Games and Economic Behavior 45 (2003) 395–409

[Link]/locate/geb

Uniqueness of equilibrium in sealed high-bid

auctions ✩
Eric Maskin a,b,∗ and John Riley c
a Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540, USA
b Princeton University, USA
c Department of Economics, UCLA, Box 951477, 8283 Bunche, Los Angeles, CA 90095, USA

Received 17 February 2003

Abstract
For the case of two buyers we show that equilibrium in the sealed high-bid auction is unique when
(i) buyers’ reservations prices are drawn independently from distributions with finite support and
positive mass at the lower endpoint; (ii) buyers have private values; and (iii) buyers’ preferences
are log supermodular. For more than two buyers, we obtain the same result under the additional
assumptions that (iv) buyers with the same reservation price have the same preferences; (v) buyers
are risk neutral or risk averse with non-increasing absolute risk aversion; and (vi) the supports of the
different buyers’ distributions of reservation prices have the same upper endpoint.
 2003 Elsevier Inc. All rights reserved.

JEL classification: D44; D82

Keywords: High-bid auctions; Uniqueness; Equilibrium existence

Introduction

Although much has been written on the theory of auctions, most of this work focuses
exclusively on the symmetric equilibrium of an auction in which bidders are ex ante the
same in the sense that the joint distribution of buyers’ types is symmetric. In previous work
(Maskin and Riley, 2000a, 2000b), we have begun exploring the theory in the absence of

✩
This paper has existed in various forms since the 1980s. The most recent working-paper version is Maskin
and Riley (1996, Mimeo)
* Corresponding author.
E-mail addresses: maskin@[Link] (E. Maskin), riley@[Link] (J. Riley).

0899-8256/$ – see front matter  2003 Elsevier Inc. All rights reserved.
doi:10.1016/S0899-8256(03)00150-7
396 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409

symmetry.1 Specifically we have examined (i) the existence of equilibrium in a sealed high-
bid auction and (ii) the differences between the equilibrium in high-bid and second-price
auctions when buyers are asymmetric ex ante.
Here we turn to the question of uniqueness. With a symmetric distribution of types, it
is well known that there is only one symmetric equilibrium (Milgrom and Weber, 1982;
Maskin and Riley, 1984). However, it is not implausible to conjecture that, even in an ex
ante symmetric setting, a particular buyer might establish a reputation as an aggressive
bidder if it is in his interest to do so. Riley (1980) provides an example of the “war of
attrition” in which this is indeed the case. In fact, there is a continuum of asymmetric
equilibria in which one buyer bids “aggressively” and the other “passively.” Furthermore,
the greater the degree of aggression, the larger is the equilibrium expected gain of the
aggressive buyer.
A second example of a continuum of equilibria occurs in a pure common-values setting,
if the item is sold by open ascending bid. As first noted by Milgrom (1981) there is always a
continuum of equilibria in the two-buyer case. Bikhchandani and Riley (1991) also present
an example in which, with n bidders, there is a continuum of equilibria.
For the symmetric high-bid auction with private values, however, we show that there
can be no asymmetric equilibrium under the assumption that reservation prices are drawn
independently from a distribution with finite support2 and positive mass at the lower
endpoint.3 That is, equilibrium is unique.
When we drop the symmetry assumption, uniqueness continues to obtain under same
assumptions if there are only two buyers. For more than two buyers, we need the
additional fairly mild assumptions that buyers with the same reservation price have the
same preferences, that absolute risk aversion is non-increasing, and that the supports of the
different buyers’ distributions of reservation prices have the same upper endpoint.
The argument that equilibrium is unique is basically an application of the fundamental
theorem of ordinary differential equations (FTODE). As we will see, the major problems
with applying this theorem are (i) ensuring that buyers’ (inverse) bid functions are
differentiable, so that they satisfy a system of differential equations; and (ii) establishing
that there exists a unique “boundary condition” for that system.
We describe the model in Section 1. In Section 2 we present characterization results. We
use these in Section 3 to derive our main theorems. Concluding remarks are in Section 4.

