Competitive Nonlinear Pricing
Competitive Nonlinear Pricing
Competitive Nonlinear Pricing
Abstract
A buyer of a divisible good faces several identical sellers. The buyer’s preferences
are her private information, and they may directly affect the sellers’ profits (common
values). Sellers compete by posting menus of nonexclusive contracts, so that the buyer
can simultaneously and privately trade with several sellers. We focus on the finite-
type case, and we provide a full characterization of pure-strategy equilibria in which
sellers post convex tariffs. All equilibria involve linear pricing. When the sellers’ cost
functions are linear and do not depend on the buyer’s type (private values), equilibria
exist and trade is efficient. Under common values, or when the sellers’ costs are strictly
convex, there is a severe form of market breakdown as at most one type of the buyer
may actively trade. Moreover equilibria exist only under restrictive conditions.
∗
We thank conference participants at the 7th ENSAI Economic Day and at the Université Paris-Dauphine
Workshop in Honor of Rose-Anne Dana for many useful discussions. Financial support from the Chaire
Marchés des Risques et Création de Valeur and the European Research Council (Starting Grant 203929–
ACAP) is gratefully acknowledged.
†
Toulouse School of Economics (IDEI, PWRI) and Università degli Studi di Roma “Tor Vergata.”
‡
Toulouse School of Economics (CNRS, GREMAQ, IDEI).
§
Toulouse School of Economics (INRA, LERNA, IDEI).
1 Introduction
Many markets for goods and services do not restrict in any way the ability of each trader
to sign secret, bilateral contracts with different partners. This prevents outside parties from
monitoring the whole of a trader’s activities. As a consequence, the formation of prices on
such nonexclusive markets is by nature a decentralized process, unlike on idealized markets
ruled by a Walrasian auctioneer. Bilateral contracts are necessarily incomplete as they
only bear on a fraction of each trader’s activity. Moreover, bilateral negotiations allow
to tailor contracts at will, at odds with contracts that are normalized for quotation. In
particular, contracts may be discriminatory, and the balance between supply and demand
may be ensured not by a single price, but by nonlinear tariffs. These tariff offers in turn are
formulated in a strategic environment in which sellers take into account both the reaction
of buyers and the other sellers’ offers. The aim of this paper is to understand the formation
of prices on nonexclusive markets. In the case of financial markets, our results shed light on
the robustness of organized exchanges such as limit-order books to trades that take place
“in the dark,” outside visible order books. As we will see, the nonexclusive nature of such
transactions is a major obstacle to the efficient functioning of these markets.
We study these issues in the context of the following model of trade under uncertainty.
There are two commodities, money and a physical good. Trade takes place between a buyer
and a finite number of sellers offering this good. The sellers first post possibly nonlinear
tariffs expressing how much they ask for any quantity of the good. The buyer then learns her
preferences and she decides which quantity to purchase from each seller. There is an arbitrary
finite number of states of nature. In each state, the buyer has strictly convex preferences.
These preferences are ordered across states according to how much she is willing to trade at
the margin, reflecting a strict single-crossing property. As for the sellers, they weakly prefer
to sell lower quantities when the buyer is more eager to trade, reflecting a reverse weak
single-crossing property. Our model thus encompasses private-value and adverse-selection
environments as special cases. In addition, sellers may have constant or increasing marginal
costs of serving the buyer in each state of nature.
In this context, we provide a complete characterization of pure-strategy equilibria in
which sellers post convex tariffs. Such tariffs can be interpreted as sequences of limit orders,
and are natural candidates to consider in nonexclusive models of trade with adverse selection
(Biais, Martimort, and Rochet (2000, 2013), Back and Baruch (2013)) or increasing marginal
costs (Biais, Foucault, and Salanié (1998)). Importantly, we allow sellers to deviate by
posting arbitrary nonconvex tariffs, so as to fit our definition of a nonexclusive market. Our
1
main result is that all equilibria must involve linear pricing. Hence competition in our model
is powerful enough to make a single equilibrium price emerge. Sellers then cannot benefit
from using nonlinear tariffs. When sellers have constant and state-independent marginal
costs, one ends up with a unique equilibrium outcome which is efficient in the strongest
sense, as it coincides with the equilibrium outcome of a perfectly competitive market.1 When
there is adverse selection or sellers have increasing marginal costs, linear-price equilibria are
such that the buyer trades in at most one state of nature, and does not trade at all in
any other state. Hence the market breaks down in a very strong sense. Moreover, in such
cases necessary conditions for the existence of an equilibrium are severe. An implication of
our analysis is that organized exchanges such as limit-order books can be destabilized by
decentralized exchanges such as over-the-counter markets.
Standard analyses of nonexclusive markets take linear pricing as a defining feature of
such markets. The opportunity to trade small quantities from several sellers, the argument
goes, allows buyers to arbitrage away any nonlinearities in the sellers’ tariffs. In line with
this intuition, Pauly (1974) analyzed a nonexclusive insurance market in which insurance
companies are restricted to post linear tariffs, and showed that equilibria then involve cross-
subsidies between sellers’ profits across states.2 Our analysis suggests that these outcomes
do not survive when strategic interactions between sellers are explicitly taken into account.
The intuition is that due to adverse selection or increasing marginal costs, the sellers face
a high demand from the buyer precisely in those states in which the cost of serving her is
high. To hedge against this risk, each seller has an incentive to deviate by proposing a limit
order specifying the maximal quantity of the good he is ready to trade at the standing price.
In these circumstances, linear pricing can be reconciled with nonexclusive competition only
if the buyer trades a positive quantity in at most one state. In contrast with this result,
Attar, Mariotti, and Salanié (2011) showed that the restriction to linear prices is without
loss of generality in a lemons market where an informed seller can trade up to a capacity
and all market participants have linear preferences. Cross-subsidies between states can then
resist limit-order deviations because, at any given unit price, and depending on the state, the
seller is either ready to trade up to the maximum quantity demanded at this price (as long
as it does not exceed her capacity) or prefers not to trade at all. By contrast, the informed
1
The existence of an efficient equilibrium in this Bertrand-like environment with private values is quite
straightforward. Still we could not find any previous work showing that no other equilibria with convex
tariffs can exist. A similar efficiency result appears in Pouyet, Salanié, and Salanié (2008), albeit in the case
of an exclusive market in which the buyer can trade with at most one seller.
2
The same restriction to linear pricing is postulated in recent analyses of the annuity market, which is
nonexclusive in many countries (Rothschild (2007), Sheshinski (2008), and Hosseini (2010)).
2
buyer in our model has strictly convex preferences and faces no capacity constraint. This
implies that, at any given unit price, the buyer typically has different aggregate demands in
different states. This in turn gives limit-order deviations their bite and destabilizes linear-
price candidate equilibria in which trade takes place in more than one state.
One may then turn to equilibria with nonlinear tariffs, in the hope that they yield more
trading under adverse selection or increasing marginal costs. Glosten (1994) proposed a
natural candidate in a framework in which the buyer faces an exogenously given tariff and
sellers have linear production costs. Specifically, he showed that there is a unique convex tariff
that resists entry. This tariff can be interpreted as a generalization of Akerlof (1970) pricing,
for marginal quantities. It specifies that each additional quantity above any quantity q is
sold at a price equal to the expected cost of serving it, conditional on the fact that the buyer
buys at least q. Under single crossing, this amounts to compute an upper-tail expectation,
namely, the expectation of the cost given that the buyer is ready to purchase at least q. In
each state, the buyer then trades exactly her demand at the tail price. An additional nice
property is that by construction such a tariff yields zero profit to the sellers.
