( PDF ) Price discrimination without single-crossing

Download PDF

ads:

IMPA

Doctor of Philosophy Dissertation

Price Discrimination without the

Single-Crossing

Sergei Vieira Silva

Supervisor: Alo´ısio Araujo Co-Supervisor: Humberto Moreira

Rio de Janeiro, 2008

ads:

Livros Grátis

http://www.livrosgratis.com.br

Milhares de livros grátis para download.

ads:

Contents

Contents i

1 Nonlinear Pricing 3

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.3 Unidimensional Screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3.1 Unidimensional Screening under the SMC . . . . . . . . . . . . . . . . . . 6

1.3.2 Unidimensional Screening without the SMC . . . . . . . . . . . . . . . . . 11

1.4 Multidimensional Screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.5 Final Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2 Nonlineard Pricing Without a Single-Crossing Tariﬀ 21

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.2.1 The Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.3 Incentive Compatibility Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.4 The Monopolist’s Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.4.1 Single-Crossing Tariﬀ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.4.2 Non Single-Crossing Tariﬀ . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.5 An Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3 Bidimensional Screening 47

ii CONTENTS

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3.3 Discrete-Continuous Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3.3.1 Optimality Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.3.2 An Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.4 Bicontinuous Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.4.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

Bibliography 69

Acknowledgements

Em primeiro lugar, eu gostaria de agradecer a meus pais pela forma¸c˜ao, pelo apoio e pelo

incentivo em minhas escolhas e decis˜oes. Estendo o agradecimento a toda a minha fam´ılia que

sempre foi muito presente e importante em minha vida.

Agrade¸co o Professor Alo´ısio Araujo, que aceitou a ´ardua tarefa de me orientar, e com o

qual tive o privil´egio de ter excelentes conversas e discuss˜oes sobre os mais variados assuntos,

inclusive sobre economia matem´atica. Agrade¸co tamb´em o Professor Humberto Moreira, que

foi meu co-orientador e meu primeiro professor no curso de ver˜ao em Economia Matem´atica no

IMPA, pela dedica¸c˜ao e energia que empenhou nesse projeto.

Gostaria tamb´em de agradecer ao IMPA que me proporcionou um excelente ambiente

acadˆemico, uma estrutura impec´avel e de bom gosto al´em de todo o apoio necess´ario para

realizar pesquisa. Agrade¸co a todos os seus professores com destaque `aqueles que me minis-

traram cursos e em especial ao saudoso Professor Carlos Isnard. Agrade¸co tamb´em a todos os

funcion´arios do IMPA, muito prestativos e competentes, essenciais no bom funcionamento de

toda a estrutura do instituto.

Gostaria de agradecer aos colegas do IMPA pela companhia, entusiasmo e por compartil-

harem comigo o gosto pela matem´atica.

Por ﬁm gostaria de agradecer ao CNPq, que forneceu o apoio ﬁnanceiro sem o qual essa

tarefa seria muito mais dif´ıcil.

Chapter 1

Nonlinear Pricing

1.1 Introduction

Nonlinear pricing schemes are widely used in many markets such as mobile telephony, ﬁxed

telephony, electricity and gas supply, postal services, air and railroad transport, cable TV and

internet, hardware and software, cars. For this reason, it has attracted a lot of attention from

both theoretical and applied economists.

Our research is about the design of an optimal nonlinear pricing mechanism for a monopolist

who has incomplete information about the characteristics of the customers. Although there are

theoretical results available in the literature, some of them are valid only under restrictive

conditions, for instance the Spence and Mirrlees condition (SMC). Our goal is to take some

steps in relaxing some of these conditions.

For a deeper understanding of these problems, however, much more have to be done. We

believe that a more general treatment requires new mathematical tools which are still to be

developed. This seems to be especially true when we are in a multidimensional context.

In this chapter we make a brief survey of the literature and the objective is to place our

research in the proper context. It is organized as follows. In Section 2 we present the model

we are using. Then, in Section 3 we study the problem in a one-dimensional context. First

we consider it under the Spence and Mirrlees condition (SMC), and we describe three diﬀerent

approaches for solving it. Then we present a model relaxing the SMC. In Section 4 we present

4 CHAPTER 1. NONLINEAR PRICING

some results from the literature of multidimensional screening problems. Finally, in Section 5

we present the directions undertaken by our research.

1.2 Model

The customer has a quasi-linear utility function depending on his type θ ∈ Θ ⊆ R

, which is

private information:

V (q, t, θ) = v(q, θ) − t,

where q ∈ Q ⊆ R

is the customer’s decision and t is the monetary transference.

The ﬁrm is a proﬁt-maximizing monopolist producing any q ∈ Q ⊆ R

and incurring in

a cost C(q). It does not observe θ and has a prior distribution P over Θ. His revenue is also

quasi-linear:

Π(q, t) = t − C(q).

We formulate the monopolist’s problem as choosing a decision function q : Θ → Q and a

monetary transfer t : Θ → R that solves:

max

{q(·),t(·)}



{t(θ) − C(q(θ))}dP (θ), (Π)

subject to the individual rationality constraints:

v(q(θ), θ) − t(θ) ≥ 0 ∀θ ∈ Θ, (IR)

and the incentive compatibility constraints:

θ ∈ arg max



∈Θ

{v(q(θ



), θ) − t(θ



)}, ∀θ ∈ Θ. (IC)

The allocation rule represented by the pair (q, t) is called a contract. We say that a contract

(q, t) is incentive compatible when it satisﬁes the IC constraints. The decision function q is

implementable if there exists a monetary transfer t such that (q, t) is incentive compatible.

Hammond [11] and Guesnerie [9] have showed that a contract is incentive compatible if and

only if there exists a nonlinear tariﬀ T : Q → R such that:

q(θ) ∈ arg max

q∈Q

{v(q, θ) − T (q)}, ∀θ ∈ Θ.

1.2. MODEL 5

This is known as the ‘Taxation Principle’ and it easily follows from the fact that t depends on

θ only through the decision q(θ).

One of the greatest diﬃculties related to the monopolist’s problem is how to deal with the IC

constraints. In general, the binding IC constraints may be determined only endogenously which

makes it a rather diﬃcult task. There are some technical results about the characterization of

incentive compatible decisions in Rochet [25] which uses the concept of cyclical monotonicity

and Carlier [6] who uses abstract convexity concepts. Without imposing further structure on

the utility function v(q, θ), however, no much can be said about implementability.

In the unidimensional (N=M=1) case, for instance, there is a condition that provides a very

simple characterization of implementable decisions:

Deﬁnition 1. (The Spence and Mirrlees Condition.) Suppose that v : Q × Θ → R is

twice diﬀerentiable. Then v satisﬁes SMC if and only if only one of the two possibilities below

is valid:







qθ

> 0 on Q × Θ (CS

)

qθ

< 0 on Q × Θ (CS

−

(SMC)

The SMC, also known as the single-crossing condition, is an standard assumption in the

literature of monopolistic screening. Under the SMC a decision function is implementable if and

only if it is monotonic.

The monotonicity of the decision function is related to the ﬁrst- and

second-order conditions of the customer’s maximization problem (IC). Consequently, under the

SMC the ﬁrst- and second-order conditions are necessary and suﬃcient for implementability.

In the multidimensional case, when the monopolist sells a single good and the customer has

multidimensional type, there is the generalized single-crossing condition deﬁned by McAfee and

McMillan [17]:

Deﬁnition 2. (The Generalized Single-Crossing Condition.) Suppose that v : Q×Θ → R

is twice diﬀerentiable. Then v satisﬁes the generalized single-crossing property if for all q, θ

and θ

there exists λ(q, θ

, θ

) > 0 such that:

(q, θ

) − v

(q, θ

) = λ∇

(q, θ

) · (θ

− θ

). (GSC)

See Rochet [25].

See Guesnerie and Laﬀont [10].

6 CHAPTER 1. NONLINEAR PRICING

Under this condition, we also have that the ﬁrst- and second-order conditions are necessary

and suﬃcient for implementability. This condition is too restrictive as it imposes that the set

of types who choose the same allocation is a linear subspace. In the third chapter, we ﬁnd an

example of optimal contract that does not satisfy the GSC.

1.3 Unidimensional Screening

In this section we present some methods for solving the monopolist’s problem (Π) under the

assumption of a single characteristic (N = 1) and a single instrument (M = 1). The type space

is Θ = [θ, θ]. The common strategy for all these methods is to derive a relaxed problem for the

monopolist. Then, if we ﬁnd a solution for the relaxed problem that satisﬁes the constraints of

the original problem, then this solution is also a solution for the original problem.

1.3.1 Unidimensional Screening under the SMC

We present three approaches for monopolist’s problem under the (SMC). We assume the fol-

lowing conditions:

H1. v(q, θ) ∈ C

H2. v

< 0 and v

> 0,

H3. v

> 0 and v

qθ

> 0,

H4. C(0) = 0, C



(q) > 0 and C



(q) < 0.

We also assume the (CS

). In this case, a decision q is implementable if and only if it is

increasing.

Mussa and Rosen’s approach:

One of the seminal papers in the ﬁeld of nonlinear pricing is by Mussa and Rosen [20].

They study monopoly pricing problems involving a quality-diﬀerentiated spectrum of goods.

We are going to reproduce their derivation for the monopolist’s problem. Suppose that (q, t) is

incentive compatible and deﬁne the informational rent by:

V (θ) = v(q(θ), θ) − t(θ).

When the signal of v

is not constant and/or when the individual rationality (IR) constraints are type

dependent, we have countervailing incentive problem. This problem was studied by Jullien [13]. In Araujo and

Moreira [1] there is a discussion about the possibility of countervailing incentives due to the absence of the SMC.

The term increasing refers to both weakly and strictly increasing functions.

1.3. UNIDIMENSIONAL SCREENING 7

If we plug V into the monopolist’s objective function of problem (Π), we get:



{v(q(θ), θ) − C(q(θ)) − V (θ)}dθ. (1.1)

By the envelope theorem (see Milgrom and Segal [18]), we have that:



(θ) = v

(q(θ), θ).

Using assumption H2 and the fundamental theorem of calculus, we get:

V (θ) =



(q(θ), s)ds. (1.2)

Plugging equation (1.2) into equation (1.1) and using an integration by parts we ﬁnally get:



{v(q(θ), θ) − C(q(θ)) +

P (θ) − 1

p(θ)

(q(θ), θ)}p(θ)dθ. (1.3)

Let us deﬁne:

f(q, θ) = v(q(θ), θ) − C(q(θ)) +

P (θ) − 1

p(θ)

(q(θ), θ). (1.4)

Dropping the global IC constraints, we can deﬁne a relaxed version of the monopolist’s problem:

max

{q(·)}



f(q(θ), θ)p(θ)dθ. (Π

)

This is a classical problem in the calculus of variations.

The Euler’s equation gives the necessary

condition for an extremum of problem (Π

(q, θ) = 0. (1.5)

Let us denote the solution of equation (1.5) by q

(θ). If q

(θ) is increasing, then it is the

solution of the monopolist’s problem (Π). If q

(θ) is not increasing, then we must use the

ironing procedure on q

(θ) to ﬁnd the solution.

In the Figure 1.1 we illustrate the ironing procedure. Notice that q

(θ) is decreasing in

the interval [θ

, θ

]. The procedure consists in optimally choosing an interval I = [θ

, θ

]

such that the optimal decision q

∗

(θ) = q

, ∀θ ∈ I. In this solution, we have a bunching of

By assumption H2, V is increasing and we can set V (θ) = 0, eliminating the IR constraints.

We suggest Petrov [22] as a basic reference for calculus of variations and optimal control.

8 CHAPTER 1. NONLINEAR PRICING

∗

(θ )

Figure 1.1: The ironing procedure.

customers choosing the same decision q

. Using optimal control, the optimal bunching can be

characterized by:



, θ)p(θ)dθ = 0.

Goldman, Leland and Sibley’s approach:

The second approach we are going to present was introduced by Goldman, Leland and

Sibley [8]. They study nonlinear pricing through quantity discounts. Remember that under the

(CS

), an incentive compatible decision q must be increasing so we can deﬁne its pseudo-inverse

ψ(q). It is called the type assignment function, as it assigns quantities q to the type θ = ψ(q)

choosing q. Notice that ψ(q) must be also increasing. In this approach, the monopolist’s

objective function will depend on ψ(q). Then the monopolist has to optimally choose ψ(q).

Consider an incentive compatible contract (q, t). Using the taxation principle, we have that

T (q(θ)) = t(θ). Using the fundamental theorem of calculus, and supposing that T (0) = C(0) =

we can write the monopolist’s objective function of problem (Π) as:



{T (q(θ)) − C(q(θ))}p(θ)dθ =



q(θ)



(q) − C



(q)}p(θ)dqdθ,

and using Fubinis’ theorem and the customer’s ﬁrst-order condition:



q(θ)



(q) − C



(q)}p(θ)dqdθ =



ψ(q)

(q, ψ(q)) − C



(q)}p(θ)dθdq.

Let us deﬁne:

F (q, ψ) =



(q, ψ) − C



(q)}p(θ)dθ.

This is a simplifying hypothesis.

1.3. UNIDIMENSIONAL SCREENING 9

Dropping the global IC constraints, now related to the monotonicity of the type assignment

function ψ(q), we can deﬁne a relaxed version of the monopolist’s problem:

max

{ψ(·)}



F (q, ψ(q))dq. (Π

)

The Euler’s equation for the problem (Π

) is given by:

(q, ψ(q)) = 0. (1.6)

Let us denote by ψ

(q) the solution of equation (1.6). If ψ

(q) is increasing, then it is

the solution of the monopolist’s problem (Π). If ψ

(q) is not increasing, then the bunching

condition for the decision function q translates into an optimal jump condition for the optimal

type assignment function ψ

∗

F (q

, ψ

∗

+)) = F (q

, ψ

∗

−)) = max

{θ∈[θ,θ]}

F (q

, θ).

∗

−)

Figure 1.2: The optimal jump condition.

We will see that this approach provides a very convenient way to deal with the monopolist’s

problem without the SMC. The diﬀerence is that we will have two type assignment functions,

because the decision functions may not be monotonic.