1. The model

Throughout we shall make the following assumptions about the auction and the buyers
participating in it. A single item is to be sold to the buyer who makes the highest non-

1 There is also a literature on efficient auctions (see (Maskin, 2003), for a survey) that eschews the symmetry
assumption.
2 If the support of the distribution is unbounded, we conjecture that there will be a continuum of asymmetric
equilibria.
3 This latter assumption is weak because it is satisfied automatically if the seller sets a reserve price that is
even marginally above the lowest possible buyer reservation price.
 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409 397

negative sealed bid. If two or more bids tie, the winner is selected at random from among
the high bidders. There are n potential buyers. Buyer i of type si obtains utility Ui (b, si ) if
he wins a bid of b, where Ui is twice continuously differentiable. We assume that
∂Ui ∂Ui
<0 and >0 for all i.
∂b ∂si
Without loss of generality, we can interpret si as buyer i’s reservation price. Hence
Ui (si , si ) = 0. Buyer i’s reservation price is drawn independently from a distribution with
support [s i , s̄i ], where s̄i > 0, and c.d.f. Fi (·). We assume that Fi is twice continuously
differentiable, that its derivative is strictly positive on [s i , s̄i ], and that Fi (s i ) > 0 (see
footnote 4).
Clearly it is a dominated strategy for a buyer to bid more than his reservation price.
Hence, we will rule this out by assumption.

Assumption 1. Bidder i never bids more than his reservation price si in equilibrium.

If a buyer i has a negative reservation price, then it is a dominated strategy for him to
bid at all, and so without loss of generality we can assume that s i
0.
Let Πi be the probability that bidder i wins. Then his expected utility is
Ei = Πi Ui (b, si ).
We shall assume throughout that the higher is a bidder’s reservation price, the “flatter”
are his indifference curves in bid-probability space. That is, the single-crossing property
holds5 . Given our assumptions, bidder i’s indifference curve depicted in Fig. 1. Specifi-
cally, at an indifference curve,


db

∂Ei /∂Π 1 Ui

= =− .
dΠ Ei =const ∂E /∂b i Π ∂U /∂b i i

Fig. 1. Single-crossing property.

5 In technical terms, this is the assumption that utility is log supermodular.

398 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409

Thus, for single-crossing, we require the following assumption.

Assumption 2 (Single-crossing).

∂Ui
− Ui (b, si )
∂b
is a decreasing function of si .

Note that if Ui takes the form Ui (b, si ) = Vi (si − b), then Assumption 2 is satisfied
provided that bidder i is risk-neutral or risk-averse, i.e., Vi  0.
As we shall see in Section 3, it will be helpful to define the “log cost” of having to bid
to win the item, rather than getting it gratis:
ci (b, si ) ≡ log Ui (0, si ) − log Ui (b, si ). (1.1)
Then,

∂ci ∂Ui
=− Ui (b, si ), (1.2)
∂b ∂b
and so Assumption 2 is equivalent to the assumption that the marginal log cost is lower
for higher reservation prices. Given this assumption, buyer i’s bidding behavior will be
monotonic in si (see Lemma 2 below).
Since it will be useful below, we note that
 
∂ 2 ci ∂ci ∂ci
= Ai (b, si ) + , (1.3)
∂b 2 ∂b ∂b
where Ai (b, si ) = (∂ 2 Ui /∂b2)/(∂Ui /∂b) is buyer i’s coefficient of absolute risk aversion.
Note that as long as a buyer is risk-neutral or risk-averse (and hence Ai (b, si )
0), ci (b, si )
is strictly convex for all b ∈ [0, si ).

2. Characterizing the equilibrium bid functions

From (Maskin and Riley, 2000a, b) we have the following two results:

Lemma 1. If Assumptions 1 and 2 hold, the distribution of winning bids in equilibrium

has a support consisting of an interval [b∗ , b∗ ] and a c.d.f. Gw (b) which is continuous on
(b∗ , b∗ ] (see Maskin and Riley, 2000b, Proposition 3).

Lemma 2 (Monotonicity). If Assumptions 1 and 2 hold, then if bi (si ) is a best response by

buyer i with reservation price si to the other buyers’ bidding strategies, it is non-decreasing
in si (see Maskin and Riley, 2000a, Proposition 1).

To understand Lemma 2 geometrically, consider Fig. 1. If (b , Π  ) is optimal for a buyer

with reservation price s  , there can be no feasible alternatives in the heavily shaded region.
 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409 399

Thus, any alternatives preferred to (b , Π  ) by the higher reservation price s  must lie in
the lightly shaded region, i.e., they must entail higher bids.
As our first preliminary result, we characterize b∗ , the lower endpoint of the support of
the distribution of winning bids.