In our setting, the question becomes whether we can find convex tariffs for the sellers
that once aggregated yield the Glosten (1994) tariff, and such that no seller can profitably
deviate by posting another tariff.3 Suppose that in equilibrium two different types of the
buyer, corresponding to two different states of nature, end up trading at two different tail
prices. Then there must exist a seller that sells more to the type trading at the highest
price than to the other type. Note that when facing this seller, the former type does not
want to deviate and choose the quantity traded by the latter type because, when tariffs
are convex, optimality conditions imply that all quantities traded by a given type with the
sellers are traded at the same price. On the other hand, the seller designs his tariff so as
to maximize his expected profit, under incentive-compatibility constraints. Given convex
tariffs and single crossing, we show that downward local incentive-compatibility constraints
must be binding at the solution of such a problem. But this contradicts the fact that the
highest type does not want to mimic the lowest type. Hence in a Glosten-like equilibrium
all trades must take place at the same price. Moreover, we show that the above logic also
applies to any convex tariff. Therefore, the only equilibria are linear-price equilibria. We are
then back to the conclusion that at most one type may trade in equilibrium under adverse
selection or increasing marginal costs.
Our results confirm those obtained by Attar, Mariotti, and Salanié (2013). That paper
3
This study was not performed in Glosten (1994), see the discussion in Glosten (1998).
3
examines the case with two states of nature, adverse selection, and constant marginal costs
in each state. A complete characterization of aggregate equilibrium allocations is provided,
with no restriction on equilibrium tariffs. It turns out that all equilibrium allocations can
be supported by linear tariffs, with at most one type trading. Focusing on equilibria with
convex tariffs, this paper shows that, strikingly, the result that the buyer may trade in at
most one state extends to an arbitrary finite number of states. We thus exhibit a new form
of market failure, characterized by a dramatic market breakdown that exceeds by far the
one first characterized by Akerlof (1970).
On the other hand, our results stand in stark contrast with those obtained in Biais,
Martimort, and Rochet (2000), who consider a parametric version of our model with a
quasilinear, quadratic utility function for the buyer, and constant marginal costs with adverse
selection for the sellers. The main difference is that the set of states is assumed to be
continuous, instead of finite as in this paper. This allows Biais, Martimort, and Rochet
(2000) to focus on equilibria with strictly convex tariffs.4 They show that such an equilibrium
exists, is unique in this class, and is symmetric across sellers. Moreover the buyer trades in
a nontrivial set of states in equilibrium, at a tariff between the perfectly competitive tariff
that would obtain under complete information and the monopoly tariff under incomplete
information. We thus exhibit in this paper a remarkable discontinuity between the finite-
state case and the continuous-state case: the equilibrium characterized in the latter case is
not a limit of equilibria in the former case as the number of types grows large.
The paper is organized as follows. Section 2 describes the model. Section 3 states and
discusses our central result, the proof of which is outlined in Section 4. Section 5 discusses
various extensions of our analysis. Section 6 concludes.
2 The Model
Our model features a buyer who can purchase nonnegative amounts of a divisible good from
several sellers. The good is homogeneous, so the buyer only cares about aggregate trade.
The possibility of adverse selection plays an important role, as in well-known models of
insurance provision, labor supply, or more generally competitive screening.
4
The buyer is privately informed of her preferences. Her type may take a finite number of
P
values in the set {1, . . . , I}, with positive probabilities mi such that i mi = 1. Each type of
the buyer only cares about the aggregate quantity Q ≥ 0 she purchases from the sellers and
the aggregate transfer T she makes in return. Type i’s preferences over aggregate quantity-
transfer bundles (Q, T ) are represented by a utility function ui defined over R+ ×R. For each
i, ui is assumed to be continuous and strictly quasiconcave in (Q, T ), and strictly decreasing
in T . The following strict single-crossing assumption is the main determinant of the buyer’s
behavior in our model, and is also used throughout the related literature.
In words, higher types are more eager to increase their purchases than lower types are.
At the end of our analysis, we shall also use an additional property that we now introduce.
For each p ∈ R, let Di (p) be type i’s demand at price p, that is, the unique solution to
The continuity and strict quasiconcavity of ui imply that Di (p) is uniquely defined and
continuous in p. Assumption 1 implies that for each p, Di (p) is nondecreasing in the buyer’s
type i. We strengthen this monotonicity property as follows.
A sufficient condition for both Assumptions 1 and 2 to hold is that the marginal rate of
substitution M RSi (Q, T ) of the good for money be well defined and strictly increasing in i
for all (Q, T ).
5
introduces adverse selection in our model: a higher type is willing to buy more, but faces
sellers that are more reluctant to sell.
To allow comparisons with the literature, we represent each seller’s preferences over trades
(q, t) by a linear profit function: if a seller provides type i with a quantity q and receives
a transfer t in return, he earns a profit t − ci q, where ci is the cost of serving type i. Our
reverse single-crossing property can thus be written as follows.
1. Sellers simultaneously post tariffs, which are mappings tk : R+ → R ∪ {∞} such that
tk (0) = 0. We let tk (q) ≡ ∞ if seller k does not offer the quantity q.
2. After privately learning her type, the buyer purchases a nonnegative quantity q k from
P
each seller k, for which she pays in total k tk (q k ).
A pure strategy for type i is a function si that maps any tariff profile (t1 , . . . , tK ) into a
quantity profile (q 1 , . . . , q K ). We let s = (s1 , . . . , sI ) be the buyer’s strategy. To ensure that
type i’s problem
( Ã !)
X X
max ui qk , tk (q k ) (1)
(q 1 ,...,q K )∈RK
+ k k
always has a solution, we require the tariffs tk to be lower semicontinuous, and the sets
{q ∈ R+ : tk (q) < ∞} to be compact. This definition is general enough to allow sellers to
offer menus containing a finite number of trades, including the (0, 0) trade. It also allows us
to use perfect Bayesian equilibrium as our equilibrium concept.
In line with Biais, Martimort, and Rochet (2000, 2013) and Back and Baruch (2013),
we focus on pure-strategy equilibria (t1 , . . . , tK , s) in which sellers post convex tariffs tk
that one can interpret as sequences of limit orders.5 Two elementary implications of this
5
By convention, all functions in Gothic letters refer to equilibrium objects.
6
restriction are worth mentioning at this stage. First, because the utility functions ui are
strictly quasiconcave, any type i has uniquely determined aggregate equilibrium demand Qi
and transfer Ti , which additionally are nondecreasing in i under Assumption 1. Second,
convexity of equilibrium tariffs is preserved under aggregation. In particular, suppose that
the buyer wishes to trade an aggregate quantity Q−k with the sellers other than k. Then
the minimum transfer she has to make in return is
( )
X 0 0 0
X 0
T−k (Q−k ) ≡ min tk (q k ) : q k ∈ R+ for all k 0 6= k and q k = Q−k . (2)
k0 6=k k0 6=k
0
The aggregate tariff T−k is the infimal convolution of the individual tariffs tk posted by the
sellers other than k, and is convex if each of them is convex (Rockafellar (1970)).
Theorem 1 Suppose that Assumptions 1–3 are satisfied, and let (t1 , . . . , tK , s) be an
equilibrium with convex tariffs. If some trade takes place in equilibrium, then
(i) All trades take place at unit price cI and each type i purchases Di (cI ) in the aggregate.
