Академический Документы
Профессиональный Документы
Культура Документы
Published by
Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
In this article, we investigate how information revelation rules affect the existence and the efficiency
of equilibria in two-round elimination contests. We establish that no symmetric separating equilib-
rium exists under the full revelation rule and find that the non-existence result is very robust. We
then characterise a partially efficient separating equilibrium under the partial revelation rule when
playersÕ valuations are uniformly distributed. We finally investigate the no revelation rule and find
that it is both most efficient and optimal in maximising the total efforts from the contestants. Within
our framework, more information revelation leads to less efficient outcomes.
Contests are widely used by various organisations wishing to select the best person or
the best team. Sports, for example, are undoubtedly always in the form of contests.
Science contests in schools are another example. We can consider a contest as an
auction in which players compete for prizes by submitting ÔbidsÕ. These ÔbidsÕ, once
submitted, have to be paid regardless of winning or losing. R&D races, political elec-
tions, competitions for promotions within certain organisations, lobbying etc., all fit in
the framework of contests. The ÔbidsÕ in contests such as R&D races, political elections
and lobbying are the amount of money spent by each agent, while in sports, science
competitions and promotion competitions, these ÔbidsÕ become the amount of effort
exerted by each contestant. Since all participants have to pay their bids regardless of
winning or not, contests are actually all-pay auctions. In the literature, Baye et al.
(1996), Hillman and Riley (1989) and Moldovanu and Sela (2006), to name a few, all
model contests as all-pay auctions.
Limited by the number of contestants in each contest, many contests are held in
multiple stages. Contestants are initially divided into a few groups and competitions
within each group are held first. The winners from each group compete again in later
stages, and the losers are eliminated from the competitions. As pointed out in
Moldovanu and Sela (2006), Ô[b]esides sports, [these] elimination contests are very
popular and widely used in the following situations: (1) in the organisation of internal
labor markets in large firms and public agencies, the sub-contests are usually regional
or divisional, and the prizes are promotions to well-defined (and usually equally paid)
positions on the next rung of the hierarchy-ladder; (2) in political competition (e.g.,
for the US presidency), candidates first spend resources to secure their party’s nomi-
nation, and later, if they are nominated, spend more resources to get elected; and (3)
science contests among university or high-school students, e.g., the Mathematics
Olympiad.Õ
* Wang’s research is supported by Social Sciences and Humanities Research Council of Canada. We thank
the editor, three anonymous referees, James Amegashie, James Bergin, Lu Jingfeng, Jan Zabojnik, partici-
pants at the 2008 Royal Economic Society Annual Conference, the 2008 Conference on Tournaments,
Contests and Relative Performance Evaluation, the 2008 Canadian Economic Theory Conference and sem-
inar participants in Economics Department at Queens University for very helpful comments.
[ 613 ]
614 THE ECONOMIC JOURNAL [MARCH
It is commonly believed that these multi-stage contests are capable of selecting the
ÔbestÕ contestants. Indeed, Moldovanu and Sela (2006) show that the contestant with
the highest willingness to pay for the prize will submit the highest bid in each stage and
thus win the contest. Their analysis is based on the assumption that no information is
released after each stage of the contest. In many contests, however, the assumption of
no information revelation is not applicable. Sometimes, the bids of the contestants may
be fully or partially revealed. In sports, for example, the performance of each team in
each stage of contest is publicly observed and cannot be concealed. In competition for
promotions within a firm, for another example, the performance of different workers
can also be partially observed by each other. In some cases, the organisers can even
control the amount of information released to the contestants. For example, the US
Federal Election Commission can have different policies on whether presidential
election candidates should reveal the amount of money they raised during election
campaigns. In schools, for another example, school officials can decide whether to
reveal the scores of the students in each round of science contests. In this article, we
investigate whether multi-stage elimination contests can still select the best contestant
in situations where some information is released in the interim stages.
We consider the simplest model possible with two rounds of contests and four ex ante
identical players. Each player has some privately informed valuation for the prize
offered by the contest. In the first round, the four players are equally divided into two
groups. The two players in the same group compete in an all-pay auction; the player
with the lower bid is eliminated from the contest and the player with the higher bid
becomes the winner in that group. In the second round, the two winners from the first
round compete again in another all-pay auction and the player with the higher bid
becomes the second round winner of the contest and receives the prize.
In order for the contest to be efficient, that is, the player with the highest valuation
always wins the prize, it must be the case that the higher valuation player submits a
higher bid in each auction. With our assumption that players are ex ante identical,
contest efficiency can be achieved only when the players use a symmetric and strictly
increasing bidding function in each auction. We find that when playersÕ bids in the first
round are publicly revealed before the second round competition, no symmetric
strictly increasing equilibrium bidding function exists and, therefore, the player with
the highest willingness to pay may not always win the prize. Hence, multi-stage elim-
ination contests with full revelation of bids are not efficient.
