Chapter 3

Static Games of Incomplete

Information

This chapter begins our study of games of incomplete information,

also called Bayesian games. Recall that in a game of complete infor-
mation the players' payoff functions are common knowledge. In
a game of incomplete information, in contrast, at least one player
is uncertain about another player' s payoff function. One common
example of a static game of incomplete information is a sealed-bid
auction: each bidder knows his or her own valuation for the good
being sold but does not know any other bidder' s valuation; bids
are submitted in sealed envelopes, so the players' moves can be
thought of as simultaneous. Most economically interesting Bayes-
ian games, however, are dynamic. As we will see in Chapter 4,
the existence of prívate information leads naturally to attempts by
informed parties to communicate (or mislead) and to attempts by
uninformed parties to learn and respond. These are inherently
dynamic issues.
In Section 3.1 we define the normal-forro representation of a
static Bayesian game and a Bayesian Nash equilibrium in such
a game. Since these definitions are abstract and a bit complex,
we introduce the main ideas with a simple example--Cournot
competition under asymmetric information.
In Section 3.2 we consider three applications. First, we provide
a formal discussion of the interpretation of a mixed strategy given
in Chapter 1: player j's mixed strategy represents player i's uncer-
tainty about j's choice of apure strategy, and j's choice depends on
 145
144 STATIC GAMES OF INCOMPLETE INFORMATION Theory

the realization of a small amount of private information. Second, choose qi(cH) to solve
we analyze a sealed-bid auction in which the bidders' valua1·0s max [(a - qi - q2) - cH]q2.
are private information but the seller's valuation is known. i- · qz
nally, we consider the case in which a buyer and a seller e h
Similarly, if firm 2's cost is low, q2(ci) will solve
have private information about their valuations (as when a ·rm
knows a worker's marginal product and the worker knows his max [(a - qi - q2) - cL]q2.
or her outside opportunity). We analyze a trading game called a qz
~ouble auction: the seller names an asking price and the buyer Finally, firm 1 knows that firm 2' s cost is high with probability 0
s1multaneously names an offer price; trade occurs at the average and should anticipate that firm 2's quantity choice will be qi_(cH)
of the two prices if the latter exceeds the former. or q (cL), depending on firm 2's cost. Thus, firm 1 chooses q1 to
In Section 3.3 we state and prove the Revelation Principie, and 2
solve
briefly suggest how it can be applied in designing games when
the players have private information.

so as to maximize expected profit.

The first-order conditions for these three optimization prob-
3.1 Theory: Static Bayesian Games and Bayesian lems are
Nash Equilibrium a - qi - CH
2
3.1.A An Example: Cournot Competition under a - q1 - CL
Asymmetric Information 2

Consider a Cournot duopoly model with inverse demand given and

* 0[a - qi(cH) - c] + (1 - 0)[(a - qi(cL) - c]
by P(Q) = a - Q, where Q = q1 + q2 is the aggregate quantity q1 = 2
on the market. Firm l's cost function is C1 (q 1 ) = cq 1 . Firm 2's
cost function, however, is C2(q2) = cHq2 with probability 0 and Assume that these first-order conditions characterize the solutions
C2(q2) = cLq2 with probability 1 - 0, where cL < cH. Furthermore, to the earlier optimization problems. (Recall from Problem 1.6
information is asymmetric: firm 2 knows its cost function and that in a complete-information Cournot duopoly, if the firms' costs
firm l's, but firm 1 knows its cost function and only that firm 2's are sufficiently different then in equilibrium the high-cost firm
m~rginal cost is CH with probability 0 and CL with probability 1- 0. produces nothing. As an exercise, find a sufficient condition to
~F1rm 2 could be a new entrant to the industry, or could have just rule out the analogous problems here.) The solutions to the three
1~.vented a new technology.) All of this is common knowledge: first-order conditions are
firm 1 knows that firm 2 has superior information, firm 2 knows a - 2cH + c 1- 0( )
that firm 1 knows this, and so on. 3 + -6- CH - CL '

Naturally, firm 2 may want to choose a different (and pre-

sumably lower) quantity if its marginal cost is high than if it is
a - 2cL +c - 0 (CH - CL ) ,
low. Firm 1, for its part, should anticipate that firm 2 may tailor
3 6
its quantity to its cost in this way. Let q2(cH) and q2(cL) denote and
firm 2's quantity choices as a function of its cost, and let qi de- * a - 2c + 0cH + (1 - 0)ci
note firm 1' s single quantity choice. If firm 2' s cost is high, it will
ql = 3
I
 147
146 STATIC GAMES OF INCOMPLETE INFORMATION eory

Compare q2(cH), q2(ci), and qi to the Cournot equilibrium un,}

der complete information with costs c1 and c2. Assuming that the.
·. .,
¡ is called player zEs ti(Ptye
(or type space) T¡. ac P 1
ª: d belongs to a set of possible types
t· corresponds to a different payoff

values of c1 and c2 are such that both firms' equilibrium quanti- : function that player i miglht hapv~~e player i has two possible pay-
ties are both positive, firm i produces q¡ = (a - 2c¡ + Cj)/3 in thi - As an abstract examp e sup . t· nd
tiO ns We would say that player i has two types, ~,1 a
complete-information case. In the incomplete-information case, i fu
"off ne · . T _ {t· t· } and that player z s two
contrast q2(cH) is greater than (a - 2cH + c)/3 and qi(ci) is 1 ss , t· that player i' s type space 1s i - z1 12 , . . ) W
i2, . ( a·t·)andu·(a1,•·•,ªn,t12• ecan
than (a - 2ci + c)/3. This occurs because firm 2 not only taifors payoff functions are U¡ a1' ... ' n, i 1 i ds to a different
its quantity to its cost but also responds to the fact that firm 1 use the idea that each of a player's types correspont the possibility
. h 1 r might have to represen
cannot do so. If firm 2' s cost is high, for example, it produces less
because its cost is high but also produces more because it knows that the p ayer mig
J
payoff fu~ct1on t . e ~~:e different sets of feasible actions, as. fol-
le that player i's set of feasible act10ns
that firm 1 will produce a quantity that maximizes its expected lows. Suppose, for examp ' .h b b'lit
1 1-q Then
nd
profit and thus is smaller than firm 1 would produce if it knew is {a, b} with pro?ability q a {a(,t~' c}n:t!. r:~;e th! probability
y that 1 has two types 11 12, { b }
firm 2's cost to be high. (A potentially misleading feature of this
example is that qi exactly equals the expectation of the Cournot
quantities firm 1 would produce in the two corresponding games
we cansa d
of til is q) an wb etcdan
for both types u e
fin:
d fine i's feasible set of actions to be a, 'c
the payoff from taking action c to be -oo