Demand Proﬁle approach:

The demand proﬁle approach was introduced by Brown and Sibley [5]. The derivation we

do here, however, can be found in Wilson [29], which provides a detailed description of this

This result was established by Nˆldeke and Samuelson [21].

10 CHAPTER 1. NONLINEAR PRICING

approach, with many examples and applications to nonlinear pricing via quantity discounts.

This approach results in an optimization problem depending on the marginal tariﬀ T



(q).

The demand proﬁle is a function representing the fraction of customers purchasing at least

q units when the monopolist chooses the tariﬀ T :

N(T (·), q) = Pr



θ ∈ [θ, θ]|∃x ∈ arg max

x≥0

{v(x, θ) − T (x)} and x ≥ q



In general N(T (·), q) may depend on the entire tariﬀ T (·) and this could make the problem

rather complicated. However, it is possible to simplify the demand proﬁle by assuming an

aditional condition.

Deﬁnition 3. (Single-Crossing Tariﬀ.)

A tariﬀ is said to be single-crossing at q if, for

any θ such that v

(q, θ) ≥ T



(q), v



, θ) ≥ T



) for q



≤ q. If T (q) is single-crossing for all

q, we term it a single-crossing tariﬀ.

When we have a single-crossing tariﬀ, the demand proﬁle depends only on the marginal

tariﬀ T



(q). In this case it is denoted by N(T



(q), q) and we have that:

N(T



(q), q) = Pr



θ ∈ Θ|v

(q, θ) ≥ T



(q)



The derivation of the monopolist’s optimization problem resembles the previous one from Gold-

man, Leland and Sibley. Using the fundamental theorem of calculus, Fubini’s theorem and

supposing that T (0) = C(0) = 0, we have that:



{T (q(θ)) − C(q(θ))}p(θ)dθ =



q(θ)



(q) − C



(q)}dqp(θ)dθ



∞



(q) − C



(q)}1

{q≤q(θ)}

dqp(θ)dθ



∞



(q) − C



(q)}Pr(q ≤ q(θ))dq.

Finally, if T is a single-crossing tariﬀ, then Pr(q ≤ q(θ)) = N(T



(q), q), so we can write the

monopolist’s objective function as:



∞



(q) − C



(q)}N(T



(q), q)dq.

This deﬁnition was taken from [8].

1.3. UNIDIMENSIONAL SCREENING 11

Again, we deﬁne the relaxed version of the monopolist’s problem:

max



(·)}



∞



(q) − C



(q)}N(T



(q), q)dq. (Π

)

The Euler’s equation for the problem (Π

) is given by:

N(T



(q), q) + {T



(q) − C



(q)}N



(q), q) = 0, (1.7)

where N

denotes the derivative of N with respect to its ﬁrst variable.

We have to check ex-post whether the optimal marginal tariﬀ T



(q) is single-crossing. In

this formulation, bunching emerges as the result of jumps in the optimal marginal tariﬀ T



(q)

obtained from equation (1.7).

1.3.2 Unidimensional Screening without the SMC

Without the SMC the problem becomes rather complex. Unlike the previous case, a deci-

sion function satisfying the ﬁrst- and second-order conditions of the customer’s maximization

problem may not be implementable.

Araujo and Moreira [1] introduced a model where the SMC is relaxed to:

AM1 v

qθ

(ξ, θ) = 0 deﬁnes a decreasing function q

: θ → R

such that:

∀θ ∈ Θ v

qθ

(ξ, θ) ≥ 0 ⇔ ξ ≥ q

(θ). (1.8)

So the curve q

divides the (θ, q) plane in two regions: CS

, where v

qθ

> 0 and CS

−

, where

qθ

< 0.

Using the ﬁrst- and second-order conditions, one can show that an implementable decision

must be increasing in CS

and decreasing in CS

−

so the typical decision function will have

a U-shaped form. This shape allows that two distinct customers θ

and θ

choose the same

decision q. Analysing the ﬁrst-order conditions for the pooling types θ

and θ

, we have another

necessary condition for implementability imposing that their marginal utility from choosing q

must be the same: v

(q, θ

) = v

(q, θ

). This identity between the marginal utilities of pooling

types is called the U-condition (UC). The derivation consists in incorporating the UC into the

monopolist’s objective function. The Goldman, Leland ans Sibley’s approach is convenient for

this task, because it allows the UC to be treated separetely for each q. The diﬀerence is that

now we will have two type assignment functions ψ

(q) and ψ

(q). We are going to adapt their

12 CHAPTER 1. NONLINEAR PRICING

−

(q)

q(θ)

Figure 1.3: The decision function q and its type assignment functions.

derivation to Araujo and Moreira’s setup. Considering and implementable decision q, we can

write the monopolist’s objective function as:



f(q(θ), θ)p(θ)dθ, (1.9)

where f(q, θ) is deﬁne by equation (1.4).

Using the fundamental theorem of calculus and the Fubini’s theorem, we can write:



f(q(θ), θ)p(θ)dθ =







q(θ)

(q, θ)dq



p(θ)dθ +



f(0, θ)p(θ)dθ



{F (q, ψ

(q)) − F (q, ψ

(q))}dq +



f(0, θ)p(θ)dθ,

where F is deﬁned as:

F (q, ψ) =



(q, θ)p(θ)dθ.

So we can deﬁne a relaxed version of the monopolist’s optimization problem as:

max

{ψ

(·),ψ

(·)}



{F (q, ψ

(q)) − F (q, ψ

(q))}dq +



f(0, θ)p(θ)dθ,

s.t. v

(q, ψ

(q)) − v

(q, ψ

(q)) ≤ 0.

(Π

AMR

)

With an abuse of notation the constraint in problem (Π

AMR

) is called UC. The Euler’s

equation of problem above depends on whether the UC constraint is binding or not:

In Araujo and Moreira there are two distinct classes of solutions, one considering the constraint v

(q, ψ

(q))−

(q, ψ

(q)) ≤ 0 and the other considering v

(q, ψ

(q)) − v

(q, ψ

(q)) ≥ 0. We restrict our description to the ﬁrst

kind of solution.

We refer the reader to Araujo and Moreira [1] for more details.

1.4. MULTIDIMENSIONAL SCREENING 13

• If UC is slack and θ < ψ

(q) < ψ

(q) = θ then:

(q, ψ

(q)) = 0,

• If UC is binding and θ < ψ

(q) < ψ

(q) < θ then:

(q, ψ

(q))

qθ

(q, ψ

(q))

(q, ψ

(q))

qθ

(q, ψ

(q))

• If the condition UC is binding and θ < ψ

(q) < ψ

(q) = θ then:

(q, ψ

(q)) = v

(q, θ).

The optimal solution will be obtained as a combination of the pointwise conditions above. In

the appendix show that the tariﬀ that implements the solutions in Araujo and Moreira [1] is

single-crossing. This means that the demand proﬁle approach could have been used extended

to this non single-crossing setup. As far as we know, this methodology has never been used for

this kind of problem. Under the demand proﬁle approach, however, the drawback is that the

computations are more complicated.

1.4 Multidimensional Screening

As we saw in the previous section, the solution of the monopolist’s optimization problem is

straightforward in the unidimensional case under the SMC. In the multidimensional case, how-

ever, the characterization of the optimal contract is a very complex problem. One of the

main diﬃculties refers to the fact that the binding IC constraints may be determined only en-

dogenously and we don’t have a satisfatory way of incorporating them into the maximization

problem. So there are only partial responses for this problem.

Matthews and Moore [16] studied the monopolistic screening problem with a two-dimensional

product space, representing the quality spectrum and the warranty. The customers are rep-

resented by a discrete parameter. Their goal was to extend Mussa and Rosen [20], and they

observed that some global IC constraints may be binding in the optimal decision.

We recommend the excellent survey in Multidimensional Screening by Rochet and Stole [27].

14 CHAPTER 1. NONLINEAR PRICING

McAfee and McMillan [17] studied the problem of a multi-product monopolist who faces a

customer with multidimensional characteristics with (M ≤ N). They introduced the General-

ized Single-Crossing (GSC)

, a condition under which the ﬁrst- and second-order conditions

for the customer’s problem are necessary and suﬃcient for implementability. They also charac-

terized the optimal contract when (M = 1, N ≥ 1) under the (GSC).

Armstrong [2] studied the problem of multiproduct monopolist who faces customer’s with

multidimensional characteristics (N ≥ 1, M ≥ 1). He gave closed form solutions for some

examples in this multidimensional context. One of his main contributions was to discover the

exclusion property. This property states that the optimal contract leaves a positive measure

set of customers excluded from consumption.

Rochet and Chon

E [26] established the existence of the optimal contract and provided the

characterization in the case of a multiproduct monopolist who faces a customer with multidi-

mensional characteristics. They assume the number of products and characteristics to be the

same (N = M ≥ 1) and also that the parametrization of customer’s preferences is linear with

types. They introduced the sweeping procedure as a generalization of the ironing procedure for

dealing with bunching in the multidimensional context.

Basov [3] introduced the Hamiltonian approach as a tentative of generalizing Rochet and

Chon

E [26] to the case when the number of products and characteristics may be diﬀerent

(N ≥ 1, M ≥ 1). Later on, these techniques were extended in his book [4] to deal with more

general customer’s preferences.

Parallel to this literature, there were some works analysing the existence of a solution for the

monopolist’s maximization problem (Π). We highlight the papers by Monteiro and Page [19]

and by Carlier [6]. The former uses compactness properties resulting from budget constraints

considerations and the latter using direct methods and concepts of abstract convexity. However,

these approaches do not actually provide any recipe for the characterization of the optimal

contract.

The GSC was deﬁned on Section 1.2.

1.5. FINAL REMARKS 15

1.5 Final Remarks

Our research concerns the problem of a single-product monopolist who faces a customer with

two dimensions of heterogeneity. We assume neither the Spence and Mirrlees condition nor the

Generalized Single-Crossing.

In the second chapter we study this problem in a particular context. In our model these two

dimensions have antagonic eﬀects in the decision q. We further assume that they are dependent

in a way that we can reduce the problem into a one-dimensional screening model. The Spence

and Mirrlees condition, however, is no longer valid because there is only one dimension to cope

with the antagonic eﬀects of the original parameters.

Comparing our results with Araujo and Moreira [1] we present some advances. The optimal

decision we found Pareto dominates theirs and besides it is not implemented by a single-crossing

tariﬀ. We were able to incorporate a new global IC constraint into the monopolist’s problem,

and this gives rise to a new optimality condition for this problem. This condition is new in the

literature of monopolistic screening.

In the third chapter we extend the techniques developed by Araujo and Moreira [1] to a

bidimensional screening context. This chapter is divided in two main sections contemplating

two cases.

In the ﬁrst one, called discrete-continuous, one of the customer’s characteristic is discrete

and the other is continuous. We derived optimality conditions analogous to the ones in [1] and

then we show how to use them to construct an optimal contract.

In the second one, called bicontinuous, both customers’ characteristics are continuous. This

case is close to McAfee and McMillan [17] generalization of Laﬀont, Maskin and Rochet [14].

The diﬀerence is that we did not assume their Generalized Single-Crossing (GSC). We solved

an example where the (GSC) is not valid.

16 CHAPTER 1. NONLINEAR PRICING

Appendix

Our goal is to prove that the optimal decisions in Araujo and Moreira [1] is implemented by

a single-crossing tariﬀ. In all that follows R(q) denotes the range of the function q. We begin

deﬁning the pseudoinverse of a monotonic function:

Deﬁnition 4. Consider a monotonic function q : [θ, θ] → R along with its upper contour set

q(·)

(q) = {θ ∈ [θ, θ] : q(θ) ≥ q}.

If q is decreasing, we deﬁne ψ

: R → [θ, θ] as:

(q) =







sup S(q), if S(q) = ∅,

θ, if S(q) = ∅.

If q is increasing, we deﬁne ψ

: R → [θ, θ] as:

(q) =







inf S(q), if S(q) = ∅,

θ, if S(q) = ∅.

In both cases, ψ

and ψ

are monotonic and they are denoted the pseudoinverses of q.

Then we establish the following result:

Lemma 1. Consider a monotonic function q : [a, b] → R along with its pseudoinverse ψ(·).

For all x in the range of q, we have that:

(i) q(θ) ≥ x ⇔ θ ≥ ψ(x), if q(·) is increasing;

(ii) q(θ) ≥ x ⇔ θ ≤ ψ(x), if q(·) is decreasing.

Proof. We will only prove (i):

(⇐) It follows from the increasingness of q.

(⇒) It follows from the deﬁnition of the pseudoinverse.

We can do the same for proving (ii).

As a direct consequence of Lemma 1, we can state:

1.5. FINAL REMARKS 17

Lemma 2. Consider q : [a, b] → R. Suppose that there exists θ

∈ (a, b), point of continuity of

q, such that the restrictions, q

= q|

[a,θ

]

and q

= q|

[θ

,b]

, are monotonic, with q

decreasing

and q

increasing. Let ψ

and ψ

be their pseudoinverses. We have:

(i) If x ∈ R(q

) ∩ R(q

), then q(θ) ≥ x ⇔ θ ≤ ψ

(x) or θ ≥ ψ

(x),

(ii) If x ∈ R(q

) ∩ R(q

)

, then q(θ) ≥ x ⇔ θ ≥ ψ

(x),

(iii) If x ∈ R(q

)

∩ R(q

), then q(θ) ≥ x ⇔ θ ≤ ψ

(x).

−

Figure 1.4: The (DD) decision function.

We divide the optimal decisions in Araujo Moreira in two kinds; the ﬁrst one is denoted the

discontinuous decision (DD) and the second one is denoted the continuous decision (CD).