Lemma 3 (Characterization of the minimum winning bid). Without loss of generality,

suppose that s n  · · ·  s 1 . If Assumptions 1 and 2 hold, then the minimum bid satisfies
s 2  b∗  s 1 . (2.1)
Furthermore, if s 2 < s 1 , then

b∗ = max arg max Fi (b)U1 (b, s 1 ). (2.2)
b
i =1

Proof. Suppose first that b∗ > s 1 . Consider a buyer with a reservation price ŝ ∈
( 12 s 1 + 12 b∗ , b∗ ). Because ŝ < b∗ , the lowest winning bid, the buyer has an equilibrium
expected payoff of zero. But there is a positive probability that all other buyers have
reservation prices less than 12 s 1 + 12 b∗ . Thus, from Assumption 1, our buyer has a strictly
positive payoff if he bids 12 s 1 + 12 b∗ , a contradiction. We conclude that b∗  s1 .
Suppose next that b∗ < s 2 . From Lemma 1, there are no mass points on (b∗ , b∗ ]. Thus,
buyers 1 and 2, regardless of their reservation prices, have strictly positive expected payoffs
from bidding just above b∗ . This means that if I = {i | buyer i bids b∗ or more with
probability 1}, then 1, 2 ∈ I . For all i ∈ I , let pi be the probability that buyer i bids b∗ .
If, for all i ∈ I, pi > 0, then bidding b∗ results in a tie with positive probability. Thus,
buyer 1 is strictly better off bidding slightly above b∗ , since this increases his probability of
winning discontinuously. Hence, for some i ∈ I , pi = 0. If i = 1, then buyer 1’s probability
of winning, and hence his expected utility, is approximately zero for bids near b∗ . But
we have already argued that buyer 1’s equilibrium expected utility is strictly positive,
a contradiction. Hence, p1 = 0. But now the same contradiction pertains to buyer 2. We
conclude that (2.1) holds.
Suppose that s 2 < s 1 . From Assumption 1, if buyer 1 with reservation price s 1 bids
b = b∗ , his expected payoff is at least i =1 Fi (b)U1 (b, s 1 ). It follows that for b∗ to be an
equilibrium bid for him,
 
Fi (b)U1 (b, s 1 )  Fi (b∗ )U1 (b∗ , s 1 ) for all b.
i =1 i =1

Hence,

b∗ ∈ arg max F1 (b)U1 (b, s 1 ).
i =1

Finally, suppose that both b and b solve this maximization problem and that b < b  .
Buyer 1 with reservation price s 1 weakly prefers b to any lower bid. Given Assumption 2,
all other buyer 1 types strictly prefer b to any lower bid. Thus the minimum bid for all
reservation prices s1 > s 1 is at least b . But then b is not the lower endpoint of the support
of the equilibrium distribution of winning bids. We conclude that (2.2) holds. ✷
400 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409

Lemma 4 (Strict monotonicity of the probability of winning). Suppose that b < b and
that b and b  are in the support of the distribution of winning bids in equilibrium. Then at
least two buyers bid in the interval (b , b ) with positive probability.

Proof. From Lemma 1, the support of Gw (b) is connected, and so all the bids in the
interval (b , b ) are also in the support. This implies that at least one buyer bids in (b , b )
with positive probability. Suppose, contradicting the lemma, that buyer i is the only one
to do so. Specifically, assume that for reservation price si buyer i bids b̂ ∈ (b , b ) in
equilibrium. But buyer i can reduce his bid to b̂ − ε ∈ (b  , b ) without diminishing his
probability of winning, a contradiction. ✷

Let (b̃1 (s1 ), . . . , b̃n (sn )) be equilibrium bidding strategies (possibly mixed strategies).
Because Gw (b) is continuous, any deterministic selection bi (si ) from b̃i (si ) is strictly
increasing at all si for which bi (si ) > b∗ . It follows that
yi (·) = b̃i−1 (·)
is a non-decreasing function that is well defined at all b > b∗ for which there exists si with
b ∈ supp b̃i (si ). Thus, for all b > b∗ we can define
 

φi (b) = sup yi b̂
b̂  b, yi b̂ defined . (2.3)
Because yi (·) is non-decreasing, φi (·) is non-decreasing and continuous for all b > b∗ .
Note, furthermore, that buyer i’s probability of winning can be written as
 
Gi (b) ≡ Fj φj (b) . (2.4)
j =i

Because φj (b) is continuous for all j , so is Gi (b). Any realization of b̃i (si ) solves

max Ei (b, si ) = max Fj (φj (b))Ui (b, si ).
b b
j =i

Equivalently, it solves:
  Ui (b, si )
max Fj φj (b) .
b Ui (0, si )
j =i

That is, the bidder maximizes the ratio of his expected utility to his utility if he is simply
given the item for free.
Define

pi (b) ≡ log Fi φi (b) . (2.5)
Then, any realization of b̃i (si ) solves
max ei (b, si ),
b
  Ui (b, si )
where ei (b, si ) = log Fj φj (b) = pj (b) − ci (b, si ),
Ui (0, si )
j =i j =i

and ci (b, si ) is given by (1.1).

 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409 401

 As a preliminary to establishing uniqueness, we now derive properties of φi (·) and

j =i pj (b). Proofs of Lemmas 5–8 can be found in Appendix A.