of complete information. This is typically not true; consider the for type til· te example consider the Cournot game in
case in which firm i's total cost is c¡q'f, for example.) As a more concre ' ·ty h ·ces
. ti The firms' actions are their quantl e 01 ,
the prdev1ousF~ec ;~as two possible cost functions and thus two
q1 an q2, irm . .
3.1.B Normal-Form Representation of Static Bayesian possible profit or payoff functions.
Games 71'2(q1,q2;ci) = [(a - q1 - q2) - ci]q2
Recall that the normal-form representation of an n-player game
of complete information is G = {S1 ... Sn; u1 ... Un}, where S¡ is and
player i' s strategy space and u¡ (s1, ... , Sn) is player i' s payoff when
the players choose the strategies (s 1, ... , sn), As discussed in Sec- Firm 1 has only one possible payoff function:
tion 2.3.B, however, in a simultaneous-move game of complete
inforrnation a strategy for a player is simply an action, so we can 71'
1(q 1 ,q 2;c) = [(a - q1 - q2) - c]q1,
write G = {A1 .. ,An;u1 ... un}, where A¡ is player i's action space . T { } and that firm l's
and U¡(a 1, ... ,an) is player i's payoff when the players choose the We say that firm 2's type space is 2 = ci,cH
actions (a1, ... ,an)- To prepare for our description of the timing of type space is T1 = {e}. · th t 1 r i
. this definition of a player's type, saymg a f aye
a static game of incomplete information, we describe the timing of G1ve1: a off function is equivalent to saymg that
a static game of complete information as follows: (1) the players si-
multaneously choose actions (player i chooses a¡ from the feasible
knows ~1s or her_ ow; ife/
e. Likewise, saying that playe~ i may
player z ~ow~ hi~ ~he oth?r,Pplayers' payoff functions is eqmvalent
set A¡), and then (2) payoffs u¡(a1, ... , an) are received. be uncertam a or . be uncertain about the types of the
We now want to develop the normal-form representation of a to saying that P ayer 1 may . t· . . . t ). We use
simultaneous-move game of incomplete information, also called a
static Bayesian game. The first step is to represent the idea that
other p~aye:s, t~:n;;::f bJi tp~s=b~:
T -i to eno e
~f ';;i~~; 1t~\:
and nwe use the
I ) t d ote player i's belief about
each player knows his or her own payoff function but may be probability distribution p¡(L¡ t~ o eln ., kn wledge of his
1 , types t . g1ven p ayer l s o
uncertain about the other players' payoff functions. Let player i's the other P ayers ' -u pplication analyzed \n Section 3.2
possible payoff functions be represented by u¡(a1, ... , an; t¡), where or her own type, t¡. 1n every a
 149
148 STATIC GAMES OF INCOMPLETE INFORMATION Theory

(and in most of the literature), the players' types are independent, ian games. First there are games in which player i has private
in which case p¡(L¡ 1 t¡) does not depend on t¡, so we can write information not only about his or her own payoff function but
player i's belief as p¡(L¡). There are contexts in which the players' also about another player's payoff function. In Problem 3.2, for
types are correlated, however, so we allow for this in our definition example, the asymmetric-information Cournot model from Sec-
of a static Bayesian game by writing player i's belief as p¡(L¡ 1 t¡). 1 tion 3.1.A is changed so that costs are symmetric and common
Joining the new concepts of types and beliefs with the familiar knowledge but one firm knows the level of demand and the other
elements of the normal-forro representation of a static game of does not. Since the level of demand affects both players' payoff
complete information yields the normal-forro representation of a functions, the informed firm's type enters the uninformed firm's
static Bayesian game. payoff function. In the n-player case we capture this possib~lity
by allowing player i's payoff to depend not only on_ the ~chons
Definition The nonnal-fonn representation of an n-player static (a 1, ... , an) but also on all the types (t1, ... , t11)- We wnte th1s pay-
Bayesian game specifies the players' action spaces A 1 , ... , An, their type off as u¡(a1, ... ,an;t1, ... , tn)-
spaces T1, ... , T n, their beliefs p1, ... , Pn, and their payofffunctions u1, ... , The second technical point involves the beliefs, p¡(L¡ 1 t¡).
Un. Player i's type, t¡, is privately known by player i, determines player We will assume that it is common knowledge that in step (1) of
i's payoff function, u¡(a1, ... , an; t¡), and is a member of the set of possible the timing of a static Bayesian game, nature draws a type vector
types, T¡. Player i's belief p¡(L¡ 1 t¡) describes i's uncertainty about the t = (t 1 , ... , tn) according to the prior probability distribution p(t).
n - 1 other players' possible types, L¡, given i's own type, t¡. We denote When nature then reveals t¡ to player i, he or she can compute the
this game by G = {A1, ... ,An; T1, ... , Tn;p1, ... ,p11 ; U1, ... , Un}- belief p¡(L¡ 1 t¡) using Bayes' rule:
2

Following Harsanyi (1967), we will assume that the timing of a p(L¡, t¡)
p(L;, t¡)
static Bayesian game is as follows: (1) nature draws a type vector
t = (ti, ... , tn), where t¡ is drawn from the set of possible types
p¡(L¡ 1 t¡) = p(t¡) L p(L¡,t¡)
L;ET_¡
T¡; (2) nature reveals t¡ to player i but not to any other player;
(3) the players simultaneously choose actions, player i choosing a¡
from the feasible set A¡; and then (4) payoffs u¡(a 1, ... , an; t¡) are Furthermore, the other players can compute the various beliefs
received. By introducing the fictional moves by nature in steps (1) that player i might hold, depending on i's type, namely p¡(L¡ 1 t¡)
and (2), we have described a game of incomplete information as a for each t; in T¡. As already noted, we will frequently assume that
game of imperfect information, where by imperfect information we the players' types are independent, in which case p¡(L¡) does not
mean (as in Chapter 2) that at sorne move in the game the player depend on t¡ but is still derived from the prior distribution p(t).
with the move <loes not know the complete history of the game In this case the other players know i's belief about their types.
thus far. Here, because nature reveals player i's type to player i
but not to player j in step (2), player j does not know the complete 3.1.C Definition of Bayesian Nash Equilibrium
history of the game when actions are chosen in step (3).
Two slightly more technical points need to be covered to com- We now want to define an equilibrium concept for static Bayesian
plete our discussion of normal-forro representations of static Bayes- games. To do so, we must first define the players' strategy spaces

1Imagine that two firms are racing to develop a new technology. Each firm's 2 Bayes' rule provides a formula for P(A I B), the (conditional) probability that

chance of success depends in part on how difficult the technology is to develop, an event A will occur given that an event B has already occurred. Let P(A), P(B),
which is not known. Each firm knows only whether it has succeeded and not and P(A, B) be the (prior) probabilities (i.e., the probabilities before either A or B
whether the other has. If firm 1 has succeeded, however, then it is more likely has had a chance to take place) that A will occur, that B will occur, and that both
that the technology is easy to develop and so also more Jikely that firm 2 has suc- A and B will occur, respectively. Bayes' rule states that P(A I B) = P(A, B)/P(B).
ceeded. Thus, firm l's belief about firm 2's type depends on firm l's knowledge That is, the conditional probability of A given B equals the probability that both
of its own type. A and B will occur, divided by the prior probability that B will occur.
 150 STATIC GAMES OF INCOMPLETE INFORMATION Theory 151

