Games and Economic Behavior 28, 238–255 (1999)

Article ID game.1998.0693, available online at [Link] on

An Asymptotic Solution for Sealed Bid

Common-Value Auctions with Bidders
Having Asymmetric Information

Michael C. Laskowski*

Department of Mathematics, University of Maryland, College Park, Maryland 20742

and

Robert L. Slonim

Department of Economics, Weatherhead School of Management,

Case Western Reserve University, Cleveland, Ohio 44106

Received September 9, 1997

We investigate a first-price common-value auction where bidders have asymmet-

ric information about an item of unknown value. We compute the unique Nash
equilibrium when the bidders are constrained to translation-invariant bid functions.
Further, this profile of bid functions is also an asymptotic Nash equilibrium (without
the constraint on the bidders’ strategies) as the a priori distribution of the true value
becomes increasingly diffuse. All bidders have positive expected profits at equilib-
rium. In the second-price analogue with two bidders there is a continuum of Nash
equilibria in which both bidders have positive expected profits. Journal of Economic
Literature Classification Number: C7. © 1999 Academic Press

1. INTRODUCTION

Sealed-bid auctions are widespread. In many applications an item auc-

tioned often has a significant common-value component for all bidders.
For example, United States sales of public land and offshore sites for ex-
ploration and development have a common-value component because the
amount of oil and gas will be about the same for everyone (Capen et al.,
1971).

* Partially supported by NSF Research Grant DMS-9403701.

238
0899-8256/99 $30.00
Copyright © 1999 by Academic Press
All rights of reproduction in any form reserved.
 asymptotic auction solution 239

Extensive research exists for common-value auctions (see Wilson, (1992

for a review). In common-value auctions bidders observe a private signal
that provides information about the common value. Although the realiza-
tion of signals is allowed to differ across bidders, researchers almost always
assume that signals are drawn from the same distribution. This information
symmetry implies, among other things; that any two bidders that receive the
same signal will have identical beliefs about the distribution of the common
value.
Whether symmetric information is a valid assumption for a particular
sealed-bid auction with a common-value component is an empirical ques-
tion, but one may reasonably assume there are many situations in which
some firms are able to employ greater resources, more experience and/or
more skills to generate a signal. In cases where firms have differential
capabilities, it may be important for them to consider these differences
when bidding and for sellers to consider the revenue ramifications of these
differences.
Two approaches have been taken to model asymmetric information
among bidders. One approach assumes that one bidder has superior in-
formation to all the other bidders (Bidder i’s information is superior to
Bidder j’s if i’s information reveals j’s). Several papers have shown, un-
der many conditions, that if one bidder has superior information, then
the Nash equilibrium is degenerate in the sense that all outsiders play
mixed strategies and earn zero expected profits (Wilson, 1967; Milgrom,
1979; Weverberg, 1979; Milgrom and Weber, 1982; Englebrecht-Wiggins
et al., 1983; and Hendricks et al., 1994). Thus, in equilibrium outsiders are
indifferent between bidding and not bidding.
The second asymmetric information approach is modeled by Kagel and
Levin (in press). They assume there are n ≥ 2 bidders, each of whom are
risk neutral, bidding on an object whose worth is a random variable V ,
whose distribution (but not its value) is known to all of the bidders. Each
of the bidders has a “margin of error” ei that is also known to all of the
bidders. In addition, each bidder receives a “private signal” Vi , which is
drawn uniformly in the interval ’V − ei ; V + ei “. The motivation is that
bidders (firms) who employ greater resources will receive better informed
signals, i.e., the value of ei will decrease.
Kagel and Levin consider the case where the distribution of V is uni-
form on some large interval ’L; H“ and show that the risk-neutral Nash
equilibrium can be described by a system of n partial differential equa-
tions. The solution to this system is nonlinear, due to the presence of the
endpoints. In their model, at equilibrium each player will make his bid in
the range ’L; H“, even if his private signal is outside this interval. Because
of this nonlinearity, they are only able to describe the Nash equilibrium
in the case of one player with perfect information (i.e., e1 = 0) and the
240 laskowski and slonim

other n − 1 players having the same margin of error (i.e., ei = ej for all i,
j > 1).
In Section 2 we exhibit a profile of n bid functions and prove that it has
two desirable properties. First, it is the unique Nash equilibrium if each of
the bidders is constrained to choose a translation-invariant bid function. A
bid function bi Vi ‘ is called translation-invariant if Vi − bi Vi ‘ is constant
(i.e., it only depends on the margins of error e1 ; : : : ; en ). As well, in Sec-
tion 2 we define the notion of an asymptotic Nash equilibrium and show that
this profile is the unique (up to almost everywhere equivalence) asymptotic
Nash equilibrium for the more general setting in which each bidder is al-
lowed to choose among all measurable bid functions.
In Section 3 we discuss a number of comparative statics under the as-
sumption that all bidders employ the profile mentioned above. These
are of immediate interest if one considers the assumption of choosing a
translation-invariant bid function to be reasonable. However, our motiva-
tion is deeper. Since the profile is also an asymptotic Nash equilibrium, it
will follow that for each of these statics the direction of each inequality will
hold provided the underlying distribution of the true signal is sufficiently
diffuse.
Finally, in Section 4 we analyze the analogous second-price auction in
which two bidders are given asymmetric information and are constrained
to choose translation-invariant bid functions. We find that there are a con-
tinuum of Nash equilibria for which each bidder has a positive expected
profit and we give a geometric characterization of these equilibria.

