Vous êtes sur la page 1sur 27

# I Know Something You Don’t Know:

## Games with Private Information 10

1. Greg is deciding whether to ask Marcia out on a date. However, Greg isn’t sure whether
Marcia likes him, and he would rather not ask if he expects to be rejected. Whether
Marcia likes Greg is private information to her. Thus, her preferences regarding Greg
constitute her type. Greg does not have any private information. Assume that there is a
25% chance that Marcia likes Greg. The Bayesian game is shown here. Should Greg ask
Marcia? Find a BNE that answers this question.

Nature

## Marcia likes Marcia doesn't

Greg like Greg
Probability = .25 Probability = .75

Greg

Do not Do not
Invite Invite
invite invite

Marcia Marcia
3 3
4 4
Yes No Yes No

Greg 10 2 8 2
Marcia 7 2 1 3

ANSWER: A strategy for Marcia is a pair of actions: what to do if she likes Greg
and he asks her out and what to do if she doesn’t like Greg and he asks her out.
Her optimal strategy is fairly clear. If she likes Greg, then by accepting his invita-
tion her payoff is 7 and that exceeds the payoff of 2 from turning him down. If she
doesn’t like Greg, then her payoff from rejecting his invitation is 3 and that exceeds
her payoff of 1 from accepting it. Hence, Marcia’s Bayes-Nash equilibrium strategy
is (accept, reject). Given this strategy, let us derive Greg’s optimal strategy. If he
asks Marcia, then his expected payoff is .25  10  .75  2  4. With probability
.25, Marcia likes him and thus accepts (according to her optimal strategy). In that
event, he gets a payoff of 10. With probability .75, Marcia doesn’t like him and thus
declines his invitation, so Greg’s payoff is 2. His expected payoff from asking
Marcia is then 4, given his beliefs of Marcia’s type and his conjecture of Marcia’s
strategy. Alternatively, he can choose not to ask her, in which case his payoff is 3.
Thus, it is optimal for him to ask her. There is then a unique Bayes-Nash equilib-
rium in which Greg invites Marcia and Marcia accepts if she likes Greg and
declines if she does not.
10-1
10-2 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

2. Consider a gunfight between Bat Masterson and William “Curly Bill” Brocius. Both of
the men have private information regarding their skill with a six-shooter. Nature moves
first by determining each gunfighter’s skill. He can have either a fast draw or a slow
draw. There is a 65% chance that Bat is fast and a 60% chance that Curly Bill is fast.
After each gunfighter learns his type—though remaining uncertain about the other gun-
fighter’s type—he chooses between draw and wait. If both wait, then the payoff is 50. If
both draw and (1) they are of the same type (either both fast or both slow), then each
has a payoff of 20; and (2) they are of different types, then the fast gunfighter has a pay-
off of 30 and the slow one of 40. If one draws and the other waits and (1) they are of
the same type, then the one who drew has a payoff of 30 and the other a payoff of 40;
(2) the one who draws is fast and the other is slow, then the one who drew has a payoff
of 30 and the other a payoff of 40; and (3) the one who draws is slow and the other is
fast, then each has a payoff of 20. If at least one chooses draw, then there is a gunfight.
a. Is it consistent with BNE for there to be a gunfight for sure? (That is, both gunfight-
ers draw, regardless of their type.)

ANSWER: Yes, as it is an equilibrium for each gunfighter to use a strategy that has
him draw regardless of his type. The equilibrium conditions for Bat are

## Fast Type (Draw): .6  20  .4  30  .6  (40)  .4  20 1 24  16

Slow Type (Draw): .6  (40)  .4  20  .6  (40)  .4  (40) 1
32  40.

## The equilibrium conditions for Curly Bill are

Fast Type (Draw): .65  20  .35  30  .65  (40)  .35  20 1 23.5  19
Slow Type (Draw): .65  (40)  .35  20  .65  (40)  .35  (40) 1
19  40.

b. Is it consistent with BNE for there to be no gunfight for sure? (That is, both gunfight-
ers wait, regardless of their type.)

ANSWER: Yes, as it is an equilibrium for each gunfighter to use a strategy that has
him wait regardless of his type. Doing so realizes a payoff of 50—as a gunfight is
avoided—and all other outcomes yield a lower payoff, so the expected payoff from
any drawing must be less. More explicitly, the equilibrium conditions for Bat are

## Fast Type (Wait): .6  50  .4  50  .6  30  .4  30 1 50  30

Slow Type (Wait): .6  50  .4  50  .6  20  .4  30 1 50  24.

## Fast Type (Draw): .65  50  .35  50  .65  30  .35  30 1 50  30

Slow Type (Draw): .65  50  .35  50  .65  20  .35  30 1 50  23.5.

## c. Is it consistent with BNE for a gunfighter to draw only if he is slow?

ANSWER: Yes. Consider a strategy profile in which each draws when slow and
waits when fast. The equilibrium conditions for Bat are

## Fast Type (Wait): .6  50  .4  20  .6  30  .4  30 1 38  30

Slow Type (Draw): .6  20  .4  20  .6  50  .4  (40) 1 20  14.

## Fast Type (Wait): .65  50  .35  20  .65  30  .35  30 3 39.5  30

Slow Type (Draw): .65  20  .35  20  .65  50  .35  (40) 3 20  18.5.
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-3

3. Consider a first-price, sealed-bid auction in which a bidder’s valuation can take one of
three values: 5, 7, and 10, occurring with probabilities .2, .5, and .3, respectively. There
are two bidders, whose valuations are independently drawn by Nature. After each bidder
learns her valuation, they simultaneously choose a bid that is required to be a positive
integer. A bidder’s payoff is zero if she loses the auction and is her valuation minus her
bid if she wins it.
a. Determine whether it is a symmetric BNE for a bidder to bid 4 when her valuation is
5, 5 when her valuation is 7, and 6 when her valuation is 10.

## ANSWER: It is a Bayes-Nash equilibrium. To see this, suppose a player has valu-

ation of 5. Then the player’s optimal action can be found in the table below.

## Probability of Payoff of Expected

Bid Winning Winning Payoff
3 0 2 0
4 .2  .5  .1 1 .1
5 .2  .5  .5  .45 0 0

By bidding 3 or less, the player can never win the item. And if his bid is greater
than or equal to 5 (his valuation), then even if he wins the payoff is nonpositive.
By bidding 4, he can win the auction only when the other player has valuation 5,
which happens with a probability .2. In this case, since the winner will be deter-
mined randomly, the actual probability of winning is .1. Hence, if the player’s
valuation is 5, he bids 4.
If his valuation is 7, then the corresponding table of bids is shown in the table
below.

## Probability of Payoff of Expected

Bid Winning Winning Payoff
4 .1 3 .3
5 .45 2 .9
6 .2  .5  .5  .3  .85 1 .85

By similar reasons as above, he will not submit a bid that is greater than or equal
to 7 or a bid that is below 4. Hence, bidding 5 gives him the highest expected pay-
off, .9.
If his valuation is 10, then the corresponding table of bids is shown in the table
below.