in such a_game. Recall from Sections 2.3.B and 2.4.B that a player's of the definition of a strategy just given, the pair (qi( CH), q2(cL))
~trategy 1s a ~omplete. plan ?f action, specifying a feasible .• action) . is firm 2's strategy and qj is firm l's strategy. It is easy to imag-
m eve:)' conbn~e~cy m which the player might be called on to · ine that firm 2 will choose different quantities depending on its
act.. G1ven the timmg of a static Bayesian game, in which nature cost. It is equally important to note, however, that firm l's sin-
begms the ~ame by dr~wing the players' types, a (pure) strategy gle quantity choice should take into account that firm 2' s quantity
for J:>layer l must spec1fy a feasible action for each of player 7s will depend on firm 2' s cost in this way. Thus, if our equilibrium
poss1ble types. concept is to require that firm l's strategy be a best response to
firm 2's strategy, then firm 2's strategy must be a pair of quantities,
1
Definition In the static Bayesian game G = {A A •T T . one for each possible cost type, else firm 1 simply cannot compute
f 1, · · · , n, 1, • • • , n,
P1, · · · , Pn; U1, • • • , Un}, a strategy for player i is a function s •(t ·) where whether its strategy is indeed a best response to firm 2' s.
for each type t¡ in T¡, s¡(t¡) specifies the action from the feasibl; s;t A· 1
that More generally, we would not be able to apply the notion of
type t¡ would choose if drawn by nature. Nash equilibrium to Bayesian games if we allowed a player's
strategy not to specify what the player would do if sorne types
ynlike (b~th static and dynamic) games of complete information, were drawn by nature. This argument is analogous to one from
m a Bayes1an ga~e the strategy spaces are not given in the normal- Chapter 2: it may have seemed unnecessary to require player i's
form representahon of the game. Instead, in a static Bayesian strategy in a dynamic game of complete information to specify a
game ~he strate~f spaces are c?nstructed from the type and action feasible action for each contingency in which player i might be
spaces. pl~yer z s set of poss1ble (pure) strategies, S¡, is the set called on to move, but we could not have applied the notion of
of ~11 poss1ble functions with domain T¡ and range A¡. In a sepa- Nash equilibrium to dynamic games of complete information if
ratz_ng strategy, for example, each type t¡ in T¡ chooses a different we had allowed a player's strategy to leave the player's actions
acbon a¡ fro°: A¡. In_a p~ol~ng ~trategy, in contrast, all types choose in sorne contingencies unspecified.
the sa~e ac~on. ~h1s d1stmc~1on be~een separating and pooling Given the definition of a strategy in a Bayesian game, we turn
str~teg1es w11l be 1mportant m our d1scussion of dynamic games next to the definition of a Bayesian Nash equilibrium. In spite
~f mcomplete information in Chapter 4. We introduce the distinc- of the notational complexity of the definition, the central idea is
hon here only to help describe the wide variety of strategies that both simple and familiar: each player's strategy must be a best
can be constructed from a given pair of type and action spaces y. response to the other players' strategies. That is, a Bayesian Nash
and A¡. ' 1
equilibrium is simply a Nash equilibrium in a Bayesian game.
. It may_ seem ~nnecessary to require player i' s strategy to spec-
ify a feas1ble achon for each of player i's possible types. After Definition In the static Bayesian game G = {A 1, ... , An; T1 , ... , T n;
all, onc~ nature has drawn a particular type and revealed it to a p 1, ... , pn; u 1, ... , Un}, the strategies s* = (si, ... , s~) are a ( pure-strategy)
player, 1t may seem that the player need not be concemed with Bayesian Nash equilibrium if for each player i and for each of i's types
the actions he or she would have taken had nature drawn sorne t¡ in T¡, s1(t;) solves
other type. On the other hand, player i needs to consider what
the oth:r players w_ill ?º'
and what they will do depends on what
they thmk player z w11l do, for each t¡ in T¡. Thus, in deciding
w~at to do once one type has been drawn, player i will have to
th1nk about what he or she would have done if each of the other That is, no player wants to change his or her strategy, even if the change
types in T¡ had been drawn.
involves only one action by one type.
. Consider the asymmetric-information Cournot game in Sec-
bon ~- l.A, for example. We argued that the solution to the game It is straightforward to show that in a finite static Bayesian game
cons1sts of three quantity choices: q2(cH), qi( cL), and qj. In terms (i.e., a game in which n is finite and (A1, ... , An) and (T1, ... , Tn)
 152
STATIC GAMES OF INCOMPLETE INFORMATION '
Applications 153
are all finite sets) there exists a Bayesian Nash equilibrium per-
haps ~ mixed strat~gies. The proof closely parallels the pr~of7 Now suppose that, although they have known each other f~r
the ex1stence_ of a mix_ed-strategy Nash equilibrium in finite gam · quite sorne time, Chris and Pat are not qui~e, sure of e~ch other s
of complete mformation, and so is omitted here. payoffs. In particular, suppose that: Chns s payoff 1f both ~t-
tend the Opera is 2 + te, where t~ is i:rivately known by _Chri~;
Pat's payoff if both attend the F1ght 1s 2 + tp, where tp 1s pn-
vately known by Pat; and te and tp are in~ependent _draws ~ro~
a uniform distribution on (O,x]. (The ch01ce of a u~for~ d1str1-
3.2 Applications bution on [O, x] is not important, but we do have m mm~ that
the values of te and tp only slightly perturb the payoffs m the
original game, so think of x as small.) _All the ?ther pay~ffs are
3.2.A Mixed Strategies Revisited the same. In terms of the abstract stabc Bayes1an game m nor-
mal form G = {Ae,Ap;Te, Tp;pe,pp;ue,Up}, the action spaces are
As we ~ent~oned in Section 1.3.A, Harsanyi (1973) suggested that Ae = Ap = {Opera, Fight}, the type spaces are Te= Tp = [O,x], the
r,layer _J s mixed strategy represents player i's uncertainty about beliefs are Pe(tp) = pp(te) = 1/x for all te and tp, and the payoffs
J s choice o_f a _pure strategy, and that j's choice in tum depends are as follows.
on the reahzahon of a small amount of prívate information. We
can now give a more precise statement of this idea: a mixed- Pat
strategy Nash equilibrium in a game of complete information can
(aln:i~st _alw~ys) be interpreted as a pure-strategy Bayesian Nash Opera Fight
eqmhbnum m a closely related game with a little bit of incom- Opera 2 + te, 1 O, O
plete information. (We will ignore the rare cases in which such an Chris
interpretation ~s not possible.) Put more evocatively, the crucial Fight O, O 1, 2 + tp
feature of a m1xed-strategy Nash equilibrium is not that player j
chooses a strategy randomly, but rather that player i is uncertain The Battle of the Sexes with Incomplete Information
about player j's choice; this uncertainty can arise either because of
~andomi~ation ?r (more plausibly) because of a little incomplete
mformahon, as m the following example. We will construct a pure-strategy Bayesian Nash equilibriu~
Recall !~at_in the Battle of the Sexes there are two pure-strategy of this incomplete-information version of ~1:e Battle of the Sexes m
Nash eqmlibna (Opera, Opera) and (Fight, Fight) and a mixed- which Chris plays Opera if te exceeds a cnbcal value, ~, _and plays
str_a~egy Nash equilibrium in which Chris plays Opera with prob- Fight otherwise and Pat plays Fight if tp exc~~ds_ a cnbca~ value,
ab1hty 2/3 and Pat plays Fight with probability 2/3. p, and plays Opera otherwise. In such an eqmlibr~um, C~ns plays
Opera with probability (x - c)/x and Pat J?lays F1ght ~1th pro~a-
bility (x - p)/x. We will show that as the mcomplete 1nJ'.or~at10_n
Pat disappears (i.e., as x approaches z~~o)~ the players' behav1?r m thi~
Opera Fight pure-strategy Bayesian Nash eqmhbnum approaches ~h~rr behav
ior in the mixed-strategy Nash equilibrium in the ongmal game
Opera 2, 1 O, O of complete information. That is, both (x - c)/x and (x - p)/x
Chris f-----+-----1
Fight O, O 1, 2 approach 2/3 as x approaches zero. . . . , '" , .
Suppose Chris and Pat play the strateg1es 1ust descnbed. E~ftc'/:,
The Battle of the Sexes a given value of x, we will d~termine val1:1~s ?f e a~d p SJ.J_: ·
that these strategies are a Bayes1an Nash eqmhbnum. G1ven Pa .
 154 STATIC GAMES OF INCOMPLETE INFORMATION Applications 155