2. THE TIN PROFILE

Suppose that n ≥ 2 risk-neutral bidders are bidding on an item of un-

known worth V . Each of the bidders has a margin of error ei that is known
to all bidders. By reindexing, we assume that 0 ≤ e1 ≤ e2 ≤ · · · ≤ en . We
also assume that e2 > 0, for otherwise the auction is degenerate. Each of
the bidders receives a private signal Vi , that is drawn uniformly from the in-
terval ’V − ei ; V + ei “. We describe one particular profile of bid functions
b1 V1 ‘; : : : ; bn Vn ‘‘, which we call the Translation-Invariant Nash (TIN)
Profile.
The TIN Profile. Let bi Vi ‘ = Vi − ei for 2 ≤ i ≤ n and let b1 V1 ‘ =
V1 − c1∗ , where, letting
φx‘ = x2e2 − x‘ · · · 2en − x‘;
c1∗
is the unique solution to φ0 x‘ = 0 in the interval ’e1 ; e2 “ when e1 = 0 and
c1∗
is the unique value in the interval ’e1 ; e2 “ satisfying φc1∗ + e1 ‘ = φc1∗ − e1 ‘
when e1 > 0.
 asymptotic auction solution 241

A bid function is translation-invariant if Vi − bi Vi ‘ is constant for all

values of Vi . If none of the bidders has any a priori hypothesis about the
distribution of V , then due to the symmetry of the bidders’ signals around
V and as the payoff to the highest bidder is V − bi Vi ‘, it is reasonable for
each of the bidders to choose a translation-invariant bid function. Under
this constraint, the TIN Profile is the only Nash equilibrium. More precisely,
in Appendix A we prove the following theorem.
Theorem 1. Suppose that n ≥ 2 risk-neutral bidders with margins of error
0 ≤ e1 ≤ e2 ≤ · · · ≤ en and e2 > 0 each select a translation-invariant bid
function. Then, regardless of the distribution of the true value V , the TIN
Profile is the unique Nash equilibrium.
It is somewhat remarkable that this profile is the unique Nash equi-
librium regardless of the distribution of the true signal V . This fact is a
byproduct of the symmetry that results from constraining each bidder to
the class of translation-invariant functions. When we relax this constraint
on the bidders’ strategies we lose this invariance. As well, the problem
of computing the unconstrained Nash equilibria when the distribution of
the true value V is known to all bidders appears intractable. For exam-
ple, in their original formulation Kagel and Levin (in press) assumed that
the true value was drawn uniformly from a large interval ’L; H“. As a con-
sequence, the Nash equilibrium no longer consists of translation-invariant
bid functions (it would be absurd for any bidder to make a bid outside
of the interval ’L; H“). This asymmetry near the endpoints has a damped
ripple effect throughout the entire interval ’L; H“, which prevented Kagel
and Levin from being able to present a closed form solution for the Nash
equilibrium. Here, we try to circumvent this problem at the endpoints by
introducing the general notion of an asymptotic Nash equilibrium. Our mo-
tivation behind the notion is to model the situation where the densities of
the true value get increasingly diffuse.
Definition. Suppose n ≥ 2 bidders are bidding on an item of unknown
worth V and let ”µk •k∈N be a sequence of densities of V . Given a fixed n
tuple of bid functions b1 ; : : : ; bn ‘, where each bi is chosen from a set Si
of allowable functions, let
µ
Ei k b1 ; : : : ; bn ‘
denote the expected value of Bidder i if the true value V has density µk .
We say that the n tuple b1 ; : : : ; bn ‘ is an asymptotic Nash equilibrium with
respect to S1 ; : : : ; Sn and ”µk • if for all 1 ≤ i ≤ n and all f ∈ Si ,
µ
µ
Ei k b1 ; : : : ; bi−1 ; f; bi+1 ; : : : ; bn ≤ Ei k b1 ; : : : ; bn ‘
for all but finitely many k.
242 laskowski and slonim

The intuition behind the proof of Theorem 2 is that as the interval ’L; H“
grows ever larger (relative to the margins of error e1 ; : : : ; en ) then the
likelihood that any bidder’s signal is near an endpoint approaches zero,
and hence can be ignored in the limit. The proof of this theorem is given
in Appendix B.

Theorem 2. Let µk be the uniform density function on ’−k; k“ and, for

each i, let Si be the set of all Lebesque measurable functions. Then the TIN
profile of bid functions is an asymptotic Nash equilibrium with respect to
S1 ; : : : ; Sn and ”µk •k . Furthermore, for each i, the choice of bi is the unique
optimal bid (up to almost everywhere equivalence) given that the opponents
bid according to the TIN Profile.

It would be desirable to extend Theorem 2 to allow other sequences of

increasingly diffuse measures ”µk •k , but at present we are not able to show
that the analogue of Theorem 2 holds for all sequences. We consider it
interesting to find a sufficient condition on the sequence ”µk •k that would
make Theorem 2 hold.

3. COMPARATIVE STATICS

In this section we state a number of comparative statics that follow when-

ever the bidders use the TIN Profile of bid functions.1 For ease of wording,
we call every bidder whose margin of error is e1 an insider and we call ev-
ery other bidder an outsider. An insider is perfect if ei = 0, i.e., his signal
Vi is equal to the true value V . As the auction with two or more perfect
insiders degenerates, we assume for the whole of this discussion that there
is at most one perfect insider, i.e., e2 > 0. Let Ei∗ be the expected profit of
Bidder i, provided that the bidders employ the TIN Profile.
First, note that in the TIN Profile, no bidder will ever bid more than
the true value of the item and will bid less with positive probability. In
particular, Ei∗ > 0 for each of the bidders, which differs from the case of
auctions with superior information where every outsider has expected value
zero at equilibrium. Also, expected profits are higher for more informed
bidders (Ei∗ > Ej∗ if and only if ei < ej ).
Next, consider the effect of a new bidder entering the auction. It follows
immediately from the TIN Profile that no bidder (except possibly the in-
sider) will change his bid with the addition of a new bidder. If there is not a
unique insider (i.e., e1 = e2 ) then no bidder will change his bid when a new

