Problem Set 4 Answers

ECON 6206, Game Theory and Experiments

March 19, 2013

Directions: Answer each question completely. If you cannot determine the answer, explaining how you
would arrive at the answer might earn you some points.

1. Consider the following game of incomplete information between Player 1 and Player 2. Player 1’s type
is known but Player 2 may be either an H type (with probability p) or an L type (with probability
1 p). The payo¤s to this simultaneous game of incomplete information are as follows:
Player 2 (H type) Player 2 (L type)
x y x y
Player 1 a 1; 3 1; 2 Player 1 a 3; 2 1; 3
b 3; 1 2; 5 b 2; 1 0; 4

a Suppose that p = 0:75. Find all pure strategy Bayes-Nash equilibria under this assumption.

Answer:
Suppose that Player 1 chooses a. The H type’s best response is to choose x and the L type’s best
response is to choose y. Player 1’s expected value of choosing a is:
3 1
E [a] = 1 +1 =1
4 4
Player 1’s expected value of choosing b is:
3 1 9
E [b] = 3 +0 =
4 4 4
Since E [b] > E [a], this is is NOT an equilibrium.
Suppose that Player 1 chooses b. The H type’s best response is to choose y and the L type’s best
response is to choose y. Player 1’s expected value of choosing b is:
3 1 6
E [b] = 2 +0 =
4 4 4
Player 1’s expected value of choosing a is:
3 1
E [a] = 1 +1 =1
4 4
Since E [b] > E [a], this is a pure strategy Bayes-Nash equilibrium. Thus, Player 1 choose b, Player 2
(H type) choose y, and Player 2 (L type) choose y is a pure strategy Bayes-Nash equilibrium.

b Find all pure strategy Bayes-Nash equilibria for each value of p (since p is a probability p 2 [0; 1]).
Hint 1: There are no values of p such that there is more than one equilibrium for that value of
p. Hint 2: It is best to …nd ranges of p for which a speci…c equilibrium exists. Hint 3: There
is a range of p for which there are no pure strategy Bayes-Nash equilibria.

1
 Answer:
If Player 1 chooses a, Player 2’s best response is x if H type and y if L type. Player 1’s expected value
of choosing a is:
E [a] = 1 p + 1 (1 p) = 1
Player 1’s expected value of choosing b is:
E [b] = 3 p + 0 (1 p) = 3p
As long as E [a] E [b] this will be a Bayes-Nash equilibrium. If
1 3p
1
p
3
then there is a pure strategy Bayes-Nash equilibrium. If p 2 0; 31 then Player 1 choose a, Player 2
(H type) choose x, and Player 2 (L type) choose y is a pure strategy Bayes-Nash equilibrium.
If Player 1 chooses b, Player 2’s best response is y if H type and y if L type. Player 1’s expected value
of choosing b is:
E [b] = 2 p + 0 (1 p) = 2p
Player 1’s expected value of choosing a is:
E [a] = 1 p + 1 (1 p) = 1
If E [b] E [a] this will be a Bayes-Nash equilibrium. If
2p 1
1
p
2
If p 2 12 ; 1 then Player 1 choose b, Player 2 (H type) choose y, and Player 2 (L type) choose y is a
pure strategy Bayes-Nash equilibrium.
1 1
If p 2 3; 2 then there is no pure strategy Bayes-Nash equilibrium.
2. Consider two …rms who compete by simultaneously choosing prices (a Bertrand game). If Firms 1
and 2 choose prices p1 and p2 , respectively, the quantity that consumers demand from each …rm is:
q1 (p1 ; p2 ) = a p1 + bp2
q2 (p1 ; p2 ) = a p2 + bp1
Firm 2 has constant marginal cost c. With probability Firm 1 has constant marginal cost cH while
with probability (1 ) Firm 1 has constant marginal cost cL . Assume that there are no …xed costs.
Prices must be nonnegative (p1 0, p2 0) and …rms wish to maximize pro…t. Important note:
Even though it is a Bertrand game, the demand functions are di¤erentiable and continuous for both
…rms. Let = 12 , a = 108, b = 12 , cH = 16, cL = 8, and c = 12.
a Find the best response functions for Firms 1 and 2.
Answer:
The key is to realize that Firm 1 has two types and so needs to specify a price if cost is cH (call it pH )
and a price if cost is cL (call it pL ). If Firm 1 has cost cH it solves:
max 1;cH = (a pH + bp2 ) pH cH (a pH + bp2 )
pH
@ 1;cH
= a 2pH + bp2 + cH
@pH
0 = a 2pH + bp2 + cH
a + bp2 + cH
pH =
2
124 + 12 p2
pH =
2