strategy, Chris's expected payoffs from playing Opera and from 3.2.B An Auction
playing Fight are
Consider the following first-price, sealed-bid auction. There are
two bidders, labeled i = 1, 2. Bidder i has a valuation V¡ for the
good-that is, if bidder i gets the good and pays the price p, then
i's payoff is v¡ -p. The two bidders' valuations are independently
and and uniformly distributed on [O, 1]. Bids are constrained to be
nonnegative. The bidders simultaneously submit their bids. The
higher bidder wins the good and pays the price ~he bid; t~e oth:r
bidder gets and pays nothing. In case of a tie, the wmner 1s
respectively. Thus playing Opera is optimal if and only if determined by a flip of a coin. The bidders are risk-neutral. All
of this is common knowledge.
X
te :2: - - 3 = C. (3.2.1) In order to formulate this problem as a static Bayesian game,
p we must identify the action spaces, the type spaces, the beliefs, and
the payoff functions. Player i's action is to submit a (nonnegative)
Similarly, given Chris' s strategy, Pat' s expected payoffs from play-
ing Fight and from playing Opera are bid, b¡, and her type is her valuation, V¡. (In terms of the abstract
game G = {A 1,A2;T1,T2;p1,p2;u1,ui}, the action space is "'.-i
[O, oo) and the type space is T¡ = [O, l].) Because the valuabons
[1- -e] ·0+-(2+tp)
X
e
X
e
= -(2+tp) X
are independent, player i believes that Vj is uniformly distributed
on [O, 1], no matter what the value of V¡. Finally, player i's payoff
and function is

[1--xe] ·1+--0=1--
e
X
e
X'
if b¡ > bj,

respectively. Thus, playing Fight is optimal if and only if if b¡ = bj,

if b¡ < bj.
X
tp :2: - - 3 = p. (3.2.2)
e To derive a Bayesian Nash equilibrium of this game, we be-
gin by constructing the players' strategy spaces. Recall that in a
Solving (3.2.1) and (3.2.2) simultaneously yields p = e and p2 + static Bayesian game, a strategy is a function from types to actions.
3p - x = O. Solving the quadratic then shows that the probability Thus, a strategy for player i is a function b¡(v¡) specifying the bid
that Chris plays Opera, namely (x - c)/x, and the probability that that each of i's types (i.e., valuations) would choose. In a Bayesian
Pat plays Fight, namely (x - p)/x, both equal Nash equilibrium, player l's strategy b1(v1) is a best response to
player 2's strategy b2 (v 2 ), and vice versa. Formally, the pair of
1 - -3 + v'9+4x strategies (b(v 1 ),b(v2 )) is a Bayesian Nash equilibrium if for each
2x ' v¡ in [O, 1], b¡(v¡) solves
which approaches 2/3 as x approaches zero. Thus, as the incom- 1
plete information disappears, the players' behavior in this pure- max (v¡ - b¡)Prob{b¡ > bj(vj)} + - (v¡ - b¡)Prob{b¡ = bj(vj)}.
b¡ 2
strategy Bayesian Nash equilibrium of the incomplete-information
game approaches their behavior in the mixed-strategy Nash equi- We simplify the exposition by looking for a linear equilibrium:
librium in the original game of complete information. b1(v1) = a1 + C1V1 and b2(v2) = a2 + c2v2. Note well that we are
 157
156 STATIC GAMES OF INCOMPLETE INFORMATION Applications

not res~ricting the players' strategy spaces to include only linear ·• a¡ s O, a¡= a¡/2, and Cj = 1/2. Combining these _two sets of results
s~ateg1es. Rather, we allow the players to choose arbitrary strate- ' then yields a¡ = aj = O and e¡ = e¡= 1/2. That 1s, b¡(v¡) = v¡/2, as
g1es but ask whether there is an equilibrium that is linear. It tufs claimed earlier.
ou~ that bec~1:1se_ the players' valuations are uniformly distribut.ed, One might wonder whether there are other Bayesian Nash
a lmear equihbnum not only exists but is unique (in a sense to b equilibria of this game, and also how equilibrium bidding changes
made precise). We will find that b¡(v¡) = v¡/2. That is, each pla er "é as the distribution of the bidders' valuations changes. Neither of
submits a bid equal to half her valuation. Such a bid reflects the these questions can be answered using the technique just applied
fundamental trade-off a bidder faces in an auction: the higher the (of positing linear strategies and then deriving the coefficients that
bid, the more likely the bidder is to win; the lower the bid, the make the strategies an equilibrium): it is fruitless to try to guess all
larger the gain if the bidder <loes win. the functional forms other equilibria of this game might have, and
Suppose that player j adopts the strategy bj(vj) = a1- + c·v·. a linear equilibrium does not exist for any other distribution of val-
1 1 For ·
.
a g1ven va 1ue o f v¡, p 1ayer z., s b est response solves uations. In the Appendix, we derive a symmetric Bayesian Nash
equilibrium,3 again for the case of uniformly distributed valua-
11),ªx (v¡ - b¡)Prob{b¡ >aj+ CjVj}, tions. Under the assumption that the players' strategies are strictly
1
increasing and differentiable, we show that the unique symmetric
where we have used the fact that Prob{b¡ = bj(vj)} = O (because Bayesian Nash equilibrium is the linear equilibrium already de-
bj_( Vj) = aj + CjVj and Vj is uniformly distributed, so bj is uniformly rived. The technique we use can easily be extended to a broad
4
d1~tributed). Since it is pointless for player i to bid below player j's class of valuation distributions, as well as the case of n bidders.
mmimum bid and stupid for i to bid above j' s maximum bid, we
s s
have aj b¡ aj + Cj, so Appendix 3.2.B
Prob{b¡ > aj + CjVj} = b·_
Prob Vj < _z
{
- _J a·} = b· - a·J
_i_ _ •
Suppose player j adopts the strategy b(·), and assume that b(·) is
strictly increasing and differentiable. Then for a given value of V¡,
Cj Cj
player i's optimal bid salves
Player i' s best response is therefore
max (v¡ - b¡)Prob{b¡ > b(vj)}·
b¡
if V¡ 2:: aj,
if V¡ < aj. Let b- 1 (bj) denote the valuation that bidder j must have in arder to
bid bj. That is, b- 1 (bj) = Vj if b¡ = b(v¡). Since v¡ is uniformly dis-
If O < aj < l then there are sorne values of V¡ such that v¡ < a·, in 1
tributed on [O, l], Prob{b¡ > b(v¡)} = Prob{b- 1 (b¡) > v¡} = b- (b¡).
which case b¡(v¡) is not linear; rather, it is flat at first and posititely The first-order condition for player i's optimization problem is
sloped later. Since we are looking for a linear equilibrium, we
therefore
therefore rule out O < a¡ < 1, focusing instead on a¡ 2:: 1 and aj s O.
But the former cannot occur in equilibrium: since it is optimal for
a higher type to bid at least as much as a lower type's optimal
3 A Bayesian Nash equilibriurn is called symmetric if the players' strategies
bid~ we have Cj 2:: O, ?ut then a¡ 2:: 1 would imply that bj(v¡) 2:: vj, are identical. That is, in a symmetric Bayesian Nash equilibrium, there is a single
wh1ch cannot be ophmal. Thus, if b¡(v¡) is to be linear, then we function b(v¡) such that player l's strategy b1 (v1 ) is b(v1) and player 2's strategy
must have aj s O, in which case b¡(v¡) = (v¡ + aj)/2, so a¡ = a¡/2 b2 (v2 ) is b(v 2 ), and this single strategy is a best response to itself. Of course,
ande¡= 1/2. since the players' valuations typically will be different, their bids typically will
We can repeat the same analysis for player j under the assump- be different, even if both use the same strategy.
4 Skipping this appendix will not hamper one' s understanding of what follows.
tion that player i adopts the strategy b¡(v¡) =a¡+ c¡V¡. This yields
 -;
158 STATIC GAMES OF INCOMPLETE INFORMATIO .·
159