## Probability of Payoff of Expected

Bid Winning Winning Payoff
4 .1 6 .6
5 .45 5 2.25
6 .85 4 3.4
7 .2  .5  .3  1 3 3

Any bid greater than 7 is strictly dominated by a bid of 7, since 7 is the small-
est bid that can win the auction. Hence, 6 is his optimal bid when his valuation
is 10.
10-4 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

b. Determine whether it is a symmetric BNE for a bidder to bid 4 when her valuation is
5, 6 when her valuation is 7, and 9 when her valuation is 10.

ANSWER: Suppose player 2 uses the described strategy. If player 1 has a valuation
of 10, by bidding 9, her expected payoff is

.3
a.2  .5  b  (10  9)  .85.
2

## (.2  .5)  (10  7)  .7  3  2.1.

Hence, bidding 9 is not optimal. Thus, the strategy profile is not a Bayes-Nash
equilibrium.

4. Consider a first-price, sealed-bid auction, and suppose there are only three feasible
bids: A bidder can bid 1, 2, or 3. The payoff to a losing bidder is zero. The payoff to a
winning bidder equals his valuation minus the price paid (which, by the rules of the
auction, is his bid). What is private information to a bidder is how much the item is
worth to him; hence, a bidder’s type is his valuation. Assume that there are only two
valuations, which we’ll denote L and H, where H  3  L  2. Assume also that each
bidder has probability .75 of having a high valuation, H. The Bayesian game is then
structured as follows: First, Nature chooses the two bidders’ valuations. Second, each
bidder learns his valuation, but does not learn the valuation of the other bidder.
Third, the two bidders simultaneously submit bids. A strategy for a bidder is a pair
of actions: what to bid when he has a high valuation and what to bid when he has a
low valuation.
a. Derive the conditions on H and L whereby it is a symmetric BNE for a bidder to bid
3 when he has a high valuation and 2 when he has a low valuation.

ANSWER: Let us first derive the equilibrium condition for a player when he is
type L. Payoff from bidding 1 is

3 1
a b  0  a b  0  0.
4 4

## The payoff from bidding 2 is

3 1 1 1
a b  0  a b a b(L  2)  a b(L  2).
4 4 2 8

## The payoff from bidding 3 is

3 1 1 5
a b a b(L  3)  a b(L  3)  a b(L  3).
4 2 4 8

## Clearly, bidding 2 is better than bidding 3, since by assumption 3  L  2. Hence,

bidding 2 is optimal if bidding 2 is better than bidding 1

1
a b(L  2)  0 1 L  2,
8

which is true by assumption. Next, we’ll consider the equilibrium condition for a
player when he is type H. The payoff from bidding 1 is

3 1
a b  0  a b  0  0.
4 4
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-5

## The payoff from bidding 2 is

3 1 1 1
a b  0  a b a b(H  2)  a b(H  2).
4 4 2 8

## The payoff from bidding 3 is

3 1 1 5
a b a b(H  3)  a b(H  3)  a b(H  3).
4 2 4 8

## Bidding 3 is better than bidding 1 when

5
a b(H  3)  0 B H  3,
8

## which is true by assumption. Bidding 3 is better than bidding 2 when

5 1 13
a b(H  3)  a b(H  2) 1 H  .
8 8 4

13
This is a Bayes-Nash equilibrium if and only if H  .
4
b. Derive the conditions on H and L whereby it is a symmetric BNE for a bidder to bid
2 when he has a high valuation and 1 when he has a low valuation.

payoff from bidding 1 is

3 1 1 1
a b  0  a b a b(L  1)  a b(L  1).
4 4 2 8

## The payoff from bidding 2 is

3 1 1 5
a b a b(L  2)  a b(L  2)  a b(L  2).
4 2 4 8

## The payoff from bidding 3 is

3 1
a b(L  3)  a b(L  3)  (L  3).
4 4

## Clearly, bidding 1 is better than bidding 3 since by assumption 3  L  2. Hence,

bidding 1 is optimal when

1 5 9
a b(L  1)  a b(L  2) 1 L  .
8 8 4

Next, consider the equilibrium condition when the player’s type is H. Payoff from
bidding 1 is

3 1 1 1
a b  0  a b a b(H  1)  a b(H  1).
4 4 2 8

## The payoff from bidding 2 is

3 1 1 5
a b a b(H  2)  a b(H  2)  a b(H  2).
4 2 4 8
10-6 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

## The payoff from bidding 3 is

3 1
a b(H  3)  a b(H  3)  (H  3).
4 4

## Bidding 2 is better than bidding 1 when

5 1 9
a b(H  2)  a b(H  1) 1 H  ,
8 8 4

## which is true by assumption. Bidding 2 is better than bidding 3 when

5 14
a b(H  2)  (H  3) 1 H  .
8 3

9 14
This is a Bayes-Nash equilibrium if and only if L  and H  .
4 3
c. Derive the conditions on H and L whereby it is a symmetric BNE for a bidder to bid
3 when he has a high valuation and 1 when he has a low valuation.

ANSWER: Consider the equilibrium condition when a player’s type is L. The payoff
from bidding 1 is

3 1 1 1
a b  0  a b a b(L  1)  a b(L  1).
4 4 2 8

## The payoff from bidding 2 is

3 1 1
a b  0  a b(L  2)  a b(L  2).
4 4 4

## The payoff from bidding 3 is

3 1 1 5
a b a b(L  3)  a b(L  3)  a b(L  3).
4 2 4 8

## Clearly, bidding 1 is better than bidding 3 since 3  L  2. Bidding 1 is best when

1 1
a b(L  1)  a b(L  2) 1 L  1  2L  4 1 3  L,
8 4

## which is true by assumption. Next, consider the equilibrium condition when a

player’s type is H. The payoff from bidding 1 is

3 1 1 1
a b  0  a b a b(H  1)  a b(H  1).
4 4 2 8

## The payoff from bidding 2 is

3 1 1
a b  0  a b(H  2)  a b(H  2).
4 4 4

## The payoff from bidding 3 is

3 1 1 5
a b a b(H  3)  a b(H  3)  a b(H  3).
4 2 4 8
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-7

## Bidding 3 is better than bidding 1 when

5 1 7
a b(H  3)  a b(H  1) 1 H  .
8 8 2

## Bidding 3 is better than bidding 2 when

5 1 11
a b(H  3)  a b(H  2) 1 H  .
8 4 3

11 7 11
Since  , this is a Bayes-Nash equilibrium if and only if H  .
3 2 3
d. Derive the conditions on H and L whereby it is a symmetric BNE for a bidder to bid
1 when he has either a high valuation or a low valuation.