1
Kagel and Levin (in press) make similar observations for the two cases they consider,
namely e1 = e2 = · · · = en and e1 = 0; e2 = · · · = en .
 asymptotic auction solution 243

bidder enters the auction, while if there is a unique insider then he will in-
crease his bid. In either case, the new bidder will have a positive expected
profit and each of the existing bidders will have smaller expected profits.
Intuitively, the new entrant increases demand, which hurts all existing bid-
ders. Note that the directional impact the new bidder has on an existing
bidder’s bid and expected profit is the same regardless of the accuracy of
the new bidder’s information.
If an existing bidder gets more accurate information (ei decreases), then
it follows from the TIN Profile that his expected bid will not decrease (the
expected bid will strictly increase in all cases except when n = 2 and i = 1).
If there were more than one insider originally, then no other bidder will
change his bid. If there were a unique insider then both he and the bidder
getting more accurate information will increase their expected bids, while
everyone else will keep their bid. When a bidder’s information gets more
accurate, all other bidders’ expected profits decrease. Intuitively, when a
bidder gets more accurate information, he will bid more aggressively, which
will hurt the other bidders.
Whether more accurate information helps a bidder himself is less intu-
itive. If an insider receives more accurate information, or if e1 = e2 and
any bidder’s information gets more accurate, then the expected profit of
the bidder getting more accurate information increases (i.e., ∂Ei∗ /∂ei < 0).
In these situations the expected profit of the bidder getting more accurate
information increases because no other bidder will change his bid. How-
ever, if an outsider gets more accurate information and there is a unique
insider, then the insider’s expected bid will also increase. Somewhat surpris-
ingly, there are examples where the outsider’s expected profit will increase,
examples where his profit will stay the same, and examples where his profit
will decrease with the more accurate information.
One situation where the outsider is always worse off with getting more
accurate information is when there are only two bidders. In this case, it
is easy to show that the outsider’s expected profit monotonically decreases
with more accurate information (i.e., ∂E2∗ /∂e2 > 0). In one limit, as e2 ap-
proaches e1 , the outsider’s expected profit approaches e1 /3. In the other
limit, as e2 approaches infinity e1 , the outsider’s expected profit approaches
e2 /4. The reason the outsider’s expected profit decreases as he receives
more accurate information is that he is becoming a bigger threat to the
insider, causing the insider to bid more, thus permitting smaller profit op-
portunities for both bidders.
Perhaps a more extreme case where better information is harmful is that
of an n-bidder auction with a single perfect insider. Each of the n bidders
has a positive expectation in such an auction. However, if any one of the
outsiders is told the true value (and everyone knows that he was told) then
all n bidders will have an expectation of zero.
244 laskowski and slonim

Expected seller revenue increases whenever a new bidder enters the auc-
tion and in almost all cases when existing bidders get better informed (ei
decreases). This is not surprising, as we noted above that each bidder’s ex-
pected bid is nondecreasing and is usually increasing. However, this does
not necessarily imply that the expected sales price is nondecreasing in all
cases. When any outsider gets better informed, expected revenue unam-
biguously increases. Expected revenue increases in this case as each of the
new bid functions is stochastically larger than the old bid function (the ran-
dom variable X is stochastically larger than Y if PX ≤ a‘ ≤ PY ≤ a‘
for all a) and the cumulative distribution function (cdf) of the sales price
is simply the product of the cdf’s of the bid functions over all bidders.
When an insider gets better informed, expected revenue typically in-
creases, but there are examples where the expected revenue stays the same
and where it decreases. For example, when n = 2, expected revenue de-
creases monotonically as e1 decreases throughout the range from the sym-
metric (e1 = e2 = e) to the perfect information case (0 = e1 < e2 = e).
As e1 decreases, both the insider’s and outsider’s expected bids remain un-
changed at V − e, but the insider’s bid distribution narrows.
However, as evidence that this behavior is anomalous, when n ≥ 3 then
the seller’s expected revenue is strictly increasing as e1 decreases in the
range 0 ≤ e1 ≤ e2 = e3 = · · · = en (i.e., there is one insider and n −
1 symmetric outsiders). Despite this result, one cannot generalize to say
that when n ≥ 3, the seller’s revenue always increases as an insider gets
better informed. For example, when n = 3 and e3 is substantially larger than
e2 , expected revenue will continue to decrease as the insider’s information
increases (this is because the presence of e3 will have essentially no effect
on the insider’s bid, and thus this case reduces to the n = 2 case).
Expected revenue will increase monotonically as all bidders simultane-
ously become better informed. For example, if the expected sales price
is V − f when the bidders have margins of error e1 ; : : : ; en , then the ex-
pected sales price will be V − αf when the bidders have margins of error
αe1 ; : : : ; αen for any 0 < α < 1.
Thus, it may be worthwhile for the seller to spend money to decrease
the uncertainty in estimating the true value in order to tell the bidders. In
contrast, it is not always in the interest of some bidders to obtain more
accurate information. In fact, it may be in the interest of some bidders to
actively avoid obtaining information.

4. THE SECOND-PRICE AUCTION

In this section we modify our model by considering a second-price auc-

tion. We find that even in the restricted context of translation-invariant bid
 asymptotic auction solution 245

