efficiency and Surplus

Bounds in Cournot Competition. Journal of Economic Theory, 113, 253-

264.

[15] Anderson, Simon P. and Régis Renault (2003b). Second degree price dis-

crimination à la Pigou. Working Paper, Université de Cergy-Pontoise.

[16] Anderson, Simon P. and Yves Schneider (2008) Price Competition with

Revenue Management and Costly Consumer Search. Mimeo, University of

Virginia.

[17] Anderson, Simon P. and Jacques-François Thisse (1988). Price Discrimi-

nation in Spatial Competitive Markets. European Economic Review, 32 ,

578-590.

60
[18] Anderson, Simon P. and Wesley W. Wilson [2008]. Spatial Competition,

Pricing, and Market Power in Transportation: A Dominant Firm Model.

Journal of Regional Science, 48(2), 367-397.

[19] Armstrong, Mark and John Vickers (2001). Competitive price discrimina-

tion. RAND Journal of Economics, 32, 579-605.

[20] Babu, Nahata, Krzysztof Ostaszewski, and P. K. Sahoo, (1990). Direction

of price changes in third-degree price discrimination. American Economic

Review, 80, 1254-1258.

[21] Ballard, Charles L., John B. Shoven and John Whalley (1985). General

Equilibrium Computations of the Marginal Welfare Costs of Taxes in the

United States. American Economic Review, 75, 128-38.

[22] Bilotkach, Volodymyr (2005). Understanding Price Dispersion in the Air-

line Industry: Capacity Constraints and Consumer Heterogeneity. Ad-

vances in Airline Economics, volume 1 (ed. Darin Lee).

[23] Boîteux, Marcel (1956). Sur la gestion des monopoles publics astreints à

l’équilibre budgétaire. Econometrica, 24, 22-40.

[24] Borenstein, Severin (1985). Price discrimination in free entry markets.

RAND Journal of Economics, 16, 380-397.

[25] Borenstein, Severin (1991). Selling costs and switching costs: explaining

retail gasoline margins. RAND Journal of Economics, 22, 354-369.

61
[26] Borenstein, Severin and Nancy Rose (1994) Competition and price disper-

sion in the U.S. airline industry. Journal of Political Economy, 102, 653-683.

[27] Brander, J. A. and A. Zhang (1990). “Market Conduct in the Airline In-

dustry: An Empirical Investigation,” RAND Journal of Economics, 21,

567-583. (Forthcoming in K. Button and P. Nijkamp (eds.), Classics in

Transport Analysis, Edward Elgar Publishing Ltd.)

[28] Brown S. and Sibley, David (1986). The theory of public utility pricing.

Cambridge University Press, Cambridge, England.

[29] Champsaur, Paul and Jean-Charles Rochet (1989). Multiproduct

Duopolists. Econometrica, 57, 533-557.

[30] Chander, Parkash and Luc Leruth (1989). The Optimal Product Mix for

a Monopolist in the Presence of Congestion Effects: A Model and Some

Results. International Journal of Industrial Organization, 7, 437-449.

[31] Clerides, Sofronis (2004). Price discrimination with differentiated products:

deÞnition and identiÞcation. Economic Inquiry, 42, 402-412.

[32] Cohen, Andrew (2002) Package Size and Price Discrimination in the Paper

Towel Market. Working Paper. University of Virginia.

[33] Corts, Kenneth S. (1998). Third-Degree Price Discrimination in Oligopoly:

All-Out Competition and Strategic Commitment. RAND Journal of Eco-

nomics, 29, 306-323.

62
[34] Cournot, Augustin (1838). Recherches sur les principes mathématiques de

la théorie des richesses. L. Hachette, Paris.

[35] De Palma, André and Robin Lindsey (1998). Information and Usage of Con-

gestible Facilities Under Different Pricing Regimes, The Canadian Journal

of Economics, 31, 666-692.

[36] De Palma, André and Robin Lindsey (2000). Private toll roads : competi-

tion under various ownership regimes. The Annals of Regional Science, 34,

13-35.

[37] Dupuit, Jules (1849). De l’inßuence des péages sur l’utilité des voies de com-

munication. Annales des ponts et chaussées, 17, mémoires et documents,

207.

[38] Gerardi, Kristopher and Shapiro, Adam Hale (2007). The Effects of Com-

petition on Price Dispersion in the Airline Industry: A Panel Analysis.

Federal Reserve Bank of Boston Working Paper 07-7.

[39] Gerstner, Eitan (1986). Peak Load Pricing in Competitive Markets. Eco-

nomic Inquiry, 24, 349-361.

[40] Hausman, Jerry A. and James Poterba (1987). Household Behavior and

the Tax Reform Act of 1986. Journal of Economic Perspectives, 1, 101-119.

[41] Holahan, William L. (1975). The welfare effects of spatial price discrimina-

tion. American Economic Review, 65, 498-503.

63
[42] Holmes, Thomas J. (1989). The effects of third degree price discrimination

in oligopoly. American Economic Review, 79, 244-250.

[43] Hoover, E. M. (1937). Spatial price discrimination. Review of Economic

Studies, 4, 182-191.

[44] Hoover, E. M. (1948). The location of economic activities. New York,

McGraw-Hill.

[45] Hotelling, Harold (1938). The general welfare in relation to problems of

taxation and of railway and utility rates. Econometrica, 6, 242-269.