The intuition behind the above result that multi-stage elimination contests with full
revelation of bids fail to always select the player with the highest willingness to pay
(valuation) as the winner is as follows. In an efficient equilibrium where the player with
the highest willingness to pay always wins, players must use a symmetric strictly
increasing bidding function in each of the two rounds of the contest. That means the
equilibrium is completely separating. With full revelation of bids right after the first
round, players in the second round have complete information regarding each other’s
valuation along the equilibrium path. However, a player’s expected payoff calculated at
the start of the first round contest is not differentiable in his bid at the value of his
equilibrium bid. This is because to ensure that a player does not want to pretend to
have a lower valuation, the bidding function in the first round needs to be relatively
flat; conversely, in order to ensure that a player does not want to pretend to have a
Ó The Author(s). Journal compilation Ó Royal Economic Society 2009
2009 ] ROLE OF INFORMATION 615
higher valuation, that bidding function needs to be relatively steep. As a result, no
bidding function can satisfy the incentive compatibility constraints in both directions in
the first round. Therefore, if an equilibrium exists, it must involve either non-mono-
tone bids, pooling of bids, or mixed strategies in the first round. This implies that the
winners in the first round contests may not have the highest valuation and entries to
the second round contest may not be efficient.
There are other examples of dynamic auctions where efficient equilibria may not
exist. In the war of attrition literature (see Bulow and Klemperer (1999), for
example), it is well known that efficient equilibrium does not exist for more than
two ex ante identical players. A war of attrition is an open-bid all-pay auction and it is
slightly different from the sealed-bid all-pay auction we consider in this article. The
intuition for the non-existence of an efficient equilibrium when there are more
than two players is as follows. Suppose that there are three players competing for
one prize. At the equilibrium quitting time, a player is indifferent between quitting
and continuing. If he quits, he gets zero. If he continues, one of his rivals may quit
and he then can enter the second competition with the remaining rival. But the fact
that he is indifferent between quitting and not quitting at that time implies that he
has the lowest valuation among the three players and thus he will surely lose the
remaining attrition even if one of his rivals quits around that time. Therefore, he
should quit earlier. This unravelling establishes the non-existence of a symmetric
separating equilibrium.1 In this case, a player’s quitting time is publicly revealed and
observed. The player with the highest valuation may not win this war of attrition in
equilibrium.
The literature on signalling in auctions is another example of non-existence of
efficient equilibrium. Goeree (2003) considers an auction with an aftermarket and
bidders compete for the advantage in the strategic interactions in the aftermarket.
The winning bid in the auction is revealed. Haile (2000, 2001, 2003) considers a
model of an auction followed by a resale auction offered by the winner of the first
auction. The bidders have some noisy private signals regarding their valuations
coming into the first auction and their bids are revealed before the resale auction.
In these papers, efficient symmetric separating Perfect Bayesian Nash Equilibria are
investigated and it is found that such separating equilibria may not exist. Goeree
(2003) states that Ôif bidders want to understate their private information, a separ-
ating equilibrium may fail to exist when the incentives to signal via a lower bid are
stronger for higher valuationsÕ. Furthermore, Haile (2003) finds that, due to sig-
nalling effects, player’s objective function is not quasi-concave, which contributes to
the failure of the existence of a separating equilibrium. In yet another recent paper,
Cai et al. (2007) investigate sequential first-price and second-price auctions of
multiple units. They find that when all bids are revealed after each auction, there
exists no equilibrium with a symmetric strictly increasing bidding function. Similarly
to our model, bidders have different incentives in increasing and decreasing their
bids, and no strictly monotone bidding function can satisfy the first order condi-
tions implied by the incentive compatibility constraints. Therefore, bidders with the
highest valuations may not win the objects.
1
We thank a referee for linking the intuitions of the two non-existence results here.
2
Zhang (2008) changes the second round all-pay auction to a lottery where a player’s winning probability
is an increasing function of his bid. The winning probability in this second round contest is a smooth
function. Under this setting, a separating equilibrium exists.
3
See McAfee and Vincent (1993), Mezzetti et al. (2008) and Weber (1983).
4
Early work on eliminating contests considers the case of complete information. See Groh et al. (forth-
coming), Horen and Riezman (1985), Hwang (1982), Rosen (1986) and Schwenk (2000).
2. Full Revelation
We first analyse the full revelation rule. All bids in the first round contest are revealed.
This is the most interesting case as we need to deal with signalling effects in the analysis.
When a player deviates from his equilibrium strategy and pretends to have a different
valuation, he is mistaken by the other player in the second round as that pretended
type and thus receives responses aimed at that type. In this way, a player can manipulate
the strategies of the other player by choosing different pretended valuations. Of course,
in equilibrium, each player will follow his equilibrium strategy.
Assume that the equilibrium bidding function in the first round is strictly increasing.
We first examine the second round interactions.
Player A bids
8 0; if w > vA and w > vB ;
>
<
F 0; if w > vA and w vB ;
bA ðw; vA ; vB Þ ¼ ð2Þ
: vB ;
> if w vA and w > vB ;
w; if w vA and w vB :
Player A’s payoff is
8
>
> vA ðw vB Þ
>
< w ; if w > vA and w > vB ;
F
PA ðw; vA ; vB Þ ¼ 0; if w > vA and w vB ; ð3Þ
>
> v vB ; if w vA and w > vB ;
>
: A
vA w; if w vA and w vB :
where I all expectations are taken on the upper case variables in the above equation,
and for the rest of the article as well. Unfortunately, as we show in the following
Proposition, no symmetric separating equilibrium exists in this game.
