EconS 503 - Advanced Microeconomics II

Handout on Bayesian Nash Equilibrium

1. MWG 8.E.1
Consider the following strategic situation. Two opposed armies are poised to seize an island.
Each army’s general can choose either "attack" or "not attack." In addition, each army is
either "strong" or "weak" with equal probability (the draws for each army are independent),
and an army’s type is known only to its general. Payo¤s are as follows: The island is worth
M if captured. An army can capture the island either by attacking when its opponent does
not or by attacking when its rival does if it is strong and its rival is weak. If two armies
of equal strength both attack, neither captures the island. An army also has a "cost" of
…ghting, which is s if it is strong and w if it is weak, where s < w: There is no cost of
attacking if its rival does not.

Identify all pure strategy Bayesian Nash equilibria of this game.

Answer:
There are four pure strategies contingent on the type of player:

AA : Attack if either weak or strong type,

AN : Attack if strong and Not attack if weak,
N A : Not attack if strong and Attack if weak,
N N : Never attack.

We can determine the expected payo¤s for each player by simple calculation. For example,
the expected payo¤ for player 1 playing the strategy AA given that player 2 also plays the
strategy AA is
Player 1 2 3
is Strong Player 2 is Strong Player 2 is Weak
z}|{ z }| { z }| {
EU1 (AAjAA) = 0:5 4 0:5 ( s) + 0:5 (M s)5 + 0:5 [0:5 ( w) + 0:5 ( w)]

M s+w
=
4 2
The remaining expected payo¤s for each pair of strategies can be easily computed and are
given in …gure 1:

1
 Player 2
AA AN NA NN

M s+w M s+w M s M s+w 3M s+w w

AA 4 2
; 4 2 4
;
2 2 2 4 4
; 2
M; 0

M s M s+w M s M s M s M w M
AN 4
;
2 2 4 4
; 4 2 4
; 4 2
;0
Player 1
w 3M s+w M w M s M w M w M
NA 2
; 4 4 4
; 2 4 4
; 4 2
;0

NN 0; M 0; M2 0; M2 0; 0

Figure 1: Normal Form Representation

Any NE of this normal form game is a Bayesian NE of the original game.

[Case 1] M > w > s; and w > M2 > s

From the above payo¤s we can see that (AA; AN ) and (AN; AA) are both pure strategy
Bayesian Nash equilibriua.

[Case 2] M > w > s; and M2 < s

From the above payo¤s we can see (AA; N N ) and (N N; AA) are both pure strategy Bayesian
Nash equilibria.

[Case 3] w > M > s; and M2 < s

From the above payo¤s we can see that (AN; AN ) ; (AA; N N ) and (N N; AA) are pure strat-
egy Bayesian Nash equilibria.

[Case 4] w > M > s; and M2 > s

From the above payo¤s we can see that (AN; AN ) is the pure strategy Bayesian Nash
equilibrium.

2. MWG 8.E.3

Consider the linear Cournot model described in Ex 8.B.5; two …rms 1 and 2; simultaneously
choose the quantities they will sell on the market, q1 and q2 : The price each receives for each
unit given these quantities is P (q1 ; q2 ) = a b (q1 + q2 ) : Their costs are c per unit sold.
Now, however, suppose that each …rm has probability of having unit costs of cL and (1 )
of having unit costs of cH ; where cH > cL : Solve for the Bayesian Nash equilibrium.

2
Answer:
A …rm of type i = H or L will maximize its expected pro…t, taken as given that the other
…rm will supply qH or qL depending whether this …rm is of type H or L: A type i 2 fH; Lg
…rm 1 will maximize:

M1 ax(1 ) [(a 2
b (qi1 + qH ) ci )qi1 ] + [(a b (qi1 + qL2 ) ci )qi1 ]
qi 0

The FOC yields:

(1 )(a b (2qi1 + qH
2
) ci ) + (a b (2qi1 + qL2 ) ci ) = 0

In a symmetric Bayesian Nash equilibrium:

1 2
qH = qH = qH and qL1 = qL2 = qL

Plugging this into the F.O.C we get the following two equations:

(1 ) [a 3bqH cH ] + [a b (2qH + qL ) cH ] = 0
(1 ) [a b (qH + 2qL ) cL ] + [a 3bqL cL ] = 0

Therefore, we obtain that

1
qH = a cH + 2 (cL cH ) 3b
;
qL = a cL + 1 2 (cH cL ) 1
3b
:

3. Auction Theory

Consider a …rst-price sealed-bid auction in which bidders simultaneously submit bids and
the object goes to the highest bidder at a price equal to his/her bid. Suppose that there are
N 2 bidders, whose types i are drawn from the cumulative distribution F (:) and receive
utility from the object from the function u( i b), where b is the amount of the bid paid by
player i.

a) Solve for s0 ( ), the derivative of the bidding function of each individual.