functions, there are many Nash equilibria. This lack of uniqueness con-
trasts with our results on the first-price auction and with the theory of
private value second-price auctions.
Throughout this section we assume there is an item to be auctioned of
some specific value V . We assume that there are two bidders and that the
high bidder wins the auction, but pays the price of the smaller of the two
bids. As in the previous section, we assume that each bidder has a mar-
gin of error ei > 0, that is known to both bidders.2 Each bidder receives a
private signal Vi that is uniformly distributed in the interval ’V − ei ; V +
ei “. As in the earlier sections, we aim to classify the Nash equilibria un-
der the constraint where each bidder is required to choose a translation-
invariant bid function of the form bi Vi ‘ = Vi − ci , where ci does not de-
pend on Vi . As in the proof of Theorem 1, it is harmless to assume that
V = 0.
We analyze the auction from the point of view of Bidder i. Once a bid
function bj Vj ‘ is chosen, Bidder j’s actual bid can be construed as a ran-
dom variable, uniformly distributed on the interval ’cj − ej ; cj + ej “. For
ease of notation, assume that Bidder j chooses an interval ’r1 ; r2 “ (neces-
sarily r2 − r1 = 2ej ). We wish to find an optimal interval ’s1 ; s2 “ for Bidder i.
As s2 − s1 = 2ei , we have only one degree of freedom.
Let us dispense with two trivial cases first. Call a bid ’r1 ; r2 “ extremely
large if r1 ≥ 0, call it extremely small if r2 ≤ 0 and reasonable if r1 < 0 < r2 .
Clearly, if ’r1 ; r2 “ is extremely large, then Bidder j is always bidding more
than the item is worth, so Bidder i never gains by competing. Any interval
’s1 ; s2 “ satisfying s2 ≤ r1 is optimal.
Similarly, if ’r1 ; r2 “ is extremely small, then as this is a second-price auc-
tion, Bidder i wants to bid enough to ensure that he always wins the auction.
That is, any interval ’s1 ; s2 “ satisfying s1 ≥ r2 will be optimal.
Curiously, it follows from these two paragraphs that any combination
of bids where one is extremely small and the other is extremely large will
be Nash. We will see below that these are the only Nash equilibria where
one bidder makes an extreme bid. These Nash solutions are degenerate as
in each case, one bidder has no chance of winning the auction, so he is
indifferent between bidding and not bidding.
Let us now turn to the case where Bidder j’s interval ’r1 ; r2 “ is rea-
sonable. Assume that Bidder i bids x. If x ≥ r2 then his expected profit
is −r1 + r2 /2‘. If r1 ≤ x ≤ r2 , then his expected profit given that he
wins the auction is −r1 + x/2‘, so his unconditional expected profit is
−r1 + x/2‘x − r1 /r2 − r1 ‘. Clearly, if x ≤ r1 then his expected profit is
0, as he cannot win the auction. Thus, his expected profit by choosing an

2
The case where one ei = 0 is similar. The case where e1 = e2 = 0 is degenerate.
246 laskowski and slonim

interval ’s1 ; s2 “ is given by

1 Z s2
Ei s1 ; s2 ‘ = 1/4ej ‘Hi x‘ dx; (1)
2ei s1
where, recalling that r2 − r1 = 2ej ,

2 2

 r1 − r2 ; if r2 ≤ xy
2
Hi x‘ = r1 − x2 ; if r1 ≤ x ≤ r2 y (2)

 0; if x ≤ r1 :
Evidently, Ei is differentiable with derivative Ei0 s1 ; s2 ‘ = kHi s2 ‘ −
Hi s1 ‘‘ for some constant k. So, if the choice of ’s1 ; s2 “ is to be optimal,
then we require that Ei0 = 0, which implies Hs1 ‘ = Hs2 ‘. To examine
what this entails, we split into cases.
• If r1 < r2 ≤ s1 < s2 then Hi s1 ‘ = Hi s2 ‘ always. However, in this
case, since 0 < r2 the strategy ’s1 ; s2 “ is dominated by the strategy ’0; 2ei “.
By choosing the latter interval Bidder i makes the same profit whenever j’s
bid is less than the true value, but loses less of the time when Bidder j’s bid
is positive. Thus, Bidder i’s optimal response cannot satisfy the hypothesis
of this case.
• If r1 ≤ s1 < r2 ≤ s2 then E 0 s1 ; s2 ‘ = ks12 − r22 ‘ for some constant k,
so E 0 s1 ; s2 ‘ = 0 and the hypothesis of this case imply s1 + r2 = 0.
• If s1 < r1 < r2 ≤ s2 then Hi s1 ‘ = 0 and Hi s2 ‘ = r22 − r12 , so
0
Ei s1 ; s2 ‘ = 0 implies r1 + r2 = 0. Thus, if r1 + r2 = 0 then any ’s1 ; s2 “
satisfying the hypotheses of this case will be optimal, while if r1 + r2 6= 0
then no ’s1 ; s2 “ in this range is optimal.
• If r1 ≤ s1 < s2 < r2 then Ei0 s1 ; s2 ‘ = 0 implies s1 + s2 = 0.
• If s1 < s2 < r1 < r2 then Hi s1 ‘ = Hi s2 ‘ = 0 as Bidder i cannot
win the auction. However, as in the first case, if r1 < 0 then this strategy
is dominated by the interval ’−2ei ; 0“, which offers a positive expectation.
Thus, there is no optimal interval satisfying the hypothesis of this case.
The results above have a nice geometric interpretation. Given that Bid-
der j makes a reasonable bid of ’r1 ; r2 “, the interval ’s1 ; s2 “ is optimal for
Bidder i if and only if the two intervals have nonempty intersection ’t1 ; t2 “
and 0 is the midpoint of ’t1 ; t2 “, i.e., t1 + t2 = 0.
Note that this condition is symmetric for the two bidders. If ’s1 ; s2 “ is
optimal for ’r1 ; r2 “, then ’r1 ; r2 “ is optimal for ’s1 ; s2 “. Consequently, the
pair of strategies is a Nash equilibrium. We summarize our results in the
theorem below.
Theorem 3. 1. For every bid function bi Vi ‘ = Vi − ci of Bidder i, there
is a bid function bj Vj ‘ = Vj − cj of Bidder j making the pair of strategies a
Nash equilibrium.
 asymptotic auction solution 247

2. Any pair of bid functions where one is extremely large and the other
is extremely small is a Nash equilibrium.
3. If one bid function is reasonable and the pair is Nash, then the other
bid function is also reasonable and the intersection of the two intervals is
centered at the true value of the item. In this case, both bidders have a positive
expected profit.

One can draw many conclusions from this theorem. Clearly, there is a
large collection of Nash equilibria, so if this auction were to actually be
held, there would be tension about which Nash equilibrium to choose. It
turns out that this auction has a temporal aspect, whose effect dominates
the relative sizes of the margins of error.
Somewhat surprisingly, it is to either bidder’s advantage to announce his
intentions first. Clearly, either bidder can trivialize the auction by announc-
ing his intention to make an extreme bid. If one bidder announces that he
will make an extremely low bid, then he will not make a profit at any Nash
solution. It is more interesting to consider the effect of Bidder i announc-
ing his intention to bid bi = Vi + ei , i.e., he is choosing the interval ’0; 2ei “.
This leaves Bidder j with no chance of making a profit. His optimal bids
are exactly the set of extremely small bids. Each of these guarantee that
Bidder i will make a profit of at least ej , while ensuring that Bidder j will
lose the auction.
Even if the rules of the auction preclude making extreme bids, Bidder i
can do well by announcing his intention to bid bi = Vi + di , where di is in-
finitessimally less than ei . In this case, Bidder j has a unique optimal bid,
namely bj = Vj + ei − di + ej . However, in this case, Bidder i’s expected
profit is infinitessimally less than ej , while Bidder j’s expected profit, al-
though positive, is infinitessimally small.