Lemma
 5 (Strict monotonicity property of bid distributions). For any b > b∗ and any i,
j =i j (b) is strictly increasing at b.
p

Lemma 6. If φi (b) is strictly increasing to the right or left at b = b̂
b∗ , then b̂ is a best

response for buyer i with reservation price ŝi = φi (b̂).

Lemma
 7. If φi (b) is strictly increasing to the right or to the left at b = b̂
b∗ , then
j =i pj (b) is correspondingly right or left continuously differentiable at b̂. Moreover, the
right or left derivative satisfies
 ∂ci  
pj b̂ = b̂, φi b̂ . (2.6)
∂b
j =i

Lemma 8. φi (b) is right or left continuously differentiable at all b
b∗ .

Define the inverse function

hi (·) ≡ (log Fi )−1 (). (2.7)
Then we can rewrite Eq. (2.6) as
∂  
pj (b) = ci b, hi pi (b) . (2.8)
∂b
j =i

We shall make important use of the following:

Lemma 9. Suppose that (p̄1 , . . . , p̄n ) and (p̂1 , . . . , p̂n ) are two solutions to the differential
equation system
∂  
pj (b) = ci b, hi pi (b) , i = 1, . . . , n, (2.9)
∂b
j =i

on the interval (b1 , b2 ]. If for some b◦ ∈ (b1, b2 ], p̄i (b◦ ) < p̂i (b◦ ) for all i, then, for all
b ∈ (b1 , b◦ ),
p̄i (b) < p̂i (b), for all i, (2.10)
and
n n
p̄j (b) > p̂j (b). (2.11)
j =1 j =1

Proof. Dividing both sides of (2.09) by n − 1 and then summing over i, we obtain
n n
1  
pj (b) = cj b, hj pj (b) . (2.12)
n−1
j =1 j =1
402 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409

Subtracting (2.9) from (2.12), we have for all i,

 
1 ∂   ∂  
pi (b) = cj b, hj pj (b) − (n − 2) ci b, hi pi (b) . (2.13)
n−1 ∂b ∂b
j =i

Suppose, contrary to (2.10), there exist i and b ∈ (b1 , b◦ ) such that p̄i (b) = p̂i (b). Let
b̂ be the biggest such b. Then
 
p̄i b̂ = p̂i b̂ (2.14)
and
p̄j (b) < p̂j (b) (2.15)
for all b ∈ (b̂, b◦ ) and j = 1, . . . , n.
Now, from the fundamental theorem for ordinary differential equations (FTODE), there
exists a unique solution (p1 , . . . , pn ) to (2.9) with the point condition pj (b̂) = p̄j (b̂) for
all j . Hence, from (2.14) and (2.15), there exists k = i such that
 
p̄k b̂ < p̂k b̂ . (2.16)
From (2.13) and (2.14)
  1 ∂        
p̄i b̂ − p̂i b̂ = cj b̂, hj p̄j b̂ − cj b̂, hj p̂j b̂ . (2.17)
n−1 ∂b
j =i

But from (2.15) and (2.16) and Assumption 2, the right-hand side of (2.17) is positive
and hence p̄i (b) > p̂i (b) for b in a right neighborhood of b̂, contradicting (2.15). We
conclude that (2.10) holds as claimed. Then (2.11) follows from (2.10), (2.12), and
Assumption 2. ✷

3. Uniqueness

When buyers are ex ante asymmetric, we do not generally obtain uniqueness of equilib-
rium bids that win zero probability. To see this, consider the following example.

Example. Suppose that n = 2, that s1 is distributed uniformly in the interval [0,1], and
that s2 is distributed uniformly in [3,4].6 One equilibrium consists of buyer 2 bidding
b2 (s2 ) = 1 for all s2 and b1 (s1 ) = s1 for all s1 . However, we can replace buyer 1’s bid
function with b̂1 (s1 ) = s12 without destroying equilibrium. Indeed, there is a continuum of
different possible equilibrium bids for buyer 1. Nevertheless, all this multiplicity occurs
below b∗ = 1, and thus pertains only to bids that have no chance of winning.

Such examples dictate that when we speak of “uniqueness of equilibrium” we will

henceforth be referring only to the portions of the equilibrium bid functions at or above b∗ .

6 Strictly speaking, this example violates our assumption that F (s ) > 0, but we could modify it slightly to
i i
satisfy the assumption without changing our conclusion.
 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409 403

Proposition 1. Suppose that n = 2. If Assumptions 1 and 2 hold, then equilibrium is unique.