(ii) If Di (cI ) > 0, then ci = cI . Thus each seller earns zero profit on each trade.
The first insight of Theorem 1 is that nonexclusive competition leads to linear pricing, at
least when attention is restricted to equilibria with convex tariffs. This shows the disciplining
role of competition in our model: although sellers are allowed to propose arbitrary tariffs,
they end up trading at the same price.
From the standard Bertrand undercutting argument, this price cannot be strictly above
the highest possible cost cI . In an equilibrium it cannot lie below neither. If it did, then
sellers would want to limit the quantities they sell to the highest types, which they can
do by posting a limit order at the equilibrium price with a well-chosen maximum quantity.
We then have a tension between zero profits in the aggregate, and the high equilibrium
price cI . In the pure private-value case in which the cost ci is independent of the buyer’s
type i, this tension is easily relaxed, and we obtain the usual Bertrand result, leading to
an efficient outcome. By contrast, in the pure common-value case in which the cost ci is
strictly increasing with the buyer’s type i, our result implies that only the highest type I
may actively trade in equilibrium, whereas all types i < I must be excluded from trade.
7
This market failure is much more dramatic than in Akerlof (1970) or Rothschild and Stiglitz
(1976), as only a single type may actively trade in equilibrium.
Additionally conditions for the existence of an equilibrium are very restrictive: from
Theorem 1(ii) one must have Di (cI ) = 0 for all i < I if an equilibrium is to exist at all.
Hence the highest type must have preferences different enough from those of other types.
4 Proof Outline
Throughout this section, we suppose the existence of an equilibrium (t1 , . . . , tK , s) with
convex tariffs, and we investigate its properties. Recall that from the viewpoint of seller k
0
the aggregate tariff T−k of the sellers other than k can be computed from the tariffs tk as in
(2). In turn T−k determines how type i evaluates any bundle (q, t) she may trade with seller
k through the following indirect utility function
z−k −k −k −k
i (q, t) ≡ max {ui (q + Q , t + T (Q ))}. (3)
Q−k ∈R+
Observe that the maximum in (3) is always attained and that the indirect utility functions
z−k
i , when their value is finite, are strictly decreasing in t and continuous in (q, t).
6
Two types of arguments are used in the proof. Some rely only on the convexity of
tariffs and preferences. Because we only assume weak convexity, given a convex function
f : R+ → R we use the notation ∂f (x), ∂ − f (x), and ∂ + f (x) to denote respectively the
subdifferential of f at x, the minimum element of ∂f (x), and the maximum element of
∂f (x). Hence ∂f (x) = [∂ − f (x), ∂ + f (x)]. Other arguments rely on single-crossing properties,
in particular when it comes to examining the buyer’s best response to a deviation. Most
often the deviations we consider correspond to finite menus, including as many options as
there are types. We denote such a menu by {(0, 0), . . . , (qi , ti ), . . .}.
Finally, we say that individual quantities are nondecreasing if, given a family of tariffs,
the quantities qik traded by each type i with each seller k are such that for any k and i < I
k
one has qik ≤ qi+1 .
8
tariffs are affine with the same slope on some intervals of quantities, then the buyer may
have multiple best responses. Still we can show the following result.
Lemma 1 Let (t1 , . . . , tk ) be a family of convex tariffs. Then the buyer has a best response
to (t1 , . . . , tk ) with nondecreasing individual quantities.
The proof of Lemma 1 introduces some notations and additional results that will be
used later on. It only relies on convexity, by showing the existence of a best response with
individual quantities that are comonotonic with aggregate quantities.
Consider next the choice problem faced by the buyer in her relationship with any seller
0
k, fixing the equilibrium tariffs tk of the sellers other than k. From these tariffs one can
build T−k as in (2), and z−k
i as in (3). The convexity of the aggregate tariff T−k crucially
implies that the indirect utility functions z−k
i inherit a weak single-crossing property from
the primitive utility functions ui .
z−k −k 0 0 −k −k 0 0
i (q, t) ≤ zi (q , t ) implies zi0 (q, t) ≤ zi0 (q , t ), (4)
z−k −k 0 0 −k −k 0 0
i (q, t) < zi (q , t ) implies zi0 (q, t) < zi0 (q , t ). (5)
In words, higher types are more eager to buy higher quantities from a given seller. As
an application, suppose that seller k deviates and posts an arbitrary tariff tk . From the
viewpoint of seller k, type i’s maximization problem amounts to
max {z−k k k k
i (q , t (q ))}. (6)
q k ∈R+
9
For each type i of the buyer to select the trade (qi , ti ) in this menu, it must be that the
following incentive-compatibility and individual-rationality constraints hold for all i and i0 :
z−k −k
i (qi , ti ) ≥ zi (qi0 , ti0 ), (7)
z−k −k
i (qi , ti ) ≥ zi (0, 0). (8)
These constraints are not sufficient to ensure that each type i will choose to trade (qi , ti ) after
the deviation. Indeed, a given type may be indifferent between two trades, thus creating
some ties. The following result shows that, as long as he sticks to nondecreasing quantities,
seller k can secure the profit he would obtain if he could break ties in his favor. Define
( I )
X
Vk (t−k ) ≡ sup mi (ti − ci qi ) (9)
i=1
over all menus {(0, 0), . . . , (qi , ti ), . . .} that satisfy (7)–(8) for all i and i0 , and that have
nondecreasing quantities qi+1 ≥ qi for all i < I.
Lemma 3 In an equilibrium (t1 , . . . , tK , s) with convex tariffs, seller k’s profit is no less
than Vk (t−k ).
Any seller k can thus control the quantities he trades with the buyer if, given the other
sellers’ tariffs, he deviates to an incentive-compatible menu that displays nondecreasing
quantities. This last requirement is not a direct consequence of (7)–(8), given that the
buyer’s preferences only satisfy the weak single-crossing property characterized in Lemma 2.
Indeed, this requirement is likely to be costly because, given Assumption 3, any seller would
prefer to sell less to higher types. However, it cannot be dispensed with as the buyer always
has a best response with nondecreasing quantities. Therefore, Vk (t−k ) is the highest payoff
that seller k may expect by deviating, if he faces a buyer who systematically selects a best
response with nondecreasing quantities.
The proof for Lemma 3 goes as follows. Consider a menu of trades that verifies the
constraints in the Lemma, and suppose that two consecutive types i and i + 1 are both
indifferent between their trade and the other type’s trade. Then seller k can modify his
menu by pooling both types on the same trade. Under Assumption 3, because qi ≤ qi+1 this
can be done without reducing the profits on the right-hand side of (9). This first step is key
to the proof, as it shows that between two neighboring types only one incentive-compatible
constraint can be binding. The proof then shows that seller k can slightly perturb the
transfers in the menus so as to make all the relevant incentive-compatibility constraints
slack. Hence the buyer has a unique best response, which guarantees that seller k gets the
profit on the right-hand side of (9).
10
4.3 Equilibria with Nondecreasing Quantities
The above results suggest that we first focus on equilibria with nondecreasing individual
quantities, that is qik ≤ qi+1
k
for all k and i < I. In this section, we characterize these
equilibria. We then show in Section 4.4 that the latter restriction on the buyer’s behavior
actually is inconsequential.