ANSWER: Consider the equilibrium condition when a player’s type is L. The payoff
from bidding 1 is

3 1 1 1 1
a b a b(L  1)  a b a b(L  1)  a b(L  1).
4 2 4 2 2

## The payoff from bidding 2 is

3 1
a b(L  2)  a b(L  2)  (L  2).
4 4

## The payoff from bidding 3 is

3 1
a b(L  3)  a b(L  3)  (L  3).
4 4

## Clearly, bidding 1 is better than bidding 3 since by assumption 3  L  2. Hence,

bidding 1 is optimal when

1
a b(L  1)  (L  2) 1 L  3,
2

## which is true by assumption. Next, consider the equilibrium condition when a

player’s type is H. The payoff from bidding 1 is

3 1 1 1 1
a b a b(H  1)  a b a b(H  1)  a b(H  1).
4 2 4 2 2

## The payoff from bidding 2 is

3 1
a b(H  2)  a b(H  2)  (H  2).
4 4

## The payoff from bidding 3 is

3 1
a b(H  3)  a b(H  3)  (H  3).
4 4

## Bidding 1 is the best when

1
a b(H  1)  (H  2) 1 H  3,
2

## which is not true by assumption. Hence, there is no such Bayes-Nash equilibrium.

10-8 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

5. Consider this Bayesian game on page 389. Nature chooses the type of player 1, where
type H occurs with probability p and type L with probability 1  p. Player 1 learns his
type and then chooses either action x or action y. Simultaneously, player 2 chooses either
action a or action b.

Nature

H L
Probability = p Probability = 1 – p

1 1

x y x y

a b a b a b a b

Player 1 3 1 2 5 2 1 3 4
Player 2 1 3 1 2 3 2 1 0

## a. Assume that p  .75. Find a BNE.

b. For each value of p, find all Bayes–Nash equilibria.

strategy for player 1 is a pair of actions: an action when his type is H and an action
when his type is L. A strategy for player 2 is a single action since he has only one
information set. There are then eight possible strategy pairs, which are shown in
the table below.

## # 1’s Action When H 1’s Action When L 2’s Strategy Equilibrium?

1 x x a No
1
2 x y a Yes, if p  3
3 y x a No
4 y y a No
5 x x b No
6 x y b No
7 y x b No
1
8 y y b Yes, if p  2

Let’s consider each of them. Strategy pair 1 is not an equilibrium, since when
player 1 is type L, he prefers to choose y and get a payoff of 3 than choose x and
get a payoff of 2. As for strategy pair 2, player 1’s strategy is optimal since, given
player 2 is to choose action a, player 1’s payoff is 3 from x and 2 from y (when he
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-9

is type H) and is 3 from y and 2 from x (when he is type L). Action a yields a higher
expected payoff than b for player 2 when

1
p  1  (1  p)  1  p  3  (1  p)  0 3  p.
3

Turning to strategy pair 3, player 1’s strategy is not optimal as it calls for action y
when he is type H and we already showed that he prefers x (given player 2 chooses
a). The same argument applies to strategy pair 4. For strategy pair 5, given player
2 chooses b, player 1 prefers y when he is type H (with a payoff of 5) to x (with a
payoff of 1). Hence, both strategy pairs 5 and 6 are not equilibria. For strategy pair
7, player 1 prefers y to x when he is type L since the former yields 4 and the latter
yields 1 (given player 2 is choosing b). Finally, it should be clear from the previous
argument that player 1’s strategy in strategy pair 8 is optimal for both types. Player
2’s strategy is optimal when

1
p  2  (1  p)  0  p  1  (1  p)  1 1 p  .
2

The last column on the right in the table above answers part (b); the answer to
part (a) is strategy pair 8.

6. Two U.S. senators are considering entering the race for the Democratic nomination for
U.S. president. Each candidate has a privately known personal cost to entering the race.
Assume that the probability of having a low entry cost, fL, is p and the probability of hav-
ing a high entry cost, fH, is 1  p. Thus, the type space has just two values. A candidate’s
payoff depends on whether he enters the race and whether the other senator enters as
well. Let v2 be a candidate’s payoff when he enters and the other senator does as well (so
that there are two candidates), v1 be a candidate’s payoff when he enters and the other
senator does not (so that there is one candidate), and 0 be the payoff when he does not
enter. Assume that

v1  v2  0,
fH  fL  0,
v2  fL  0  v2  fH,
v1  fH  0.

a. Derive the conditions whereby it is a symmetric BNE for a candidate to enter only
when she has a low personal cost from doing so.

## ANSWER: It is a symmetric Bayes-Nash equilibrium for a senator to enter only

when she has low personal cost if and only if

## This condition holds when p is close to 1.

b. Derive the conditions whereby it is a symmetric BNE for a candidate to enter for sure
when she has a low personal cost and to enter with some probability strictly between
0 and 1 when she has a high personal cost.

ANSWER: Consider a symmetric strategy profile such that (1) if the candidate has
low cost, she enters and (2) if she has high cost, she enters with probability q.
10-10 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

## pv2  (1  p)v1  fH (1  p)(v1  v2 )  v2  fH

1q 3q . [SOL10.6.2]
(1  p)(v1  v2 ) (1  p)(v1  v2 )

With the high type, she must be indifferent between entering and not entering.
For this to be an equilibrium, the derived value for q must lie between 0 and 1:

(1  p)(v1  v2 )  v2  fH
0 1
(1  p)(v1  v2 )
1 (1  p)(v1  v2 )  v2  fH  0 1 pv2  (1  p)v1  fH. [SOL10.6.3]

## If we substitute (SOL10.6.2) into (SOL10.6.1), the expression becomes

fH (v1  v2 )  fL (v1  v2 ), which always holds true. This means (SOL10.6.2) implies
(SOL10.6.1), so the only condition we need is (SOL10.6.2). If pv2  (1  p)v1  fH,
then there is an equilibrium in which a low-cost Senator enters and a high-cost
one randomizes. If fH  pv2  (1  p)v1  fL, then, as we know from part (a),
there is an equilibrium in which a low-cost Senator enters and a high-cost one
does not.

c. Find some other BNE distinct from those described in (a) and (b).

## ANSWER: There is also an asymmetric equilibrium in which Senator 1 always

enters and Senator 2 enters only when she has low cost. Given Senator 2’s strategy,
it is always optimal for Senator 1 to enter if and only if

## Low type: p(v2  fL )  (1  p)(v1  fL )  0 1 pv2  (1  p)v1  fL

High type: p(v2  fH )  (1  p)(v1  fH )  0 1 pv2  (1  p)v1  fH.

Given Senator 1’s strategy, it is optimal for Senator 2 to enter only when she has
low personal cost if and only if

Low type: v2  fL  0
High type: v2  fH  0.

The last two inequalities obviously hold. Hence, this equilibrium exists if
pv2  (1  p)v1  fH.

7. Assume that two countries are on the verge of war and are simultaneously deciding
whether or not to attack. A country’s military resources are its type, and their relevance
is summarized in a parameter which influences the likelihood that they would win a war.
Suppose the type space is made up of two values: p
and p–, where 0 p
p– 1. A
country is type p– with probability q and type p
with probability 1  q. Consider a coun-
try of type p (which equals either p
or p–). If it chooses to attack and it attacks first, then
it believes it’ll win the war with probability xp, where x takes a value such that p xp 1.
If the two countries both attack, then the probability that a type p country wins is p. If
a type p country does not attack and the other country does attack, then the probability
of victory for the type p country is yp, where y takes a value such that 0 yp p. Finally,
if neither country attacks, then there is no war. A country is then more likely to win the
war the higher is its type and if it attacks before the other country. A country’s payoff
when there is no war is 0, from winning a war is W, and from losing a war is L. Assume
that W  0  L.
a. Derive the conditions for it to be a symmetric BNE for a country to attack regardless
of its type.
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-11

ANSWER: If one country expects the other to attack for sure (that is, if it is type
p¿ or p–), then it is always optimal to attack. Doing so results in an expected
payoff of pW  (1  p)L and not doing so results in pyW  (1  py)L, which is
lower. Attacking regardless of one’s type is then a symmetric Bayes-Nash
equilibrium.

b. Derive the conditions for it to be a symmetric BNE for a country to attack only if its
type is p–.