This first-order condition is an implicit equation for bidder i' s best,

product and the worker knows his or her outside opportunity. _See
resp~nse to t~e s~rategy b( ·) played by bidder j, given that bid-\
der z ~ valuahon 1s V¡. If the strategy b(-) is to be a symmetric"' Problem 3.8.) We analyze a trading game called a double auchon.
~ayes1an Nash equilibrium, we require that the solution to th Toe seller names an asking price, Ps, and the buyer simultaneou~ly
first-order_ condi!ion b~ b(v¡): that is, for each of bidder i's pos - names an offer price, Pb· If Pb ~ p5 , then trade occurs at pnce
ble va~uahons, b1~der z ~oes not wish to deviate from the strat gy p = (pb + Ps)/2; if Pb < Ps, then no trad: occurs._ , .
b(), g1ven that b1dder J plays this strategy. To impose this re- •· The buyer's valuation for the seller s good 1s vb, the seller s 1s
qmrement, we substitute b¡ = b(v¡) into the first-order condition v5 • These valuations are prívate information and are drawn from
yielding: ' independent uniform distributio1!s on_ ~O, l_l. If the b~yer get~ the
good for price p, then the buyer s uhhty 1s vb - p; 1f there 1s no
1
-b- (b(v¡)) + (v¡- b(v¡)) a!.b- 1 (b(v¡)) = O. trade, then the buyer' s utility is zero. If the selle~ sells the good
l for price p, then the seller's utility is p-vs; ~-there 1s ~o trade, then
the seller's utility is zero. (Each of these uhhty funchons measur~s
Of ?ourse, b- _(b(v¡)) ~~ simply v¡. Furthermore, d{b- 1 (b(v¡))}/db¡ =
1
the change in the party' s utility. If there is no trade, then there 1s
1/b (v¡). That 1s, d{b (b¡)}/db¡ measures how much bidder i's val- no change in utility. lt would make no differ~nce to defii:ie, say,
~ation must change to produce a unit change in the bid, whereas the seller' s utility to be p if there is trade at pnce p and Vs 1f there
b (V¡) m~asures how ~uch the bid changes in response to a unit is no trade.)
c~ange ~n the valuahon. Thus, b( ·) must satisfy the first-order In this static Bayesian game, a strategy for the buyer is a func-
d1fferenhal equation
tion Pb(vb) specifying the price the buyer will offer for each of _the
1 buyer' s possible valuations. Lik_ewise, a strate~y for the seller 1s a
-v¡ + (v¡ - b(v¡)) b'(v¡) = O, function Ps (v 5 ) specifying the pnce the seller w1ll demand for e~ch
of the seller's valuations. A pair of strategies {pb(vb),ps(vs)} is a
which is ~ore conveniently expressed as b'(v¡)v¡ + b(v¡) = V¡. The Bayesian Nash equilibrium if the following two conditions hold.
left-han~ s1de of t~s differential equation is precisely d{b(v¡)v¡} /dv¡. For each vb in [O, 1], Pb(vb) salves
Integrahng both s1des of the equation therefore yields
max [vb - Pb + E[ps(Vs) 1 Pb ~ Ps(Vs)]] Prob {Pb ~ Ps(Vs)}' (3.2.3)
b(v¡)v¡ = 1 v'f + k, Pb 2
2
where k is a constant of integration. To eliminate k, we need where E[p5 (v 5 ) 1 Pb ~ p5 (v5 )] is the expected price the seller will
a bo~ndary condition. Fortunately, simple economic reasoning demand, conditional on the demand being less than the buyer s 1

prov1des one: no player should bid more than his or her valuation. offer of Pb· For each V5 in [O, 1], Ps(Vs) salves
Thus, we r_equin~ b(v¡) :S V¡ for every V¡. In particular, we require
b(O) ~ O. Smce b1ds are constrained to be nonnegative, this implies
that b(O) = O, so k = O and b(v¡) = v¡/2, as claimed.
where E[pb(vb) 1 Pb(vb) ~ Ps] is the expected price the buyer will of-
3.2.C A Double Auction
fer, conditional on the offer being greater than the seller's demand
We next consider the case in which a buyer and a seller each have of Ps· .. . .
prívate information about their valuations, as in Chatterjee and There are many, many Bayesian Nash eqmhbna of this gam_e.
Samuelson (19~3). (In Hall and Lazear [1984], the buyer is a firm Consider the following one-price equilibrium, for example, m
and the seller 1s a worker. The firm knows the worker's marginal which trade occurs at a single price if it occurs at all. For any
value x in [O, 1], let the buyer's strategy be to offer x if vb ~ x and
160 STATIC GAMES OF INCOMPLETE INFORMATION Applications 161

linear equilibrium has interesting efficiency properties, which we

describe later.
Suppose the seller's strategy is Ps(Vs) = as + CsVs- Toen Ps is
1 uniformly distributed on [as, as+ es], so (3.2.3) becomes

~!X [Vb - ½{rb + as ; Pb } ] Pb : as

TRADE
X
the first-order condition for which yields
2 1 (3.2.5)
Pb = 3 vb + 3as.
Thus, if the seller plays a linear strategy, then the buyer' s best
response is also linear. Analogously, suppose the buyer's strategy
is Pb(vb) = ab+cbvb. Then Pb is uniformly distributed on [ab,ab+cb],
so (3.2.4) becomes

X 1
the first-order condition for which yields

Figure 3.2.1. 2 1
3vs + 3 (ab + cb)-
Ps = (3.2.6)

Thus, if the buyer plays a linear strategy, then the seller' s best
to offer zero otherwise, and let the seller' s strategy be to demand
response is also linear. If the players' linear strategies are to be
x if Vs :S x and to demand one otherwise. Given the buyer' s strat-
best responses to each other, (3.2.5) implies that cb = 2/3 and
egy, the seller's choices amount to trading at x or not trading, so
ab = a5 /3, and (3.2.6) implies that Cs = 2/3 and as = (ab + cb) /3.
the seller' s strategy is a best response to the buyer' s because the
Therefore, the linear equilibrium strategies are
seller-types who prefer trading at x to not trading do so, and vice
versa. The analogous argument shows that the buyer's strategy
is a best response to the seller' s, so these strategies are indeed a (3.2.7)
Bayesian Nash equilibrium. In this equilibrium, trade occurs for
the (vs, vb) pairs indicated in Figure 3.2.1; trade would be efficient and
for all (vs, vb) pairs such that vb 2 Vs, but <loes not occur in the (3.2.8)
two shaded regions of the figure.
We now derive a linear Bayesian Nash equilibrium of the dou- as shown in Figure 3.2.2.
ble auction. As in the previous section, we are not restricting the Recall that trade occurs in the double auction if and only if
players' strategy spaces to include only linear strategies. Rather, Pb 2 Ps• Manipulating (3.2.7) and (3.2.8) shows that trade occurs
we allow the players to choose arbitrary strategies but ask whether in the linear equilibrium if and only if vb 2 Vs + (1/4), as shown in
there is an equilibrium that is linear. Many other equilibria exist Figure 3.2.3. (Consistent with this, Figure 3.2.2 reveals that seller-
besides the one-price equilibria and the linear equilibrium, but the types above 3/4 make demands above the buyer's highest offer,
 163
162 STATIC GAMES OF INCOMPLETE INFORMATION •· Applications