Proposition 1. In the elimination contest with full revelation rule,no symmetric separating
equilibrium exists in which all players adopt the same strictly increasing bidding function in the
first round.
8
< w þ /A vB /B þ bB ;
> if w þ /A vB þ /B ;
GB ðbB Þ ¼ w þ /A ð4Þ
>
: bB ; if w þ /A vB þ /B .
w þ /A
Player A bids
8
> 0; if w > vA and w þ /A > vB þ /B ;
<
0; if w > vA and w þ /A vB þ /B ;
bAF ðw; vA ; vB ; /A ; /B Þ ¼ ð5Þ
> vB þ /B ;
: if w vA and w þ /A > vB þ /B ;
w þ /A ; if w vA and w þ /A vB þ /B .
Given the above Lemma, we can formulate a player’s problem at the beginning of the
first round and determine the conditions under which a separating equilibrium must
satisfy. Again, we obtain the non-existence result.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2009
2009 ] ROLE OF INFORMATION 623
Proposition 2. In the elimination contest with full revelation rule and publicly observed
interim valuation shocks, no symmetric separating equilibrium exists in which all players adopt the
same strictly increasing bidding function in the first round.
Note that when the random shocks are equal to zero, the analysis is exactly the
same as in Proposition 1. The intuition for these random shocks not being able
to smooth the kinks is similar to the intuition in the first modification. For each
realisation of the random shock, there is a kink in the payoff function. Taking
the expectation at the beginning of the first round does not iron out these
kinks.
Making the above random shocks the private information of the players would
not smooth out the kinks either. The reason for the kink is because of the dis-
continuity in the deviated first round winner’s best response function in the special
deviated continuation game in the second round. This feature remains the same
here. Suppose finalist i has under-reported his type in the first round, then it is
optimal to bid zero in the second round when the shock is sufficiently low. If he
over-reports his type in the first round, then it is optimal to bid the upper bound of
the bidding function in the second round when the shock is sufficiently high. Both
situations occur with strictly positive probabilities and it makes the best response
function discontinuous in his reported type in the first round. Therefore, kinks
remain in the payoff function.
As we argued in the introduction, the non-existence of a symmetric separating
equilibrium is common in dynamic games with discontinuous probability of winning
functions and full revelation of bids in the interim stages. Zhang (2008) assumes that
the second round is a lottery in a two-round elimination contest. The first round is still
an all-pay auction. The second round is a lottery, with a player’s winning probability
increasing in his own bid but decreasing in other playersÕ bids. He is able to establish
the existence of a symmetric separating equilibrium in this game. The analysis is
complex but the intuition is quite straightforward. Consider an extreme case. If the
second round is a pure lottery so that the playersÕ winning probabilities do not depend
on their bids in the second round, then it is quite obvious that a separating equilibrium
exists since first round bids no longer have any signalling effects. The reward for
entering into the second round is fixed and the first round contest becomes a one-shot
all-pay auction with incomplete information. Changing the pure lottery to a lottery with
endogenous winning probability transforms the fixed reward to a variable but still
smooth reward. Since there is no kink in the payoff functions, a symmetric separating
equilibrium can be characterised.
3. Partial Revelation
In this Section, we analyse the case where the designer can conceal the efforts
(bids) of the winners but not those of the losers. Therefore, after the first round of
contests, the bids of the eliminated players are revealed and the bids of the
winners remain concealed. Below, we analyse how this change in information rev-
elation affects playersÕ behaviour and whether a symmetric separating equilibrium
exists.
Ó The Author(s). Journal compilation Ó Royal Economic Society 2009
624 THE ECONOMIC JOURNAL [MARCH
3.1. Second Round Strategy
We first examine a non-deviated continuation game in the second round; all players
follow their equilibrium strategy in the first round. According to the rule, the bids
of the eliminated players are revealed. After seeing these bids, the finalists (win-
ners), player A and player B, update their beliefs regarding each other’s valuation.
Since the bidding function in the first round is strictly increasing, they can calculate
the losersÕ valuations va and vb by inverting the bidding function and thus va and vb
are fully revealed. Meanwhile, it is common knowledge that players A and B are the
winners in their own groups. Using BayesÕ rule, player A believes that player BÕs
valuation is distributed on ½vb ; v with c.d.f. [F(vB) F(vb)]/[1 F(vb)] and player B
believes that player AÕs valuation is distributed on ½va ; v with c.d.f. [F(vA) F(va)]/
[1 F(va)]. Since prob(va ¼ vb) ¼ 0, players are in an asymmetric all-pay auction
with probability one. There are many papers investigating asymmetric auctions.6 It is
impossible to obtain an analytical solution for asymmetric auctions with general
distributions. Nevertheless, an analytical solution can be obtained for uniform
distributions.7 Define
The following Lemma characterises the equilibrium bidding strategy under the
assumption that F(v) is uniformly distributed.