## ANSWER: Consider a symmetric strategy profile in which a country attacks only

if its type is p–. The two equilibrium conditions (for each country) are

## Type p–: q 3p–W  (1  p–)L 4  (1  q) 3xp–W  (1  xp–)L 4

 q 3yp–W  (1  yp–)L 4  (1  q)0
Type p¿: q 3yp¿W  (1  yp¿)L 4  (1  q)0
 q 3p¿W  (1  p¿)L 4  (1  q) 3xp¿W  (1  xp¿)L 4.

The first condition assures that a country prefers to attack when it is type p–. The
second condition assures that it prefers not to attack when it is type p¿.

8. Consider a first-price, sealed-bid auction with three bidders. The payoff to a bidder is
0 when he does not win the item at auction and is the value of the item less his bid
(which is the price he pays) when he is the winner. If two or more bidders submit
the highest bid, then the winner is randomly determined. Assume that the item has
the same value to all three bidders, but they receive different signals as to its value.
Nature determines the true value, which is denoted v and can take three possible
1
values: 4, 5, and 6. Each of these values is chosen with probability 3 . The signals sent
to the three bidders are v  1, v, and v  1; that is, one bidder receives a signal that
is too low (v  1), another receives a signal that is too high (v  1), and the third
receives a signal that is “just right” (v). Each bidder learns only his own signal, which
is the bidder’s type. If v  4, then one bidder is given a signal of 3, another bidder is
given a signal of 4, and the last bidder is given a signal of 5. If v  5, then one bidder
is given a signal of 4, another bidder is given a signal of 5, and the last bidder is given
a signal of 6. If v  6, then one bidder is given a signal of 5, another bidder is given
a signal of 6, and the last bidder is given a signal of 7. Thus, if a bidder’s signal is,
say, 5, then he doesn’t know if the true value is 4 (in which case he has the highest
signal), 5 (in which case he has the accurate signal), or 6 (in which case he has the
lowest signal). Given the value, each bidder has an equal chance of receiving one of
the three signals. Assume that the minimum increment in bidding is 1, so that the set
of feasible bids is {0, 1, . . . , 10}.
a. Show that the following symmetric strategy profile is a BNE:

Signal Bid
3 4
4 4
5 4
6 4
7 5

## ANSWER: By symmetry, it is sufficient to show this strategy is optimal for bidder

1 given the other two bidders use it. Let si be the signal received by bidder i. If
s1  3, then bidder 1 knows the true value is 4, as a signal of 3 is sent only when
the true value is 4. This means that the other two bidders are bidding 4—as one
receives a signal of 4 and, according to the strategy he’ll bid 4, and similarly so for
the other bidder, who receives a signal of 5. By bidding 4, bidder 1’s payoff is 0
1
because either she wins (which occurs with probability 3 since there is a three-way
tie) and pays a price equal to the item’s value, or he loses (which occurs with prob-
2
ability 3 ). If he bids below 4, then he loses for sure and thus gets a payoff of 0. By
10-12 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

bidding above 4, he wins for sure, but gets a negative payoff since he pays a price
in excess of the item’s value. It is then optimal to bid 4 when the bidder’s type is 3.
Now suppose s1  4. In this case, bidder 1 remains uncertain of the true value.
It could be either 4 (in which case he has the accurate signal) or 5 (in which case
he has the lowest signal). In the former case, the other bidders’ signals are 3 and 5,
which implies they both bid 4. In the latter case, the other bidders’ signals are 5
and 6, in which case they bid 4 and 5, respectively. If bidder 1 bids less than 4, than
he loses for sure and thereby receives a payoff of 0. If he bids 4, then again his
payoff is 0 (though he might win when the item’s value is 4). If he bids 5, then his
payoff is 0 when the true value is 5 (but here he might win) and is negative when
the true value is 4. Thus, bidder 1’s expected payoff is negative when he bids 5 (as
it is the average of something negative and 0). Finally, if he bids above 5, then he
wins for sure but he pays a price in excess of its value. In sum, his expected payoff
is negative by bidding above 4 and is 0 by bidding at or below 4. Thus, a bid of 4
is indeed optimal when his signal is 4.
Next, suppose s1  5; bidder 1 doesn’t know whether the true value is 4, 5, or 6.
When the true value is 4, the other bidders’ signals are 3 and 4, in which case they
both bid 4. When the true value is 5, the other bidders’ signals are 4 and 6, in which
case they bid 4 and 5, respectively. Last, when the true value is 6, the other bidders’
signals are 6 and 7, in which case they both bid 5. If bidder 1 bids less than 4, his
expected payoff is 0, since he loses for sure. If he bids 4, then again his payoff is 0
(though he might win when the item’s value is 4). If he bids 6, then his expected
payoff is negative, since he wins for sure when the item’s value is 4 or 5 but
pays a price in excess of its value, and when the item’s value is 6 he might win (with
1
probability 2 ) but pays a price equal to the value of the item. If he bids greater than
6, then he wins the auction for sure but he pays a price in excess of its value. The
last case is bidding 5. His expected payoff is

1 1 1 1 1 1 1 2
 1  (1)    0    1      0.
3 3 2 3 3 3 9 9

The first term in the first summation is his expected payoff when the true value is
1
4, which happens with probability 3 , in which case he wins for sure but pays 5,
resulting in a loss of 1. The second and the third terms correspond with the case
where the true values are 5 and 6, respectively. Hence, it is optimal to bid 4 when
the signal is 5.

Next, suppose s1  6. Bidder 1 then knows that the true value is 5 or 6, and the
other bidders are bidding 4 or 5. If bidder 1 bids 4, his expected payoff is

1 1 1 1
 1 02 .
2 3 2 6

1 1 1 1
10  1 .
2 2 2 4

## Hence, bidding 5 is optimal when the signal is 6.

Finally, suppose s1  7. Bidder 1 then knows the true value is 6, in which case
the other bidders will bid 4 (if the signal is 5) and 5 (if the signal is 6). Bidding 4
will result in a payoff of 0, since he loses for sure. Bidding 5 results in an expected
1 1
payoff of ( 2 )(6  5)  2 since, given there are two bidders who bid 5, bidder 1 has
a 50% chance of winning the item. If he bids 6, then his payoff is 0. Hence, he
optimally chooses 5 if the signal is 7.

b. Show that there is no symmetric BNE in which a bidder’s bid is strictly increasing.
That is, if b(s) is a bidder’s bid, given that her signal is s, and if

## b(7)  b(6)  b(5)  b(4)  b(3),

then this strategy is not a symmetric BNE.
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-13

nicity means

## b(7)  b(6)  b(5)  b(4)  b(3).

Suppose a bidder receives a signal of 3. She knows she’ll lose for sure by bidding
b(3) as the other bids will be b(4) and b(5). This can be optimal only if by bidding
b(5) she pays too much for the item if she wins:

b(5)  4 [SOL10.8.1]