1 1

TRADE
3/4

1/4

1
1/4 3/4 1

Figure 3.2.3.
Figure 3.2.2.

ian Nash equilibria of the double auction (incl~ding but far ~rom
Pb(1) = 3/4, and buyer-types below 1/4 make offers below the limited to the one-price equilibria). This imphes that there IS no
seller's lowest offer, p5 (0) = 1/4.) Bayesian Nash equilibrium of the double auction_ in which trade
_comp~re Figures 3.2.1 and 3.2.3-the depictions of which val- occurs if and only if it is efficient (i.e., if and only If Vb ~ Vs). They
~ation pairs trade in the one-price and linear equilibria, respec- also show that this latter result is very general: if Vb is continuously
hv~. In both cases, the most valuable possible trade (namely, distributed on [xb,Yb] and v 5 is continuously distributed on [xs,Ys],
Vs. - O and vb = 1) does occur. But the one-price equilibrium
where Ys > xb and Yb > Xs, then th.ere is no bargaining ?ame the
m~sses sorne valuable trades (such as v 5 = O and vb = x - E, where buyer and seller would willingly play that ha~ ª. ~ayesia~ Nash
: 1s small) and achieves sorne trades that are worth next to noth- equilibrium in which trade occurs if and only if 1t ~s :fficient. In
mg (such a~ Vs = x - e and vb = x + e). The linear equilibrium, in the next section we sketch how the Revelation Prmciple can be
contrast, misses all trades worth next to nothing but achieves all used to prove this general result. We conclude this section by
trades wo~th at least 1/4. ~his suggests that the linear equilibrium translating the result into Hall and Lazear's employment model:
m~y dommate the orn~-pnce equilibria, in terms of the expected if the firm has private information about the worker' s marginal
gams the players receive, but also raises the possibility that the product (m) and the worker has private ~ormatio~ ~bout his
players might do even better in an alternative equilibrium. or her outside opportunity (v), then there IS no bargammg game
Myerson and Satterthwaite (1983) show that, for the uniform that the firm and the worker would willingly play that produces
v~luatio~ distributions considered here, the linear equilibrium employment if and only if it is efficient (i.e., m ~ v).
y1elds higher expected gains for the players than any other B,ayes-
 165
164 STATIC GAMES OF INCOMPLETE INFORMATION The Revelation Principle

3.3 The Revelation Principie The second way the seller can use the Revela~ion P:inc~pl~ is
to restrict attention to those direct mechanisms m which It 1s a
The Revelation Principie, due to Myerson (1979) in the context Bayesian Nash equilibrium for each bidder to tell the truth-that
of Bayesian games (and to others in related contexts), is an im- is, payment and probability functions {x1 (T1, ... , Tn), • .. , x~,( T1, • • •.,
portant tool for designing games when the players have privat Tn); q1(Ti, ... , Tn), ... , qn (T1 , •.• , Tn)} such that each player z s e_qm-
information. It can be applied in the auction and bilateral-tradi g librium strategy is to claim T¡(t¡) = t¡ for e~ch t¡ m T¡. 1:-. d1_rect
problems described in the previous two sections, as well as ih a mechanism in which truth-telli:ng is a Bayesian Nash eqmhbnum
wide variety of other problems. In this section we state and prove is called incentive-compatible.
the Revelation Principie for static Bayesian games. (Extending the Outside the context of auction design, the Revelation Principie
proof to cover dynamic Bayesian games is straightforward.) Be- can again be used in these two ways. Any Bayesian Nash equi~ib-
~ore doi:~1g so, how~ver, we sketch the way the Revelation Principie rium of any Bayesian game can be represented by a new_ Bayesian
1s used m the aucbon and bilateral-trading problems. Nash equilibrium in an appropriately chosen new Bay~s1an gam~,
Consider a seller who wishes to design an auction to maxi- where by "represented" we mean that for each poss1ble comb1-
mize his or her expected revenue. Specifying- the many different nation of the players' types (t1, ... , tn), t~e players' a~tions and
auctions the seller should consider could be an enormous task. In payoffs ih the new equilibrium are identical to those m the ':'ld
the auction in Section 3.2.B, for example, the highest bidder paid equilibrium. No matter what the ':'riginal game, the new Bayes~a~
money to the seller and received the good, but there are many game is always a direct mechamsm; no matter what ~he ong1-
other possibilities. The bidders might have to pay an entry fee. nal equilibrium, the new equilibrium in the new game 1s always
More generally, sorne of the losing bidders might have to pay truth-telling. More formally:
money, perhaps in amounts that depend on their own and others'
bids. Also, the seller might set a reservation price--a floor below Theorem (The Revelation Principie) Any Bayesian Nash equilib-
which ~ids will not be accepted. More generally, the good might rium of any Bayesian game can be represented by an incentive-compatible
stay with the seller with sorne probability, and might not always direct mechanism.
go to the highest bidder when the seller does release it.
. For~nately, the seller can use the Revelation Principie to sim- In the auction analyzed in Section 3.2.B we assumed that the
phfy th1s problem dramatically, in two ways. Fi:rst, the seller can bidders' valuations are independent of each other. We also as-
restrict attention to the following class of games: sumed (i:mplicitly, in the definition of the bidders' va~uation_~)
1. The bidders simultaneously make (possibly dishonest) claims that knowing bidder j' s valuation would not change b1dder 1., s
about their types (i.e., their valuations). Bidder i can claim valuation (although such knowledge typically would change z s
to be any type T¡ from i' s set of feasible types T¡, no matter bid). We characteri:ze these two assumptions by saying that the
what i's true type, t¡. bidders have i:ndependent, private values. For this case, My-
erson (1981) determines which direct mechanisms have a truth-
2. Given the bidders' claims (T1, ... , Tn), bidder i pays x¡( Ti, ... , telling equilibrium, and which of thes~ equi~ib~ia maximizes the
Tn) and receives the good with probability q¡(T1, ... , Tn)- For seller's expected payoff. The Revelaban Prmc1ple ~~en_ guaran-
each possible combination of claims (T1 , •• . , Tn), the sum of tees that no other auction has a Bayesian Nash equilibnum that
the probabilities q1(T1,•••,Tn) + ··· +qn(T1, ... ,Tn) must be yields the seller a higher expected payoff, because such an equilib-
less than or equal to one. rium of such an auction would have been represented by a truth-
Games of this ki:nd (i.e., static Bayesian games in which each telling equilibrium of a direct mechanism, ~nd all such incentive-
player' s only action is to submit a claim about his or her type) compatible di:rect mechanisms were cons1dered ... Myerson also
are called direct mechanisms. shows that the symmetric Bayesian Nash eqmhbnum of the
166 STATIC GAMES OF INCOMPLETE INFORMATION The Revelation Principie 167