2
If Firm 1 has cost cL it solves:
max 1;cL = (a pL + bp2 ) pL cL (a pL + bp2 )
pL
@ 1;cL
= a 2pL + bp2 + cL
@pL
0
= a 2pL + bp2 + cL
a + bp2 + cL
pL =
2
116 + 21 p2
pL =
2
Firm 2 only has one type and so only maximizes one (expected) pro…t function. Firm 2 solves:
max 2 = ((a p2 + bpH ) p2 c (a p2 + bpH )) + (1 ) ((a p2 + bpL ) p2 c (a p2 + bpL ))
p2
@ 2
= (a 2p2 + bpH + c) + (1 ) (a 2p2 + bpL + c)
@p2
0 = (a 2p2 + bpH + c) + (1 ) (a 2p2 + bpL + c)
0 = a 2p2 + c + bpH + (1 ) bpL
a + c + bpH + (1 ) bpL
p2 =
2
120 + 12 12 pH + 21 12 pL 120 + 14 (pH + pL )
p2 = =
2 2
b Find a pure strategy Bayes-Nash equilibrium to this game.
Answer:
The easiest way to solve is to substitute pH and pL into Firm 2’s best response function. Then we
have:
1 1 124+ 12 p2 1 1 116+ 12 p2
120 + 2 2 2 + 2 2 2
p2 =
2
1 124 + 12 p2 1 116 + 12 p2
2p2 = 120 + +
4 2 4 2
124 + 12 p2 116 + 12 p2
8p2 = 480 + +
2 2
1 1
16p2 = 960 + 124 + p2 + 116 + p2
2 2
15p2 = 1200
p2 = 80
Given that p2 = 80, we have:
124 + 12 p2
pH =
2
124 + 40
pH =
2
pH = 82
And:
116 + 12 p2
pL =
2
116 + 40
pL =
2
pL = 78

3
 So the Bayes-Nash equilibrium is for Firm 1 to choose p1 = 82 if cost is cH and p1 = 78 if cost is cL ,
while Firm 2 chooses p2 = 80.
3. It is typical for the government to allocate construction contracts, such as repaving a highway, by
holding an auction for the contract. The auction rules are as follows. Each bidder is to submit a
sealed bid. The lowest bidder will win the contract, and the winning bidder will be paid an amount
equal to the second lowest bid. Suppose that each bidder draws a cost, ci , of completing the job from
the uniform distribution over the interval [c; c]. The cost draws are made independently of each other.
All bidders are aware of the common distribution of costs as well as the fact that cost draws are made
independently of one another.

a Suppose that the bidders are risk-neutral. Find a Bayes-Nash equilibrium for this auction.

Answer:
Note that this is a 2nd -price sealed bid procurement auction. Since it is a 2nd -price auction, the weakly
dominant strategy will be to submit a bid equal to cost. Thus, bi (ci ) = ci . One can use the same
arguments from class to show that a bidder in this auction would not choose to submit a bid greater
than or less than cost.

b Suppose that the bidders are risk-neutral. Suppose we changed the format so that the winning
bidder, who is still the lowest bidder, now receives a payment equal to his bid if he wins (instead
of a payment equal to the 2nd lowest bid as before). Find a Bayes-Nash equilibrium for this new
set of auction procedures.1

Answer:
This problem is a little di¤erent than the traditional 1st -price auction. For starters, in the traditional
1st -price auction bidders only care about those bidders who have values lower than them, because
bidders with values higher than them should win in equilibrium. In the procurement auction, bidders
only care about those bidders with costs above their own. The maximization problem for a bidder i
is:
max U = Pr (win) (bi ci )
bi

The key is how to set up the probability of winning. I will use the uniform distribution on [0; 1] for
solving this problem and show where to input the c and c limits of the range at the end. The key is
that now we have a Pr (win) where the bidder wants the LOWEST bid. So now, if we have N bidders,
we have:
N 1
Pr (win) = (1 F ( (b)))
because the bidder wants the lowest bid. This is where most people made their error on the homework.
So the problem is:
N 1
max U = (1 F ( (b))) (bi ci )
bi
@U N 2 N 1 N 2
= (N 1) (1 F ( (b))) ( 1) (F 0 ( (b))) b + (1 F ( (b))) (N 1) (1 F ( (b))) ( 1) (F 0 (
@bi
N 2 N 1 N 2
0 = (N 1) (1 F ( (b))) ( 1) (F 0 ( (b))) b + (1 F ( (b))) (N 1) (1 F ( (b))) ( 1) (F 0 (
0 0
0 = (N 1) ( 1) (F ( (b))) b + 1 F ( (b)) (N 1) ( 1) (F ( (b))) ci
0 = (N 1) (F 0 ( (b))) ci + 1 F ( (b)) (N 1) (F 0 ( (b))) b
0 = (N 1) 0 (b) ci + 1 (b) (N 1) 0 (b) b

If you would have gotten to that point it would have been great. The next step would be to substitute
in (b) = ci , so:
0 = (N 1) 0 (b) (b) + 1 (b) (N 1) 0 (b) b
1 If you cannot …nd one for the distribution of values [c; c] try the distribution [0; 1].