5. CONCLUDING REMARKS

Models of first-price common-value auctions with asymmetric informa-

tion have been able to fully characterize the Nash equilibria for a broad
class of distributions for both the a priori true value and the private infor-
mation signals. In order to characterize the Nash equilibria, earlier models
assume one bidder has superior information that reveals the information
of all other bidders. In the current paper we introduce a model in which
all bidders have private information. We are able to fully characterize the
unique Nash equilibrium when the bidders are constrained to translation-
invariant functions and we give an asymptotic solution as the a priori dis-
tribution of the true value becomes sufficiently diffuse.
248 laskowski and slonim

The nature of the auction at equilibrium in our model differs in two im-
portant respects from previous models that assume a bidder with superior
information. First, in equilibrium all bidders earn positive expected profits
in our model whereas in previous models all bidders other than the bidder
with superior information earn zero expected profit. Second, the seller’s
expected revenue increases if any bidder (except possibly the insider) be-
comes more informed (and all bidders know he becomes more informed) in
our model whereas in previous models the seller’s revenue decreases when
a bidder becomes more informed (and all bidders know he becomes more
informed).
Analysis of the analogous second-price common-value auction with two
bidders reveals that there are a continuum of Nash equilibria in which
both bidders have positive expected profits. In contrast, earlier models of
second-price, common-value auctions (e.g., Milgrom, 1979) that assume the
existence of a bidder with superior information result in every other bidder
receiving an expected profit of zero at the equilibrium.

APPENDIX A: PROOF OF THEOREM 1

Throughout this section, we assume that there is an item to be auctioned

of some specific value V that is drawn according to a density function that
is known to all bidders. Due to the constraints we place on the bid functions
of the bidders, the conditional expected value of a profile of bid functions
will be independent of the value of V for every bidder, so it will not matter
what the density of V is.
Each of the n ≥ 2 bidders has an associated margin of error ei . By rein-
dexing the bidders, we assume 0 ≤ e1 ≤ e2 ≤ · · · ≤ en . As the case e2 = 0 is
degenerate, we assume throughout that e2 > 0. The values of e1 ; : : : ; en are
known to all n bidders. Additionally, it is known to all that each bidder will
receive a private signal Vi , whose distribution is uniform on ’V − ei ; V + ei “.
Finally, in this section we constrain each bidder to choose a translation-
invariant bid function bi Vi ‘ = Vi − ci for some constant ci depending only
on the margins of error e1 ; : : : ; en .
Before embarking on the proof, note that in this situation we cannot
appeal to any general theorem to conclude either existence or uniqueness
of a Nash equilibrium. Indeed, we saw that there are many distinct Nash
equilibria for a second-price auction in this context.
To ease notation, let the n tuple c1 ; : : : ; cn ‘ denote the profile of bid
functions b1 V1 ‘; : : : ; bn Vn ‘‘, where bi Vi ‘ = Vi − ci . We wish to calculate
the optimal translation-invariant bid function for Bidder i, given that the
opponents have chosen their (translation-invariant) bid functions. If each
opponent has chosen a bid function bj = Vj − cj , the value of bj is a uni-
 asymptotic auction solution 249

formly distributed random variable on ’V − ej − cj ; V + ej − cj “. (If ej = 0

then bj = V − cj .) Let Gi x Ž V ‘ denote the conditional probability that all
of the opponents’ bids are at most x, given thatQthe true value is V . As
V1 ; : : : ; Vn are drawn independently, Gi x Ž V ‘ = j6=i Pbj ≤ x Ž V ‘; where


 0; if x ≤ V − cj − ej y
2ej Pbj ≤ x Ž V ‘ = x + ej + cj ; if V − cj − ej ≤ x ≤ V + ej − cj y

 2e ;
j if x ≥ V + ej − cj ;
whenever ej > 0 and

1; if x ≥ V − cj y
Pbj ≤ x Ž V ‘ =
0; otherwise
when ej = 0.
If ei > 0 and Bidder i bids bi = y − b (where y is drawn uniformly in
’V − ei ; V + ei “), then he wins b − y if bi is larger than his opponents’ bids
and he wins 0 otherwise. Thus, his expected value given that the true value
is V is given by
1 Z V +ei
Ei b Ž V ‘ = b − y‘Gi y − b Ž V ‘ dy: (3)
2ei V −ei
Note that since Gi x Ž V ‘ = Gi x + V 0 − V Ž V 0 ‘, the expected value
Ei b Ž V ‘ does not depend on V ! Thus, regardless of the density of V , the
expected value of Bidder i, Ei b‘, is equal to the conditional probabilities
Ei b Ž V ‘ for some (any) V . Because of this invariance, in what follows we
assume that the true value V = 0 although obviously the bidders do not
know this. For brevity, we write Gi x‘ for Gi x Ž 0‘.
So, setting V = 0 and using the change of variables u = b − y we obtain
1 Z b+ei
Ei b‘ = uGi −u‘ du: (4)
2ei b−ei
If ei = 0 this reduces to Ei b‘ = bGi −b‘.
Evidently, Ei b‘ is differentiable with derivative
1  
Ei0 b‘ = b + ei ‘Gi −b − ei ‘ − b − ei ‘Gi ei − b‘ (5)
2ei
when ei > 0 and Ei0 b‘ = Gi −b‘ − bG0i −b‘ when ei = 0.
We now deduce a number of conditions that are necessary for a profile
of bid functions c1 ; : : : ; cn ‘ to be a Nash equilibrium. First note that if
Bidder i employs the bid function bi Vi ‘ = Vi − ei , he will never lose money,
so his expectation is non-negative. Consequently, every bidder’s expectation
is non-negative at a Nash equilibrium. Next, we claim that if c1 ; : : : ; cn ‘ is
Nash, then every cj > −ej . To see this, first suppose Bidder j has ej > 0 and
250 laskowski and slonim