So suppose that such an equilibrium (t1 , . . . , tK , s) exists. The equilibrium trades of seller
k then verify all the constraints in program (9). An immediate consequence of Lemma 3 is
thus that these trades must be solution to this program, and that the equilibrium profit of
seller k is equal to Vk (t−k ). Considering program (9), it is clear that for each type i at least
one constraint must bind, for, otherwise, one could slightly increase ti . Our next result relies
on Lemma 2 to determine which constraints are binding.
Therefore, from the perspective of each seller, the individual-rationality constraint binds
at the bottom, or more generally for all types below a threshold, and the downward local
incentive-compatibility constraints bind for all other types. This result is reminiscent of
those obtained under monopolistic screening, with the difference that they are formulated
in terms of the indirect utility functions z−k
i instead of the primitive utility functions ui .
Under monopolistic screening, the aim is to characterize Pareto-optimal allocations, which
implies that ties are broken in the most favorable way to the monopolist.7 In our competitive
setting, Lemma 3 offers a condition under which the seller can break ties as desired, namely,
that quantities are nondecreasing. This allows us to proceed without introducing further
restrictions on the buyer’s behavior.
Our next result builds on Lemma 4 to show that equilibria with convex tariffs and
nondecreasing quantities actually feature linear pricing if trade takes place at all.
11
The proof of Lemma 5 goes as follows. When sellers offer convex tariffs, every best
response of each type i is such that she buys the last unit of the good at some price pi ,
independently of the sellers she trades with. Because the corresponding aggregate quantity
is nondecreasing in the type, it is easily shown that one must have pi ≥ pi−1 . Consider now
an equilibrium, and suppose that type i trades at a price pi > pi−1 . Clearly, it is not optimal
k
for type i to mimic type i − 1 and trade the quantity qi−1 with seller k, as this would imply
trading at a marginal price different from pi . Hence the downward local incentive constraint
from type i to type i − 1 cannot bind. A fortiori, it is not optimal for type i to trade a zero
quantity with seller k. Hence the individual rationality constraint of type i cannot bind.
But these results contradict Lemma 4.
We now show that each equilibrium trade must yield zero profit to the seller who makes
it. The intuition is simple. Under linear pricing, sellers collectively have to share a risky
demand Di (p). Under Assumption 2, we know that DI (p) > DI−1 (p) if some trade takes
place at all, so the price p must be high enough to convince some of the sellers to provide
additional quantities to the highest type. In fact, one must have p ≥ cI , otherwise a seller
k
could deviate by posting a limit order with unit price p and maximum quantity qI−1 . On the
other hand, aggregate profits cannot be positive, by a standard Bertrand argument. Because
cI is the highest possible cost, we get the following result.
12
another lower bound by imposing in program (9) the additional constraint that the transfers
ti must be computed using the equilibrium schedule tk . So define
( I )
X
Vk (t1 , . . . , tK ) ≡ sup mi [tk (qi ) − ci qi ] (10)
i=1
z−k k −k k
i (qi , t (qi )) ≥ zi (qi0 , t (qi0 )), (11)
z−k k −k
i (qi , t (qi )) ≥ zi (0, 0), (12)
and such that qi+1 ≥ qi for all i < I. By Lemma 3, we therefore have
for all k. Now, recall from Lemma 1 that the buyer has at least one best response with
nondecreasing individual quantities. Choose one such best response, and let v0k be the
resulting profit for seller k. Because the corresponding trades for seller k verify the constraints
in the above program, one must have
for all k. Finally, given the convexity of the tariffs (t1 , . . . , tK ), the aggregate quantities Qi
and the aggregate transfers Ti are the same for any best response of the buyer. Due to
P P P
the linearity of the sellers’ profits, we get k vk = i mi [Ti − ci Qi ] = k v0k . Using the
inequalities (13)–(14), we finally obtain vk = Vk (t−k ) = Vk (t1 , . . . , tK ) = v0k for all k.
This proves in particular that, in any equilibrium, each seller k earns Vk (t−k ). Therefore,
no seller can get more than the profit he could secure by sticking to nondecreasing quantities.
If we now specify that the buyer’s strategy must select nondecreasing quantities whenever
possible, it is easily understood that with this new strategy we have built an equilibrium
with nondecreasing individual quantities. This last result is proven more formally in the
Appendix.
Note that the aggregate equilibrium quantities Qi and the indirect utility functions z−k
i
are the same in the initial and the final equilibrium. Combining Lemmas 5, 6, and 7 then
shows that Theorem 1 applies to all equilibria with convex tariffs.
13
5 Extensions
So far, we have assumed that sellers have constant and possibly type-dependent marginal
costs. An examination of the proof of Lemmas 1–5 reveals that we can handle much more
general cases. We now endow each seller k with a profit function vik (q, t), which we take to
be continuous and strictly increasing in t, and such that the following generalized reverse
single-crossing assumption holds.
Each seller k therefore weakly prefers to sell lower quantities to higher types. Then the
following result holds.
Corollary 1 Under Assumptions 1–2 and 4, any equilibrium with convex tariffs and
nondecreasing individual quantities such that some trade takes place in equilibrium displays
linear pricing: there exists p ∈ R such that all trades take place at unit price p, and each type
i purchases Di (p) in the aggregate.
Extending this result to equilibria with quantities that may be decreasing requires some
additional structure. Assume that each seller’s cost of providing type i with a quantity q is
ci (q), where ci : R+ → R+ is now a strictly convex cost function, with ci (0) = 0. In this
setting, the analogue of Assumption 4 can be stated in terms of the one-sided derivatives of
these cost functions.
∂ − ci0 (q 0 ) ≥ ∂ + ci (q).
Theorem 2 Suppose that Assumptions 1–2 and 5 are satisfied, and let (t1 , . . . , tK , s) be an
equilibrium with convex tariffs. If some trade takes place in equilibrium, then there exists
p ∈ R solution to
µ ¶
DI (p)
p ∈ ∂cI (15)
K
and such that:
14
(i) All trades take place at unit price p and each type i purchases Di (p) in the aggregate,
and Di (p)/K from each seller.
When there is a single type I, this result states that any equilibrium is competitive in
the sense that the equilibrium price equalizes type I’s demand and the sum of the sellers’
supplies. Equilibrium outcomes are hence first-best efficient, as in the case of linear costs.
The introduction of multiple types does not affect this property, the only change being that
all types below I must demand a zero quantity at the equilibrium price.
The structure of the proof of Theorem 2 is similar to that of Theorem 1. First, given
Corollary 1, one has to show that the result holds for all equilibria with convex tariffs and
nondecreasing individual quantities.
The result that no trade may take place except perhaps at the top of the buyer’s type
distribution now holds whether or not the environment features common values. As in the
linear cost case, sellers collectively have to share a risky demand Di (p), but under convex
costs the precise sharing now matters. Under Assumption 2, we know that DI (p) > DI−1 (p),
so the price p must be high enough to convince some of the sellers to provide additional
quantities to the highest type. In fact, one must have p ≥ ∂ − cI (qIk ) for all k, otherwise seller
k could deviate by posting a limit order with a unit price p and a maximum quantity slightly
below qIk . But, at such a high price, sellers are willing to sell high quantities to lower types,
which is consistent with equilibrium only if all these types demand a zero quantity.