Proof. Recall from Lemma 3 that s 2  b∗  s 1 . Suppose first that b∗
s̄2 . But then,
from Lemma 3, b∗ = s̄2 , and buyer 1 with reservation price s 1 maximizes his payoff by
bidding s̄2 . Clearly, the same is true for all other types of buyer 1, and so b1 (s1 ) = s̄2 for
all s1 , i.e., equilibrium is unique at or above b∗ .
Thus, suppose that b∗ < s̄2 . Then, from Lemma 1, for any equilibrium there exists
b∗ > b∗ such that the distribution of winning bids has support [b∗ , b∗ ] with continuous
c.d.f. Gw (·). From Lemma 4, both bidders bid with strictly positive probability in any
subinterval of (b∗ , b∗ ]. Hence, from Lemma 8, if (b̃1 , b̃2 ) is an equilibrium, the transforms
(p1 , p2 ) of the inverse bid functions (φ1 , φ2 ) are differentiable everywhere and satisfy the
differential equation system (2.9).
Now suppose that there exist equilibria (p̄1 , p̄2 ) and (p̂1 , p̂2 ) such that the support of
the former is [b∗ , b̄∗ ] and that of the latter is [b∗ , b̂∗ ], where b̄∗ > b̂∗ . Then, for i = 1, 2,
  
1 = p̄i b̄∗ = p̂i b̂∗ > p̄i b̂∗ . (3.1)

Because both equilibria satisfy (2.9) on the interval (b∗ , b̂∗ ], Lemma 9 and (3.1) imply that,
for all b ∈ (b∗ , b̂ ∗ ],

2 2
p̄j (b) > p̂j (b). (3.2)
j =1 j =1

Integrating (3.2) and using the fact that p̄j and p̂j are continuous at b∗ , we obtain

2 2
    ∗
p̄j b̂∗ − p̄j (b∗ )
p̂j b̂ − p̂j (b∗ ) . (3.3)
j =1 j =1

Hence, from (3.1) and (3.3), we have

2 2
p̂j (b∗ ) > p̄j (b∗ ). (3.4)
j =1 j =1

But from Lemma 3, p̂1 (b∗ ) = p̄1 (b∗ ) = log F1 (s 1 ) and p̂2 (b∗ ) = p̄2 (b∗ ) = log F2 (b∗ ),
which contradicts (3.4). We conclude that b̄∗ = b̂∗ = b∗ , and so uniqueness follows from
FTODE with boundary condition p1 (b∗ ) = p2 (b∗ ) = 1. ✷

The proof of Proposition 1 applies the FTODE to the upper endpoint of the distribution
of winning bids. With two buyers, the upper endpoint is the same for both buyers, but with
three or more buyers, not everyone need share the same maximum bid. To guarantee that
they do, we shall impose two more fairly mild assumptions:
404 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409

Assumption 3 (Equal upper endpoints). The upper endpoint of the support of the
distribution of reservation prices is the same for all buyers, i.e.,7 s̄1 = · · · = s̄n = s̄.

We also assume that when bidders have the same reservation price, then they have the
same preferences. Formally, we have:

Assumption 4 (Identical reservation prices imply identical preferences). For all i and j , if
si = sj , then Ui (·, si ) = Uj (·, sj ).

Note that Assumption 4 is satisfied if buyers are risk-neutral, as is often assumed in the
auctions literature. We can now state:

Lemma 10. If Assumptions 3 and 4 hold, then the upper endpoints in the supports of all
buyers’ equilibrium bid distributions are the same.

Proof. Suppose that we index the buyers according to the upper endpoints of their
equilibrium bid distributions: b1∗
· · ·
bn∗ . Since equilibrium bidding is monotonic, b1∗
is a best reply for bidder 1 when his type is s̄ (by leaving the subscript off s̄, we are
invoking Assumption 3). Using the logarithmic transformation of buyer 1’s expected utility,
it follows that
n n
      
e1 bn∗ , s̄ = pj bn∗ − c bn∗ , s̄  pj b1∗ − c b1∗ , s̄ = −c b1∗ , s̄ = e1 b1∗ , s̄ ,
j =2 j =2

where we have used the fact that pj (b1∗ ) = log Fj (s̄j ) = 0, and we have invoked
Assumption 4 by leaving the subscript off c1 . Suppose that bn∗ < b1∗ . Since bn∗ is in the
support of buyer 1’s distribution of winning bids, p1 (bn∗ ) < 0 = pn (bn∗ ). Substituting for
pn (bn∗ ), we have, from the above inequality,
n−1
    
en bn∗ , s̄ = pj bn∗ − c bn∗ , s̄ < −c b1∗ , s̄ = en b1∗ , s̄ .
j =1

Thus bn∗ is not a best response for buyer n after all, a contradiction. We conclude that
bn∗ = b1∗ . ✷

The proof of Proposition 1 also relies on the property that, with just two buyers,
equilibrium bid functions are continuous above b∗ . But with three or more buyers, our
assumptions so far do not suffice to rule out the possibility that some buyer i has a
“gap” [b , b ] in the support of his equilibrium bid distribution. Still, we require only one
additional weak condition to rule out such gaps.