[46] Ivaldi, Marc and David Martimort (1994). Competition under Nonlinear

Pricing. Annales d’Economie et de Statistique, 34, 71-115.

[47] Laffont, Jean-Jacques, and Jean Tirole (1993). A theory of incentives in

procurement and regulation. MIT Press, Cambridge, Mass.

[48] Lederer, Phillip J. and Arthur P. Hurter, (1986) Competition of Þrms:

discriminatory pricing and location. Econometrica, 54, 623-40.

[49] Leslie, Phillip, (2003) Price Discrimination in Broadway Theatre. RAND

Journal of Economics, 35, 520-541.

[50] Levinson, D. and A. Odlyzko (2007). Too expensive to meter: The

inßuence of transaction costs in transportation and communication.

http://rational.ce.umn.edu.

[51] Lott, John R. and R. D. Roberts (1991). A guide to the pitfalls of identifying

price discrimination. Economic Inquiry, 29, 14-23.

64
[52] McAfee, R. Preston, Hugo M. Mialon, and Sue H. Mialon (2006). Does

large price discrimination imply great market power? Economic Letters, 3,

360-367.

[53] McManus, Brian (2007) Nonlinear Pricing in an Oligopoly Market: the

Case of Specialty Coffee. RAND Journal of Economics, 38(2), 512-532.

[54] Meade, James (1944). Price and output policy of state enterprise. Economic

Journal, 54, 321-328.

[55] Mohring, Herbert [1976] Transportation Economics. Cambridge, MA,

Ballinger.

[56] Mougeot, Michel and Florence Naegelen, (1994) La discrimination par les

prix. Economica, Paris, France.

[57] Mussa, Michael, and Sherwin Rosen, (1978). Monopoly and product qual-

ity, Journal of Economic Theory, 18, 301-317.

[58] Nahata, Babu, Krzysztof Ostaszewski, and P. K. Sahoo (1990). Direction of

Price Changes in Third-Degree Price Discrimination. American Economic

Review, 80, 1254-1258.

[59] Oi, Walter, (1971). A Disneyland dilemma: Two-part tariffs for a Mickey

Mouse monopoly, Quarterly Journal of Economics, 85, 77-90.

[60] Phlips, Louis (1983). The economics of price discrimination. Cambridge

University Press, Cambridge, UK.

65
[61] Pigou, Arthur Cecil (1938). The economics of welfare. Macmillan and Co.,

London. 4th edition, 1932, reprinted 1938.

[62] Ramsey, Frank (1927). A contribution to the theory of taxation. Economic

Journal, 37, 47-61.

[63] Rietveld, Piet, and Roberto Roson [2002] Direction dependent prices in

public transport: A good idea? The back haul pricing problem for a mo-

nopolistic transport Þrm. Transportation, 29, 397-417.

[64] Rochet, Jean-Charles and Lars A. Stole (2001). Nonlinear pricing with

random participation. Review of Economic Studies, 69, 277-311.

[65] Schmalensee, Richard (1981). Output and welfare implications of monop-

olistic third degree price discrimination. American Economic Review, 71,

242-247.

[66] Seade, J. E. (1985). ProÞtable cost increases and the shifting of taxation:

Equilibrium responses of markets in oligopoly. University of Warwick Dis-

cussion paper no. 260.

[67] Shepard, Andrea (1991). Price discrimination and retail conÞguration.

Journal of Political Economy, 99, 30-53.

[68] Sinsou, Jean-Paul (1999). Yield & Revenue Management: Optimisation

de la recette dans les transport de passagers. Les presses de l’institut du

transport aerien, Paris.

66
[69] Stigler, George (1963). A note on block booking. Supreme Court Review.

Reprinted in Organization of Industry (University of Chicago Press).

[70] Stole, Lars A. (1995). Nonlinear pricing and oligopoly. Journal of Eco-

nomics and Management Studies, 4, 529-562.

[71] Stole, Lars A. (2007). Price Discrimination in Competitive Environments.

Handbook of Industrial Organization, Vol. 3, Chapter 34. Eds. Mark Arm-

strong and Robert H. Porter, North-Holland, Amsterdam.

[72] Tirole, Jean (1988). The Theory of Industrial Organization. MIT Press,

Cambridge, Mass.

[73] Tirole, Jean (1993). La théorie de l’organisation industrielle, volume 1.

Editions Economica, Paris, France.

[74] Thisse, Jacques-François and Xavier Vives (1988). On the strategic choice

of spatial price policy. American Economic Review, 78, 122-137.

[75] Varian, Hal (1989). Price discrimination. Handbook of Industrial Organi-

zation, Vol. 1, Chapter 10. Eds. Richard Schmalensee and Robert Willig,

North-Holland, Amsterdam.

[76] Varian, Hal (2000) Introduction à la microéconomie. DeBoeuck Université.

[77] Vickrey, William S. (1969). Congestion theory and transport investment.

American Economic Review (Papers and Proceedings), 59, 251-261.

[78] Walras, Léon (1897). L’Etat et les chemins de fer. Manuscript written 1875.

Revue du Droit Public et de Science Politique. Mai-juin et juillet-août 1897.

67
[79] Wilson, Robert (1992). Non-linear pricing. Oxford University Press, Ox-

ford, England.

[80] Yin, Xiangkang (2004). Two part tariff competition in duopoly. Interna-

tional Journal of Industrial Organization, Elsevier, 22(6), 799-820.

68