The discontinuous decision (DD) is characterized by:

(1a) q(θ) < q(θ),

(1b) ∃ θ

∈ (θ, θ) continuity point of q, such that:

(θ, q(θ)) ⊆ CS

−

if θ ≤ θ ≤ θ

(θ, q(θ)) ⊆ CS

if θ

≤ θ ≤ θ,

= q|

[θ,θ

]

is decreasing,

= q|

[θ

,θ]

is increasing,

(1c) q is discontinuous at θ

∈ (θ

, θ),

(1d) ψ

: [q(θ

), q(θ)] → [θ, θ

] is continuous,

18 CHAPTER 1. NONLINEAR PRICING

(1e) ψ

: [q(θ

), q(θ)] → [θ

, θ] is continuous,

(1f) v

(q, ψ

(q)) = v

(q, ψ

(q)) ∀q ∈ [q(θ

), q(θ)],

(1g) v

(q, θ) ≤ v

(q, ψ

(q)) ∀q ∈ (q(θ), q(θ)].

−

Figure 1.5: The (CD) decision function.

The second is denoted the continuous decision (CD), characterized by:

(2a) q(θ) > q(θ),

(2b) ∃ θ

∈ (θ, θ) continuity point of q, such that:

(θ, q(θ)) ⊆ CS

−

if θ ≤ θ ≤ θ

(θ, q(θ)) ⊆ CS

if θ

≤ θ ≤ θ,

= q|

[θ,θ

]

is decreasing,

= q|

[θ

,θ]

is increasing,

(2c) q is continuous,

(2d) ψ

: [q(θ

), q(θ)] → [θ, θ

] is continuous except on q(θ),

(2e) ψ

: [q(θ

), q(θ)] → [θ

, θ] is continuous except on q(θ),

(2f) v

(q, ψ

(q)) = v

(q, ψ

(q)) ∀q ∈ [q(θ

), q(θ)],

(2g) v

(q, ψ

(q)) ≤ v

(q, θ) ∀q ∈ (q(θ), q(θ)].

1.5. FINAL REMARKS 19

We have the following result for the optimal decisions in Araujo and Moreira: For a DD

solution, we have the followinf result.

Proposition 1. Let q be a decision and T : R(q) → R be the tariﬀ that implements it. If q is

(DD) or (CD), then T is a single-crossing tariﬀ.

Proof. We will only prove the case (DD). The other case is analogous. If we consider the agent’s

maximization problem, the ﬁrst-order conditions gives:



(q) = v

(q, ψ

(q)).

Assumption (AM1), items (1f) and (1g) imply that:

∀q ∈ [q(θ

), q(θ)] v

(q, θ) ≥ T



(q) ⇔ θ ≤ ψ

(q) or θ ≥ ψ

(q),

∀q ∈ (q(θ), q(θ)] v

(q, θ) ≥ T



(q) ⇔ θ ≥ ψ

(q).

Finally, by Lemma 2, we conclude that:

q(θ) ≥ q ⇔ v

(q, θ) ≥ T



(q)

And the result follows.

Chapter 2

Nonlineard Pricing Without a

Single-Crossing Tariﬀ

2.1 Introduction

Nonlinear pricing schemes are widely used in many markets such as mobile telephony, ﬁxed

telephony, electricity and gas supply, postal services, air and railroad transport, cable TV

and internet. For this reason, it has attracted a lot of attention over the past decades from

economists. One of the seminal papers in this ﬁeld is by Mussa and Rosen [20], who analysed

nonlinear prices for the provision of quality-diﬀerentiated goods. Other contributions were made

by Goldman, Leland and Sibley [8], who studied price discrimination via quantity discounts

and by Maskin and Riley [15] who observed that all these models belong to the same class of

principal-agent problems.

On the other hand, the demand proﬁle approach is an alternative formulation for the mo-

nopolistic screening problem, introduced in the literature by Brown and Sibley [5] and then

thoroughly expounded in Wilson [29]

. The basis for this approach is the requirement that the

marginal tariﬀ T



(q) cuts the customer’s demand curve once from below. When satisﬁed, we

have that the customer’s maximization problem is quasiconcave so its demand behavior can be

In a survey on multidimensional screening, Rochet and Stole [27] make a distinction between the approach

adopted by Mussa and Rosen, and Maskin and Riley, called parameter utility approach and the other one called

demand proﬁle approach.

22 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

deduced completely from the marginal beneﬁt of consumption. In Goldman, Leland and Sibley,

a tariﬀ satisfying this requirement is called a single-crossing tariﬀ, a terminology that we will

also use in this chapter.

All these methods work ﬁne and equivalently when the Spence and Mirrlees condition (SMC)

is satisﬁed. The main reason is that it provides a natural ordering for the customer’s demand

curves. As a consequence, the monopolist is able to induce any monotonic decision for the

customers by selecting a compatible tariﬀ.

We study a nonlinear pricing model where the monopolist has incomplete information about

the slope and the intercept of the customer’s demand curve, represented by α and β. In our

model these two dimensions have antagonic eﬀects in the decision q which should increase with

α and decrease with β. We further assume that these two dimensions are dependent in a way

that we can reduce it into a one-dimensional screening model with parameter θ. Now a bigger θ

will represent at the same time as a bigger α and β and the monopolist does not know which of

the antagonic eﬀects will dominate the other. By having only one dimension summarizing the

eﬀects of those two parameters, it ends up in a model where the SMC is not satisﬁed anymore.

Without the SMC the problem becomes rather complex, in part due to the absence of this

natural ordering. As a consequence the decision q may be nonmonotonic. Besides the local

incentive compatibility conditions may not be suﬃcient as new global conditions arise. This

problem without the SMC was studied by Araujo and Moreira [1]. Although they relax the

Spence and Mirrlees condition, the optimal contract proposed there is compatible with the

demand proﬁle approach, because it is implemented by a single-crossing tariﬀ and consequently

the customers face a quasiconcave problem.

Our goal is to show that in our monopolistic screening model without the SMC, the contract

resulting from a single-crossing tariﬀ, compatible with the demand proﬁle approach may be

suboptimal for the monopolist. We present another contract that provides a larger proﬁt for

the monopolist. This contract is not compatible with the demand proﬁle approach because

the tariﬀ is not single-crossing. Moreover the optimal contract is discontinuous and induces a

separation in the quality spectrum between two groups: one who buys low quality goods, and

the other who buys high quality goods. The tariﬀ should be adjusted in a way that one group

does not envy the other.

2.2. MODEL 23

The chapter is organized as follows. In Section 2.2 we deﬁne the model we are using. Then,

in Section 2.3 we present some results related to the incentive compatibility condition. In

Section 2.4 we derive the monopolist’s optimization problem using an approach adapted from

Goldman, Leland and Sibley, we establish the pointwise optimality conditions for the monop-

olist’s problem and ﬁnally we solve some examples. In Section 2.6 we present the conclusions.

All proofs are given in the Appendix.

2.2 Model

We use the Principal-Agent framework to analyse the monopolistic screening problem. In this

model, each customer has a quasi-linear preference:

V (q, t, θ) = v(q, θ) − t,

where t represents the monetary transfer. The parameter θ is a random variable with a positive

and continuous density representing the customer’s type. The ﬁrm is a proﬁt-maximizing

monopolist which can produce any quality q ∈ Q ⊆ R

incurring in a cost C(q). Q represents

the quality spectrum. The monopolist’s revenue is given by:

Π(q, t) = t − C(q).

Using the ‘Revelation Principle’

the monopolist’s problem can be stated as choosing the

allocation rule (q, t) : Θ → R

× R that solves:

max

{q(·),t(·)}



Π(q(θ), t(θ))p(θ)dθ, (2.1)

subject to the individual-rationality constraints:

v(q(θ), θ) − t(θ) ≥ 0 ∀θ ∈ Θ, (IR)

and the incentive compatibility constraints:

θ ∈ arg max



∈Θ

{v(q(θ



), θ) − t(θ



)}, ∀θ ∈ Θ. (IC)

The ‘Revelation Principle’ has been enunciated in Gibbard [7].

24 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

Remark 1. The ‘Taxation Principle’

states that any allocation (q, t) satisfying the IC con-

straints can be implemented by a nonlinear tariﬀ T : Q = q(Θ) → R where:

T (q(θ)) = t(θ), ∀θ ∈ Θ.

2.2.1 The Model Setup

We consider a monopolist who produces a single good and is uncertain about the slope and

the intercept of the customer’s demand curves. The utility function of each customer is

parametrized according to:

v(q, α, β) = αq − β

and the individual inverse demand curve is given by:

p(q) = α − βq.

The cross derivatives of v satisfy: v

qα

> 0 and v

qβ

≤ 0. These conditions imply that an incentive

compatible decision q must increase with α and decrease with β (see Rochet [25]). Hence the

eﬀects of these parameters on q are antagonic.

We analyse the case when the random variables α and β are dependent in the following

manner:







α = θ,

β = bθ

with θ uniformly distributed on [θ, θ] = [a, a + 1]. Now we have then only one parameter θ to

cope with the eﬀects of α and β. A bigger θ means bigger α and β, so the decision will increase

or decrease depending on which eﬀect is dominant. This results in the following setup:











v(q, θ) = θq − bθ

C(q) = c

and we assume that a, b and c are nonnegative constants.

The inverse demand function is now given by

p(q) = θ − bθ

This principle can be found in Guesnerie [9], Hammond [11] and Rochet [24].

2.2. MODEL 25

and the elasticity of demand is

(q, θ) = 1 −

bθq

The demand function for types θ

and θ

are drawn in Figure 2.1. For θ-type customer, the

ratio q = 1/(bθ) measures the market size and the intercept θ represents the minimum price

such that this customer’s demand is zero.

bθ

p(q, θ

)

p(q, θ

)

Figure 2.1: Demand curves for θ

and θ

customers.

When b = 0 the speciﬁcation of this model is the same as Mussa and Rosen [20], but when

b > 0 the Spence and Mirrlees condition

is no longer satisﬁed and the solution of v

qθ

(q, θ) = 0

determines a decreasing function given by

(θ) =

2bθ

. (2.2)

This function gives a limit curve that divides the (θ, q) plane in two regions denoted by CS

where v

qθ

> 0 and CS

−

, where v

qθ

< 0. The consequence is the lack of an ordering of the

demand curves as we can see in Figure 2.1, where we have the following inequalities for the

marginal utility: v

, θ

) < v

, θ

) and v

, θ

) > v

, θ

Unlike the SMC case, it is not possible to know a priori which customer will consume

more, as this is determined endogenously as a response to the nonlinear tariﬀ adopted by the

monopolist. The decision now may be nonmonotonic, increasing in CS

and decreasing in CS

−

and these conditions suggest a bell-shaped form for the decision function q.

26 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

2.3 Incentive Compatibility Conditions

Without the Spence and Mirrlees condition there is not a simple rule for characterizing an incen-

tive compatible allocation and the problem becomes very complex. There are some necessary

conditions, however, that should be taken into account.

In all that follows, we assume that the allocation rule (q, t) is bounded and incentive com-

patible. The informational rent V : Θ → R

is given by:

V (θ) = v(q(θ), θ) − t(θ), (2.3)

and T : q(Θ) → R is the tariﬀ resulting from the ‘Revelation Principle’.

We begin by establishing a basic property regarding the Lipschitz continuity of both T and

V .

Lemma 3. (Diﬀerentiability Condition.) The tariﬀ T and the informational rent V are

Lipschitz continuous.

Notice that Lemma 3 guarantees that both T and V are absolute continuous and then a.e.

diﬀerentiable and now we are able to deduce the following envelope conditions involving their

derivatives.

Lemma 4. (Envelope Conditions for V and T .)

(i) If V is diﬀerentiable at θ ∈ int(Θ) and q ∈ q(θ), then



(θ) = v

(q, θ). (2.4)

(ii) If T is diﬀerentiable at q ∈ q(θ) ∩ int(q(Θ)), then



(q) = v

(q, θ). (2.5)

Notice that Lemma 4 (ii) can be understood as the ﬁrst order condition of the θ-customer

maximization problem:

max

q∈Q

{v(q, θ) − T (q)}.

The Spence and Mirrlees condition states that the cross derivative v

qθ

does not change its signal. When

satisﬁed the incentive compatibility is equivalent to the monotonicity of the decision function q. See Rochet [25],

for instance.

2.3. INCENTIVE COMPATIBILITY CONDITIONS 27

With the SMC a decision is incentive compatible if and only if is monotonic. However,

without the SMC we may have a nonmonotonic incentive compatible decision. The next result

is related to the local monotonicity conditions on the decision q.

Lemma 5. (Local Monotonicity Conditions.) Suppose that q is continuous at θ

. Then:

(i) If q(θ

) > q

(θ

) then q is increasing at θ

(ii) If q(θ

) < q

(θ

) then q is decreasing at θ

where q

is deﬁned in equation (2.2).

The new possibilities for q coming from Lemma 5 motivates us to deﬁne the bell-shaped

decisions.

Deﬁnition 5. (Bell-Shaped Decisions.) The decision q : Θ → R

is bell-shaped if there

exists a continuity point θ

∈ Θ such that the restrictions:

[θ,θ

]

is increasing and q|

[θ

,θ]

is decreasing.

For a bell-shaped decision, we may have a discrete pooling situation, when types θ

and

choose the same decision q. Using Lemma 4 (ii), we can easily deduce a new necessary

condition for incentive compatibility relating the marginal valuations of the pooled types.

Corollary 1. (Bell-Shaped Condition.) Suppose that T is diﬀerentiable at q ∈ q(θ

) ∩q(θ

)

an interior point of Q = q(Θ). Then

(q, θ

) = v

(q, θ

). (BS)

This marginal utility equality in Corollary 1 will be denoted BS-condition, and it says that

pooled types must pay the same marginal tariﬀ for the chosen decision q.

28 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

2.4 The Monopolist’s Problem

Suppose ﬁrst that the alocation rule (q, t) is incentive compatible. Let us now deduce the

monopolist’s maximization problem, using the same derivation as Mussa and Rosen [20]. From

the deﬁnition of the informational rent V (equation (2.3)) we can write the monetary transfer

as t(θ) = v(q(θ), θ) − V (θ) and then substitute it in equation (2.1). The result is the following

problem:

max

{q(·)}



{v(q(θ), θ) − C(q(θ)) − V (θ)}dθ. (2.6)

Now using Lemma 4 and an integration by parts procedure, we can rewrite the problem above

max

{q(·)}



f(q(θ), θ)dθ, (Π

)

where f(q, θ) is given by:

f(q, θ) = v(q(θ), θ) − C(q(θ)) + (θ − a − 1)v

(q(θ), θ).