Lemma 4. Consider an asymmetric all-pay auction with two bidders I,J. Their valuations vI
and vJ are independently distributed. Bidder I Õs valuation is distributed uniformly on ½vi ; v, and
bidder J Õs valuation is distributed uniformly on ½vj ; v, where vi vj. The following strategies
consist of a Bayesian Nash Equilibrium.
v vi vj
PI ðvI ; vi ; vj Þ ¼ HI vI þ Hi ; ð8Þ
v vj v vj
6
Amann and Leininger (1996) consider the asymmetric all-pay auction with two players. Maskin and Riley
(2000) discuss asymmetric auctions from the aspect of revenue under more general settings.
7
One good example can be found in Krishna (2002).
Bid
bJ
bI
v vj vi v−vj vi v Valuation
v( ) v−vi
v
8 vvj
< v vj H vi v þ H ; if vðv =
i v Þ i vJ v ;
vv
PJ ðvJ ; vi ; vj Þ ¼ v v i
J
v vi
J i
ð9Þ
vvj
:
0; if vj vJ < vðvi =
v Þvvi .
Figure 1 illustrates the important features of the two bidding functions above. When
vj < vi, we define player J as the weak player and I as the strong player. For the weak
player, if he is sufficiently weak, he bids 0. The cutoff valuation for the weak player,
who is indifferent
vv
between bidding and not bidding, lies between vj and vi, i.e.
j
v Þv vi vi .8 Furthermore, the weak player always bids more aggressively than
vj vðvi =
the strong player.
When all the players followed their equilibrium strategies in the first round, the
equilibrium in its non-deviated continuation game in the second round is just a
straight-forward application of Lemma 4. Note that either player A or player B can
be the strong player, depending on the revealed valuations of the eliminated
players.
In order to characterise the equilibrium bidding strategies in the first round, we
need to examine one more continuation game in the second round. In this continu-
ation game, only one player deviated and did not following his equilibrium bidding
function in the first round, but he was able to enter the second round. Without loss of
generality, let player 1 be the deviated player. Suppose that player 1 bids bP(w) instead
of his equilibrium bid bP(v1). If he fails to enter the second round (i.e. w < v2), then
the case is irrelevant to our analysis. However, if he is able to enter the second round
(i.e. w > v2) and becomes player A, then he competes with player B (the winner of
group B) in the second round.
In the second round, player B believes that player AÕs valuation is uniformly
distributed on ½va ; v, while player A believes that player BÕs valuation is uniformly
distributed on ½vb ; v. However, player BÕs belief could be wrong, i.e. player AÕs true
valuation vA could lie outside of ½va ; v, as player A deviated in the first round. Define
8
< HB Hb ; if va < vb and vb vB v
>
vb
v;
F
bB ðvB ; va ; vb Þ ¼ H B Ha ; if v a vb and
v ðv a =
v Þ vva v
B v ; ð10Þ
>
: vvb
0; if va vb and vb vB vðva =vÞvva .
bA ðvA ; va ; vb Þ ¼ HA Hb ;
F if va vb and vðvb =
v Þvvb < vA v; ð11Þ
>
> 0; if va > vb and v vA va ;
>
:
HA Ha ; if va > vb and va < vA v.
Player A’s payoff is given by:
8 vva
>
>
> P1 ðvA ;va ;vb Þ¼ 0; if va vb and v vA vðvvb Þvvb ;
>
>
>
> v va vb vA vva
if va vb and vðvvb Þvvb < vA v;
< P2 ðvA ;va ;vb Þ¼ v vb HA v vb þHb ;
>
PFA ðvA ;va ;vb Þ vvb
v v
v vva
a
>
> vb
>
> P3 ðvA ;va ;vb Þ¼ vA ; if va > vb and v vA va ;
>
>
v v
>
>
b
: P4 ðvA ;va ;vb Þ¼ v va HA vb vA þHa ; if va > vb and va < vA v.
v v b v v b
ð12Þ
There are two important features about this continuation game. First, Player BÕs
strategy does not depend on player AÕs pretended valuation, simply because he cannot
observe player AÕs bid in the first round. Second, and because of the first feature, AÕs
strategy also does not depend on his own pretended valuation.
We now have the necessary continuation games in the second round to analyse the
first round strategies.
The first four terms represent the expected gain if he wins the prize, which depends
on the realisation of the eliminated bids. The last term is the cost of bidding. In
equilibrium, it is optimal for the players to truthfully report their valuations. The
following Proposition characterises the equilibrium bidding strategy in the first round
as well as the second round.
Proposition 3. In elimination contests with partial revelation, the symmetric separating
PBNE is as follows:
In the first round, all players adopt the same strictly increasing bidding function:
Z v ( "
vv # )
P 2n vð
v nÞ n vn n n2 v 2
b ðvÞ ¼ dn: ð14Þ
v ðv vÞ3 ln ðn= vÞ v v 2
In the second round, player A believes that player B’s valuation is uniformly distributed on
½vb ; v, while player B believes that player A’s valuation is uniformly distributed on ½va ; v. They
use the bidding strategy described in Lemma 4 in the second round.