Now consider a bidder whose signal is 7. Then the following condition should be
satisfied:

## By monotonicity, b(7) has to be at least as great as b(6)  1 and there is no point

in having b(7) exceed b(6)  1, since that doesn’t change the probability of win-
ning but it does decrease the payoff of winning. If bidder 1 bids b(6)  1, then
her payoff is 6  (b(6)  1). If instead she bids b(6), then her expected payoff is
1
2
(6  b(6)). Thus, bidding b(6)  1 is optimal only if

1
6  (b(6)  1)  (6  b(6) ) 1 b(6)  4 [SOL 10.8.3]
2

## (SOL10.8.1) and (SOL10.8.3) imply b(5)  b(6)  4, which is a contradiction.

9. The Newlywed Game was a popular game show broadcast from 1966 to 1974. On this
show, four recently married couples would be queried about one another to see how well
each spouse knew the other. Each of the husbands was asked a series of questions while
his wife was offstage in a soundproof room. The wives would then return, and each
would be asked the same questions. The objective was for the answers of the husband
and wife to match. In the second round, they would reverse roles. The couple with the
most matched answers would win. Questions asked ran along these lines: What animal
ho hum, so so, or ooh la la? In what room of the house does your spouse most like mak-
ing whoopee? Let us now suppose the husband is asked the question, What drink would
best describe your wife on your wedding night: a sloe gin fizz, a Shirley Temple, or a
zombie? The husband responds with one of those three choices, and when she is asked
the same question, the wife responds with one of the same three choices. If their choices
match, a husband–wife pair has a payoff of 100; if they don’t match, the pair has a pay-
off of 0. In casting this as a Bayesian game, suppose the husband’s type is which one he
believes is true: sloe gin fizz, Shirley Temple, or zombie, where the associated probabil-
1 1 1
ities are 4 , 2 , and 4 , respectively. The wife’s type is which one she believes is true: sloe gin
1
fizz, Shirley Temple, or zombie. The sloe gin fizz occurs with probability 2 , the Shirley
1 1
Temple with probability 4 , and the zombie with probability 4 . Players’ types are indepen-
dently drawn.*
a. Find a BNE.

ANSWER: This is a bit of a trick question, as this is a coordination game and pay-
offs depend only on whether their answers match and not on the types. Here are
three Bayes-Nash equilibria: (1) the husband and wife choose fizz regardless of
type; (2) the husband and wife choose Temple regardless of type; and (3) the husband
and wife choose Zombie regardless of type. In each case, a player’s expected payoff
is 100 for each type and that is the maximal payoff possible, so there cannot be a
better strategy.

*This is a simplifying, but not very reasonable, assumption, because one would think that if the husband found his wife
to be, say, a sloe gin fizz, then it is more likely that the wife found herself to be that way as well.
10-14 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

## b. Find a BNE in which the husband always announces the truth.

ANSWER: This is definitely a trick question as, in fact, there are no Bayes-Nash
equilibria in which the husband always announces the truth. Consider the hus-
band choosing the action corresponding to his type. If his wife was a fizz, he says
fizz; if she was a Temple, he says Temple; if she was a Zombie, he says Zombie.
Given this strategy, what is the optimal strategy for the wife? Her type is irrelevant
in forming beliefs on the husband’s type, and her objective is solely to match what
her husband said. Since he is most likely to think she was a Temple—his probability
1
of that is 2 —then he is most likely to say that (according to his strategy). Thus, she
should say Temple regardless of his type. Her expected payoff from doing so is 50
1
(probability 2 of matching answers), while it is 25 from fizz and 25 from Zombie.
But if she is expected to say Temple for sure, then so should the husband. His
strategy from saying Temple is 100—regardless of his type. While his strategy is
optimal when his type is Temple, it is not optimal when it is fizz or Zombie.

10. Players 1, 2, and 3 are involved in a game requiring some coordination. Each chooses
among three options: A, B, and C. Nature determines which of these options is the best
1
one to coordinate on, where equal probabilities of 3 are assigned to A, B, and C’s being
the best one. If all three choose the option that Nature deems best, then each receives a
payoff of 5. If all three choose the same option, but it is not the one that Nature deems
best, then each receives a payoff of 1. If players do not all choose the same option, then
each has a zero payoff. Player 1 learns which option is best (i.e., she learns Nature’s
choice). The three players then simultaneously choose an option. Find a BNE.

## ANSWER: A strategy for player 1 is a 3-tuple of actions that prescribes an action

for each of the three possible moves of Nature. A strategy for player 2 or 3 is a
single action. A strategy profile that has player 1 choose A for each move of Nature
and players 2 and 3 choose A is a Bayes-Nash equilibrium. Given her type, player
1’s payoff from choosing A is 5 or 1 (the former if Nature chooses A), while it is
zero from either B or C. Thus, player 1’s strategy is optimal for all types. Given the
other two players are going to choose A, player 2’s optimal strategy is to choose A
as well. This is true for player 3, too. By analogous arguments, a strategy profile in
which all players choose B is an equilibrium, as is one in which all choose C.

11. Consider the Gunfight game when p is the probability the stranger is a gunslinger and 1  p
is the probability he is a cowpoke. Assume 0 p 1/2. Find a BNE in mixed strategies
in which Wyatt Earp randomizes and the stranger randomizes when he is a cowpoke.

ANSWER: Let q denote the probability that Earp chooses wait and r denote the prob-
ability that the stranger chooses wait when he is a cowpoke. We know that a BNE
must have the gunslinger choose draw. For Earp to randomize as part of an equilib-
rium, he must be indifferent between choosing draw and wait, which requires:

Draw: p  2  (1  p)  (r  4  (1  r)  5)  p  1  (1  p)  (r  8 
(1  r)  6): Wait

Solving for r,
1  2p
r
3(1  p)
1
which is between 0 and 1 because 0 p 2 . For the cowpoke to randomize as
part of an equilibrium, he must be indifferent between choosing draw and wait,
which requires:

Draw: q  3  (1  q)  2  q  4  (1  q)  1: Wait
1 1
Solving for q, we find that q  2 . In sum, if 0 p 2 then there is a BNE in which
1  2p
Earp chooses wait with probability 3(1  p) and the stranger chooses wait with
1
probability 2 when he is a cowpoke and with probability 0 when he is a gunslinger.
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-15

12. An instructor walks into her classroom and says “What’s in your wallet?” No, this is not a
Capital One commercial; it’s a game (as if it could be anything else, in this book). She
selects two students to engage in the following contest. Students simultaneously submit a
nonnegative integer. Whoever submits the higher number is to pay an amount in dollars to
the instructor equal to the other student’s chosen number, and the instructor is to give that
student an amount of money in dollars equal to the sum of the money in both students’
wallets. The student who submitted the smaller number neither pays nor receives. Let si
denote the number submitted by student i and wi be the amount of money in his wallet.
According to the rules just described, if s1  s2 then bidder 1 wins the contest and receives
a net payment of w1  w2  s2, as the instructor pays him an amount equal to w1  w2 and
he pays the instructor an amount equal to s2. In the event of a tie (s1  s  s2), each student
receives half of what is in each student’s wallet and pays half of the number submitted;
hence, each student has a net payment of (w1  w2  s)/2. Each student knows how much
is in his wallet but does not know how much is in the other student’s wallet. A player’s type
is the amount of money in his wallet and assume it can take any integer from 1 to 100, with
each value chosen by Nature with equal probability. Assume a player’s payoff equals the
amount of money received. Note that this is a common value auction in that the value
attached to winning is the same for both students, and students differ in the information
they have about that value. Show that it is a symmetric BNE for a student to submit a
number equal to twice what is in his wallet.