To complete the proof of Theorem 2, there thus only remains to show that the restriction
to equilibria with nondecreasing individual quantities is innocuous. To this end, consider
an equilibrium (t1 , . . . , tK , s) with convex tariffs. Denote by vk the equilibrium profits of
seller k. Replacing the linear cost functions in the definitions (9) and (10) of Vk (t−k ) and
Vk (t1 , . . . , tK ) by the now convex cost functions, we formally get the lower bound (13) for
the profits vk . On the other hand, the sum of these profits cannot exceed the value they
would reach if the buyer were to break ties in favor of the coalition of sellers. Formally,
15
define
( )
XX
V0 (t1 , . . . , tK ) ≡ sup mi [tk (qi ) − ci (qi )] (17)
k i
over all (q1 , . . . , qI ) ∈ RI+ that satisfy (11)–(12). Note that we do not impose the constraint
that quantities be nondecreasing. We thus have
X
V0 (t1 , . . . , tK ) ≥ vk . (18)
k
under the same constraints, as the aggregate transfer chosen by the buyer is uniquely defined
given the tariffs. The proof of Lemma 1 shows that such a risk-sharing problem admits a
solution with nondecreasing individual quantities: this is the efficient manner to share risk.
P
Let v0k be the associated profit for seller k; note that k v0k = V0 (t1 , . . . , tK ). Moreover, in
such a solution, each seller k trades a family of quantities that are nondecreasing, and thus
his associated profit v0k must be no more than Vk (t1 , . . . , tK ). Summarizing, we get from
(13) and (18) that
X X X X X
Vk (t1 , . . . , tK ) ≥ v0k = V0 (t1 , . . . , tK ) ≥ vk ≥ Vk (t−k ) ≥ Vk (t1 , . . . , tK ),
k k k k k
and thus these inequalities are in fact equalities. In particular, this implies for every k that
vk = Vk (t−k ). We can then apply Lemma 7 without changes.
16
Appendix
Proof of Lemma 1. Recall that given a family (t1 , . . . , tK ) of convex tariffs, the aggregate
equilibrium demand Qi of type i is uniquely defined and nondecreasing in i. Given Qi , type
i’s utility-maximization problem (1) reduces to minimizing total payment for Qi :
( )
X X
min tk (q k ) : q k ∈ R+ for all k and q k = Qi .
k k
This is a convex program, so that by the Kuhn–Tucker theorem one can associate to any
solution (q 1 , . . . , q K ) a Lagrange multiplier pi such that pi ∈ ∂tk (q k ) for all k. If there are
two different solutions (q 1 , . . . , q K ) and (q 01 , . . . , q 0K ) with different multipliers pi < p0i , then
because each tariff is convex one obtains q k ≤ q 0k for all k, and because both solutions
sum to the same Qi they must be identical, a contradiction. This shows that two different
solutions must share the same pi . Consequently one can associate to each type i a price pi
such that whatever the solution (q 1 , . . . , q K ) to type i’s problem, one has pi ∈ ∂tk (q k ) for all
k. Moreover, by the same argument as above, pi is nondecreasing in i.
For each i and each k, one can thus build the nonempty set {q : pi ∈ ∂tk (q)}. Let ski
be its minimum element, and let ski be its maximum. Both ski and ski are nondecreasing
in i. The interval [ski , sik ] is in fact the set of quantities that are provided by seller k at a
marginal price equal to pi . If this interval is nontrivial, then tk is affine over it, with slope
pi . Consequently solutions (qi1 , . . . , qiK ) to type i payment minimization problem must verify
X
qik = Qi and ski ≤ qik ≤ ski for all k, (19)
k
and these conditions are in fact sufficient, as all tariffs have the same slope pi for quantities
in these intervals. Our problem thus reduces to find a family of nondecreasing quantities
verifying (19). We in fact prove a stronger result, which will be useful for future reference.
Choose a family of strictly convex functions (f1 , . . . , fI ), and consider the following family
of problems, indexed by i:
( )
X
min fi (qik )
k
subject to (19). By strict convexity of the functions fi , each such problem admits a unique
solution. We show below that the family of these solutions must display nondecreasing
individual quantities. This naturally implies the existence of a family with nondecreasing
individual quantities verifying (19), and shows the lemma.
17
To do so, proceed by contradiction and suppose that a family of solutions has qik > qi+1
k
,
for some k and i < I. Under (19), this implies
Because the intervals for i and i + 1 have a nontrivial intersection, it must be that pi = pi+1 .
0 0 0 0
Therefore, for any seller k 0 we have ski = ski+1 and ski = ski+1 . Moreover, because qik > qi+1
k
0 0
and Qi ≤ Qi+1 , we know that there exists k 0 6= k such that qik < qi+1
k
. Using the equalities
we have just shown, this implies
0 0 0 0 0 0
k
ski = ski+1 ≤ qik < qi+1 ≤ ski = ski+1 . (21)
0
Given (20)–(21), one can slightly reduce qik and increase qik by the same amount, so that
P
(19) is still verified. Because (qi1 , . . . , qiK ) is assumed to minimize k fi (qik ), it must be that
0
at the margin −∂ − fi (qik ) + ∂ + fi (qik ) ≥ 0. Because fi is strictly convex, this implies that
0 0
qik ≤ qik . Alternatively, one could slightly increase qi+1
k k
, and reduce qi+1 by the same amount.
0
Once more, it must be that at the margin ∂ + fi+1 (qi+1
k
) − ∂ − fi+1 (qi+1
k
) ≥ 0. Because fi+1 is
k 0 k
strictly convex, this implies that qi+1 ≤ qi+1 .
0 0
Overall we thus have shown that qik ≤ qik < qi+1
k k
≤ qi+1 , in contradiction with our
assumption that qik > qi+1
k
. This concludes the proof. ¥
Proof of Lemma 2. Fix some k, i < I, q < q 0 , t, and t0 . Let T(Q) ≡ t+T−k (Q−q), defined
for Q ≥ q. Similarly, let T0 (Q) ≡ t0 + T−k (Q − q 0 ), defined for Q ≥ q 0 . According to (3),
computing z−k
i (q, t) amounts to maximize ui (Q, T(Q)) with respect to Q ≥ q. Let Qi ≥ q be
the solution to this problem; it is unique as ui is strictly quasiconcave and strictly decreasing
in aggregate transfers, and T(Q) is convex in Q. Similarly, computing z−k 0 0
i (q , t ) amounts to
maximize ui (Q, T0 (Q)) with respect to Q ≥ q 0 . Let Q0i ≥ q 0 be the unique solution to this
problem. Suppose that
z−k −k 0 0
i (q, t) < zi (q , t ) (22)
and let i0 > i. Because Qi0 ≥ q is an admissible candidate in the problem that defines
z−k
i (q, t), we must have
z−k 0 0 0 −k 0 0
i0 (q, t) = ui0 (Qi0 , T(Qi0 )) < ui0 (Qi , T (Qi )) ≤ zi0 (q , t ),
18
where the last inequality stems from the fact that Q0i ≥ q 0 is an admissible candidate in the
problem that defines z−k 0 0
i0 (q , t ). This shows (5) in this case.
Otherwise we have Qi0 ≥ Q0i ≥ q 0 . Then Qi0 is an admissible candidate in the problem
that defines z−k 0 0
i0 (q , t ), and we get
z−k 0 −k 0 0
i0 (q, t) = ui0 (Qi0 , T(Qi0 )) < ui0 (Qi0 , T (Qi0 )) ≤ zi0 (q , t ),
z−k 0 0 0 0 0 −k 0 0
i (q, t) ≥ ui (Qi , T(Qi )) ≥ ui (Qi , T (Qi )) = zi (q , t ),
in contradiction with (22). Hence we have shown (5). The proof of (4) follows by continuity.