4
 Now is where the "guessing" comes in, or you can solve the di¤erential equation if you like. If you
"guess" that:
1
b = ci + (1 ci )
N
then you would have made a good guess. We can check using the di¤erential equation. First, let’s
…nd (b) and 0 (b). We have:
1
b = ci +
(1 ci )
N
1
b = (b) + (1 (b))
N
N b = N (b) + 1 (b)
N b 1 = (N 1) (b)
Nb 1
= (b)
N 1
0
Note that we also need (b), so:
0 N
(b) =
N 1
Now plugging these terms into our di¤erential equation we see:
0
0 = (N 1)
(b) ci + 1 (b) (N 1) 0 (b) b
N Nb 1 Nb 1 N
0 = (N 1) +1 (N 1) b
N 1 N 1 N 1 N 1
N Nb 1
0 = (N b 1) + 1 Nb
N 1 N 1
Nb 1 Nb 1
0 = N + 1 Nb
N 1 N 1
Nb 1
0 = (N 1) + 1 Nb
N 1
0 = Nb 1 + 1 Nb
0 = 0

So, each …rm, based on its cost, would bid:

1
bi = ci + (1 ci )
N
for values distributed uniformly on [0; 1]. For values on [c; c], we would have:
1
bi = ci + (c ci )
N
If you look at the bid function, the bidders are now bidding their cost plus some amount above their
cost. If N = 1 (so that there is only one bidder) then the bidder places a bid equal to c, or the top of
the cost distribution.
4. Recall that in the Battle of the Sexes game there are 2 players who are simultaneously deciding whether
or not to attend a boxing match or an opera performance. However, now both players are uncertain
about the value that the other player places on his or her favorite event. Thus, the normal form is:
Player B
Opera Boxing
Player A Opera 2 + tA ; 1 0; 0
Boxing 0; 0 1; 2 + tB

5
where tA is privately known by Player A and tB is privately known by Player B. Both t1 and t2 are
independent draws on a uniform distribution from [0; x].
Let A be the critical value such that if tA A then Player A plays Opera and if tA < A then Player
A plays Boxing. Let B be the critical value such that if tB B then Player B plays Boxing and if
tB < B then Player B plays Opera. In this equilibrium, Player A plays Opera with probability x xA
and Player B plays Boxing with probability x xB .

a Find the critical values A and B. They will be a function of x (recall that x is the upper bound of
the uniform distribution).

Answer:
Player A should play Opera if EA [Opera] EA [Boxing]. We have:

B x B
EA [Opera] = (2 + tA ) + 0
x x
B
EA [Opera] = (2 + tA )
x

Now, EA [Boxing] is:

x B B
EA [Boxing] = (1) + 0
x x
x B
EA [Boxing] = (1)
x

So we need:
B x B
(2 + tA )
x x
For Player B, Player B should play Boxing if EB [Boxing] EB [Opera]. We have:

A x A
EB [Boxing] = (2 + tB ) + 0
x x
A
EB [Boxing] = (2 + tB )
x
Now EB [Opera] is:

x A A
EB [Opera] = 1+ 0
x x
x A
EB [Opera] = 1
x

So we need:
A x A
(2 + tB )
x x
We now have 2 equations and 2 unknowns:

B x B
(2 + tA )
x x
A x A
(2 + tB )
x x

6
The key is to realize that A and B are just some signal tA and tB respectively. Also, we can remove
the inequalities since we are trying to …nd the single critical value, so we now have:
tB x tB
(2 + tA ) =
x x
tA x tA
(2 + tB ) =
x x
Now we just need to solve for tA and tB .
x tB
(2 + tA ) =
tB
x tA
(2 + tB ) =
tA
Or:
x tB
tA = 2
tB
x tB 2tB
tA =
tB tB
x 3tB
tA =
tB

By the same method we get:

x 3tA
tB =
tA
And now we just plug in tA :
x 3tB
x 3 tB
tB = x 3tB
tB

Now it’s just some algebra:

xtB 3x+9tB
tB
tB = x 3tB
tB
xtB 3x + 9tB
tB =
x 3tB
2
tB x 3 (tB ) = xtB 3x + 9tB
2
3 (tB ) = 3x + 9tB
2
(tB ) = x 3tB
2
(tB ) + 3tB x = 0

Now we can just use the quadratic equation to …nd:

p
3 9 4 1 ( x)
tB =
p 2
3 9 + 4x
tB =
2
The last piece to realize is that we need tB 0 since tB 2 [0; x], so:
p
3 + 9 + 4x
tB =
2

7
We can then plug this in to …nd tA , or we can just state “by symmetry”since the equations for tA and
tB are the same to …nd: p
3 + 9 + 4x
tA =
2
b Explain how the proposed equilibrium satis…es the criteria for a Bayes-Nash equilibrium.

Answer:
In a Bayes-Nash equilibrium, each player needs to specify an action for each potential type he could
be. Look at the equilibrium as it is described. Consider Player A. For any signal tA < A, Player A
plays boxing, and for any signal tA A Player A plays opera. For Player B, for any tB B Player B
plays Boxing and for any tB < B Player B plays Opera. Thus, an action is speci…ed for any potential
type for both players. Part a of this problem shows how both are using best responses to the others
choice of strategy.