This is called the relaxed version of the monopolist’s maximization problem, because the

monotonicity constraints are not considered in this problem. The inconvenience of the derivation

above is that in many situations the solution of Problem (Π

) is far from being incentive

compatible. One reason is that it does not consider the (BS) condition. We are going to derive

the monopolist’s optimization problem in a way that the (BS) necessary condition given by

Corollary 1 can be taken into account. We need the following deﬁnitions:

Deﬁnition 6. Consider a monotonic function q : [θ, θ] → R along with its upper contour set

S(q) = {θ ∈ [θ, θ] : q(θ) ≥ q}.

If q is decreasing, we deﬁne ψ

: R → [θ, θ] as:

(q) =







sup S(q), if S(q) = ∅,

θ, if S(q) = ∅.

If q is increasing, we deﬁne ψ

: R → [θ, θ] as:

(q) =







inf S(q), if S(q) = ∅,

θ, if S(q) = ∅.

In both cases, ψ

and ψ

are monotonic and they are denoted the pseudoinverses of q.

We are implicitily assuming that V



(θ) ≥ 0. For a more general case we refer to Jullien [13].

2.4. THE MONOPOLIST’S PROBLEM 29

For a bell-shaped decision, using a procedure similar to Goldman, Leland and Sibley [8], we

can reformulate the problem (Π

). First we use the fundamental theorem of calculus to write:



f(q(θ), θ)dθ =



q(θ)

(q, θ)dqdθ.

Then, we use Fubini’s theorem, and the following problem results:

max

{ψ

,ψ

}



{F (q, ψ

(q)) − F (q, ψ

(q))}dq +



f(q, θ)dθ.

(2.7)

F is given by the following equation:

F (q, ψ) =



(q, θ)dθ, (2.8)

and besides ψ

is the pseudoinverse of the increasing part of q and ψ

is the inverse of the

decreasing part of q. They are denoted type assignment functions, because they relate q with the

type who purchases it. Notice that the integration variable on the left hand side of equation (2.7)

is q, and now we can treat the constraint imposed by the condition (BS) condition for each q

separately.

2.4.1 Single-Crossing Tariﬀ

First let us deﬁne a single-crossing tariﬀ

Deﬁnition 7. (Single-Crossing Tariﬀ.) A tariﬀ is said to be single-crossing at q if, for any

θ such that v

(q, θ) ≥ T



(q), v



, θ) ≥ T



) for q



≤ q. If T (q) is single-crossing for all q, we

term it a single-crossing tariﬀ schedule.

If the tariﬀ is single-crossing and continuous, then the customer’s maximization problem is

quasiconcave.

Using the concept of a single-crossing tariﬀ, we will be able to deﬁne the following optimiza-

tion problem for the monopolist, when q(θ) < q(θ)











max

{ψ

,ψ

}



{F (q, ψ

(q)) − F (q, ψ

(q))}dq +



f(q, θ)dθ,

s.t. v

(q, ψ

(q)) − v

(q, ψ

(q)) ≤ 0.

(Π

)

This deﬁnition was taken from Goldman, Leland and Sibley [8].

The other case, when q(θ) > q(θ), is treated analogously, and we only have to reverse the inequality in

problem Π

30 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

Notice that the problem Π

is coherent with a single-crossing tariﬀ T . In this case, when

q(θ) < q(θ) and q < q(θ), then v

(q, ψ

(q)) ≤ v

(q, ψ

(q)), because T



(q) = v

(q, ψ

(q)) and

(q, θ) ≥ T



(q).

Once we ﬁnd the optimal type assignment functions ψ

and ψ

we can invert them and ﬁnd

the optimal decision q.

Now let us establish the optimality conditions for Π

Theorem 1. The solution of Π

is characterized by the following conditions:

(i) If the condition BS is not binding and θ < ψ

(q) < ψ

(q) = θ then:

(q, ψ

(q)) = 0, (2.9)

(ii) If the condition BS is binding and θ < ψ

(q) < ψ

(q) < θ then:

(q, ψ

(q))

qθ

(q, ψ

(q))

(q, ψ

(q))

qθ

(q, ψ

(q))

, (2.10)

(iii) If the condition BS is binding and θ < ψ

(q) < ψ

(q) = θ then:

(q, ψ

(q)) = v

(q, θ). (2.11)

−

Figure 2.2: A typical solution for Π

In Figure 2.2, we show the typical solution for the optimization problem Π

resulting from

Theorem 1. For q < q

, we use condition (i), for q ∈ (q

, q

) we use condition (iii) and ﬁnally for

2.4. THE MONOPOLIST’S PROBLEM 31

q ∈ (q

, q) we use condition (ii). As we can see there is a problem related with the monotonicity

condition resulting from Lemma 5.

We have two alternatives for ﬁxing this lack of monotonicity. The ﬁrst one is a vertical

ironing compatible with a single-crossing tariﬀ T and the next proposition shows how restrictive

this condition is.

Proposition 2. If the tariﬀ T is single-crossing then the decision q is convex valued.

As a consequence of Proposition 2, if the decision q jumps at θ

, then the marginal tariﬀ

for q ∈ [q(θ

−), q(θ

+)] is given by T



(q) = v

(q, θ

) and this aﬀects the decision for customers

with types θ close to θ. The reason is simple: with a single-crossing tariﬀ, the marginal

net beneﬁt determines the customer’s choice and when a jump occurs, the marginal tariﬀ is

automatically deﬁned for all the intermediate quantities and of course this may aﬀect some

customer’s decisions.

−

q(θ

−)

Figure 2.3: The vertical ironing procedure.

The decision function that emerges from this vertical ironing is given by:

(θ) =











(θ) , if θ < θ

(θ) , if θ

≤ θ ≤

(θ, θ

) , if

≤ θ ≤

θ,

(2.12)

32 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

where q

and q

result from Theorem 1 (i) and (ii), respectively. On the other hand q

results

from solving the equation:

(q, θ

) = v

(q, θ).

Notice that the customers with type θ ≥

choose q ∈ [q(θ

−), q(θ

+)] and their decision is

aﬀected by the fact that for this q the marginal tariﬀ is T



(q) = v

(q, θ

). The function q

captures this eﬀect. Finally, we get

from solving q

(θ

) = q

(

The point θ

must be chosen optimally and is characterized by the following:

Theorem 2. The ﬁrst order condition for the optimal θ

is given by:

f(q

(θ

), θ

) − f(q

(θ

), θ

) =



(θ, θ

), θ)

∂q

∂θ

(θ, θ

)dθ. (2.13)

Notice that this vertical ironing ﬁxes the monotonicity problem by keeping the type assign-

ment functions ψ

and ψ

continuous. This kind of solution was also proposed in Araujo and

Moreira [1]. It can also be obtained by using the demand proﬁle approach of Wilson, but the

computations are more diﬃcult than the one we presented here.

2.4.2 Non Single-Crossing Tariﬀ

Now we propose an alternative for this vertical ironing where ψ

and ψ

may be discontinuous.

This is illustrated in Figure 2.4. More than that, our procedure will result in a non single-

crossing tariﬀ.

We will impose the global incentive compatibility constraint relating θ

and θ customers.

We are asking that the θ-type customer does not envy θ

-type customer. In other words we

impose:

v(q(θ), θ) − t(θ) ≥ v(q(θ

), θ) − t(θ

). (2.14)

The following result will be useful to restate the condition above

Lemma 6. Consider an allocation rule (q, t) : Θ → R

× R. If (q, t) is incentive compatible

then:



θ)

qθ

(˜q,

θ)d˜qd

θ ≥ 0, ∀θ,

θ ∈ Θ. (2.15)

This condition is actually equivalent to incentive compatibility, as shown by Araujo and Moreira [1].

2.4. THE MONOPOLIST’S PROBLEM 33

q(θ)

(q)

q(θ)

Figure 2.4: An alternative ironing procedure.

Notice that Lemma 6 relates incentive compatibility with an area condition represented by

a double integral.

The same procedure used in the proof of Lemma 6 allows us to write the inequality (2.14)

as:



θ)

q(θ

)

qθ

(˜q,

θ)d˜qd

θ ≥ 0. (2.16)

Notice that:



θ)

q(θ

)

qθ

(˜q,

θ)d˜qd

θ =



qθ

(˜q,

θ)d˜qd

θ +



θ)

qθ

(˜q,

θ)d˜qd

θ. (2.17)

Using Fubini’s theorem, we can write the second integral on the right hand side of equation

(2.17) as:



θ)

qθ

(˜q,

θ)d˜qd

θ =



(q)

qθ

(˜q,

θ)d

θd˜q



(q, ψ

(q)) − v

(q, ψ

(q))}dq.

However v

(q, ψ

(q)) = v

(q, ψ

(q)) for q ∈ (q

, q] which implies that this part is null.

Using Fubini’s theorem again we can write the ﬁrst integral on the right hand side of equation

(2.17) as:



qθ

(˜q,

θ)d˜qd

θ =



(q, ψ

(q)) − v

(q, θ

)}dq. (2.18)

34 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

and the condition in equation (2.16) will be given by:



(q, ψ

(q)) − v

(q, θ

)}dq ≥ 0. (ISO)

In the interval [q

, q

], we will optimally choose ψ

(q) such that the condition (ISO) is fulﬁlled.

So we have the following isoperimetric problem:











max

{ψ

}



{F (q, ψ

(q)) − F (q, θ

)}dq

s.t.(ISO).

(Π

ISO

)

Theorem 3. The solution of Π

ISO

is characterized by the following condition:

(q, ψ

(q)) + λv

qθ

(q, ψ

(q)) = 0, (2.19)

where λ is chosen to satisfy the condition (ISO) with equality.

Now we use conditions given by Theorem 1 and Theorem 3 to build a candidate for the

monopolist’s optimization problem. The cited theorems give the pointwise conditions for the

three pieces that constitute the candidate we are proposing. Looking again at Figure 2.4, we

see that the only change is for q ∈ [q

, q

], where we are doing this alternative ironing. The

values q

, q

and θ

are endogenously determined as a result of an optimization process. Let us

describe the type assignment functions ψ

and ψ

(q) =











(q) , if q

≤ q ≤ q

(q) , if q

≤ q ≤ q

(q) , if q

≤ q ≤ q,

(2.20)

where we have that ψ

= θ, ψ

(q) is given by equation (2.19) and ψ

is given by equation

(2.10).

(q) =











(q) , if q ≤ q ≤ q

(q) , if q

≤ q ≤ q

(q) , if q

≤ q ≤ q,

(2.21)

where ψ

= θ

, we deﬁne ψ

from equation (2.9) and ψ

from equation (2.10).

Now the monopolist’s problem will depend on the variables q

, q

and θ

, and we can write

its expected proﬁt as Π(q

, q

, θ

) where q

, q

and θ

satisfy the condition (ISO). Our last step

is to optimally choose these parameters:

2.5. AN EXAMPLE 35

Theorem 4. The optimal choices for the parameters q

, q

and θ

are characterized by:

(i)

F (q

, ψ

)) − F (q

, ψ

)) + F (q

, θ

) − F (q

, ψ

))

, ψ

)) − v

, θ

)

= λ,

(ii)

F (q

, ψ

)) − F (q

, θ

) + F (q

, ψ

)) − F (q

, ψ

))

, θ

) − v

, ψ

))

= λ,

(iii)



(q, θ

) + λv

qθ

(q, θ

)}dq = 0,

where λ is the Lagrangian multiplier associated with the isoperimetric constraint.

Remark 2. Considering the isoperimetric condition, we have four equations and four param-

eters: q

, q

, θ

and λ. We get the optimal candidate by solving this system of equations.

2.5 An Example

This example illustrates the diﬀerence between the two ironing procedures used for ﬁxing the

monotonicity problem. The ﬁrst one is compatible with the demand proﬁle approach because

the resulting tariﬀ is single-crossing. The second one uses the isoperimetric condition and results

in a non single-crossing tariﬀ.

Let us choose the following values for the parameters in our model: a = 2, b = 1 and c = 6.

With these parameters, we have that θ is uniformly distributed in [2, 3] and:











v(q, θ) = θq − θ

C(q) = 3q

Using Theorem 1, we can derive the solution for Π

(q) =











3q+1−

√

−9q

−3q+1

, if

≤ q ≤

102



15 +

√



1−3q

, if

102



15 +

√



≤ q ≤



9 +

√



3q−

√

−24q

, if



9 +

√



≤ q ≤

√

(q) =











3, if

≤ q ≤

102



15 +

√



3, if

102



15 +

√



≤ q ≤



9 +

√



3q+

√

−24q

, if



9 +

√



≤ q ≤

√

As we can see in Figure 2.5 ψ

is nonmonotonic. We are going to derive the solutions using

the two ironing procedures developed before.

36 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

(q)

Figure 2.5: The solutions ψ

and ψ

for Π

a) Standard Ironing Procedure:

The ﬁrst step is to write the solution depending on the parameter θ

, as we did in equation

(2.12):

(θ) =











(θ) =

3−2θ

−3θ

+6θ−6

, if θ < θ

(θ) =

3θ+

√

3θ

−12

2(3θ

+6)

, if θ

≤ θ ≤

(θ, θ

) =

θ+θ

, if

≤ θ ≤ θ,

where

= 2θ

−

√



− 4.

Using Theorem 2, we can ﬁnd the optimal θ

∗

and

∗

≈ 2.164282,

∗

≈ 2.89596.

The decision fucntion resulting from this ironing procedure is denoted q

(θ). We can ﬁnd

the tariﬀ T

(q) that implements the decision by integrating the marginal tariﬀ given by:



(q) = v

(q, ψ

(q)),

where ψ

is the pseudoinverse of q

(θ)|

[θ,

√

. In Figure 2.6, we plot the decision function and

the marginal tariﬀ. Notice that the marginal tariﬀ is continuous so the tariﬀ is continuously

diﬀerentiable.