auction analyzed in Section 3.2.B is equivalent to this payoff-maxi- new action spaces are simple. Player i's feasible actions in the di-
mizing truth-telling equilibrium (as are the symmetric equilibria rect mechanism are (possibly dishonest) claims about i' s possible
of severa! other well-known auctions). types. That is, player i's action space is T¡. The new payoff -~nc-
As a second example of the Revelation Principie in action, con- tions are more complicated. They depend not only on the ongmal
sider the bilateral trading problem described in Section 3.2.C. We game, G, but also on the original equilibrium in that game, s*.
analyzed one possible trading game the buyer and seller could The crucial idea is to use the fact that s* is an equilibrium in G to
play-the double auction. In that game, if there is trade then the ensure that truth-telling is an equilibrium of the direct mechanism,
buyer pays something to the seller, while if there is no trade then as follows.
there is no payment, but there are again many other possibili- Saying that s* is a Bayesian Nash equilibrium of G means that
ties. There could be payments (from the buyer to the seller, or for each player i, s'( is i's best response to the other players: strate-
vice versa) even if there is no trade, and the probability of trade gies (si, ... , s'(_ 1 , s'f+ 1 , ••• , s~). More specifically, for each of t's types
could be strictly between zero and one. Also, the rule for deter- t¡ in T¡, s'((t¡) is the best action for i to choose from A¡, given that the
mining whether trade is to occur could require that the buyer's other players' strategies are (si, ... , s'(_ 1 , s'f+ 1 , ... , s~). Thus, if i's
offer exceed the seller' s demand by a certain (positive or negative) type is t¡ and we allow i to choose an action from a subset of A¡ that
amount; this amount could even vary depending on the prices the includes s'( (t¡), then i's optimal choice remains s¡(t¡), again assum-
parties name. ing that the other players' strategies are (si, ... , s'(_ 1 , s'f+ 1 , ••• , s~).
We can capture these possibilities by considering the follow- The payoff functions in the direct mechanism are chosen so as to
ing class of direct mechanisms: the buyer and the seller simul- confront each player with a choice of exactly this kind.
taneously make claims about their types, Tb and Ts, after which We define the payoffs in the direct mechanism by substituting
the buyer pays the seller x(Tb, Ts), which could be positive or neg- the players' type reports in the new game, T = (T1, ... , Tn), into
ative, and the buyer receives the good with probability q(Tb, T5 ). their equilibrium strategies from the old game, s*, and then sub-
Myerson and Satterthwaite determine which direct mechanisms stituting the resulting actions in the old game, s*(T) = (si(T1), ... ,
have a truth-telling equilibrium. They then impose the constraint s~(Tn)), into the payoff functions from the old game. Formally, i's
that each type of each party be willing to play the game (i.e., that payoff function is
each type of each party have an equilibrium expected payoff no
less than the payoff that type could achieve by refusing to play-
namely, zero for each buyer type and t5 for the seller type ts).
Finally, they show that none of these incentive-compatible direct where t = (t 1 , ... , tn)- One could imagine these payoffs occurring
mechanisms have trade with probability one if and only if trade because a neutral outsider approaches the players and makes the
is efficient. The Revelation Principie then guarantees that there following speech:
is no bargaining game the buyer and seller would willingly play
that has a Bayesian Nash equilibrium in which trade occurs if and I know you already know your types and were about
only if it is efficient. to play the equilibrium s* in the game G. Here is a
To give a formal statement and proof of the Revelation Princi- new game to play-a direct mechanism. First, each
pie, consider the Bayesian Nash equilibrium s* = (si, ... , s~) in the of you will sign a contract that allows me to dictate
static Bayesian game G = {A1, ... ,An;T1, ... ,Tn;p1,-•·,Pn;u1,--·, the action you will take when we later play G. Sec-
un}- We will construct a direct mechanism with a truth-telling ond, each of you will write down a claim about your
equilibrium that represents s*. The appropriate direct mechanism type, T¡, and submit it to me. Third, I will use each
is a static Bayesian game with the same type spaces and beliefs player' s type report in the new game, T¡, together with
as G but with new action spaces and new payoff functions. The the player's equilibrium strategy from the old game, s'(,
to compute the action the player would have taken in
 169
168 STATIC GAMES OF INCOMPLETE INFORMATION Problems

the equilibrium s* if the player's type really were T¡- 3.5 Problems
namely, s¡'(T¡). Finally, I will dictate that each of you
to take the action I have computed for you, and you Section 3.1
will receive the resulting payoffs (which will depend 3.1. What is a static Bayesian game? What is a (pure) str~~egy in
on these actions and your true types). such a game? What is a (pure-strategy) Bayesian Nash equihbnum
in such a game?
We conclude this section (and the proof of the Revelation Prin- 3.2. Consider a Cournot duopoly operating in a market with in-
cíple) by showing that truth-telling is a Bayesian Nash equilib- verse demand P(Q) = a - Q, where Q = q1 + q2 is the aggregate
rium of this direct mechanism. By claiming to be type T¡ from quantity on the market_- B?t~ fir?1s have total c_osts c¡(q¡) .=: cq¡,
T¡, player i is in effect choosing to take the action s¡' (T¡) from but demand is uncertam: 1t 1s h1gh (a = aH) with p~obabih~ 0
A¡. If all the other players tell the truth, then they are in ef- and low (a= aL) with probability 1- 0. Furthermore, mformahon
fect playing the strategies (s~, ... , s¡'_ 1, si+l, ... , s~)- But we argued is asymmetric: firm 1 knows whether demand is high or low,_ but
earlier that if they play these strategies, then when i's type is t¡ firm 2 does not. All of this is common knowledge. The two firms
the best action for i to choose is s¡'(t¡). Thus, if the other play- simultaneously choose quantities. What are the strategy spaces for
ers tell the truth, then when i's type is t¡ the best type to claim the two firms? Make assumptions concerning aH, aL, 0, ande s~ch
to be is t¡. That is, truth-telling is an equilibrium. More for- that all equilibrium quantities are positive. What is the Bayesian
mally, it is a Bayesian Nash equilibrium of the static Bayesian Nash equilibrium of this game?
game {T1, ... , Tn; T1, ... , Tn; P1, ... , Pn; V1, ... , Vn} for each player i
to play the truth-telling strategy T¡(t¡) = t¡ for every t¡ in T¡. 3.3. Consider the following asymmetric-information mod~l o~
Bertrand duopoly with differentiated products. De~and for flrm l
is q¡(p¡, Pj) = a - p¡ - b¡. Pj· Costs ar~ zero_ for_ bo!h firm~- The sen-
sitivity of firm i' s demand to firm ;' s pnce 1s e1ther high or ~ow.
That is b· is either btt or bL, where btt > bL > O. For each firm,
1

3.4 Further Reading b· = b ~ith probability 0 and b¡ = bL with probability 1 - 0, in-

depen1:ient of the realization of bj. Each firm knows its own b¡ but
not its competitor' s. All of this is common knowledge: W~at a~e
Myerson (1985) offers a more detailed introduction to Bayesian the action spaces, type spaces, beliefs, and utility fu11:ct1ons 11: th1s
games, Bayesian Nash equilibrium, and the Revelation Principle. game? What are the strategy spaces? Wh~t- c~ndibons _define ~
See McAfee and McMillan (1987) for a survey of the literature on symmetric pure-strategy Bayesian Nash eqmhbnum of th1s game.
auctions, including an introduction to the winner's curse. Búlow
Solve for such an equilibrium.
and Klemperer (1991) extend the auction model in Section 3.2.B to
produce an appealing explanation of rational frenzies and crashes 3.4. Find all the pure-strategy Bayesian Nash equilibria in the
in (say) securities markets. On employment under asymmetric following static Bayesian game:
information, see Deere (1988), who analyzes a dynamic model in
which the worker encounters a sequence of firms over time, each l. Nature determines whether the payoffs are as in Game 1 or
with its own privately known marginal product. For applications as in Game 2, each game being equally likely.
of the Revelation Principle, see Baron and Myerson (1982) on reg-
ulating a monopolist with unknown costs, Hart (1983) on implicit
2. Player 1 learns whether nature has drawn Game 1 or Game 2,
contracts and involuntary unemployment, and Sappington (1983)
on agency theory. but player 2 <loes not.
 170 STATIC GAMES OF INCOMPLETE INFORMATION Problems 171