3
Answer:
We start by setting up the utility maximization problem for player i,

max E i [vi (b; s i ( i ); i )j i ] = [F (s 1 (b))]N 1

u( i b)
b 0

Taking …rst-order conditions,

@s 1 (b)
(N 1)[F (s 1 (b))]N 2
f (s 1 (b)) u( i b) [F (s 1 (b))]N 1 0
u( i b) = 0
@b
for s( ) to be an optimal bidding function, it should be optimal for the bidder not to pretend
to have a valuation di¤erent from his real one, i . Hence s( ) = b is the optimal solution for
the above …rst-order condition, and we have that s 1 (b) = , implying

(N 1)[F ( )]N 2 f ( )u( s( ))

[F ( )]N 1 0
u( s( )) = 0
s0 ( )
Simplifying,
(N 1)f ( )u( s( ))
= F ( )u0 ( s( ))
s0 ( )
and solving for s0 ( ) yields
u( s( )) f( )
s0 ( ) = (N 1)
u0 ( s( )) F( )

b) Assume that consumers are risk neutral, i.e. u(x) = x. Derive the bidding function
s( ).

Answer:
Using what we derived in part (a), we have
f( )
s0 ( ) = ( s( )) (N 1)
F( )
rearranging some terms,

f ( )s( )(N 1) + F ( )s0 ( ) = f ( )(N 1)

we can multiply both sides by [F ( )]N 2

to obtain

[F ( )]N 2
f ( )s( )(N 1) + [F ( )]N 1 0
s ( ) = [F ( )]N 2
f ( )(N 1)
d([F (x)]N 1 s( ))
Note that the left side is now just d
. Substituting,

d([F (x)]N 1
s( ))
= [F ( )]N 2
f ( )(N 1)
d

4
and integrating both sides
Z
N 1
[F (x)] s( ) = [F (x)]N 2
xf (x)(N 1)dx

let

h(x) = x g 0 (x) = [F (x)]N 2 f (x)(N 1)dx

h0 (x) = dx g(x) = [F (x)]N 1

and applying integration by parts, we have

Z
N 1 N 1
[F (x)] s( ) = [F (x)] [F (x)]N 1
dx

and solving for s( ) yields the bidding function

Z
1
s( ) = [F (x)]N 1
dx
[F (x)]N 1

c) Now assume that consumers are risk averse, i.e. u(x) = x where 0 < < 1. Derive
the bidding function s( ).

Answer:
Using what we derived in part (a), we have
1 f( )
s0 ( ) = ( s( )) (N 1)
F( )
rearranging some terms,
1 1
f ( )s( ) (N 1) + F ( )s0 ( ) = f ( ) (N 1)
1
(N 1) 1
we can multiply both sides by [F ( )] to obtain
1
(N 1) 1 1 1
(N 1) 0 1
(N 1) 1 1
[F ( )] f ( )s( ) (N 1) + [F ( )] s ( ) = [F ( )] f ( ) (N 1)
1 (N 1)
d([F (x)] s( ))
Note that the left side is now just d
. Substituting,
1
(N 1)
d([F (x)] s( )) 1
(N 1) 1 1
= [F ( )] f ( ) (N 1)
d
and integrating both sides
Z
1
(N 1) 1
(N 1) 1 1
[F (x)] s( ) = [F (x)] xf (x) (N 1)dx

5
let
1 1
h(x) = x g 0 (x) = [F (x)] (N 1) 1
f (x) (N 1)dx
1
h0 (x) = dx g(x) = [F (x)] (N 1)

and applying integration by parts, we have

Z
1 1 1
(N 1) (N 1) (N 1)
[F (x)] s( ) = [F (x)] [F (x)] dx

and solving for s( ) yields the bidding function

Z
1 1
(N 1)
s( ) = 1
(N 1)
[F (x)] dx
[F (x)]