The monopolist’s expected proﬁt from q

(θ) is:

Π(θ

) = 0.200272 (2.22)

2.5. AN EXAMPLE 37

−

q(θ

−)

√



(q)

q(θ

−)

(θ)

Figure 2.6: The decision function and the marginal tariﬀ.

b) Isoperimetric Ironing Procedure:

Now we use the isoperimetric procedure. Let us describe the type assignment functions ψ

and ψ

according to equations (2.20) and (2.21):

(q) =











(q) = 3 , if

≤ q ≤ q

(q) =

1−q(λ−3)+

√

(λ

−6λ−9)q

+(λ−3)q+1

, if q

≤ q ≤ q

(q) =

3q+

√

−24q

, if q

≤ q ≤

√

(q) =











(q) =

3q+1−

√

−9q

−3q+1

, if

≤ q ≤ q

(q) = θ

, if q

≤ q ≤ q

(q) =

3q−

√

−24q

, if q

≤ q ≤

√

Using Theorem 4 and the isoperimetric condition, we get the following values:

≈ 0.183112, q

≈ 0.202431, θ

≈ 2.201349, λ ≈ 0.067841.

When we invert the type assignment functions, we get the following decision function:

NSC

(θ) =











3−2θ

−3θ

+6θ−6

, if θ ≤ θ

3−2θ

−3θ

+6θ

−6

, if θ

≤ θ ≤ θ

3θ

√

3θ

−12

(

3θ

)

, if θ

≤ θ ≤ θ

or if θ

≤ θ ≤ θ

3θ+

√

3θ

−12

2(3θ

+6)

, if θ

≤ θ ≤ θ

1−q(λ−3)+

√

(λ

−6λ−9)q

+(λ−3)q+1

, if θ

≤ θ ≤ 3.

38 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

where:

≈ 2.117953, θ

≈ 2.286691, θ

≈ 2.653259, θ

≈ 2.863469.

−

√



NSC

(q)

NSC

(θ)

q(3)

Figure 2.7: The decision function and the marginal tariﬀ..

We can ﬁnd the tariﬀ T

NSC

(q) that implements the decision q

NSC

(θ) by integrating the

marginal tariﬀ given by:



NSC

(q) =











(q, ψ

(q)), if

≤ q ≤ q

(q, ψ

(q)), if q

≤ q ≤ q

(q, ψ

(q)), if q

≤ q ≤

√

We can see in Figure 2.7 the decision function q

NSC

(θ) and the marginal tariﬀ T



NSC

(q). Notice

that jumps in the marginal tariﬀ correspond to bunching in the decision function graphic.

The monopolist’s expected proﬁt from q

NSC

(θ) is:

Π(q

, q

, θ

) = 0.200314 (2.23)

Comparing the equations (2.22) and (2.23), we see that the isoperimetric ironing procedure

results in a larger expected proﬁt for the monopolist. We have an increase of approximately

0.021%. This shows the suboptimality the decision function q

NSC

(θ).

In Figure 2.8 we make a superposition of the decision function and the marginal tariﬀ for

ironing procedures (a) and (b). Notice that the latter results in a smaller quality spectrum

than the former.

The fundamental diﬀerence between the two decision functions is that q

(θ) is implemented

by the single-crossing tariﬀ T

(q) and q

NSC

(θ) is implemented by the non single-crossing tariﬀ

2.6. CONCLUSION 39



(q)



NSC

(q)

−

√

(θ)

NSC

(θ)

Figure 2.8: The decision an the marginal tariﬀ for both procedures (a) and (b) .

NSC

(q). In fact, when the tariﬀ is T

NSC

(q), the maximization problem for customer θ

is not

quasiconcave as it has two local maxima q

and q

so the decision q

NSC

(θ) is not convex valued.

In Figure 2.9 we plot the customer θ

utility depending on his choice of q under the non

single-crossing tariﬀ T

NSC

(q).

V (θ

)

q(θ

)

q(θ

v(q, θ

) − T

NSC

(q)

Figure 2.9: The customer θ

’s utility under the non single-crossing tariﬀ T

NSC

(q).

2.6 Conclusion

In our model of nonlinear pricing without the Spence and Mirrlees condition, we showed that

the basic assumption of the demand proﬁle approach may not be valid and that the demand

proﬁle approach can lead to a suboptimal solution for the monopolist’s optimization problem.

40 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

The reason is that the requirement of a single-crossing tariﬀ is restrictive, as it imposes a convex

valued decision.

In a multidimensional setting we also have the same situation described here, since there is

not a natural ordering for the customer’s demand curves. Then one has to investigate whether

the restrictions imposed by the demand proﬁle approach may aﬀect or not the monopolist’s

welfare. Without a criterion based on the fundaments of the model that allows the monopolist to

anticipate whether his maximization problem results or not in a single-crossing tariﬀ, we cannot

know if the contract resulting from the demand proﬁle approach is a suboptimal contract. This

poses a serious restriction in knowing which problems are adequate for this kind of approach.

Appendix

We need the concept of cyclical monotonicity

Deﬁnition 8. A correspondence q : Θ → Q is v-cyclic-monotone if and only if for all ﬁnite

cycles θ

, θ

, . . . , θ

n+1

= θ

∈ Θ,



i=0

{v(q(θ

), θ

i+1

) − v(q(θ

), θ

)} ≤ 0.

Theorem 5. (Rochet, 1987) A decision q : Θ → Q is implementable if and only if q is v-cyclic-

monotone.

We also need some abstract convexity results related to incentive compatibility that we will

state here.

Deﬁnition 9. Let V : Θ → R. The v-subdiﬀerential of V , ∂

V : Θ → Q, is deﬁned by:

∂

V (θ) =



q ∈ Q : V (θ



) − V (θ) ≥ v(q, θ



) − v(q, θ) ∀θ



∈ Θ



Deﬁnition 10. V is said to be v-subdiﬀerentiable in Θ if and only if ∂

V (θ) = ∅ ∀θ ∈ Θ.

Deﬁnition 11. Let V : Θ → R. We deﬁne the v-conjugated of V by:

: Q → R V

(q) = sup

θ∈Θ

{v(q, θ) − V (θ)};

and the v-biconjugate of V by:

: Θ → R V

(θ) = sup

q∈Q

{v(q, θ) − V

(q)}.

This can be found in Rochet [25].

These results can be found in Carlier [6] and Rachev and R¸schendorf [23].

2.6. CONCLUSION 41

Proposition 3. Let V : Θ → R , then:

V is v-convex ⇔ V = V

Proposition 4. Let V : Θ → R , then:

q ∈ ∂

V (θ) ⇔ V (θ) + V

(q) = v(q, θ).

Theorem 6. (Carlier, 2001) A decision q : Θ → R

is implementable if and only if there

exists a v-convex and v-subdiﬀerentiable function V : Θ → R such that:

q(θ) ∈ ∂

V (θ) ∀θ ∈ Ω.

Remark 1. So when the informational rent V is v-convex, the conjugate V

plays the role of

the nonlinear tariﬀ T that implements the decision q and the correspondence ∂

V (θ) is identiﬁed

with q(θ).

Proof. of Lemma 3 The v-conjugation result allows us to write:

V (θ) = max

q∈Q

{v(q, θ) − T (q)},

and

T (q) = max

θ∈Θ

{v(q, θ) − V (θ)}. (2.24)

Let us prove ﬁrst that V is Lipschitz continuous. By the incentive compatibility constraint

we have that:

v(q(θ), θ) − t(θ) ≥ v(q(θ



), θ) − t(θ



). (2.25)

Using the deﬁnition of V , we can rewrite equation (2.25) as:

V (θ) − V (θ



) ≥ v(q(θ



), θ) − v(q(θ



), θ



). (2.26)

Changing the roles of θ and θ



in equation (2.26), we have that:

V (θ



) − V (θ) ≥ v(q(θ), θ



) − v(q(θ), θ). (2.27)

42 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

Equations (2.26) and (2.27) can be combined into:

v(q(θ), θ) − v(q(θ), θ



) ≥ V (θ) − V (θ



) ≥ v(q(θ



), θ) − v(q(θ



), θ



As q is bounded and v(q, θ) is C

, we conclude that there is M > 0 such that:



V (θ



) − V (θ)



≤ M



θ − θ





Now let us prove that T is Lipschitz continuous. Supposing that q

∈ q(θ

) and q

∈ q(θ

by Proposition 4 and equation (2.24), we can write:

T (q

) = v(q

, θ

) − V (θ

) ≥ v(q

, θ

) − V (θ

), (2.28)

T (q

) = v(q

, θ

) − V (θ

) ≥ v(q

, θ

) − V (θ

). (2.29)

Equations (2.28) and (2.29) results in:

v(q

, θ

) − v(q

, θ

) ≥ T (q

) − T (q

) ≥ v(q

, θ

) − v(q

, θ

As q is bounded and v(q, θ) is C

, we conclude that there is M > 0 such that:

|T (q

) − T (q

)| ≤ M |q

− q

Proof. of Lemma 4(i ) We know that θ is an interior point of Θ, then by the diﬀerentiability

of V at θ, we have that:

V (θ + h) = V (θ) + V



(θ) · h + o(h). (2.30)

Using equation (2.26) we have that:

V (θ + h) ≥ V (θ) + v(q(θ), θ + h) − v(q(θ), θ) = V (θ). (2.31)

Combining equations (2.30), (2.31) and using the diﬀerentiability of v(q, θ), we get:



(θ) · h −

∂v

∂θ

(q(θ), θ) · h ≥ o(h). (2.32)

Taking the limit in equation (2.32) ﬁrst when h ↑ 0 and then when h ↓ 0 we get that:



(θ) =

∂v

∂θ

(q(θ), θ).

2.6. CONCLUSION 43

Proof. of Lemma 4(ii ) We know that q is an interior point of Q, then by the diﬀerentiability

of T at q ∈ q(θ), we have that:

T (q + h) = T (q) + T



(q) · h + o(h). (2.33)

And equation (2.24) gives us that:

T (q + h) ≥ v(q + h, θ) − V (θ) = v(q + h, θ) − v(q, θ) + T (q), (2.34)

and so:

T (q + h) − T (q) ≥ v(q + h, θ) − v(q, θ). (2.35)

Combining equations (2.33), (2.35) and using the diﬀerentiability of v(q, θ), we have that:



(q) · h −

∂v

∂q

(q, θ) · h ≥ o(h). (2.36)

Taking the limit in equation (2.36) ﬁrst when h ↑ 0 and then when h ↓ 0 we get that:



(q) =

∂v

∂q

(q, θ).

Proof. of Lemma 5 We will only prove (i). For (ii), the proof is analogous. The sets CS

given by:

= {(θ, q) ∈ Θ × R

: q(θ) < q

(θ)}.

If q is continuous at θ

and (θ

, q(θ

)) ∈ CS

, it is possible to build an interval I = [θ

−δ, θ

+δ]

such that (θ, q(θ)) ∈ CS

for all θ ∈ I. Let us take θ

∈ I and deﬁne:

∆(θ

, θ

) = [v(q(θ

), θ

) − v(q(θ

), θ

)] + [v(q(θ

), θ

) − v(q(θ

), θ

)].

By Theorem 5, we must have ∆(θ

, θ

) ≤ 0. Notice that we can write:

∆(θ

, θ

) = −



q(θ

)

q(θ

)

qθ

(q, θ)dqdθ.

The region of integration is a subset of I, so the integrand v

qθ

is always positive. This gives us

that:

∆(θ

, θ

) ⇔ (q(θ

) − q(θ

))(θ

− θ

) ≥ 0.

So we conclude that q must be increasing in θ

44 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

Proof. of Lemma 6 We must have:

v(q(θ), θ) − t(θ) ≥ v(q(

θ), θ) − t(

θ).

Using the informational rent V , we can write the inequality above as:

V (θ) − V (

θ) ≥ v(q(

θ), θ) − v(q(

θ),

θ). (2.37)

By Lemma 4(i), we have that V



(θ) = v

(q(θ), θ), and equation (2.37) becomes:



(q(

θ),

θ)d

θ ≥



(q(

θ),

θ)d

θ.

And using the fundamental theorem of calculus, we get the following equation:



θ)

qθ

(˜q,

θ)d˜qd

θ ≥ 0.

Proof. of Theorem 1

We have a variational problem with inequality constraint. In this case the extremum is

achieved on a composite curve of pieces of extremals of the unconstrained problem and pieces

of extremals with binding constraints.

(i) As the constraint (BS) is not binding, we have an unconstrained maximization problem

for ψ

. The Euler equation for problem Π

implies that:

(q, ψ

(q)) = 0,

and using the deﬁnition of F in equation (2.8) the result follows.

(ii) BS is binding and θ < ψ

(q) < ψ

(q) < θ. By introducing a Lagrangian multiplier λ for

the constraint, we get:

L(q, ψ

, ψ

) = F (q, ψ

(q)) − F (q, ψ

(q)) + λ(q)(v

(q, ψ

(q)) − v

(q, ψ

(q))).

The ﬁrst order condition for the Lagrangian L results in :

λ(q) =

(q, ψ

(q))

qθ

(q, ψ

(q))

(q, ψ

(q))

qθ

(q, ψ

(q))

And the result follows from the deﬁnition of F in equation (2.8).

Petrov [22] derives the optimality conditions for this kind of problems.

2.6. CONCLUSION 45

(iii) BS is binding, but now we have a corner solution, as ψ

= θ. So the characterization

follows from solving the following constraint equation:

(q, ψ

(q)) = v

(q, θ).

Proof. of Theorem 2 The expected proﬁt coming from the decision q(θ, s) is given by:

Π(s) =



f(q

(θ), θ)dθ +



A(s)

f(q

(θ), θ)dθ +



A(s)

f(q

(θ, s), θ)dθ, (2.38)

where A(s) solves:

(s), s) = v

(A(s)), A(s)).

Diﬀerentiating equation (2.38) with respect to s gives us the result.