3. P_layer 1 chooses either T or B; player 2 simultaneously chooses

cording to the strictly positive density f (v¡) on [O, l]. Compute a
j e1ther L or R.
symmetric Bayesian Nash equilibrium for the two-bidder case.
4. Payoffs are given by the game drawn by nature.
3.8. Reinterpret the buyer and seller in the double auction ana-
lyzed in Section 3.2.C as a firm that knows a worker' s marginal
L R L R product (m) anda worker who knows his or her outside opportu-
T 1,1 0,0 T nity (v), respectively, as in Hall and Lazear (1984). In this context,
0,0 0,0
trade means that the worker is employed by the firm, and the price
B 0,0 0,0 B 0,0 2,2 at which the parties trade is the worker's wage, w. If there is trade,
) then the firm' s payoff is m - w and the worker' s is w; if there is
Gamel Game2 no trade then the firm's payoff is zero and the worker's is v.
Suppose that m and v are independent draws from a uniform
distribution on [O, l], as in the text. For purposes of comparison,
Section 3.2
compute the players' expected payoffs in the linear equilibrium
3.5. Recall f:r-om Sect~on 1.3 that Matching Pennies (a static game of the double auction. Now consider the following two trading
of complete m~ormahon) has no pure-strategy Nash equilibrium games as alternatives to the double auction.
but has one m1xed-strategy Nash equilibrium: each player plays Game I: Before the parties learn their prívate information, they
H with probability 1 /2. sign a contract specifying that if the worker is employed by the
firm then the worker's wage will be w, but also that either side
can escape from the employment relationship at no cost. After the
Player 2
parties learn the values of their respective pieces of prívate infor-
H T mation, they simultaneously announce either that they Accept the
H 1, -1 wage w or that they Reject that wage. If both announce Accept,
-1, 1
Player 1 then trade occurs; otherwise it does not. Given an arbitrary value
T -1, 1 1, -1 of w from [O, 1], what is the Bayesian Nash equilibrium of this
game? Draw a diagram analogous to Figure 3.2.3 showing the
type-pairs that trade. Find the value of w that maximizes the sum
Provide a pure-strategy Bayesian Nash equilibrium of a corre-
of the players' expected payoffs and compute this maximized sum.
sponding game of incomplete information such that as the incom-
Game II: Before the parties learn their prívate information, they
plete info~a~ion disappears, the players' behavior in the Bayesian
sign a contract specifying that the following dynamic game will
Nash equ~l~b~um ~pproaches their behavior in the mixed-strategy
be used to determine whether the worker joins the firm and if so
Nash eqmhbnum m the original game of complete information.
at what wage. (Strictly speaking, this game belongs in Chapter 4.
3.6. Consider a first-price, sealed-bid auction in which the bid- We will anticípate the spirit of Chapter 4 by arguing that this
ders' valuations are independently and uniformly distributed on game can be solved by combining the lessons of this chapter with
[O, 1]. Show that if there are n bidders, then the strategy of bid- those of Chapter 2.) After the parties learn the values of their
ding (n-1)/n times one's valuation is a symmetric Bayesian Nash respective pieces of prívate information, the firm chooses a wage
equilibrium of this auction. w to offer the worker, which the worker then accepts or rejects.
Try to analyze this game using backwards induction, as we did
3.7. Consider a first-price, sealed-bid auction in which the bid- for the analogous complete-information games in Section 2.1.A, as
ders' valuations are independently and identically distributed ac- follows. Given w and v, what will the worker do? If the firm
172 STATIC GAMES OF INCOMPLETE INFORMATION

anticipates what the worker will do, then given m what will the
firm do? What is the sum of the players' expected payoffs?

3.6 References
Baron, D., and R. Myerson. 1982. "Regulating a Monopolist with Chapter 4
Unknown Costs." Econometrica 50:911-30.
Bulow, J., and P. Klemperer. 1991. '~Rational Frenzies and
Crashes." Stanford University Graduate School of Business Re-
search Paper #1150. Dynamic Games of
Chatterjee, K., and W. Samuelson. 1983. "Bargaining under
Incomplete Information." Operations Research 3;1-:~35-51. .
Deere, D. 1988. "Bilateral Trading asan Effic1ent Auchon over
Incomplete Information
Time." Journal of Political Economy 96:100-15.
Hall, R., and E. Lazear. 1984. "The Excess Sensitivity of Layoffs
and Quits to Demand." Journal of Labor Economics 2:233-57. In this chapter we introduce yet another equilibrium concept-
Harsanyi, J. 1967. "Games with Incomplete Inform~tion Played perfect Bayesian equilibrium. This makes four equilibrium concel?ts
by Bayesian Players Parts I II and III." Management Science 14:159- in four chapters: Nash equilibrium in static games of complete m-
82, 320-34, 486--502. formation, subgame-perfect Nash equilibrium in dynamic games
- - - · 1973. "Games with Randomly Disturbed Payoffs: A of complete information, Bayesian Nash equilibrium in static games
New Rationale for Mixed Strategy Equilibrium Points." Interna- of incomplete information, and perfect Bayesian equilibrium in
tional Journal of Game Theory 2:1-23. . dynamic games of incomplete information. It may seem that we
Hart, O. 1983. "Optimal Labour Contracts under Asymmetnc invent a brand new equilibrium concept for each class of games
Information." Review of Economic Studies 50:3-35. we study, but in fact these equilibrium concepts are closely re-
McAfee, P., and J. McMillan. 1987. "Auctions and Bidding." lated. As we consider progressively richer games, we progres-
Journal of Economic Literature 25:699-738. . sively strengthen the equilibrium concept, in arder to rule out im-
Myerson, R. 1979. "Incentive Compatability and the Bargain- plausible equilibria in the richer games that would survive if we
ing Problem." Econometrica 47:61-73. . applied equilibrium concepts suitable for simpler games. In each
_ _ _ , 1981. "Optimal Auction Design." Mathematics of Oper- case, the stronger equilibrium concept differs from the weaker con-
ations Research 6:58-73. cept only for the richer games, not for the simpler games. In partic-
___ . 1985. "Bayesian Equilibrium and Incentive Compat- ular, perfect Bayesian equilibrium is equivalent to Bayesian Nash
ibility: An Introduction." In Social Goals and. Social Organiz~tion. equilibrium in static games of incomplete information, equivalent
L. Hurwicz, D. Schmeidler, and H. Sonnenschem, eds. Cambridge: to subgame-perfect Nash equilibrium in dynamic games of com-
Cambridge University Press. plete and perfect information (and in many dynamic games of
Myerson, R., and M. Satterthwaite. 1983._ "Efficient Mecha- complete but imperfect information, including those discussed in
nisms for Bilateral Trading." Journal of Economic Theory 28:265-81. Sections 2.2 and 2.3), and equivalent to Nash equilibrium in static
Sappington, D. 1983. "Limited Li~bility Contracts between games of complete information.
Principal and Agent." Journal of Economic Theory 29:1-21. Perfect Bayesian equilibrium was invented in arder to refine
(i.e., strengthen the requirements of) Bayesian Nash equilibrium

1 '7'l