Indeed, assume that z−k −k 0 0 −k −k
i (q, t) = zi (q , t ). Then, for each ε > 0, zi (q, t + ε) < zi (q, t) and
thus z−k −k −k
i0 (q, t + ε) < zi0 (q, t) from (5). Because zi0 is continuous, one can take limits as ε
Step 1 Pick a menu µ = {(0, 0), . . . , (qi , ti ), . . .} that satisfies the incentive-compatibility
and individual-rationality constraints (7)–(8) for all i and i0 , and that has nondecreasing
quantities qi+1 ≥ qi for all i < I. We build a new menu µ0 = {(0, 0), . . . , (qi0 , t0i ), . . .}
by applying the following algorithm. At each step n ≥ 0 of the algorithm, let µ(n) =
© ¡ (n) (n) ¢ ª
(0, 0), . . . , qi , ti , . . . be the current menu, with µ(0) ≡ µ by convention. If there
(n) (n)
exists i < I such that qi < qi+1 and the following local incentive-compatibility constraints
both bind:
¡ (n) (n) ¢ ¡ (n) (n) ¢
z−k
i q i , ti = z−k
i qi+1 , ti+1 ,
¡ (n) (n) ¢ ¡ (n) (n) ¢
z−k q ,
i+1 i+1 i+1 t = z −k
i+1 i , ti
q ,
then take the smallest such i, i(n) , and pool types i(n) and i(n) + 1 on the same trade
¡ (n+1) (n+1) ¢ ¡ (n+1) (n+1) ¢ ¡ (n) (n) ¢ ¡ (n) (n) ¢
qi(n) , ti(n) = qi(n) +1 , ti(n) +1 equal to either qi(n) , ti(n) or qi(n) +1 , ti(n) +1 according to
19
the maximum value it gives to the profit on the (i(n) , i(n) + 1) pair
h i h i
(n+1) (n+1) (n+1) (n+1)
mi(n) ti(n) − ci(n) qi(n) + mi(n) +1 ti(n) − ci(n) +1 qi(n) .
Otherwise, the algorithm stops, and µ0 ≡ µ(n) . Note that the algorithm stops in a finite
number of steps as there are finitely many types. Moreover, applying the algorithm only
affects the way ties are broken. Therefore, the menu µ0 remains incentive compatible and
0
individually rational. Moreover, by construction, it has nondecreasing quantities qi+1 ≥ qi0
for all i < I. Finally, at each step of the algorithm, seller k’s profit cannot be decreased.
(n) (n)
Indeed, the algorithm is active at step n ≥ 0 only if qi(n) < qi(n) +1 . In that case, either
(n) (n) (n) (n) (n) (n)
ti(n) − ci(n) qi(n) < ti(n) +1 − ci(n) qi(n) +1
¡ (n) (n) ¢
and then seller k’s profit is increased by pooling i(n) and i(n) + 1 on qi(n) +1 , ti(n) +1 , or
(n) (n) (n) (n) (n) (n)
ti(n) − ci(n) qi(n) ≥ ti(n) +1 − ci(n) qi(n) +1 ,
(n) (n)
so that, as ci(n) ≤ ci(n) +1 by Assumption 3 and qi(n) < qi(n) +1 by construction, seller k’s profit
¡ (n) (n) ¢
cannot be decreased by pooling i(n) and i(n) + 1 on qi(n) , ti(n) . As a result,
I
X I
X
mi (t0i − ci qi0 ) ≥ mi (ti − ci qi ), (23)
i=1 i=1
Step 2 We may now proceed to the second step of the proof. Let ε > 0 be given.
We are going to modify transfers (t01 , . . . , t0I ) into transfers (t001 , . . . , t00I ) such that the menu
µ00 = {(0, 0), . . . , (qi0 , t00i ), . . .} satisfies the following incentive-compatibility and individual-
rationality constraints for all i and i0 :
z−k 0 00 −k 0 00
i (qi , ti ) ≥ zi (qi0 , ti0 ), (24)
00
z−k 0 −k
i (qi , ti ) ≥ zi (0, 0), (25)
where now these inequalities are strict as soon as, respectively, qi0 6= qi00 and qi0 6= 0. Moreover,
we will perform this modification in such a way that transfers remain almost the same:
Suppose this modification performed. Then for each ε > 0 seller k could deviate to the menu
µ00 . Because of the above properties, each type i must then choose to trade (qi0 , t00i ) with seller
k. Hence by playing so seller k can secure a profit
I
X I
X I
X
mi (t00i − ci qi0 ) ≥ mi (t0i − ci qi0 ) − ε ≥ mi (ti − ci qi ) − ε,
i=1 i=1 i=1
20
where the first and second inequalities follow from (26) and (23). As ε can be made arbitrarily
small, this shows that seller k’s equilibrium profit is at least (9), and the result follows.
To conclude the proof, there only remains to modify the transfers as announced above.
We now turn to this task. Because the quantities (q10 , . . . , qI0 ) are given, and because the
functions z−k
i are continuous and strictly decreasing in transfers, we can define two families
of (extended) real-valued functions γik and δik for i < I such that, for each t,
z−k 0 −k 0 k −k 0 −k 0 k
i (qi , t) = zi (qi+1 , γi (t)) and zi+1 (qi , t) = zi+1 (qi+1 , δi (t)). (27)
Here γik (t) = −∞ or δik (t) = −∞ by convention if there exists no solution to the relevant
equation, which may occur for t below some threshold. Both γik and δik are continuous
and strictly increasing where they are finite. If qi0 = qi+1
0
, then clearly γik (t) = δik (t) = t.
If qi0 < qi+1
0
, then, according to (4) along with the fact that the functions z−k
i are strictly
decreasing in transfers, γik (t) ≤ δik (t) for all t ≥ 0. Finally, by construction of the menu µ0 ,
if qi0 < qi+1
0
, then γik (t0i ) ≤ t0i+1 ≤ δik (t0i ) with at least one strict inequality. (One may have
γik (t0i ) = −∞, but δik (t0i ) is necessarily finite.)
Given ε > 0, we can recursively construct a family of strictly positive real numbers
(ε1 , . . . , εI ) as follows. Let εI ≡ ε. Then, for each i < I, consider εi+1 > 0 as given. If
qi0 = qi+1
0
, choose εi such that 0 < εi < εi+1 , which is clearly feasible. If qi0 < qi+1
0
, choose εi
such that 0 < εi < εi+1 /2 and such that
εi+1 εi+1
γik (t) < δik (t) and γik (t) − < t0i+1 < δik (t) + (28)
2 2
for all t that satisfy |t − t0i | < εi . This is feasible because if εi+1 > 0, all these properties
hold for t = t0i , and because the functions γik and δik are continuous at t0i . Observe that the
family (ε1 , . . . , εI ) is strictly increasing.
We now recursively construct a family of transfers (t001 , . . . , t00I ) such that |t00i − t0i | < εi for
all i. Set t001 ≡ t01 if q10 = 0, and set t001 ≡ t01 − ε1 /2 otherwise. Note that |t001 − t01 | < ε1 . Suppose
next that |t00i − t0i | < εi for some i < I, and define t00i+1 as follows:
0
(i) If qi+1 = qi0 , set t00i+1 ≡ t00i . Note that because t0i+1 = t0i in this case, we then have
|t00i+1 − t0i+1 | = |t00i − t0i | < εi < εi+1 , as required.