Proof. of Proposition 2 Consider q

, q

in q(θ) with q

< q

. We have that:





(q, θ) − T



(q)



dq = v(q

, θ) − T (q

) − (v(q

, θ) − T (q

)) = 0. (2.39)

As T is single crossing, if v

, θ) − T



) ≥ 0 then:

(q, θ) − T



(q) ≥ 0 ∀q ∈ [q

, q

] . (2.40)

Now, by equations (2.39) and (2.40) we conclude that

(q, θ) − T



(q) = 0 ∀q ∈ [q

, q

] a.s.

and remembering that for q ∈ [q

, q

] we have





(x, θ) − T



(x)



dx = 0.

Then we have:

v(q, θ) − T (q) = v(q

, θ) − T (q

) ∀q ∈ [q

, q

] ,

and so q is convex valued.

46 CHAPTER 2. NONLINEARD PRICING WITHOUT A SINGLE-CROSSING TARIFF

Proof. of Theorem 3 This problem is known as an Isoperimetric Problem. It is solved by

introducing a Lagrangian multiplier for the (ISO) condition, and appending the constraint with

the multiplier to the original problem. Then the multiplier λ should be adjusted to cope with

the isoperimetric constraint. The resulting problem is:

max

{ψ

}



{F (q, ψ

(q)) − F (q, θ

)}dq + λ



(q, ψ

(q)) − v

(q, θ

)}dq.

The Euler equation for the problem above is exactly:

(q, ψ

(q)) + λv

qθ

(q, ψ

(q)) = 0,

and remembering the deﬁnition of F in equation (2.8) we see that F

(q, ψ

(q)) = f

(q, ψ

(q)).

Finally, we have to adjust λ in order that the condition (ISO) is satisﬁed.

Proof. of Theorem 4 The way we have built our optimal candidate allows us to write the

monopolist’s expected proﬁt as:

Π(q

, q

, θ

) =



{F (q, ψ

(q)) − F (q, ψ

(q))}dq +



{F (q, ψ

(q)) − F (q, ψ

(q))}dq +



{F (q, ψ

(q)) − F (q, θ

) + λ[v

(q, ψ

(q)) − v

(q, θ

)]}dq.

Diﬀerentiating Π(q

, q

, θ

) with respect to q

, q

and θ

results in conditions (i), (ii) and (iii),

respectively.

We refer the interested reader to the book by Petrov [22]. It covers all the material from calculus of variations

that we use here.

Chapter 3

Bidimensional Screening

3.1 Introduction

Most of the work available in the literature of hidden information is based on one-dimensional

screening models. In this case there is a condition on the fundamentals, the Spence and Mirrlees

condition (SMC), that allows an easy characterization of an incentive compatibility allocation

rules. The reason is that the SMC provides a natural ordering in the type space which implies

in an equivalence between monotonicity and implementability of the agent’s decision function.

Therefore the principal can restrict his attention to monotonic decision functions and then

derive the optimal contract.

However, in the case of multidimensional screening models there is not an analogous con-

dition

on the fundamentals. As a consequence, it is very diﬃcult to deal with the incentive

compatibility constraints. Unlike the previous case, it is not possible to know a priori the

directions of the binding constraints. The result is that the principal may have too many con-

straints to take into account which makes the full characterization of the optimal contract a

rather diﬃcult problem.

This chapter is a tentative to extend the techniques developed by Araujo and Moreira [1] to

a bidimensional screening context. We study a nonlinear pricing model where the monopolist

sells a single product and the customer has a bidimensional characteristic. We study two cases.

The generalized single-crossing property deﬁned by McAfee and McMillan [17] is too restrictive.

48 CHAPTER 3. BIDIMENSIONAL SCREENING

In the ﬁrst one, called discrete-continuous, one of the customer’s characteristic is discrete and

the other is continuous. In the second one, called bicontinuous, both customer’s characteristics

are continuous. In the latter case, this chapter is close to McAfee and McMillan [17] and Laﬀont,

Maskin and Rochet [14]. But we do not assume the generalized single-crossing property as in

[17], and we generalize the example solved by [14].

In both setups, we derive a new necessary condition for implementability, resulting from

the customer’s maximization problem. This condition states that if two customers choose the

same decision they must pay the same marginal tariﬀ which implies that their marginal utility

must be the same. This condition will permit us to ﬁnd the optimality conditions that we use

for solving the monopolist’s maximization problem.

The chapter is divided in two main sections, one treating the discrete-continuous setup, and

the other one treating the bicontinuous setup.

3.2 Model

We use the Principal-Agent framework to analyse the bidimensional monopolistic screening

problem. In this model, each customer has a quasi-linear preference:

V (q, t, a, b) = v(q, a, b) − t,

where q ∈ R

is the customer’s decision, t is the monetary transfer and (a, b) ∈ Θ ⊆ R

is the

customer’s characteristics or types which are private information. The distribution function

P (a, b) is common knowledge. We assume that the customer’s utility function v(q, a, b) is C

The ﬁrm is a proﬁt-maximizing monopolist which can produce any quality q ∈ Q ⊆ R

incurring in a cost C(q). The monopolist’s revenue is given by:

Π(q, t) = t − C(q),

where C(q) is a C

, nonnegative, increasing and convex.

Using the ‘Revelation Principle’

the monopolist’s problem can be stated as choosing the

allocation rule (q, t) : Θ → R

× R that solves:

max

{q(·),t(·)}



Π(q(a, b), t(a, b))dP (a, b), (Π)

The ‘Revelation Principle’ has been enunciated in Gibbard [7].

3.3. DISCRETE-CONTINUOUS TYPE 49

subject to the individual-rationality constraints:

v(q(a, b), a, b) − t(a, b) ≥ 0 ∀(a, b) ∈ Θ, (IR)

and the incentive compatibility constraints:

(a, b) ∈ arg max



)∈Θ

{v(q(a



, b



), a, b) − t(a



, b



)}, ∀(a, b) ∈ Θ. (IC)

Remark 3. An allocation rule (q, t) satisfying the IC constraints is called incentive compatible.

A decision q is implementable if there exists a monetary transfer t such that the pair (q, t) is

incentive compatible.

Remark 4. The ‘Taxation Principle’

states that any allocation (q, t) satisfying the IC con-

straints can be implemented by a nonlinear tariﬀ T : Q = q(Θ) → R where:

T (q(a, b)) = t(a, b), ∀(a, b) ∈ Θ.

3.3 Discrete-Continuous Type

In this section we study a bidimensional monopolist’s screening model where the customer’s

characteristics are represented by two parameters. One is a discrete variable and the other one

is a continuous variable. This speciﬁcation for the parameters can be understood as a limit

case when their values are b

and b

, as shown in Figure 3.1.

We have two distinct groups of customers, one with lower b and the other one with higher

b. We assume that for a ﬁxed b, the Spence and Mirrlees condition (SMC) is valid. By this

assumption, an incentive compatible decision function must be monotonic in the variable a.

However, the monotonicity is no longer a suﬃcient condition for implementability. The reason

is that the ordering of the binding IC constraints among customers with lower and higher b is

endogenous. In particular, there may exist two types (lower and higher b) choosing the same q.

From the customer’s maximization problem, we derive a new necessary IC condition. If two

customers choose the same decision they must pay the same marginal tariﬀ. Consequently their

marginal utility must be the same. We will refer to it as the U-condition (UC).

This principle can be found in Guesnerie [9], Hammond [11] and Rochet [24].

50 CHAPTER 3. BIDIMENSIONAL SCREENING

Figure 3.1: The parameter b may take the values b

and b

We have two optimization problems: the relaxed problem and the UC problem which takes

the U-condition into account. Consequently we have two kinds of optimal solutions, depending

on whether the U-condition is binding or not. The solution we propose is a composite curve of

pieces of the relaxed problem’s solution and the UC problem’s solution.

We assume that the variable a is distributed in [a, a] with a continuous and positive density

p(a) and distribution function given by P (a). The variable b is discrete, taking values in

, b

}, with a probability function given by Pr(b = b

) = p

. These variables are independent

and their joint distribution is given by P (a, b

) = p

P (a). The type space is then given by

Θ = [a, a] × {b

, b

We assume the following condition:

Deﬁnition 12. (The Spence and Mirrlees condition.) For a ﬁxed b = b

the sign of the

cross derivative of v(q, a, b) with respect to the decision q and type a is positive:

(q, a, b

) > 0 on R

× [a, a]. (SMC)

With the SMC one can show that an incentive compatible decision must be increasing

with a.

This condition, however, is no longer suﬃcient and we derive now another necessary

condition for implementability. By Remark 3 for all incentive compatible allocation rule (q, t),

The proof is exactly the same as the one used for the one-dimensional screening problem with the SMC.

See Rochet [25].

3.3. DISCRETE-CONTINUOUS TYPE 51

we can ﬁnd a tariﬀ T that implements the decision q. This results in the following maximization

problem for the customer (a, b

max

q∈q(Θ)

{v(q, a, b

) − T (q)}.

The ﬁrst-order condition for an interior solution of the problem above is:

(q, a, b

) = T



(q).

So we can state the following result as a direct consequence of the ﬁrst-order condition.

Proposition 5. (U-condition.) Suppose that T is diﬀerentiable at an interior point of Q =

q(Θ), q ∈ q(a, b

) ∩ q(a



, b

). Then

(q, a, b

) = v

(q, a



, b

). (UC)

This marginal utility identity between the pooled customers (a, b

) and (a



, b

) is called the

U-condition (UC). It simply states that these customers must pay the same marginal tariﬀ at

their same decision. We can solve this marginal identity for a



, which results in a relation be-

tween the higher and lower b customers that will be useful for deriving the optimality conditions

when the UC is binding.

3.3.1 Optimality Conditions

In this subsection we derive the pointwise optimality conditions for the monopolist’s problem.

There are two maximization problems, one when the UC constraint is not binding and the

other when it is binding. The former is the relaxed problem and the latter is the UC problem.

The pointwise conditions derived give the relaxed solution and the UC solution q

and q

respectively.

Consider an incentive compatible allocation rule (q, t). Then we deﬁne the customer’s

informational rent by:

V (a, b

) = v(q(a, b

), a, b

) − t(a, b

52 CHAPTER 3. BIDIMENSIONAL SCREENING

We can plug the equation above into the monopolist’s objective function in the problem (Π)

and the result is:



i=1



{v(q(a, b

), a, b

) − C(q(a, b

)) − V (a, b

)}p(a)da, (3.1)

Using the envelope theorem (see Milgrom and Segall [18]) and an integration by parts procedure

we can rewrite the monopolist’s objective function as:



i=1



g(q(a, b

), a, b

)da,

where g is deﬁned as:

g(q, a, b

) = p



v(q, a, b

) − C(q) +



P (a) − 1

p(a)



(q, a, b

)



p(a). (3.2)

By dropping the global incentive compatibility constraints we can deﬁne a relaxed version of

problem (Π) which is called the monopolist’s relaxed problem:

max

q(·)



i=1



g(q(a, b

), a, b

)da. (Π

)

We have the following imediate result for the problem (Π

Proposition 6. (The Relaxed Decision.) The interior solution of (Π

) must satisfy the

following ﬁrst-order condition:

(q, a, b

) = 0. (3.3)

Notice that this gives pointwise conditions for a solution of problem (Π

). The relaxed

decision q

(a, b

) is exactly the solution of equation (3.3) for each q. The relaxed solution may

not satisfy the necessary UC condition. So we may have to incorporate the UC constraint into

the monopolist’s problem.

Suppose that we have an interval [r

, r

] of lower b customers such that the UC constraint

is binding for them. By using the UC identity and the implicit function theorem we can deﬁne

implicitly a function a(q, r) by:

(q, r, b

) = v

(q, a(q, r), b

3.3. DISCRETE-CONTINUOUS TYPE 53

Let φ(r) denote the decision function associated with the (r, b

) customer. Then we have:

(φ(r), r, b

) = v

(φ(r), a(φ(r), r), b

). (3.4)

Notice that for each r ∈ [r

, r

], the customers (r, b

) and (a(φ(r), r), b

) choose the same

decision φ(r).

Considering equation (3.4), we can write the monopolist’s objective function on this interval

as:



g(φ(r), r, b

)dr + p



g(ψ(s), s, b

)ds, (3.5)

where:











s = a(φ(r), r),

ψ(s) = φ(r),

= a(φ(r

), r

= a(φ(r

), r

(3.6)

Using the equations in (3.6) we can make a change of variables in expression (3.5), resulting in:



g(φ(r), r, b

) + p

g(φ(r), a(φ(r), r), b

)[a

(φ(r), r)φ



(r) + a

(φ(r), r)]}dr.

Notice that the UC constraints are incorporated in the expression above. Let us deﬁne:

F (r, x, y) = p

g(x, r, b

) + p

g(x, a(x, r), b

)[a

(x, r)y + a

(x, r)].

Then we can deﬁne a less constrained version of the monopolist’s maximization problem taking

into account only the UC constraints:

max

φ(·)



F (r, φ(r), φ



(r))dr. (3.7)

We then establish the pointwise optimality conditions for (3.7):

Proposition 7. (UC Decision.) If φ

∗

is the optimal decision for problem (3.7), then:

(q, r, b

)

(q, r, b

)

= −p

(q, a(q, r), b

)

(q, a(q, r), b

)

, (3.8)

where q = φ

∗

(r).

This condition is analogous to the U-shaped condition in Araujo and Moreira [1] and to the

Bell-shaped condition from Chapter 2.

54 CHAPTER 3. BIDIMENSIONAL SCREENING

Finally we suggest a method for determining the optimal decision. We construct a composite

decision combining pieces of the relaxed and the UC decisions. It is important to optimally

choose the point where you change from one to another. In what follows we solve an example

that illustrates this method.

3.3.2 An Example

In this example, a is uniformly distributed in [0, 1], b

= 0 and b

= 1, with p

= p

= 1/2. The

customer’s utility function is given by:

v(q, a, b

) = aq − (1 + b

)

and for simplicity we assume that the monopolist’s production costs are zero.

Hence:

g(x, a, b

) =



(2a − 1)q(a, b

) − (b

+ 1)

q(a, b

)



Using Proposition 6, we can ﬁnd the relaxed decision:







(a, 0) = {2a − 1}

(a, 1) = {

2a−1

}

The relaxed solution is not incentive compatible, because it does not satisfy the necessary UC

condition.