ANSWER: The unique symmetric BNE is si  2wi; that is, a student submits a
number equal to twice what is in his wallet. We will not prove uniqueness but will
show that it is a BNE. Furthermore, you’ll see that it does not depend on the pos-
sible values that wi can take or the probabilities that Nature assigns to those values.
Suppose that student 2 uses this strategy, so that s2  2w2, and student 1 submits
s1. If student 1 wins then his payoff is w1  w2  2s2  w1  w2  2w2  w1  w2
which is positive only when w1  w2. (Note that student 1’s submission does not
affect his net payment when he loses; it only affects it when he wins.) Thus, student
1 would like to submit a value for s1 so that he wins if and only if he has more money
in his wallet than student 2. Given that s2  2w2, student 1 can do so by setting
s1  2w1 as then student 1 wins (s1  s2 ) if and only if 2w1  2w2 or w1  w2.
This exercise is from Paul Klemperer, “Auctions with Almost Common Values: The
‘Wallet Game’ and Its Applications,” European Economic Review, 42 (1998), 757–769.

13. A seller has an object to sell and is deciding at what price to sell it. There is one buyer
who values it at \$10 with probability q and \$2 with probability 1  q. The seller chooses a
price not knowing how much the buyer values it. The price can be any number between
0 and 20. After observing the seller’s price, the buyer decides whether or not to buy. The
buyer’s payoff is zero if she does not buy the item and is the valuation attached to the
item less the price paid if she buys it. The seller’s payoff equals zero if he does not sell
the item and equals the price it sold for if he sells it.
a. Find a BNE in which the buyer’s strategy has her act optimally for any price of the
seller (and not just the price the seller selects in equilibrium). Assume that when the

ANSWER: If the buyer is the low type (i.e., she values it at 2) then it is optimal for
her to buy as long as the price (denoted p) does not exceed 2 and if she is the high
type (i.e., values it at 10) then it is optimal for her to buy as long as the price does
not exceed 10. Given that strategy for the buyer, a seller’s payoff is p when p  2
(as both buyer types buy), q  p when 2 p  10 (as only a buyer with valuation
10 buys), and is zero when p  10 (as neither buyer type buys). The seller’s payoff
is strictly increasing in price when p  2 which implies p  2 is preferred to any
lower price. For 2 p  10, the seller’s payoff is strictly increasing in price which
implies p  10 is preferred to any price lower than 10 and at least 2. Finally, a price
no higher than 10 is strictly preferred to any price above 10. Therefore, the optimal
price is either 2 or 10. A price of 2 yields a payoff of 2 for the seller and a price of
10 yields a payoff of 10q. A price of 2 is then optimal if and only if 2  10q or
q  .2, and a price of 10 is optimal if and only if 10q  2 or q  .2. In sum, if q  .2
then there is a BNE in which the seller prices at 2 and if q  .2 then there is a BNE
in which the seller prices at 10. With both BNE, the buyer’s strategy has her buy
when p  2 and her valuation is 2 and when p  10 and her valuation is 10.
10-16 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

## b. Find a BNE different from that in part (a).

ANSWER: Consider a strategy for the buyer which has her buy when p  p¿ and her
valuation is 2 and when p  p– and her valuation is 10. Assume p¿  2  p–  10.
The optimal strategy for the seller is to price either at p¿ and earn p¿, or at p– and earn
q  p–. p¿ is optimal if and only if p¿  qp– or q  p¿/p–, and a price of p– is optimal
if and only if p¿  qp– or q  p¿/p–. If q  p¿/p– then the seller prices at p¿ and both
buyer types buy which is optimal since p¿  2. If q  p¿/p– then the seller prices at p–
and only the high valuation buyer buys which is optimal since 2  p–  10. Note that
a low type buyer is not acting optimally if the seller were to offer a price between p¿
and 2 but, given the seller does not offer a price between p¿ and 2, it is consistent with
the buyer’s strategy being part of a BNE. Similarly, a high type buyer is not acting
optimally if the seller were to offer a price between p– and 10. This equilibrium is dif-
ferent from that in part (a) as long as p¿ 2 and/or p– 10.

14. Consider two pharmaceutical companies investing in R&D. Each company can invest
either at a low (L), medium (M), or high (H) rate. For any pair of investment rates, com-
pany 1’s profit is known but company 2’s profit is private information because only
company 2 knows the cost that it incurs. Company 2 either has low cost or high cost. If
it is low cost then the payoff matrix is the top figure and if it is high cost then the payoff
matrix is the bottom figure. Note that the payoffs for company 1 are the same in the two
matrices because its payoff depends only on the investment rates and not on company
2’s type. In contrast, company 2’s payoffs are higher when it is low cost, holding fixed
the investment rates of the two companies. Assume the probability that company 2 is
low cost is p and is high cost is 1  p. Nature chooses company 2’s type and then the two
companies simultaneously choose investment rates. Find all BNE.

## Company 2 has Low Cost

Company 2
L M H
L 7,12 5,15 4,16
Company 1 M 9,10 6,12 3,11
H 8,8 4,9 0,8

## Company 2 has High Cost

Company 2
L M H
L 7,3 5,4 4,0
Company 1 M 9,2 6,0 3,4
H 8,0 4,3 0,8

ANSWER: First note that M strictly dominates H for company 1. Thus, BNE must
have company 1 choosing either L or M. Consider a strategy profile that has company
1 choose L. Company 2 will then choose H when it is a low type and M when it is a
high type. Given that strategy for company 2, company 1’s expected payoff from L is

p  4  (1  p)  5  5  p

and from M is

p  3  (1  p)  6  6  3p.

## Hence, L is optimal if and only if

1
5  p  6  3p 1 p  .
2
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-17

1
In conclusion, if p  2 then there is a BNE in which company 1 chooses L and
company 2 chooses H when a low type and M when a high type.
Now consider a strategy profile in which company 1 chooses M. The optimal
strategy for company 2 is to choose M when a low type and L when a high type.
Given that strategy for company 2, company 1’s expected payoff from L is

p  5  (1  p)  7  7  2p

and from M is

p  6  (1  p)  9  9  3p.

Hence, M is always optimal; in fact, M will yield a higher payoff than L for com-
pany 1 for both types of company 2. In conclusion, it is always a BNE for company 1
to choose M and for company 2 to choose M when it is a low type and L when it
is a high type.