0
(ii) If qi0 < qi+1 , then, as |t00i − t0i | < εi by assumption, we know from the first part of (28)
that γik (t00i ) < δik (t00i ). Choose any ε̂ such that 0 < ε̂ < min {εi+1 /2, δik (t00i ) − γik (t00i )},
and consider the following three subcases. If t0i+1 ≥ δik (t00i ), set t00i+1 ≡ δik (t00i ) − ε̂. If
t0i+1 ≤ γik (t00i ), set t00i+1 ≡ γik (t00i ) + ε̂. Otherwise, set t00i+1 ≡ t0i+1 . The second part of
21
(28) ensures that in each of these three subcases |t00i+1 − t0i+1 | < ε̂ + εi+1 /2 < εi+1 , as
required.
By construction, we have γik (t00i ) < t00i+1 < δik (t00i ) if qi0 < qi+1
0
. This shows that the local
incentive-compatibility constraints
z−k 0 00 −k 0 00
i (qi , ti ) ≥ zi (qi+1 , ti+1 ),
z−k 0 00 −k 0 00
i+1 (qi+1 , ti+1 ) ≥ zi+1 (qi , ti )
0
are satisfied, with strict inequalities if qi0 < qi+1 . Similarly, our choice of t001 ensures that the
individual rationality constraint for i = 1,
z−k 0 00 −k
1 (q1 , t1 ) ≥ z1 (0, 0),
is also a strict inequality if q10 6= 0. Given the single-crossing property (4)–(5), a standard
argument can be used to establish that this set of local constraints implies that the menu µ00
satisfies the incentive-compatibility and individual-rationality constraints (24)–(25) for all i
and i0 , with strict inequalities as soon as, respectively, qi0 6= qi00 and qi0 6= 0. Finally, for each
i we have |t00i − t0i | < εi < εI = ε, which yields (26). The result follows. ¥
z−k k k k −k k k k −k k k k −k
j (qj , t (qj )) > zj (qj−1 , t (qj−1 )) and zj (qj , t (qj )) > zj (0, 0) (29)
for some j ≥ 2. Because the quantities (q1k , . . . , qIk ) are given, we can define a family of
(extended) real-valued functions δik for i < I as in (27). Using this notation, the first
inequality in (29) equivalently says that tk (qjk ) < δj−1
k
(tk (qj−1
k
)). Now, choose ε such that
k
0 < ε < δj−1 (tk (qj−1
k
)) − tk (qjk ) and define a family of transfers (t01 , . . . , t0I ) as follows: for
i < j, set t0i ≡ tk (qik ); for i = j, set t0j ≡ tk (qjk ) + ε; for i > j, define recursively t0i ≡
k
max {tk (qik ), δi−1 (t0i−1 )}.
The menu {(0, 0), . . . , (qik , t0i ), . . .} has three noticeable features:
(i) It has the same nondecreasing quantities and higher—sometimes strictly higher—
transfers as the equilibrium allocation.
(ii) It satisfies the incentive-compatibility constraints (7). This is obvious for types i <
j, because their transfers are unchanged, whereas the transfers of types i0 ≥ j are
(weakly) increased. As for type j, observe that she cannot be better off mimicking
type j − 1 because t0j = tk (qjk ) + ε, t0j−1 = tk (qj−1
k
), and ε has been chosen so that
22
z−k k k k −k k k k k k
j (qj , t (qj ) + ε) > zj (qj−1 , t (qj−1 )). Using the fact that the quantities (q1 , . . . , qj )
are nondecreasing along with the single-crossing property (4), it follows that type j
cannot be better off mimicking any type i < j − 1 either. Finally, type j cannot
be better off mimicking any type i > j. Indeed, suppose by way of contradiction
that i > j is the first type such that z−k k 0 −k k 0 −k k 0
j (qj , tj ) < zj (qi , ti ). Then zj (qi−1 , ti−1 ) =
z−k k 0 −k k 0 k k
j (qj , tj ) < zj (qi , ti ) so that, from qi−1 ≤ qi along with the single-crossing property
z−k k 0 −k k k 0 k
i (qi−1 , ti−1 ) < zi (qi , δi−1 (ti−1 )), in contradiction with the definition (27) of δi−1 .
The claim follows. The proof that no type i > j can be better off mimicking any type
i0 6= i is similar, and is therefore omitted.
(iii) It satisfies the individual-rationality constraints (8). This is obvious for types i <
j, because their trade is unchanged. Thus in particular z−k k 0 −k
j−1 (qj−1 , tj−1 ) ≥ zi (0, 0).
i0 ≥ j, from which the claim follows as the menu {(0, 0), . . . , (qik , t0i ), . . .} is incentive
compatible.
Given (i)–(iii), we can apply Lemma 3 to conclude that this menu must give seller k at most
his equilibrium profit. This however contradicts (i) above. The contradiction establishes
that for each j ≥ 2, at least one inequality in (29) must bind. The proof that the individual
rationality constraint (8) binds at i = 1 is similar, and is therefore omitted. ¥
23
We can then iterate the reasoning until reaching type 1, in contradiction with i ≥ 2.
We can now complete the proof of Lemma 5. Suppose that for some i > 1 we have
pi > pi−1 . Then ski ≥ ski−1 ≥ qi−1
k
for all k. If for k the individual-rationality constraint (8) of
type i binds in equilibrium, then it must be that ski = 0, and therefore that qi−1
k
= 0 = ski−1 .
If for k the individual-rationality constraint (8) of type i is slack in equilibrium, then from
Lemma 4 the incentive-compatibility constraint constraint (7) for i0 = i − 1 must bind in
equilibrium, which implies ski = qi−1
k
= ski−1 . Hence for all k we have qi−1
k
= ski−1 . Applying
our preliminary result, we have shown that pi > pi−1 implies that Qi−1 = 0. Because the
aggregate quantity is nondecreasing, it must thus be that all trades take place at the same
price p. Moreover, once more applying our preliminary result, a type i who actively trades
cannot exhaust the aggregate supply at price p, and thus can freely choose her most preferred
quantity at price p, which is Di (p). The result follows. ¥
Proof of Lemma 6. Any seller k could deviate to a menu that would allow types i < I to
buy the equilibrium quantity qik at price p, whereas type I would be asked to buy only qI−1
k
at
price p. Such an offer is incentive compatible and individually rational, with nondecreasing
quantities. From Lemma 3, the variation in the deviator’s profit must be at most zero,
k
(p − cI )(qI−1 − qIk ) ≤ 0.
Summing on k yields (p − cI )[DI−1 (p) − DI (p)] ≤ 0, and under Assumption 3 this implies
that p ≥ cI . On the other hand, if aggregate profits are strictly positive then, because
the functions Di are continuous, any buyer k could claim almost all profits for himself by
charging a uniform unit price slightly below p. So aggregate profits are zero, and under
p ≥ cI we get that p = ci = cI for any type i who actively trades. The result follows. ¥
Proof of Lemma 7. From Lemma 1 we know that the buyer has a best response q ∈ RIK
+
with nondecreasing individual quantities. Construct a strategy ŝ for the buyer as follows:
(i) When the buyer faces the tariff profile (t1 , . . . , tK ), she trades the quantities in q.
(ii) When the buyer faces a unilateral deviation (tk , t−k ) from the tariff profile (t1 , . . . , tK )
by seller k, we know as a consequence of Lemma 2 that there exists a solution to (6)
that is nondecreasing in i. Given the tariffs (tk , t−k ), this corresponds to a solution to
the buyer’s utility-maximization problem that has nondecreasing quantities for seller
k. Let ŝ select this solution.