Using the U-condition (UC), we have that:

a(q, r) = r + q,

and, using Proposition 7, we can ﬁnd the UC decision:







(a, 0) = {4a − 2}

(a, 1) = {

4a−2

}

In Figure 3.2 we plot the relaxed and the UC decision functions.

Our strategy now is to construct a decision function combining the relaxed and the UC

decision, depending on the parameter a

, which is the point where we change from one decision

3.3. DISCRETE-CONTINUOUS TYPE 55

0.2 0.4 0.6 0.8 1

0.2

0.4

0.6

0.8

0.2 0.4 0.6 0.8 1

0.2

0.4

0.6

0.8

(a, 0)

(a, 1)

Figure 3.2: The relaxed and the UC decision functions.

to the other:

q(a, 0, a

) =











0 if 0 ≤ a ≤

(a, 0) if

≤ a ≤ a

, 0) if a

≤ a ≤ a

(a, 0) if a

≤ a ≤ 1.

(3.9)

q(a, 1, a

) =











0 if 0 ≤ a ≤

, 1) if

≤ a ≤ ˆa

(ˆa

, 1) if ˆa

≤ a ≤ 1,

(3.10)

where ˆa

= a(q

, 0), a

) and a

is such that q

, 0) = q

, 0). In Figure 3.3 we plot

this decision function which combines the relaxed decision and the UC decision.

0.2 0.4 0.6 0.8 1

0.2

0.4

0.6

0.8

Figure 3.3: The composite solution in thicker black.

From the objective function in (Π

), we can determine the monopolist proﬁt as a function

56 CHAPTER 3. BIDIMENSIONAL SCREENING

of a

(see Figure 3.4):

Π(a

) =

28a

− 18a

23a

−

Now we have to optimally choose the parameter a

By solving the ﬁrst-order condition

∗

Π(a

)

0.586 0.588 0.59 0.592 0.594 0.596 0.598

0.06102

0.06104

0.06106

0.06108

Figure 3.4: The expected proﬁt depending as a function of a



) = 0, we ﬁnd the optimal choice for a

∗

(18 −

√

2).

So the solution we propose results from equations (3.9) and (3.10), choosing the parameter

= a

∗

. This solution is implemented by the following tariﬀ:

T (q) =







q(4 − 3q), if 0 ≤ q ≤

(4 −

√

2),

196

(−9 + 4

√

2 + 98q − 49q

), if

(4 −

√

2) ≤ q ≤ 1,

which is determined by integrating the marginal utility.

Comparing the relaxed decision q

(a, b) with q(a, b, a

∗

), we have a decrease of approximately

2.3% in the expected proﬁt.

3.4 Bicontinuous Type

In this section we show how to use the marginal utility condition to solve nonlinear pricing

model where the customer has two dimensions of private information. Unlike the case treated

The optimal a

∗

can also be found through an ironing procedure. See Araujo and Moreira [1].

3.4. BICONTINUOUS TYPE 57

in the previous section, where we have a continuous and a discrete parameter for the type,

now we have that both dimensions are continuous, that is, the type space is the rectangle

Θ = (0, α) × (0, β), with α, β > 0.

As the decision q is a function deﬁned from Θ ⊆ R

into R

, we expect that its level

sets are one-dimensional curves with images contained in Θ. The level sets for q represents a

continuous pooling of types, and the marginal condition states that pooling types must have

identical marginal utility at the same chosen decision. The result is that both q(a, b) and

(q(a, b), a, b) should have common level sets. This principle may be expressed in a convenient

way by using partial diﬀerential equation (PDE), as we will see in what follows.

Analysing the customer’s maximization problem we derive a partial diﬀerential equation

(PDE) whose solution can be interpreted as resolving this relation among the level sets. This

is a quasilinear ﬁrst-order partial diﬀerential equation and we use the method of characteristic

curves to solve it. This method is adequate for our problem because the characteristic curves

for this particular PDE is exactly the level sets of both q(a, b) and v

(q(a, b), a, b). In the end

it provides us a natural change of variables that reduces the problem from two dimensions to

only one dimension. Then we apply variational calculus techniques to ﬁnd the solution of this

one-dimensional problem and ﬁnally we go back to the original variables.

The parameter representing the customer’s type (a, b) ∈ Θ has a positive and continuous

density p(a, b). For simplicity, we assume that the monopolist’s production costs are zero.

We also assume that the monetary transference t is positive. Consequently the tariﬀ T is

positive for all units consumed q. This means that the monopolist ever has positive proﬁts when

selling to any customer (a, b). Consider an incentive compatible allocation rule (q, t). From the

(IC) condition, each customer has the following maximization problem:

max



)∈Θ

{v(q(a



, b



), a, b) − t(a



, b



)}.

The ﬁrst-order conditions for the maximization problem above give us an analogous UC condi-

tion of the discrete-continuous problem

Proposition 8. (The Quasilinear Equation.) Suppose that (q, t) : Θ → R

×R is incentive

Actually the UC condition is equivalent to the PDE (3.11). The PDE formulation is more convenient to

the bicontinuous model.

58 CHAPTER 3. BIDIMENSIONAL SCREENING

compatible and twice diﬀerentiable in an open set Ω ⊂ Θ. Then we have that:

(q(a, b), a, b)q

(a, b) = v

(q(a, b), a, b)q

(a, b). (3.11)

Equation (3.11) is classiﬁed as a ﬁrst-order partial diﬀerential equation (PDE)

. We can

interpret it geometrically by observing that the gradients ∇

a,b

and ∇q have the same direction

and, consequently, both q and v

have the same level set.

Although Proposition 8 is a general result, we will work with a particular speciﬁcation for

the customer’s utility function. This allow us to show how to use the PDE derived to solve the

monopolist’s problem. Hence we assume that the customer’s utility function is given by:

v(q, a, b) = aq − (c + w(b))

− t,

where a and b are distributed independently and uniformly on Θ; w(0) = 0 and w



(b) > 0;

γ > 1, q represents the consumption choice and t is the monetary transfer. The monopolist’s

production cost C(q) is assumed to be zero.

For an incentive compatible allocation rule (q, t), the customer’s informational rent is given

by V (a, b) = v(q(a, b), a, b) − t(a, b). As the production costs are now zero, if we plug the

informational rent in equation (Π) we can rewrite the monopolist’s objective function as:

αβ



{v(q(a, b), a, b) − V (a, b)}dadb.

Using the envelope theorem (see [18]), and an integration by parts procedure, we can derive

the following result about the monopolist’s expected proﬁt:

Proposition 9. (The Monopolist’s Expected Proﬁt.) Let q be an implementable decision.

Then the monopolist’s expected proﬁt is given by:

αβ



{(2a − α)q(a, b) − (c + w(b))

}dadb. (3.12)

See Fritz John [12] for a complete analysis of this kind of PDE and the method of characteristic curves for

solving it.

3.4. BICONTINUOUS TYPE 59

Let us deﬁne G(q, a, b) = (2a − α)q(a, b) − (c + w(b))

From Proposition 8, we can establish the PDE (3.11) for our speciﬁc model:



(b)q

γ−1

+ q

= 0. (3.13)

We now formulate the following Cauchy problem for this PDE:









(b)q

γ−1

+ q

= 0,

q(r, 0) = φ(r),

(3.14)

where we assume that φ is a continuous and increasing function such that φ(0) = 0.

We now use the method of characteristic curves to ﬁnd the solution of (3.14). The method

works in the following way: along the characteristic curves the PDE equation becomes an

ordinary diﬀerential equation (ODE) with initial condition given by q(r, 0) = φ(r). So, for each

ﬁxed r, we have an ODE with initial value and we then invoke the existence and uniquiness

theorems to solve it. We choose the curve (r, 0) to give the initial value and it is important

that the characteristic curves and the initial value curve to be not tangent. The idea is that we

can ﬁnd the characteristic curves and relate the value of q in Θ to the value of q in (r, 0). In

Figure 3.5, the projections of the characteristic curves are plotted. As we can see in the ﬁgure,

they are not tangent to the line (r, 0) and the value of q is constant along these curves.

Figure 3.5: Characteristic curves in Θ.

So using the method of characteristic curves, we can establish the following result:

60 CHAPTER 3. BIDIMENSIONAL SCREENING

Proposition 10. (The Cauchy Problem.) The solution for the Cauchy problem (3.14) is

given by:

q(a, b) = q(a(r, s), b(r, s)) = φ(r(a, b)),

where we get r by inverting the following equations:







a(r, s) = r + w(s)φ(r)

γ−1

b(r, s) = s.

For each r we have characteristic curves parametrized as (a(r, s), b(r, s), φ(r)). With them

we have a natural change of variables that we will plug in equation (3.12). This results in the

following system of variables:







a = A(x, r, s) = r + w(s)x

γ−1

b = B(r, s) = s,

where the Jacobian is simply A



(r) + A

, such that A

∂A

∂x

and A

∂A

∂r

Using these new variables, we can rewrite the monopolist’s expected proﬁt as

αβ



L(φ(r),r)

{G(φ(r), A(φ(r), r, s), s)





(r) + A



}dsdr,

where, for each r, L is deﬁned as the s coordinate of the intersection between the characteristic

curve and the boundary of Θ. Notice that there should exist an intermediate r

such that for

r < r

, L is constant with L = β and for r > r

, L solves A(x, r, L(x, r)) ≡ α. In Figure 3.5, L

is represented by dot points.

Then we can deﬁne a less constrained version of the monopolist’s maximization problem,

by dropping the IC constraints and considering only the constraints following from the EDP

given in Proposition 8:

max

φ(·)

αβ



L(φ(r),r)

{G(φ(r), A(φ(r), r, s), s)





(r) + A



}dsdr. (3.15)

Now we can establish our main result:

Proposition 11. The Euler’s equation for the monopolist’s optimization problem (3.15) is

given by:



L(φ(r),r)

(φ(r), A(φ(r), r, s), s)ds = 0. (3.16)

3.4. BICONTINUOUS TYPE 61

3.4.1 Examples

Here we solve some examples of bidimensional contracts with one instrument.

Example 1. (Laﬀont, Maskin and Rochet, 1987.) This example was taken from [14]. We

set α = β = c = 1, γ = 2 and w(b) = b in our model.

We have that:











G(q, a, b) = (2a − 1)q −

(b + 1)q

A(x, r, s) = r + sx,

B(x, r, s) = s.

We can determine L as:

L(x, r) =







1 , if r < r

1−r

, if r > r

By Proposition 11, the Euler equation is given by:



L(φ(r),r)

(φ(r), A(φ(r), r, s), s)A

(φ(r), r, s)ds = 0.

This gives us:



{(2(r + sφ) − 1 − (s + 1)φ)}ds = (2r − 1 − φ)L +

φ = 0.

Solving the equation above for φ results in:

φ(r) =











0 , if 0 ≤ r ≤

4r − 2 , if

≤ r ≤

3r−1

, if

≤ r ≤ 1.

Now we have to go back to the original variables. Solving the equation A(φ(r), r, s) = r + sφ(r)

for r in terms of a and b, we get:

r(a, b) =











a , if a ≤

a+2b

1+4b

, if

≤

a+2b

1+4b

≤

2a+b

2+3b

, if

≤

2a+b

2+3b

≤ 1.

We have that q(a, b) = φ(r(a, b)). Making the substitution we can ﬁnd the optimal decision

q(a, b):

q(a, b) =











0 , if a ≤

4a−2

1+4b

, if

≤

a+2b

1+4b

≤

3a−1

2+3b

, if

≤

2a+b

2+3b

≤ 1,

62 CHAPTER 3. BIDIMENSIONAL SCREENING

and also the nonlinear tariﬀ

T that implements q:

T (q) =







−

, if q ≤

−

, if q ≥

Example 2. This example is an slight variation of the previous one. The fundamental dif-

ference is that now the characteristic curves are not straight lines for r > 1/2. We set

α = β = c = 1, γ = 2 and w(b) = b

in our model.

We have that:











G(q, a, b) = (2a − 1)q −

+ 1)q

A(x, r, s) = r + s

B(x, r, s) = s.

We can determine L as:

L(x, r) =







1 , if r ≤ r



1−r

, if r ≥ r

By Proposition 11, the Euler equation is given by:



L(φ(r),r)

(φ(r), A(φ(r), r, s), s)A

(φ(r), r, s)ds = 0.

This gives us:





(2(r + s

φ) − 1 − (s

+ 1)φ)



ds = (2r −1 − φ)L +

φ = 0.

Solving the equation above for φ results in:

φ(r) =











0 , if 0 ≤ r ≤

6r−3

, if

≤ r ≤

5r−2

, if

≤ r ≤ 1.

Now we have to go back to the original variables. And solving the equation A(φ(r), r, s) =

r + s

φ(r) for r in terms of a and b, we get:

r(a, b) =











a , if a ≤

2a+3b

2+6b

, if

≤

2a+3b

2+6b

≤

3a+2b

3+5b

, if

≤

3a+2b

3+5b

≤ 1.

We ﬁnd the nonlinear tariﬀ T by integrating the marginal tariﬀ given by v

(q, a, 0).

3.4. BICONTINUOUS TYPE 63

We have that q(a, b) = φ(r(a, b)). Making the substitution we can ﬁnd the optimal decision

q(a, b):

q(a, b) =











0 , if a ≤

6a−3

2+6b

, if

≤

2a+3b

2+6b

≤

5a−2

3+5b

, if

≤

3a+2b

3+5b

≤ 1,

and also the nonlinear tariﬀ T that implements q:

T (q) =







−

, if q ≤

−

160

, if q ≥

Example 3. (Basov, 2005.) This is a generalization of the ﬁrst one and was proposed by

Basov [4]. We set w(b) = b and β < 2c in our model.

We have that:











G(q, a, b) = (2a − α)q −

(b + c)q

A(x, r, s) = r + sx

γ−1

B(x, r, s) = s.

We can determine L as:

L(x, r) =







β , if r ≤ r

α−r

γ−1

, if r ≥ r

By Proposition 11, the Euler equation is given by:



L(φ(r),r)

(φ(r), A(φ(r), r, s), s)A

(φ(r), r, s)ds = 0.