15. Major League Baseball has hired a former NYPD officer as an inspector to keep ballplay-
ers clean of steroids. The inspector has her eye on a particular athlete. Conducting an
inspection costs the inspector an amount c and the value to a type i inspector from
catching someone on steroids is vi. There are three types of inspectors who vary in terms
of the value of catching a cheater: v1, v2, and v3 where v1  v2  v3  c 0. Thus, if an
inspection is conducted by a type i inspector and the athlete tests positive for steroids
then the inspector’s payoff is vi  c and if he tests negative then her payoff is c. If no
test is conducted then the payoff is zero. As for the athlete, the benefit to him from tak-
ing steroids is bj when he is type j, and the cost of testing positive for steroids is d. There
are three types, b1  b2  b3  0, and it is assumed d  b1. The athlete’s payoff when he
does not take steroids is zero (regardless of whether he is tested for steroids), bj  d if
he took steroids and is caught, and bj if he took steroids and is not caught. Nature
chooses the inspector’s type and the athlete’s type where each type is chosen indepen-
dently with probability 1/3. After the inspector learns her type, she decides whether or
not to conduct an inspection. After the athlete learns his type, he decides whether or not
to take steroids. The inspector and athlete move simultaneously. Find all BNE.

ANSWER: First note that if it is optimal for a type 3 inspector to test for steroids then
it is also optimal for a type 2 since the value of catching the athlete is higher for type 2,
and, by analogous logic, if it is optimal for a type 2 inspector to test for steroids then
it is also optimal for a type 1. Hence, the only strategies for an inspector that could
be part of a BNE are: 1) all types test; 2) only types 1 and 2 test; 3) only type 1 tests;
and 4) no types test. By the same line of reasoning, the only strategies for an athlete
that could be part of a BNE are: 1) all types take steroids; 2) only types 1 and 2 take
steroids; 3) only type 1 takes steroids; and 4) no types take steroids.
If the inspector’s strategy has her never test then the optimal strategy for the
athlete is to take steroids for all types since the payoff is bi  0. But if the athlete’s
strategy means he is assured of taking steroids then the optimal strategy for the
inspector is to always test since the payoff is vi  c  0. Hence, there is no BNE
in which the inspector never inspects. By the same logic, there is no BNE in which
the athlete never takes steroids, or the inspector always inspects, or the athlete
always takes steriods.
This leaves us with 4 possible Bayes Nash equilibria: 1) type 1 and 2 inspectors
test, and type 1 athlete uses steroids, 2) types 1 and 2 inspectors test, and types 1
and 2 athletes use steroids, 3) type 1 inspector tests, and type 1 athlete uses steri-
ods, 4) type 1 inspector tests, and types 1 and 2 athletes use steroids. We will
consider each of them in turn.
Let Ii denote the probability that the inspector is type i and Ai denote the prob-
ability that the athlete is type i. Consider the candidate BNE in which the inspector
inspects when she is type 1 or 2, and athlete takes steriods when he is type 1 or 2.
This strategy is optimal for the athlete if and only if

## (I1  I2 )(b2  d)  (1  I1  I2 )b2  0  (I1  I2 )(b3  d)  (1  I1  I2 )b3

b2 2 b3
3b2  2d  0  3b3  2d 1  
d 3 d
10-18 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

The first inequality ensures that the type 2 athlete prefers to take steroids (which
will also imply that it is preferred by the type 1 athlete), and the second inequality
ensures that the type 3 athlete prefers not to take steroids. The inspector’s strategy
is optimal if and only if

## (A1  A2 )(v2  c)  (1  A1  A2 )(c)  0  (A1  A2 )(v3  c)  (1  A1  A2 )(c)

v2 3 v3
2v2  3c  0  2v3  3c 1  
c 2 c
2 3
Thus, this strategy pair is a BNE if and only if b2/d  3  b3/d and v2/c  2  v3/c.
Next consider the candidate BNE in which the inspector inspects when she is
type 1 or 2, and the athlete takes steroids when he is type 1. It is optimal for the
athlete if and only if

## (I1  I2 )(b1  d)  (1  I1  I2 )b1  0  (I1  I2 )(b2  d)  (1  I1  I2 )b2

b1 2 b2
3b1  2d  0  3b2  2d 1  
d 3 d

## A1 (v2  c)  (1  A1 )(c)  0  A1 (v3  c)  (1  A1 )(c)

v2 v3
v2  3c  0  v3  3c 1 3
c c
2
Thus, this strategy pair is a BNE if and only if b2/d  3  b3/d and v2/c  3  v3/c.
Next consider the candidate BNE in which the inspector inspects when she is
type 1, and the athlete takes steroids when he is type 1 or 2. It is optimal for the
athlete if and only if

## I1 (b2  d)  (1  I1 )b2  0  I1 (b3  d)  (1  I1 )b3

b2 1 b3
3b2  d  0  3b3  d 1  
d 3 d

## (A1  A2 )(v1  c)  (1  A1  A2 )(c)  0  (A1  A2 )(v2  c)  (1  A1  A2 )(c)

v1 3 v2
2v1  3c  0  2v2  3c 1  
c 2 c
1 3
Thus, this strategy pair is a BNE if and only if b2/d  3  b3/d and v1/c  2  v2/c.
Finally, consider the candidate BNE in which the inspector inspects when she
is type 1, and the athlete takes steroids when he is type 1. It is optimal for the
athlete if and only if

## I1 (b1  d)  (1  I1 )b1  0  I1 (b2  d)  (1  I1 )b2

b1 1 b2
3b1  d  0  3b2  d 1  
d 3 d

## A1 (v1  c)  (1  A1 )(c)  0  A1 (v2  c)  (1  A1 )(c)

v1 v2
v1  3c  0  v2  3c 1 3
c c
1
Thus, this strategy pair is a BNE if and only if b1/d  3  b2/d and v1/c  3  v2/c.

16. Consider the game on page 394. Nature chooses player 1’s type which is L1 with prob-
ability p1 and H1 with probability 1  p1, and chooses player 2’s type, which is L2 with
probability p2 and H2 with probability 1  p2. After players learn their type, they
simultaneously choose actions; player 1 chooses between a and b and player 2
between c and d.
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-19

Nature

Type L1 Type H1
Probability = p1 Probability = 1  p1

Nature Nature

## Type L2 Type H2 Type L2 Type H2

Probability = p2 Probability Probability = p2 Probability
= 1  p2 = 1  p2

1 1

a b a b a b a b

c d c d c d c d c d c d c d c d

Player 1 2 3 4 1 3 2 0 0 0 5 1 4 3 3 2 3
Player 2 5 1 4 0 2 6 3 4 3 4 6 3 5 0 2 0

a. If player 1 chooses a for either type, derive an optimal strategy for player 2. (Note:

ANSWER: For the answers to parts (a)-(c), a strategy for player 1 will be repre-
sented as x/y where x refers to the action selected when player 1 is type L1 and y
refers to the action when he is type H1. Thus, the four possible strategies for player 1
are a/a, a/b, b/a, and b/b. A strategy for player 2 is of the form x/y where x refers to
the action selected when player 2 is type L2 and y refers to the action when she is
type H2. Thus, the four possible strategies for player 2 are c/c, c/d, d/c, and d/d.
Suppose player 1 uses a/a. For her two possible actions, the expected payoff to
player 2 when she is type L2 is:

c: p1  5  (1  p1 )  3  3  2p1
d: p1  1  (1  p1 )  4  4  3p1
1
c is optimal if 3  2p1  4  3p1 or p1 
5

## and when she is type H2 is:

c: p1  2  (1  p1 )  5  5  3p1
d: p1  6  (1  p1 )  0  6p1
5
c is optimal if 5  3p1  6p1 or p1
9
1 1
If 0  p1 5 then d/c is player 2’s optimal strategy. If p1  5 then d/c and c/c are
1 5
both optimal strategies for player 2. If 5 p1 9 then c/c is player 2’s optimal
5
strategy. If p1  9 then c/c and c/d are both optimal strategies for player 2. If
5
9
p1  1 then c/d is player 2’s optimal strategy.
10-20 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

b. Derive conditions on p1 and p2 such that it is a BNE for player 1 to choose a for either
type and player 2 to choose c for either type.