cj ≤ −ej . In this case, his bid bj = Vj − cj would always be at least the true
value of 0, and would be positive with positive probability. Consequently,
the sum of the expectations of the bidders’ profits over all n bidders would
be negative. But this implies that some bidder has negative expectation,
so c1 ; : : : ; cn ‘ cannot be Nash. Thus, every bidder with ej > 0 will bid
cj > −ej . In particular, there is a positive probability that the maximum
bid among those bidders with ej > 0 will be negative. So, in the remaining
case where e1 = 0, there is a small b > 0 so that if Bidder 1 bids b1 = −b,
his expectation will be positive. In particular, if c1 ; : : : ; cn ‘ is Nash, then
c1 > 0. Note that this argument breaks down if there are two bidders with
ei = ej = 0.
Since cj > −ej , Pbj ≤ 0‘ > 0 for all j. It follows that Gi 0‘ > 0 for all i.
Substituting into (2), Ei ei ‘ > 0 whenever ei > 0 and there is some c such
that E1 c‘ > 0 whenever e1 = 0. As c1 ; : : : ; cn ‘ was assumed to be Nash,
Ei ci ‘ > 0 for all i. Further, as each Ei b‘ is differentiable, we must have
Ei0 ci ‘ = 0. However, since Gi 0‘ > 0 and Gi is nondecreasing, Ei0 b‘ > 0
for all b satisfying −ei < b < ei , so ci ≥ ei .
To proceed further, let us focus on one way that two of the conditions
above could fail to be satisfied. Assume that we are looking at Bidder i and
that each of the opponents have chosen their bid functions, so in particular
Gi x‘ is defined. We say that b is flawed with respect to Gi if Ei b‘ = 0
or Ei0 b‘ < 0. We call a profile c1 ; : : : ; cn ‘ stable if each ci ≥ ei and no ci
is flawed with respect to Gi . From above, if c1 ; : : : ; cn ‘ is Nash then it is
stable. Also, it follows from (3) that e1 ; : : : ; en ‘ is stable.
The motivation for these definitions is that stability, unlike the notion
of being a Nash equilibrium, is preserved under passing to subauctions. In
the first lemma we explicitly compute the conditions under which a profile
c1 ; c2 ‘ can be stable for a two-bidder auction. The second lemma implies
that stability is preserved under subauctions.
Lemma A.1. If c1 ; c2 ‘ is stable in a two-bidder auction then c2 = e2 and
e1 ≤ c1 ≤ e2 .

Proof. We prove this in the case where e1 > 0 and leave the simpler
case to the reader. As there are only two bidders, Gi x‘ = Pb3−i ≤ x‘ for
i = 1; 2.
First, suppose by way of contradiction that c1 ≥ c2 and c1 > e2 . We claim
that c1 is flawed with respect to G1 . Due to the trichotomy of G1 , we argue
by cases.
If c1 ≥ e1 + e2 + c2 then E1 c1 ‘ = 0 as the integrand is always 0.
If e2 − e1 + c2 ≤ c1 < e1 + e2 + c2 , then for all b in this interval
Z c2 +e2
4e1 e2 E1 b‘ = ue2 + c2 − u‘ du:
b−e1
 asymptotic auction solution 251

By computation, E10 b‘ is a convex, quadratic polynomial in b with zeros at

e1 and e1 + e2 + c2 , so E10 c1 ‘ < 0.
If c2 ≤ c1 < e2 − e1 + c2 , then for all b in this interval
Z b+e1
4e1 e2 E1 b‘ = ue2 + c2 − u‘ du:
b−e1

In this case, E1 b‘ is a quadratic in b taking its maximum at e2 + c2 ‘/2 <
c1 . Thus, E10 c1 ‘ < 0.
Next, suppose c2 > e2 . From our condition above this implies c2 > c1 . We
show that c2 is flawed with respect to G2 in two cases. If c2 ≥ e1 + e2 + c1
then E2 c2 ‘ = 0 as the integrand is always 0. Also, if c1 < c2 < e1 + e2 + c1
then for all b in this interval
Z e1 +c1
4e1 e2 E20 b‘ = ue1 + c1 − u‘ du
b−e2

is a convex, quadratic polynomial with zeros at e2 and e1 + e2 + c1 , so

E20 c2 ‘ < 0.
Thus, c2 = e2 as c2 ≥ e2 is immediate from the definition of stability. But
then, e1 ≤ c1 ≤ e2 follows from stability and the first clause above.

Lemma A.2. If b is flawed with respect to Hi and Ki is differentiable and

nondecreasing, then b is flawed with respect to Gi = Hi Ki .
Proof. Let EH b‘ and EG b‘ be the expectation of Bidder i in the re-
spective auctions and assume b is flawed with respect to Hi . On one hand,
if Gi ei − b‘ = 0 then EG b‘ = 0 as the integrand of (2) is always 0, so b
would be flawed with respect to Gi . On the other hand, if Gi ei − b‘ > 0,
then Hi ei − b‘ and Ki ei − b‘ are both positive. Since b was flawed with
0
respect to Hi , it must be that EH b‘ < 0. Thus, if ei > 0 then
0
2ei EG b‘ = ei − b‘Gi ei − b‘ + ei + b‘Gi −ei − b‘
= ei − b‘Ki ei − b‘Hi ei − b‘ + ei + b‘Ki  − ei − b‘Hi  − ei − b‘
≤ ei − b‘Ki ei − b‘Hi ei − b‘ + ei + b‘Ki ei − b‘Hi −ei − b‘
0
= Ki ei − b‘EH b‘ < 0

and if ei = 0 then
0
EG b‘ = Gi −b‘ − bG0i −b‘
= Ki −b‘Hi −b‘ − bKi Hi ‘0 −b‘
0
≤ Ki −b‘EH b‘ < 0:

In either case, b is flawed with respect to Gi .