7 Assumption 3 is weak in the sense that, for any vector of distributions (F , . . . , F ), there exists another
1 n
1 , . . . , F
vector (F n ) that is arbitrarily close to (F1 , . . . , Fn ) and satisfies the assumption. Moreover, our method
of proof can be extended readily to the case of different upper endpoints.
 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409 405

Assumption 5 (Non-increasing absolute risk-aversion). For all i, the coefficient of absolute

risk aversion, Ai (b, si ) = (∂ 2 Ui /∂b2)/(∂Ui /∂b), is non-negative and non-increasing in si .

We can now establish our final preliminary result.

Lemma 11. If Assumptions 1, 2, 4, and 5 hold, the support of each buyer i’s equilibrium
bid distribution is an interval [b∗ , b1∗ ].

Remark. We ignore bids that have no chance of winning for the reasons illustrated by the
example at the beginning of the section.

Proof. Suppose, to the contrary, that some buyer i’s equilibrium bid distribution has
a “gap” [b ◦ , b◦◦]. That is, there exists some reservation price si◦ = φi (b◦ ) for which both
b◦ and b ◦◦ are best replies, and φi (b) = si◦ for all b ∈ [b ◦, b◦◦ ]. Buyer i with reservation
price si◦ chooses b to maximize
 
ei b, si◦ = pj (b) − ci b, si◦ . (3.5)
j =i

Thus, at b◦ ,
∂ei  ∂ci  ◦ ◦
= pj b◦ − b , si  0.
∂b ∂b
j =i

Let b̂ be the biggest bid in [b◦, b◦◦ ] such that

∂ei ∂ci  ◦
= pj (b) − b, si  0 (3.6)
∂b ∂b
j =i

for all b ∈ [b◦ , b̂]. Suppose that m of the equilibrium bid functions are strictly increasing
at b◦ . Without loss of generality, let these be the bid functions of bidders 1 to m and suppose
that they are increasing throughout the interval [b◦ , b̂] (if not, we can conduct the following
argument on each subinterval of strictly increasing bid functions). Then, from (3.6),
m
∂ei ∂ci 
= pj (b) − b, φi (b)  0, (3.7)
∂b ∂b
j =i

and from (2.6),

m
∂ck 
pj (b) − b, φk (b) = 0, k = 1, . . . , m. (3.8)
j=1
∂b
j =k

Comparing (3.7) with (3.8), we obtain

∂ci ∂ck
> for all k = 1, . . . , m. (3.9)
∂b ∂b
Hence, from Assumptions 2 and 4,
 
φi (b) < φk (b), k = 1, . . . , m, for all b ∈ b ◦ , b̂ . (3.10)
406 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409

Summing (3.8) over k, we have

m m
∂cj 
(m − 1) pj (b) = b, φj (b) . (3.11)
∂b
j =1 j =1

Differentiating (3.11) by b, we obtain, using Assumption 2 and (1.3),

m m m m  
 ∂ 2 cj ∂cj ∂cj
(m − 1) pj (b) < = Aj +
∂b 2 ∂b ∂b
j =1 j =1 j =1 j =1
m m  2
∂cj ∂cj
< Ai + , (3.12)
∂b ∂b
j =1 j =1

where the last inequality follows from Assumptions 4 and 5 and (3.10), and where the
fact that pj (b) is twice differentiable at b follows from our assumptions about Fj and the
FTODE.
From (3.7) and (3.11),
 m 
∂ei  1 ∂cj  ∂ci 
b, φi (b) = b, φj (b) − (m − 1) b, φi (b) . (3.13)
∂b m−1 ∂b ∂b
j =1

Also, from (3.7),

m m  
∂ 2 ei ∂ 2 ci ∂ci ∂ci 2
= pj − = pj − Ai −
∂b 2 ∂b 2 ∂b ∂b
j =1 j =1
m m  2  
Ai ∂cj 1 ∂cj ∂ci ∂ci 2
< + − Ai − (from (3.12))
m−1 ∂b (m − 1) ∂b ∂b ∂b
j =1 j =1
 m

Ai ∂cj ∂ci
< − (m − 1)
(m − 1) ∂b ∂b
j =1

 m    
1 ∂cj 2 ∂ci 2
+ − (m − 1)
(m − 1) ∂b ∂b
j =1
 m     
∂ei 1 ∂cj 2 ∂ci 2
< Ai + − (m − 1) from (3.13). (3.14)
∂b (m − 1) ∂b ∂b
j =1