As we can show in the proof, a separating equilibrium always exists. In this equilib-
rium, efficient of entry to the second round is achieved. However, since the second
round is an asymmetric all-pay auction, the allocation of the prize may not be efficient.
If we examine (13), we find that the problem is no longer equivalent to a standard
all-pay auction, since we cannot separate the valuations from the winning probabilities.
With partial information revelation, the equilibrium is partially efficient.
4. No Revelation
We finally analyse the case that the organiser chooses not to reveal any information
after the first round of contest. The finalists know only who won and who lost in the
first round. This case has been analysed by Moldovanu and Sela (2006). We summarise
their results in the following Proposition.
Proposition 4. (Moldovanu and Sela, 2006) In the elimination contest with no revela-
tion, the symmetric separating PBNE is as follows.
In the first round, players bid according to the strictly increasing function:
In the second round, the finalists believe that their rivalsÕ valuations are drawn from the
distribution F()2 and bid according to the strictly increasing function:
Z v
b F ðvÞ ¼ ndF ðnÞ2 : ð16Þ
v
If we define the value of entering the second round as a player’s new valuation, then
the first round contest is equivalent to a standard all-pay auction. In this auction, a
player’s payoff is equal to his valuation of the prize multiplied by the winning probab-
ility in the second round, and minus his bid. However, under the other two informa-
tion revelation rules, this is not the case. A player’s valuation and winning probability
are not separable.
The existence of a separating equilibrium guarantees the most efficient entry to the
second round. Since the finalists are using symmetric bidding function in the second
round, the prize goes to the player with the highest valuation. As a result, the outcome
is efficient and the player with the highest valuation always wins the prize.
5. Conclusion
In this article, we consider several commonly observed information revelation rules and
address how they affect the existence of separating equilibria and the allocation of the
prize in elimination contests. In general, different revelation rules imply different
allocations of the prize. Only the no revelation rule induces efficient entry to the
second round and efficient allocation of the prize in the second round. Furthermore, it
maximises the total efforts from the players. The partial revelation rule introduces
distortion of efficiency in the second round because of asymmetric information in that
round, even though efficient entry to the second round is still achieved. The full
revelation rule cannot warrant efficient entry to the second round, since no separating
equilibria exists with strictly increasing bidding function in the first round. Any equi-
librium under the full revelation rule would necessarily involve inefficient entry to the
second round competition. We show that the non-existence result is very robust with
full information revelation. Our analysis confirms the findings of non-existence of
separating equilibrium by other researchers in other auction models with full interim
information revelation.
Within the framework of our model, we can conclude that more information
revelation in the interim stages reduces the efficiency of the contest. The no revelation
rule guarantees that the player with the highest valuation eventually wins the prize. It is
also the best rule to maximise player total efforts. Even though these conclusions are
drawn from a four-player model with two rounds of competitions, the qualitative results
should remain valid in a model with any arbitrary number of players and any arbitrary
number of rounds of competitions. As long as the objective is to facilitate efficient entry
to later rounds of competitions and to ensure that the player with the highest valuation
wins the prize, the no revelation rule is optimal.
9
The effort levels are implied by the incentive compatibility constraints.
with boundary condition bðvÞ ¼ 0. The expected gain function G(v,v1): ½v; v ½v; v ! R is twice
continuously differentiable and depends on both the true valuation v1 and the reported valuation Rv v. Assume
that G1(v1,v1) > 0, 8v1 2 ðv; v, and G12(v,v1) > 0, 8v; v1 2 ðv; v. Then bðv1 Þ ¼ v 1 G1 ðn; nÞdn is
the equilibrium bidding function. This bidding function is strictly increasing and fully valuation revealing.
Proof. Suppose there exists a strictly increasing function b(v) such that truthfully reporting
(v ¼ v1) is optimal, then incentive compatibility implies that
Z v1
G1 ðv1 ; v1 Þ b 0 ðv1 Þ ¼ 0 ) b 0 ðv1 Þ ¼ G1 ðv1 ; v1 Þ ) bðv1 Þ ¼ G1 ðn; nÞdn: ð18Þ
v
0
Note that b ðv1 ÞR ¼ G1 ðv1 ; v1 Þ > 0; 8v1 2 ðv; v. So it is true that b(v1) is strictly increasing.
v
Given bðv1 Þ ¼ v 1 G1 ðn; nÞdn, we have
dU ðv; v1 Þ
¼ G1 ðv; v1 Þ b 0 ðvÞ ¼ G1 ðv; v1 Þ G1 ðv; vÞ: ð19Þ
dv
Since G12 ðv; v1 Þ > 0 8v; v1 2 ðv; v, then G1(v,v1) is a strictly increasing function of v1. Hence,
dU(v,v1)/dv ¼ G1(v,v1) G1(v,v) > 0 if v1 > v; and dU(v,v1)/dv ¼G1(v,v1) G1(v,v) < 0 if
v1 < v. Thus, v ¼ v1 is optimal.
Proof of Lemma 2
We divide the second round continuation game into the following four situations: (1) w > vA,
w > vB, (2) w > vA, w vB, (3) w vA, w > vB, and (4) w vA, w vB.