(iii) In all other cases, the strategy ŝ simply selects a best response.
24
It remains to show that (t1 , . . . , tK , ŝ) is an equilibrium. First, note that for ŝ is indeed a best
response for the buyer. Second, we have shown in the text that each seller k earns exactly
Vk (t−k ). Third, if seller k were to deviate, then according to ŝ the buyer would react by
trading with k a family of nondecreasing quantities. Then the resulting profit cannot exceed
Vk (t−k ), by definition of this threshold. The result follows. ¥
Proof of Corollary 1. Lemmas 1–2 remain valid, as they are only concerned with the
buyer’s behavior, and not with the sellers’ profits. Our tie-breaking result, Lemma 3, needs
obvious modifications in the definition of Vk (t−k ), which becomes the supremum of
I
X
mi vik (qi , ti )
i=1
over the relevant set. Step 1 in the proof of Lemma 3 can easily be extended thanks to
Assumption 4, as this condition is enough to ensure that when qi < qi+1 a seller cannot loose
by pooling both types on either (qi , ti ) or (qi+1 , ti+1 ). Indeed, one will choose to pool both
types on (qi+1 , ti+1 ) when
which allows to pool both types on (qi , ti ) without reducing profits. Step 2 of the proof
requires no modification, as it only relies on modifications of transfers, leaving quantities
and hence costs unaffected. Lemma 4 then follows immediately: the proof is the same as in
the constant-marginal-cost case. Finally Lemma 5 also goes through. The result follows. ¥
Proof of Lemma 8. Fix an equilibrium with convex tariffs and nondecreasing individual
quantities in which all trades take place at price p and accordingly each type i purchases
Di (p) in the aggregate. We first prove that (16) holds if at least one type actively trades in
equilibrium, that is, if DI (p) > 0. Fix some k and let
© ª
C1 = (q1 , . . . , qI ) ∈ RI+ : there exists (t1 , . . . , tI ) such that (7)–(8) hold for all (i, i0 ) ,
© ª
C2 = (q1 , . . . , qI ) ∈ RI+ : qi+1 ≥ qi for all i < I ,
C3 = [0, D1 (p)] × . . . × [0, DI (p)].
25
for all k, where the equality in (30) follows from Lemma 3 given that the equilibrium has
nondecreasing individual quantities. In particular, vk is no larger than
( I )
X
k
v ≡ max mi [pqi − ci (qi )] : (q1 , . . . , qI ) ∈ C2 ∩ C3 . (31)
i=1
The relaxed problem (31) is compact and strictly convex, and therefore admits a unique
solution (q1k , . . . , qIk ). Using the strict convexity of the cost functions ci and Assumption
5, we get that for each i < I, qik < qi+1
k
implies that qik = Di (p). As a result, we have
qik = min {qIk , Di (p)} for all i. Consider now the menu {(0, 0), . . . , (qik , pqik ), . . .}. By the
single-crossing property (4), this menu satisfies the incentive-compatibility and individual-
rationality constraints (7)–(8); moreover, it has nondecreasing quantities at most equal to
DI (p). Thus (q1k , . . . , qIk ) ∈ C1 ∩ C2 ∩ C3 . Because the above menu yields a profit vk to
seller k, we get from (30)–(31) that vk = vk and, by the strict concavity of the common
objective function in (30)–(31), that (q1k , . . . , qIk ) is the unique solution to (30). Because
the equilibrium quantities traded by seller k must solve (30), we thus get that they are
exactly given by (q1k , . . . , qIk ). Moreover, because (q1k , . . . , qIk ) solves (31), which is independent
of k, so must be (q1k , . . . , qIk ), which we can thus write as (q1 , . . . , qI ). All sellers hence
trade the same quantities with each type in equilibrium, so that Kqi = Di (p) for all i. As
qi = min {qI , Di (p)} for all i, two cases may now arise. If qi = Di (p), then necessarily
Di (p) = 0. Alternatively, if qi < Di (p), then qi = qI and thus Qi = DI (p). By Assumption
2, (16) must thus hold as soon as DI (p) > 0, as claimed. To prove that (15) must then hold,
simply observe by (31) that qI = DI (p)/K > 0 maximizes pq − cI (q). The result follows. ¥
26
References
[1] Akerlof, G.A. (1970): “The Market for “Lemons”: Quality Uncertainty and the Market
Mechanism,” Quarterly Journal of Economics, 84(3), 488–500.
[2] Aliprantis, C.D., and K.C. Border (2006): Infinite Dimensional Analysis: A Hitchhiker’s
Guide, Berlin, Heidelberg, New York: Springer.
[3] Attar, Andrea, Thomas Mariotti, and François Salanié (2011): “Nonexclusive Competi-
tion in the Market for Lemons,” Econometrica, 79(6), 1869–1918.
[4] Attar, Andrea, Thomas Mariotti, and François Salanié (2013): “Nonexclusive Competi-
tion under Adverse Selection,” Theoretical Economics, forthcoming.
[5] Back, K., and S. Baruch (2013): “Strategic Liquidity Provision in Limit Order Markets,”
Econometrica, 81(1), 363–392.
[6] Biais, B., T. Foucault, and F. Salanié (1998): “Floors, Dealer Markets and Limit Order
Markets,” Journal of Financial Markets, 1(3–4), 253–284.
[7] Biais, B., D. Martimort, and J.-C. Rochet (2000): “Competing Mechanisms in a Common
Value Environment,” Econometrica, 68(4), 799–837.
[8] Biais, B., D. Martimort, and J.-C. Rochet (2013): “Corrigendum to “Competing Mech-
anisms in a Common Value Environment”,” Econometrica, 81(1), 393–406.
[9] Chade, H. and E. Schlee (2012): “Optimal insurance with adverse selection”, Theoretical
Economics, 7(3), 571–607.
[10] Glosten, Lawrence R. (1994): “Is the Electronic Open Limit Order Book Inevitable?”
Journal of Finance, 49(4), 1127–1161.
[11] Glosten, Lawrence R. (1998): “Competition, Design of Exchanges and Welfare,” Un-
published Manuscript, Columbia Business School.
[13] Hosseini, R. (2010): “Adverse Selection in the Annuity Market and the Role for Social
Security,” Unpublished Manuscript, Arizona State University.
27
[14] Pauly, M.V. (1974): “Overinsurance and Public Provision of Insurance: The Roles of
Moral Hazard and Adverse Selection,” Quarterly Journal of Economics, 88(1), 44–62.
[15] Pouyet, J., B. Salanié, and F. Salanié (2008): “On Competitive Equilibria with Asym-
metric Information,” The B.E. Journal of Theoretical Economics, 8(1), Article 13.
[16] Rockafellar, R.T. (1970): Convex Analysis, Princeton, NJ: Princeton University Press.
[17] Rothschild, C. (2007): “Adverse Selection, Linear Pricing and Front-Loading in Annuity
Markets,” Unpublished Manuscript, Middlebury College.
[18] Rothschild, M., and J.E. Stiglitz (1976): “Equilibrium in Competitive Insurance Mar-
kets: An Essay on the Economics of Imperfect Information,” Quarterly Journal of Eco-
nomics, 90(4), 629–649.
[19] Sheshinski, E. (2008): The Economic Theory of Annuities, Princeton, NJ: Princeton
University Press.
[20] Topkis, D.M. (1998): Supermodularity and Complementarity, Princeton, NJ: Princeton
University Press.
28