If ∂ei /∂b  0 it follows from (3.13) that mj =1 ∂cj /∂b − (m − 1)∂ci /∂b  0. Hence,
 m   
∂ci ∂cj ∂ci 2
− (m − 1)  0,
∂b ∂b ∂b
j =1

and so, from (3.9), the bracketed expression on the right-hand side of (3.14) is negative.
Thus, for all b ∈ [b◦, b◦◦ ),
∂ei /∂b  0 ⇒ ∂ 2 ei /∂b2 < 0.
 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409 407

It follows that b̂ = b◦◦ , and so ei (b, si◦ ) is strictly decreasing over b◦ = b◦◦ , a contradiction
of our hypothesis that bidder i with reservation price si◦ is indifferent between bidding b◦
and b◦◦ . Thus there can be no such “gap” after all. ✷

Proposition 2 (Uniqueness with n buyers). If Assumptions 1–5 hold, equilibrium is unique.

Proof. Lemmas 8 and 11 imply that equilibrium-inverse bid functions are differentiable,
and Lemma 10 implies that, in equilibrium, each buyer makes the same maximum bid.
Hence, we can apply Lemma 9, as in the proof of Proposition 1, to show that the maximum
bid b ∗ is the same in any equilibrium. Uniqueness then follows from FTODE. ✷

4. Concluding remarks

We have limited our attention to the case of “independent private values,” in which
a buyer’s reservation price does not depend on other buyers’ private information, and
reservation prices are independently distributed. Note that, for this case, our arguments
also establish equilibrium existence without the need to invoke existence theorems for
discontinuous games such as Dasgupta and Maskin (1986), Simon and Zame (1990), and
Reny (1999) (existence results for high-bid auctions that do use these theorems include
Lebrun, 1996; Maskin and Riley, 2000b; Bresky, 1999; Jackson and Swinkels, 2001; and
Reny and Zamir, 2002).
When there are only two buyers, Lizzeri and Persico (2000) relax the independence
and private-values assumptions and establish uniqueness (and existence) under affiliation
and certain forms of interdependent values. We believe that our methods can be adapted
to accommodate such relaxations when there are more than two buyers, but this avenue
remains to be explored (Bajari, 2001, establishes uniqueness when there are more than two
buyers under the assumption the inverse bid functions are everywhere differentiable).

Acknowledgments

We thank B. Lebrun for helpful comments and the NSF for research support.

Appendix A

Lemma 5 (Strict monotonicity property of bid distributions). For any b > b∗ and any i, j =i pj (b) is strictly
increasing at b.

Proof. Choose ε > 0. From Lemma 4, there must buyer k = i who bids in [b − ε, b] with positive
 be at least one 
probability. Hence pk (b − ε) < pk (b) and so j =i pj (b − ε) < j =i pj (b). ✷

Lemma 6. If φi (b) is strictly increasing to the right (or left) at b = b̂
b∗ , then b̂ is a best response for buyer i
with reservation price ŝi = φi (b̂).
408 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409

Proof. Since both cases are handled in the same way, we consider only the case in which φi (b) is strictly
increasing to the right. If φi (b) is also strictly increasing to the left, then φi (b̂) = yi (b̂), and so the lemma follows.
Thus for some δ > 0, suppose that φi (b) = si∗ for all b ∈ [b̂ − δ, b̂]. That is, for some b∗ ∈ [b̂ − δ, b̂], yi (b∗ ) = si∗ .
Because φi (b) is strictly increasing to the right at b̂, there exists a decreasing sequence {b1 , . . . , bt , . . .} converging
to b̂ such that sequence {yi (b1 ), . . . , yi (bt ), . . .} converges to si∗ .
Since b is optimal for reservation price yi (bt ), we have
      
ei bt , yi bt = pj bt − ci bt , yi (bt )
pj b∗ − ci b∗ , yi b∗ , for all t. (A.1)
j =i j =i

From Lemma 5, it follows that j =i pj (b) = log Gi (b) is continuous. Also ci (b, si ) is continuous. Therefore
we have, in the limit,
   
pj b̂ − ci b̂, si∗
pj b∗ − ci b∗ , si∗ . (A.2)
j =i j =i

From (A.2) it follows that buyer i with reservation price si∗ is at least as well off choosing b̂ as b∗ . ✷


Lemma 7. If φi (b) is strictly increasing to the right (or to the left) at b = b̂ > b∗ , then j =i pj (b) is
correspondingly right (or left) continuously differentiable at b̂. Moreover, the right (left) derivative satisfies
 ∂ci  
pj b̂ = b̂, φi b̂ . (A.3)
∂b
j =i