Player A overbids in the first round and meets a rival weaker than his pretended valuation
in the second round. Then from Lemma 1, player B randomises his bid in the interval [0,vB]
according to c.d.f. GB(bB) ¼ (wvBþbB)/w. Knowing player BÕs reaction function as well as his
own true valuation, player AÕs optimisation problem becomes
w vB þ bA
max PFA ¼ max vA GB ðbA Þ bA ¼ max vA bA : ð20Þ
bA bA bA w
The first order condition gives us @PFA =@bA ¼ ðvA =wÞ 1 < 0. Note that player A is the
strong player and always wins in a tie. Therefore, it is optimal for him to bid 0 and get an
expected payoff of vA(w vB)/w.
Player A overbids in the first round and meets a rival stronger than his pretended valuation in
the second round. Then from Lemma 1, player B randomises his bid in the interval [0,w]
according to c.d.f. GB(bB) ¼ bB/w. Then player AÕs optimisation problem becomes:
bA
max PFA ¼ max vA GB ðbA Þ bA ¼ max vA bA : ð21Þ
bA bA bA w
Player A underbids in the first round and meets a rival weaker than his pretended valu-
ation in the second round. Then from Lemma 1, player B randomises his bid in the interval
[0,vB] according to c.d.f. GB(bB) ¼ (wvB þ bB)/w. Then player AÕs optimisation problem
becomes:
w vB þ bA
max PFA ¼ max vA GB ðbA Þ bA ¼ max vA bA : ð22Þ
bA bA bA w
FOC: @PFA =@bA ¼ ðvA =wÞ 1 > 0, then it is optimal for him to bid vB and get an expected
payoff of vA vB.
Case 4. w vA, w vB
Player A underbids in the first round and meets a rival stronger than his pretended
valuation in the second round. Then from Lemma 1, player B randomises his bid on the
interval [0,w] according to c.d.f. GB(bB) ¼ bB/w. Then player AÕs optimisation problem
becomes:
bA
max PFA ¼ max vA GB ðbA Þ bA ¼ max vA bA : ð23Þ
bA bA bA w
FOC: @PFA =@bA ¼ ðvA =wÞ 1 > 0, then it is optimal for him to bid w and get an expected
payoff of vA w. We have now covered all possible cases.
Proof of Proposition 1
If player 1 chooses to overbid in the first round (w v1), then player 1Õs payoff is given by:
Z w
v1 ðw vB Þ
PP1 ðw; v1 Þ ¼ F ðwÞ dF ðvB Þ2 b P ðwÞ
v w
Z
F ðwÞv1 w
¼ v1 F ðwÞ3 vB dF ðvB Þ2 b P ðwÞ:
w v
" Z # Z
v1
0 F ðv1 Þ v1
) b P ðv1 Þ v1 F ðv1 Þ2 f ðv1 Þ f ðv1 Þ vB dF ðvB Þ2 þ vB dF ðvB Þ2 : ð25Þ
v v1 v
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
0
Proof of Proposition 2
If player 1 chooses to overbid in the first round (w v1), then player 1Õs payoff is given by:
Z / Z / Z wþ/1 /B
ðv1 þ /1 Þðw þ /1 vB /B Þ
PP1 ðw; v1 Þ ¼ F ðwÞ dF ðvB Þ2
/ / v w þ /1 ð29Þ
dH ð/1 ÞdH ð/B Þ b P ðwÞ:
Proof of Lemma 4
Suppose that in the equilibrium players I and J follow increasing bidding strategies bI() and bJ(),
respectively.
Proof. Suppose that bJ(vj) > 0. Then for player I with any valuation, bidding 0 dominates
bidding b 2 (0,bJ(vj)). This is because bidding 0 gives him payoff 0 while bidding b 2 (0,bJ(vj))
gives him a negative payoff. Thus, player I will not bid b 2 (0,bJ(vj)). However, in this case, player J
with valuation vj will not bid bJ(vj), since bidding a little bit lower is better. A similar argument can
be applied to show that bI(vi) ¼ 0.
Claim 2 bJ ð v Þ ¼ b
v Þ ¼ bI ð
Proof. Suppose that bJ ðvÞ > bI ðvÞ. Then player J with valuation v is always better off to bid a
little bit less, since his winning probability does not change while he saves on the cost. A similar
argument applies to bJ ðvÞ < bI ðvÞ.
Suppose we now assume additionally that the bidding functions take the following forms. bI(vI)
is strictly increasing on its support. bJ(vJ) ¼ 0 for vj vJ v and is strictly increasing on
v vJ v. Denote /J ¼ bJ1 for v vJ v and /I ¼ bI1 , respectively.
Player IÕs optimisation problem is given by:
/J ðbÞ vj
max vI Probfb > bJ ðvJ Þg b max vI ProbfvJ < /J ðbÞg b ¼ max vI b: ð30Þ
b b b v vj
The first order condition gives us ½1=ð v vj Þ/0J ðbÞvI ¼ 1. In equilibrium vI ¼ /I(b). Thus
this FOC becomes
/0J ðbÞ/I ðbÞ ¼ v vj : ð31Þ
Since bI ð
v Þ ¼ bJ ð we determine the constant term in (34) as A ¼ v2 ð2
v Þ ¼ b,
v vi vj Þb.