252 laskowski and slonim

Using our two lemmas we claim that if c1 ; : : : ; cn ‘ is stable then e1 ≤

c1 ≤ e2 and ci = ei for 2 ≤ i ≤ n. To see this, fix i > 1. Clearly, c1 ≥ e1 and
ci ≥ ei . As well, if c1 were flawed with respect to H1 x‘ = Pbi ≤ x‘, then
by Lemma A.2, c1 would be flawed with respect to G1 , contradicting the
stability of c1 ; : : : ; cn ‘. Similarly, ci is not flawed with respect to Hi x‘ =
Pb1 ≤ x‘. Thus, c1 ; ci ‘ is stable for the two-player subauction, so ci = ei
and e1 ≤ c1 ≤ ei by Lemma A1.
We are now able to classify the Nash equilibria. Suppose c1 ; : : : ; cn ‘
were Nash. Then it would be stable, so ci = ei for 2 ≤ i ≤ n and e1 ≤ c1 ≤
e2Q. As c2 ; : : : ; cn are known, G1 x‘ is determined. Specifically, G1 x‘ =
k ni=2 x + 2ei ‘, where 1/k = 2n−1 e2 e3 · · · en . Consider the function
φx‘ = x2e2 − x‘2e3 − x‘ · · · 2en − x‘:
By direct computation, φ0 x‘ = 0 has exactly one solution c in the interval
0; 2e2 ‘ and c can be seen to satisfy the equality
1 1 1
= + ··· + :
c 2e2 − c 2en − c
If e1 = 0 then E1 b‘ = bG1 −b‘ = φb‘ is maximized at c1 , so c1 must
be this c. If e1 > 0, then as E10 c1 ‘ = k0 ’φc1 + e1 ‘ − φc1 − e1 ‘“ for some
constant k0 , c1 must satisfy φc1 − e1 ‘ = φc1 + e1 ‘. Since 0 ≤ c1 − e1 ; c1 +
e1 ≤ 2e2 and φ0 x‘ = 0 has only one solution in 0; 2e2 ‘, there is only
one value of c1 in the interval ’e1 ; e2 “ where φc1 − e1 ‘ = φc1 + e1 ‘. (The
solution c to φ0 x‘ = 0 will be in the interval ’c1 − e1 ; c1 + e1 “, however, it
need not be the midpoint.)
It remains to show that c1 ; e2 ; : : : ; en ‘ is in fact a Nash equilibrium.
By the remarks above, c1 is the only place in the interval 0; 2e2 ‘ where
E10 b‘ = 0. However, if b ≤ 0 then E1 b‘ ≤ 0, which is not optimal and
if b > e2 then look at the two-bidder subauction b; e2 ‘. From the proof
of Lemma A.1, b is flawed with respect to H1 x‘ = Pb2 ≤ x‘, so by
Lemma A.2, b is flawed with respect to G1 . Thus, c1 is the only place
where E1 ci ‘ > 0 and E10 c1 ‘ = 0, so E1 b‘ is optimized at c1 .
Similarly, fix i > 1. Let Gi denote the distribution induced by the oppo-
nents’ profile. Since Gi 0‘ > 0 and Gi is nondecreasing, Ei0 b‘ > 0 for all
b < ei , so no b < ei can be optimal. Further, for any b > ei , it follows from
Lemma A.1 that c1 ; b‘ is flawed in the two-bidder subauction. Hence, by
Lemma A.2, b would be flawed with respect to Gi . That is, ei is the only
value where Ei b‘ > 0 and Ei0 b‘ = 0, so Ei b‘ is maximized at ei .

APPENDIX B: PROOF OF THEOREM 2

Let b1 ; : : : ; bn ‘ be the profile of bid functions prescribed by the TIN

Profile. Fix an i and we concentrate on the optimal bid function for ith
 asymptotic auction solution 253

bidder, assuming that the other bidders employ the TIN Profile. For any
bid function g, let
µ

E k g‘ = Ei k b1 ; : : : ; bi−1 ; g; bi+1 ; : : : ; bn :
Similarly, let E k y Ž Vi ‘ denote the conditional expected value of Bidder i
assuming he bids y upon seeing a private signal of Vi , where V is drawn
uniformly in the interval ’−k; k“ (and the opponenents employ the TIN
Profile). We begin with two observations.
• As the density of V is uniform on ’−k; k“ and the conditional density
of Vi Ž V is uniform on ’V − ei ; V + ei “, the conditional density of V Ž Vi is
uniform whenever Vi lies in the interval ’−k + ei ; k − ei “.
• Thus, it follows from Theorem 1 that for any k and for any Vi in
the interval ’−k + ei ; k − ei “, E k y Ž Vi ‘ attains a unique maximal value at
y = bi Vi ‘.
In fact, as E k is continuous, there is a continuous function T ‘ such that
T 0‘ = 0 and for all  > 0,
y − bi Vi ‘ > T ‘ implies E k y Ž Vi ‘ − E k bi Vi ‘ Ž Vi ‘ < −: (6)
Now let g be any measurable function that does not agree with bi almost
everywhere (i.e., g and bi disagree on a set of positive measure). We will
show that E k g‘ < E k bi ‘ for all but finitely many k. First, since the true
value V is always within en of the signal Vi , we may assume that Žgx‘ −
bi x‘Ž ≤ 3en for all x. In what follows, we write E k g Ž Vi ‘ for the more
precise E k gVi ‘ Ž Vi ‘ (and we write E k bi Ž Vi ‘ for E k bi Vi ‘ Ž Vi ‘). Note that
Z Z k−ei
E k g‘ = E k g Ž Vi ‘hVi ‘ dVi + E k g Ž Vi ‘hVi ‘ dVi ;
J −k+ei