Proof. Since the two cases are handled in the same way, we consider only the case in which φi (b) is strictly
increasing to the right. We know that φi (b) is continuous. Thus at b̂ there exists a decreasing sequence
{b1 , . . . , bt , . . .} converging to b̂ such that yi (bt ) converges to si∗ = φi (b̂) monotonically from above. Because
bt is optimal for buyer i with reservation price sit = yi (bt ), we have
     
pj b̂ − ci b̂, yi bt  pj bt − ci bt , yi bt .
j =i j =i

Rearranging, we obtain
pj (bt ) − pj (b̂) ci (bt , yi (bt )) − ci (b̂, yi (bt ))

. (A.4)
j =i bt − b̂ bt − b̂

By Lemma 6, b̂ is optimal for buyer i with reservation price φi (b̂). Thus,

     
pj b̂ − ci b̂, φi b̂
pj bt − ci bt , φi b̂ for all t.
j =i j =i

Rearranging, we obtain
pj (bt ) − pj (b̂) ci (bt , φi (b̂)) − ci (b̂, φi (b̂))
 . (A.5)
j =i bt − b̂ bt − b̂

In the limit as bt → b̂, the right-hand sides of (A.4) and (A.5) equal ∂/∂b ci (b̂, φi (b̂)), which is continuous in b̂.

Thus j =i pj (b) is right continuously differentiable at b̂, and its right derivative satisfies (A.3). ✷

Lemma 8. φi (b) is right (left) continuously differentiable at all b > b∗ .

Proof. Suppose φ1 (b), . . . , φk (b) are strictly increasing to

the right at b̂ and that φk+1 (b), . . . , φn (b) are constant
to the right at b̂. By assumption, i  k. By Lemma 7, kj =i pj (b) is right differentiable at b̂, i = 1, . . . , k.
Summing over i and dividing by k − 1, we conclude that
k k k
1
pj (b) = pj (b)
k−1
j=1 i=1 j =1
j =i
 E. Maskin, J. Riley / Games and Economic Behavior 45 (2003) 395–409 409

is also right-differentiable at b̂. Since the difference between these last two expressions is just pi (b), i = 1, . . . , k,
this too is right-differentiable at b̂. But pi (b) = Fi (φi (b)). Thus φi (b) is right-differentiable at b̂. ✷

References

Bajari, P., 2001. Comparing competition and collusion in procurement auctions: a numerical approach. Econ.
Theory 18, 187–205.
Bikhchandani, S., Riley, J.G., 1991. Equilibrium in open common value auctions. J. Econ. Theory 49.
Bresky M., 1999. Equilibrium in multi-unit auctions. Mimeo. CERGE.
Dasgupta, P., Maskin, E.S., 1986. Existence of equilibrium in discontinuous games, I: Theory. Rev. Econ. Stud. 5,
1–26.
Jackson, M., Swinkels, J., 2001. Existence of equilibrium in single and double private value auctions. Mimeo.
Lebrun, B., 1996. Existence of an equilibrium in first-price auctions. Econ. Theory 7, 421–423.
Lizzeri, A., Persico, N., 2000. Uniqueness and existence of equilibrium in auctions with a reserve price. Games
Econ. Behav. 30, 83–114.
Maskin, E., 2003. Auctions and efficiency. In: Dewatripont, M., Hansen, L., Turnovsky, S. (Eds.), Advances in
Economic Theory (invited lectures from the 8th World Congress of the Econometric Society). Cambridge
Univ. Press, pp. 1–24.
Maskin, E.S., Riley, J.G., 1984. Optimal auctions with risk averse buyers. Econometrica 52, 1473–1518.
Maskin, E.S., Riley, J.G., 2000a. Asymmetric auctions. Rev. Econ. Stud. 67, 413–438.
Maskin, E.S., Riley, J.G., 2000b. Equilibrium in sealed high-bid auctions. Rev. Econ. Stud. 67, 439–454.
Milgrom, P., 1981. Rational expectations, information acquisition and competitive bidding. Econometrica 50,
1089–1122.
Milgrom, P., Weber, R.J., 1982. A theory of auctions and competitive bidding. Econometrica 50, 1081–1122.
Reny, P., 1999. On the Existence of pure and mixed strategy Nash equilibrium in discontinuous games.
Econometrica 67, 1029–1056.
Reny, P., Zamir, S., 2002. On the Existence of pure strategy monotone equilibria in asymmetric first-price
auctions. Mimeo.
Riley, J., 1980. Strong evolutionary equilibrium and the war of attrition. J. Theoret. Biol. 82, 383–400.
Simon, L., Zame, W., 1990. Discontinuous games and endogenous sharing rules. Econometrica 58, 861–872.