Thus, we have:
/I ðbÞ/J ðbÞ ¼ ð2 þ v2 :
v vi vj Þðb bÞ ð35Þ
From bI ð we have B ¼ v
v Þ ¼ b, v vi vj
2
. Thus, (37) becomes
2vv
vvi
v vi vj
2
i vj
/I ðbÞ ¼ v ðb bÞ þ 1 : ð38Þ
v2
v2 1 vi
vvi
v
) b ¼ : ð40Þ
v vi vj
2
v vi vj
2 v v vi vj
2
vvi
i v j
i
/I ðbÞ ¼ v b þ ð41Þ
v2 v
" #
v2 v 2vv i vj
vvi
v 2vv i vj
vvi
I i
) bI ðvI Þ ¼ : ð42Þ
v vi vj
2 v v
v vi vj
2 v v vi vj
2
vvi
i v j
i
/J ðbÞ ¼ v b þ ð43Þ
v2 v
" #
v2 v 2vv i vj v 2vv i vj
J vvj i vvi
) bJ ðvJ Þ ¼ : ð44Þ
v vi vj
2 v v
vvv j
If vðvi =
v Þv vi vJ v, then
/I ½bJ ðvJ Þ vi
PJ ðvJ Þ ¼ vJ bJ ðvJ Þ
v vi
v 2vv i vj v v vi vj
2 ð48Þ
v2 ð
vv Þ j J
vv j vi v2 i vvi
¼ vJ þ :
v vi Þð2
ð v vi vj Þ v v vi v vi vj v
2
Proof of Lemma 5
For player B, he has followed his equilibrium strategy in the first round. Meanwhile, he believes
that all the other players have also followed their equilibrium strategies in the first round. Thus,
he believes that player AÕs valuation is uniformly distributed on ½va ; v with c.d.f.
ðvA va Þ=ð v va Þ, and that player A considers him as a player with a valuation uniformly
distributed on ½va ; v with c.d.f. ðvB vb Þ=ð v vb Þ. Note that two situations may occur in the
second round: one is va < vb and the other is va vb. Hence, player B’s strategy can be derived
directly from Lemma 4.
8 " #
> 2vv a vb 2vv a vb
>
> v2 v B vvb v
vb
vvb
if va < vb and vb vB v
>
> v va vb
2 v
>
>
>
< " #
2vv a vb 2vva vb vvv
vb
bBF ðvB ; va ; vb Þ ¼ v2 vB vvb va vva if v v and
v va a
vB v
>
> v va vb
2 v v a b
v
>
>
>
> vvv
>
>
vb
:0 if va vb and vb vB v va
a
.
v
ð49Þ
An important feature is that player BÕs strategy does not depend on player AÕs pretended
valuation in the first round simply because player B does not know it.
For player A, he acts optimally in the second round, taking player BÕs bidding function as given
and knowing his pretended valuation w and true valuation vA. Given that player 1 is in the second
round and becomes player A, his optimisation problem in the second round is given by
max vA Prob½b > bBF ðVB ; va ; vb Þ b: ð50Þ
b
The second step is from the fact that it is optimal for the player with valuation va to bid zero
and get payoff 0.
Second, if player A is the strong player va > vb,
vv
vvb
2
v v a vb
va 2vv a vb 2 a vb
v 2 b þ v vva
vb
v
ð50Þ ¼ max vA b:
b v vb
vA 2 v va vb v 2vv a vb 2
vva
a vb
vA va vva b
v va v
2 a vb
vA
a
¼ bþ 1 < 1 < 1 < 0: ð51Þ
v v2 v v v va
Note that player A is the strong player and he wins whenever there is a tie. Therefore, it is
v vb
ðva =
v Þ vva vb
optimal for him to bid 0 and get a payoff of vA . Lemma 4 summarises all the
v vb
situations together.
Proof of Proposition 3
The following claims will be used several times in the proof.
We will use Theorem 1 to determine the equilibrium. Player 1Õs problem is given by:
b P ðwÞ ð58Þ
where P1(v1;V2,Vb), P2(v1;V2,Vb), P3(v1;V2,Vb) and P4(v1;V2,Vb) are defined in Lemma 5. V2 has
c.d.f. F ðv2 Þ ¼ ðv2 vÞ=ð v vÞ. We know that Vb is the loser in group B, whose c.d.f. is given by
2F(vb) F(vb)2 , where F ðvb Þ ¼ ðvb vÞ=ðv vÞ.
Note that term (54) is always 0, since P1(v1;V2,Vb) ¼ 0. vV2
In order to pin down term (55), we need to investigate the properties of v1 ¼ vðVb = v ÞvVb .
We can rearrange it and write V2 as a function of Vb: V2 ¼ hðVb ; v1 Þ ¼ v ð v Vb Þ
½lnðv1 =
v Þ=lnðVb =v Þ. Figure 2 characterises the important features of function h(Vb;v1).