where J = ’−k − ei ; −k + ei “ ∪ ’k − ei ; k + ei “ and hVi ‘ is the density

function of Vi , given that V is uniformly distributed on ’−k; k“. It is readily
seen that 0 ≤ hVi ‘ ≤ 1/2k for all Vi and hVi ‘ = 1/2k for all Vi in ’−k +
ei ; k − ei “. So, in order to establish that E k g‘ < E k bi ‘, it suffices to show
that
Z
E k g Ž Vi ‘ − E k bi Ž Vi ‘‘hVi ‘ dVi
J
Z k−ei
+ E k g Ž Vi ‘ − E k bi Ž Vi ‘‘hVi ‘ dVi (7)
−k+ei

is negative.
Since g disagrees with bi on a set of positive measure, we can choose M0 ,
δ > 0, and a set A ⊆ ’−M0 ; M0 “ of measure δ such that Žgx‘ − bi x‘Ž > δ
for all x ∈ A. Choose γ such that T γ‘ < δ, let  = γδ/21en ‘2 ‘ and let
B = ”x: Žgx‘ − bi x‘Ž ≥ •. The argument now splits into two cases.
254 laskowski and slonim

Case 1. The measure of B is finite.

Choose M > M0 such that the measure of B \ ’−M; M“ is less than 
and let k > M0 + ei . As in the proof of Theorem 1, let Gy Ž V ‘ be the
conditional probability that all of the opponents’ bids are less than y, given
that the true value is V . Since Gy Ž V ‘ is given piecewise by polynomials,
there is a number R > 1/en such that ŽGx Ž V ‘ − Gy Ž V ‘Ž ≤ RŽx − yŽ for
all x; y and Vi .
Now fix Vi and let d = V − bi Vi ‘. Note that ŽdŽ ≤ 2en . Let lV ‘ be the
density function of V Ž Vi . Then
E k g Ž Vi ‘ − E k bi Ž Vi ‘
Z Vi +ei  
= V − gVi ‘‘GgVi ‘ Ž V ‘ − V − bi Vi ‘‘Gbi Vi ‘ Ž V ‘ · lV ‘ dV
Vi −ei
Z Vi +ei  
≤ ŽdŽ · GgVi ‘ Ž V ‘ − Gbi Vi ‘ Ž V ‘ + Žbi Vi ‘ − gVi ‘Ž · lV ‘ dV
Vi −ei
Z Vi +ei
≤ 2en R + 1‘ bi Vi ‘ − gVi ‘ · lV ‘ dV
Vi −ei

≤ 3en R‘ bi Vi ‘ − gVi ‘ :

So, letting C = ”x ∈ J: Žgx‘ − bi x‘Ž ≥ •, our assumption that ŽgVi ‘ −
bi Vi ‘Ž ≤ 3en and our choice of B entails
Z
E k g Ž Vi ‘ − E k bi Ž Vi ‘‘hVi ‘ dVi
J
3en Z
≤ b V ‘ − gVi ‘ dVi
2k J i i
  (8)
3e Z Z
≤ n bi Vi ‘ − gVi ‘ dVi + bi Vi ‘ − gVi ‘ dVi
2k C J\C

3en 21e2n  γδ
3en ‘ + 4en ‘‘ =
≤ < :
2k 2k 2k
On the other hand, since the density of V Ž Vi is uniform for all Vi in
’−k + ei ; k − ei “, the property of T given in (6) implies
Z k−ei  
E k g Ž Vi ‘ − E k bi Ž Vi ‘ · hVi ‘ dVi
−k+ei

1 Z  k 
≤ E g Ž Vi ‘ − E k bi Ž Vi ‘ · hVi ‘ dVi (9)
2k A
1 Z −γδ
≤ −γ‘ dVi =
2k A 2k
Thus, E k g‘ < E k bi ‘ follows from the inequalities (7), (8), and (9).
 asymptotic auction solution 255

Case 2. The measure of B is infinite.

In this case, first choose ρ > 0 such that T ρ‘ <  and then choose M
such that the set C = B ∩ ’−M; M“ has measure greater than 8e2 ei /ρ. Now
fix k > M + ei . Here, since the opponents are sure to bid at least V − 2e2 ,
the maximum expected profit is E k g Ž Vi ‘ − E k bi Ž Vi ‘ ≤ 2e2 , so
Z

E k g Ž Vi ‘ − E k bi Ž Vi ‘ hVi ‘ dVi
J
Z 4ei 4e e
≤ 2e2 ‘hVi ‘ dVi ≤ 2e2 ‘ = 2 i;
J 2k k
while by our choice of C,
Z k−ei

E k g Ž Vi ‘ − E k bi Ž Vi ‘ hVi ‘ dVi
−k+ei

1 Z

≤ E k g Ž Vi ‘ − E k bi Ž Vi ‘ dVi
2k C
1 −4e2 ei
≤ mC‘−ρ‘ < ;
2k k

so again E k g‘ < E k bi ‘.

REFERENCES

Capen, E. C., Clapp, R. V., and Campbell, W. M. (1971), Competitive bidding in high-risk
situations, J. Petroleum Technol. 23 641–653.
Englebrecht-Wiggins, R., Milgrom, P., and Weber, R. (1983), Competitive bidding with pro-
prietary information, J. Math. Econ. 11, 161–169.
Hendricks, K., Porter, R. H., and Wilson, C. A. (1994), Auctions for oil and gas leases with
an informed bidder and a random reservation price, Econometrica 62, 1415–1444.
Kagel, J., and Levin, D. (in press), Bidding in common value auctions with asymmetric infor-
mation: An experimental analysis of inexperienced and super-experienced bidders, Econo-
metrica.
Milgrom, P. (1979), The Structure of Information in Competitive Bidding, New York: Garland.
Milgrom, P., and Weber, R. (1982), A theory of auctions and competitive bidding, Economet-
rica 50, 1089–1122.
Weverberg, M. (1979), Competitive bidding with asymmetric information reanalyzed, Manage-
ment Science 25, 291–294.
Wilson, R. (1967), Competitive bidding with asymmetric information, Management Science 13,
816–820.
Wilson, R. (1992), Handbook of Game Theory, Vol. 1, Amsterdam: Elsevier Science Publishers,
228–279.