This gives us:





(2(r + sφ

γ−1

) − α − (s + c)φ

γ−1

)



ds = (2r −α − cφ

γ−1

)L +

γ−1

= 0.

Solving the equation above for φ results in:

φ(r) =











0 , if 0 ≤ r ≤



4r−2α

2c−β



γ−1

, if

≤ r ≤ α

2c+β

2c+3β



3r−α



γ−1

, if α

2c+β

2c+3β

≤ r ≤ α.

64 CHAPTER 3. BIDIMENSIONAL SCREENING

Now we have to go back to the original variables. And solving the equation A(φ(r), r, s) =

r + s

φ(r) for r in terms of a and b, we get:

r(a, b) =











a , if a ≤

a(2c−β)+2αb

2c−β+4b

, if

≤

a(2c−β)+2αb

2c−β+4b

≤ α

2c+β

2c+3β

2ac+αb

2c+3b

, if α

2c+β

2c+3β

≤

2ac+αb

2c+3b

≤ α.

We have that q(a, b) = φ(r(a, b)). Making the substitution we can ﬁnd the optimal decision

q(a, b):

q(a, b) =











0 , if a ≤



4a−2α

2c−β+4b



γ−1

, if

≤

a(2c−β)+2αb

2c−β+4b

≤ α

2c+β

2c+3β



3a−α

2c+3b



γ−1

, if α

2c+β

2c+3β

≤

2ac+αb

2c+3b

≤ α,

and also the nonlinear tariﬀ T that implements q:

T (q) =











αq

−

(β + 2c)q

4γ

, if q ≤

2α

3β+2c

αq

−

3γ

2α

18β + 12c

−

3β + 2c

12γ



2α

3β + 2c



, if q ≥

2α

3β+2c

3.5 Conclusion

In this chapter we showed that the techniques by Araujo and Moreira [1] can be extended

to the bidimensional context at least for some particular speciﬁcations. The marginal utility

identity plays a fundamental role in our analysis. In a sense, it allows us to reduce the problem

to only one dimension. Then we could derive the optimality conditions for the monopolist’s

maximization problem.

In the discrete-continuous setup we derived the pointwise optimality conditions, and then

we gave an example illustrating how to use them to build a solution for the monopolist’s max-

imization problem in a way similar to Araujo and Moreira [1]. In the bidimensional setup, we

also derived optimality conditions and then we were able to solve examples from the litera-

ture, and to solve a new example that does not satisfy the generalized single-crossing property

assumed by [17].

3.5. CONCLUSION 65

Appendix

Proof. of Proposition 5 The customer (a, b

) has the following maximization problem:

max

q∈q(Θ)

{v(q, a, b

) − T (q)}.

If q = q(a, b

) = q(a, b

) is an interior point of Q = q(Θ) and T is diﬀerentiable at q, deriving

the ﬁrst-order condition for the maximization problem above results in:

(q, a, b

) = T



(q) = v

(q, a, b

Proof. of Proposition 6 The Euler’s equation

for problem (Π

) is exactly g

(q, a, b

) = 0.

Proof. of Proposition 7 Consider the following functional:



F (r, φ(r), φ



(r))dr.

The Euler’s equation for the functional above is:

(r, φ(r), φ



(r)) =

(r, φ(r), φ



(r)).

If we expand the Euler’s equation, we get:

p1g

(φ(r), r, b

) + p2g

(φ(r), a(φ(r), r), b

(φ(r), r) = 0.

But the function a(q, r) is implicitly deﬁned as the solution of:

(q, r, b

) = v

(q, a(q, r), b

Diﬀerentiating the equation above with respect to q results in:

(q, r) =

(q, r, b

)

(q, a(q, r), b

)

and the result follows.

See Petrov [22].

66 CHAPTER 3. BIDIMENSIONAL SCREENING

Proof. of Proposition 8 For each customer (a, b), the ﬁrst-order conditions from the (IC) give

that:

= v

(q(a, b), a, b)q

(a, b),

= v

(q(a, b), a, b)q

(a, b).

By taking the cross derivatives in both equations above, we get:

= (v

(q, a, b)q

+ v

(q, a, b))q

+ v

(q, a, b)q

= (v

(q, a, b)q

+ v

(q, a, b))q

+ v

(q, a, b)q

As t is twice diﬀerentiable at (a, b), using Schwarz’ theorem we have that the cross derivatives

are equal and the result follows.

Proof. of Proposition 9 Consider an incentive compatible (IC) allocation rule (q, t). For each

customer (a, b) we should have:

V (a, b) = v(q(a, b), a, b) − t(a, b) ≥ v(q(a



, b



), a, b) − t(a



, b



), ∀(a, b), (a



, b



and using the envelope theorem (see Milgrom and Segall [18]), we have that:

(a, b) = v

(q(a, b), a, b) = q(a, b) and V

(a, b) = v

(q(a, b), a, b) = −w



(b)

q(a, b)

The fundamental theorem of calculus allows us to write:

V (a, b) = V (0, b) +



q(θ, b)dθ,

and

V (0, b) = V (0, 0) −





(θ)

q(0, θ)

dθ.

We have that v(q, 0, b) ≤ 0 and, by assumption, t is always positive. As a consequence we must

have q(0, b) = 0. Otherwise, customer (0, b) would have a negative utility. This results in:

V (0, b) = 0 and V (a, b) =



q(θ, b)dθ.

Using the informational rent we can replace t(a, b) in the monopolist’s objective function in

the problem (Π):

αβ



{v(q(a, b), a, b) − V (a, b)}dadb. (3.17)

3.5. CONCLUSION 67

Using an integration by parts procedure we have that:



V (a, b)dadb =



q(θ, b)dθdadb







q(θ, b)dθ



−



aq(θ, b)da





(α − a)q(a, b)dadb.

Plugging the equation above in equation (3.17) we get the result.

Proof. of Proposition 10 The characteristic curves for equation (3.11) emerge as the solution

of the following system of ordinary diﬀerential equations











= −v

(z(r, s), a(r, s), b(r, s)) = w



(b)z

γ−1

, with a(r, 0) = r;

= v

(z(r, s), a(r, s), b(r, s)) = 1, with b(r, 0) = 0;

= 0, with z(r, 0) = φ(r),

and the solution for the system of ordinary diﬀerential equations above is given by:











a(r, s) = r + w(s)φ(r)

γ−1

;

b(r, s) = s;

z(r, s) = φ(r).

We have that q(a, b) = z(r(a, b), s(a, b) = φ(r(a, b)) and we only have to invert the variables to

ﬁnd the solution.

Proof. of Proposition 11 Deﬁne the function:

J() =



L(φ(r)+h(r),r)

{G(φ(r) + h(r), A(φ(r) + h(r), r, s), s)

(φ(r) + h(r), r, s), s)(φ



(r) + h



(r)) + A

(φ(r) + h(r), r, s), s)]}dsdr,

See Fritz John [12].

68 CHAPTER 3. BIDIMENSIONAL SCREENING

where h(r) is an increasing function such that h(0) = h(α) = 0. Diﬀerentiating J with respect

to , and evaluating at  = 0, we get:







L(φ(r),r)



+ G

][A



+ A

] + G[A



+ A

]



+ G(φ, A(φ, r, L), L)[A

(φ, r, L)φ



+ A

(φ, r, L)]



h(r)dr +







L(φ(r),r)





(r)dr.

Rewriting the expression above and imposing that J



(0) = 0 results in:





(r)h(r) + I

(r)h



(r)



dr = 0.

Using a classical result from calculus

, we conclude from equation above that:

(r) = I



(r). (3.18)

Using equation (3.18), we get:



L(φ(r),r)

ds + G(A

− A

) = 0, (3.19)

where











G = G(φ, A(φ, r, L), L),

= A

(φ, r, L),

= A

(φ, r, L).

Now, we observe that if r < r

, then L(x, r) = β and so L

= L

= 0; if r > r

, L solves

A(x, r, L(x, r)) ≡ 1 and deriving this equation with respect to x and r, we get:



+ A

= 0, A

+ A

= 0,

(3.20)

and this gives us:







= −A

(3.21)

Finally, we conclude that A

− A

= 0 and we can simplify equation (3.19):



L(φ(t),r)

ds = 0, (3.22)

and recalling that A(x, r, s) = r + w(s)x

γ−1

, we have that A

= 1, and the result follows.

See Troutman [28], pg. 98.

Bibliography

[1] A. Araujo and H. Moreira. Adverse selection problem without the spence-mirrlees condi-

tion. 2007.

[2] M. Armstrong. Multiproduct nonlinear pricing. Econometrica, 64:51–75, 1996.

[3] S. Basov. Hamiltonian approach to multi-dimensional screening. Journal of Mathematical

Economics, 36:77–94, 2001.

[4] S. Basov. Multidimensional Screening. Springer., 2005.

[5] S. Brown and D. Sibley. The theory of public utility pricing. Cambridge University Press.,

1986.

[6] G. Carlier. A general existence result for the principal-agent problem with adverse selection.

Journal of Mathematical Economics, 35:129–150, 2001.

[7] A. Gibbard. Manipulation for voting schemes. Econometrica, 41:587–601, 1973.

[8] M. Goldman, H. Leland, and D. Sibley. Optimal nonuniform pricing. Review of Economic

Studies, 51:305–320, 1984.

[9] R. Guesnerie. On taxation and incentives: further reﬂections on the limit of redistribution.

Mimeo., 1981.

[10] R. Guesnerie and J.-J. Laﬀont. A complete solution to a class of principal-agent problems

with an application to the control of a self-managed ﬁrm. Journal of Public Economics,

25:329–369, 1984.

70 BIBLIOGRAPHY

[11] J.P. Hammond. Straightforward individual incentive compatibility in large economies.

Review of Economic Studies, 46:263–282, 1979.

[12] F. John. Partial Diﬀerential Equations. Springer Verlag, 1981.

[13] B. Jullien. Participation constraints in adverse selection models. Journal of Economic

Theory, 93:1–47, 2000.

[14] J.J. Laﬀont, E. Maskin, and J.C. Rochet. Optimal nonlinear pricing with two dimensional

characteristics. University of Minnesota Press, 1987.

[15] E. Maskin and J. Riley. Monopoly with incomplete information. Rand Journal of Eco-

nomics, 15(2):171–196, 1984.

[16] S. Matthews and J. Moore. Monopoly provision of quality and warranties: An exploration

in the theory of multidimensional screening. Econometrica, 55(2):441–468, 1987.

[17] R.P. McAfee and J. McMillan. Multidimensional incentive compatibility and mechanism

design. Journal of Economic Theory, 46:335–354, 1988.

[18] P. Milgrom and I. Segal. Envelope theorems for arbitrary choice sets. Econometrica,

70(2):583–601, 2002.

[19] P.K. Monteiro and F.H. Page Jr. Optimal selling mechanisms for multiproduct monopolists:

Incentive compatibility in the presence of budget constraints. Journal of Mathematical

Economics, 30, 1998.

[20] M. Mussa and S. Rosen. Monopoly and product quality. Journal of Economic Theory,

18:301–307, 1978.

[21] G. Nˆldeke and L. Samuelson. Optimal bunching without optimal control. Journal of

Economic Theory, 134:405–420, 2007.

[22] Iu. P. Petrov. Variational Methods in Optimum Control Theory. Academic Press, 1968.

[23] S.T. Rachev and L. R¸schendorf. Mass Transportation Problems.Volume I. Springer, 1988.

BIBLIOGRAPHY 71

[24] J.C. Rochet. The taxation principle and multitime hamilton-jacobi equations. Journal of

Mathematical Economics, 14:113–128, 1985.

[25] J.C. Rochet. A necessary and suﬃcient conditions for rationalizability in a quasi-linear

context. Journal of Mathematical Economics, 16:191–200, 1987.

[26] J.C. Rochet and P. Chon

E. Ironing, sweeping, and multidimensional screening. Econo-

metrica, 66(4):783–826, 1998.

[27] J.C. Rochet and L. Stole. The economics of multidimensional screening, in M. Dewa-

tripont, L.P. Hansen, and S.J. Turnovsky, eds. Advances in Economics and Econometrics,

Cambridge. The Press Syndicate of the University of Cambridge, UK, 2003.

[28] J. L. Troutman. Variational Calculus and Optimal Control: Optimization with Elementary

Convexity. Springer Verlag, 1995.

[29] R.B. Wilson. Nonlinear Pricing. Oxford University Press, 1993.

Livros Grátis
( http://www.livrosgratis.com.br )
 
Milhares de Livros para Download:
 
Baixar livros de Administração
Baixar livros de Agronomia
Baixar livros de Arquitetura
Baixar livros de Artes
Baixar livros de Astronomia
Baixar livros de Biologia Geral
Baixar livros de Ciência da Computação
Baixar livros de Ciência da Informação
Baixar livros de Ciência Política
Baixar livros de Ciências da Saúde
Baixar livros de Comunicação
Baixar livros do Conselho Nacional de Educação - CNE
Baixar livros de Defesa civil
Baixar livros de Direito
Baixar livros de Direitos humanos
Baixar livros de Economia
Baixar livros de Economia Doméstica
Baixar livros de Educação
Baixar livros de Educação - Trânsito
Baixar livros de Educação Física
Baixar livros de Engenharia Aeroespacial
Baixar livros de Farmácia
Baixar livros de Filosofia
Baixar livros de Física
Baixar livros de Geociências
Baixar livros de Geografia
Baixar livros de História
Baixar livros de Línguas

Baixar livros de Literatura
Baixar livros de Literatura de Cordel
Baixar livros de Literatura Infantil
Baixar livros de Matemática
Baixar livros de Medicina
Baixar livros de Medicina Veterinária
Baixar livros de Meio Ambiente
Baixar livros de Meteorologia
Baixar Monografias e TCC
Baixar livros Multidisciplinar
Baixar livros de Música
Baixar livros de Psicologia
Baixar livros de Química
Baixar livros de Saúde Coletiva
Baixar livros de Serviço Social
Baixar livros de Sociologia
Baixar livros de Teologia
Baixar livros de Trabalho
Baixar livros de Turismo
 
 