First, it is a decreasing function:
v
@hðVb ; v1 Þ ln v1 ln vv1 1 vln vv1 Vvb ln Vvb þ 1 Vvb
¼ þ ð
v Vb Þ 2 ¼ 2 <0 ð59Þ
@Vb ln Vvb ln Vvb
Vb
Vb ln Vvb
The last step comes from the fact that ðVb = v ÞðlnVb = v Þ þ 1 Vb =
v > 0 , which is implied by
Claim 3, and lnðv1 = v Þ < 0.
Second, it crosses the line V2 ¼ Vb exactly once at V2 ¼ Vb ¼ V1.
Note that we can also express Vb as a function of V2, Vb ¼ h1(V2; v1), an inverse function of
h(Vb; v1).
Let us first examine the case in which player
R w R h11ðvunderbids in the first round, i.e., w < v1.
;v Þ
From Figure 3, we obtain term (55) ¼ v f v2 2 1 P2 ðv1 ; v2 ; vb Þd½2F ðvb Þ F ðvb Þ2 g dF ðv2 Þ.
It is easy to see that term (56) ¼ 0, since Ifw>V2,V2 > Vb,v1 < V2g ¼ 0.
V2
V2 = Vb
v1
V2 = h(Vb;v1)
Vb
v1
V2
V2 = Vb
v1
w V2 = h(Vb;v1)
Vb
v1
V2
V2 = Vb
v1
V2 = h(Vb;v1)
Vb
v1
Rw Rv
Term (57) ¼ v f v 2 P4 ðv1 ; v2 ; vb Þd½2F ðvb Þ F ðvb Þ2 g dF ðv2 Þ.
Second, we examine the case in whichR player 2 over bids in the first round, i.e., w v1.
v R h 1 ðv ;v Þ
From Figure 4, we have Term (55) ¼ v 1 f v 2 1 P2 ðv1 ; v2 ; vb Þd½2F ðvb Þ F ðvb Þ2 g dF ðv2 Þ.
It is easy to see
R wthat:
Rv
Term (56) ¼ Rv1 f Rv 2 P3 ðv1 ; v2 ; vb Þd½2F ðvb Þ F ðvb Þ2 g dF ðv2 Þ
v1 v
Term (57) ¼ v f v 2 P4 ðv1 ; v2 ; vb Þd½2F ðvb Þ F ðvb Þ2 g dF ðv2 Þ:
Therefore, G(w,v1) becomes(for w < v1, )
Z w Z h1 ðv2 ;v1 Þ
2
Gðw; v1 Þ ¼ P2 ðv1 ; v2 ; vb Þd½2F ðvb Þ F ðvb Þ dF ðv2 Þ
v v2
Z (Z )
w v2
2
þ P4 ðv1 ; v2 ; vb Þd½2F ðvb Þ F ðvb Þ dF ðv2 Þ
v v
From Theorem 1, if RG(w,v1) satisfies the conditions: G1 ðv1 ; v1 Þ > 0; G12 ðw; v1 Þ > 0; 8w;
v
v1 2 ðv; v, then b P ðv1 Þ ¼ v 1 G1 ðn; nÞ dn constitutes an equilibrium.
For w < v1, we have
Z h1 ðv1 ;v1 Þ
G1 ðv1 ; v1 Þ ¼ f ðv1 Þ P2 ðv1 ; v1 ; vb Þd½2F ðvb Þ F ðvb Þ2
v1
Z v1
þ f ðv1 Þ P4 ðv1 ; v1 ; vb Þd½2F ðvb Þ F ðvb Þ2
v
Z v1
¼ f ðv1 Þ P2 ðv1 ; v1 ; vb Þd½2F ðvb Þ F ðvb Þ2
v1 ð61Þ
vvb
Z v1 v v1 vv1 vb
þ f ðv1 Þ v1 v d½2F ðvb Þ F ðvb Þ2
v
v v b
vvb
Z v1 v v1 vv1 vb
¼ f ðv1 Þ v1 v 2f ðvb Þ½1 F ðvb Þdvb :
v v vb
v w v vvb
w vb
@P3 ðv1 ; w; vb Þ v vb vb
¼ > ¼ 0;
@v1 v vb v vb
where the inequality holds because of Claim 4 and w > vb.
Third, for v1 > w,
v v1 vwb v w vwb
vv vv
@P4 ðv1 ; w; vb Þ vb vb
¼
@v1 v vb v v vb v vb v v vb
v vb vvbb
vv
vb
0;
v vb v v vb
where the second inequality holds because of Claim 4 and w vb.
Finally,
@h 1 ðw; v1 Þ 1
¼ 2 3 > 0;
@v1 vb ln vvb
v w 4 5
v vb 1 þ v vb
8 "
vv # 9
Z <vð
v1
2n v nÞ n vn n n2 v 2 =
) b P ðv1 Þ ¼ dn: ð66Þ
v v vÞ3 : ln n
ð v v 2 ;
v
This gives us the equilibrium bidding function in the first round. Note that the second round
strategy is characterised in Lemma 4.
Queen’s University
Queen’s University and Zhejiang University, China