ANSWER: Suppose player 1 uses a/a. For her two possible actions, the expected
payoff to player 2 when he is type L2 is:

c: p1  5  (1  p1 )  3  3  2p1
d: p1  1  (1  p1 )  4  4  3p1
1
c is optimal if p1 
5

## and when she is type H2 is:

c: p1  2  (1  p1 )  5  5  3p1
d: p1  6  (1  p1 )  0  6p1
5
c is optimal if p1 
9

Turning to player 1, for his two possible actions, the expected payoff to player 1
when he is type L1 is:

a: p2  2  (1  p2 )  3  3  p2
b: p2  4  (1  p2 )  0  4p2
3
a is optimal if p2 
5

## and when he is type H1 is:

a: p2  0  (1  p2 )  3  3  3p2
b: p2  1  (1  p2 )  2  2  p2
1
a is optimal if p2 
2

1 5 1
So the conditions we need are 5
 p1  9 and p2  2 .

## c. Assume p1  .4 and p2  .3. Find all of the BNE.

ANSWER: In deriving the BNE, we’ll consider each possible strategy for player
1— a/a, a/b, b/a, b/b—where the first action is what he does when he is type L1,
derive the optimal strategy for player 2, and then determine whether the assumed
strategy for player 1 is optimal. If it is then the strategy pair is a BNE and if it is
not then there is no BNE with the assumed strategy for player 1.
Suppose player 1 uses a/a. For her two possible actions, the expected payoffs to
player 2 when she is type L2 are:

c: p1  5  (1  p1 )  3  3  2p1  3.8
d: p1  1  (1  p1 )  4  4  3p1  2.8

Thus, c is optimal. When she is type H2, the expected payoffs are:

c: p1  2  (1  p1 )  5  5  3p1  3.8
d: p1  6  (1  p1 )  0  6p1  2.4

Thus, c is optimal. Given c/c is player 2’s optimal strategy when player 1 uses a/a,
let us examine the optimality of that strategy for player 1. The expected payoff to
player 1 when he is type L1 is:

a: p2  2  (1  p2 )  3  3  p2  2.7
b: p2  4  (1  p2 )  0  4p2  1.2
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-21

## Thus, a is optimal. And when he is type H1:

a: p2  0  (1  p2 )  3  3  3p2  2.1
b: p2  1  (1  p2 )  2  2  p2  1.7

## Thus, a/a is optimal for player 1. One BNE is (a/a, c/c).

Now suppose player 1 uses a/b. For her two possible actions, the expected
payoffs to player 2 when she is type L2 are:

c: p1  5  (1  p1 )  6  6  p1  5.6
d: p1  1  (1  p1 )  3  3  2p1  2.2

## Thus, c is optimal. And when she is type H2:

c: p1  2  (1  p1 )  2  2
d: p1  6  (1  p1 )  0  6p1  2.4

Thus, d is optimal. c/d is player 2’s optimal strategy. Given that strategy for player 2,
the expected payoffs to player 1 when he is type L1 from his two possible
actions are:

a: p2  2  (1  p2 )  2  2
b: p2  4  (1  p2 )  0  4p2  1.2

## Thus, a is optimal. And when he is type H1:

a: p2  0  (1  p2 )  3  3  3p2  2.1
b: p2  1  (1  p2 )  3  3  2p2  2.4

Thus, b is optimal. As this was the original strategy specified for player 1 then we
have another BNE and it is (a/b, c/d).
Now suppose player 1 uses b/a. For her two possible actions, the expected pay-
offs to player 2 when she is type L2 are:

c: p1  4  (1  p1 )  3  3  p1  3.4
d: p1  0  (1  p1 )  4  4  4p1  2.4

## Thus, c is optimal. And when she is type H2:

c: p1  3  (1  p1 )  5  5  2p1  4.2
d: p1  4  (1  p1 )  0  4p1  1.6

Hence, c/c is optimal for player 2. Given that strategy for player 2, the expected
payoffs to player 1 when he is type L1 are:

a: p2  2  (1  p2 )  3  3  p2  2.7
b: p2  4  (1  p2 )  0  4p2  1.2

Thus, a is optimal which runs counter to the strategy originally specified for player 1,
b/a. Hence, b/a is not part of a BNE.
Finally, suppose player 1 uses b/b. For her two possible actions, the expected
payoffs to player 2 when she is type L2 are:

c: p1  4  (1  p1 )  6  6  2p1  5.2
d: p1  0  (1  p1 )  3  3  3p1  1.8

## Thus, c is optimal. And when she is type H2:

c: p1  3  (1  p1 )  2  2  p1  2.4
d: p1  4  (1  p1 )  0  4p1  1.6
10-22 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

So c/c is optimal for player 2. Turning to player 1, the expected payoffs to player 1
when he is type H1 are:

a: p2  0  (1  p2 )  3  3  3p2  2.1
b: p2  1  (1  p2 )  2  2  p2  1.7
a is optimal if p2  0.3

As a is preferred then b/b is not an optimal strategy for player 1. Therefore, b/b is
not part of a BNE.
In sum, there are two Bayes-Nash equilibria: (a/a, c/c) and (a/b, c/d).

17. Consider the game on the top of page 395. Nature chooses player 1’s type which is t1 with
probability .25, t2 with probability .25, and t3 with probability .5. After player 1 learns
his type, players 1 and 2 move simultaneously with player 1 choosing between a and b
and player 2 between x, y, and z. Find all of the BNE.

Nature

Type t1 Type t3
Probability = .25 Type t2 Probability = .5
Probability = .25

1 1 1

a b a b a b

x y z x y z x y z x y z x y z x y z

Player 1 2 1 4 1 2 3 3 1 21 2 3 2 4 1 5 2 4
Player 2 1 3 4 5 2 3 0 4 22 2 0 1 3 2 5 2 0

ANSWER: Consider a strategy profile that has player 2 choose x. When he is type
t1, player 1’s optimal action is a as it delivers a payoff of 2 (given player 2 is choos-
ing x) compared to 1 from action b. When he is type t2, player 1’s optimal action
is a, and when he is type t3 then player 1’s optimal action is b. Thus, given player 2’s
strategy is x, player 1’s optimal strategy is a/a/b. Given player 1’s strategy is a/a/b,
player 2’s expected payoff from x is

## p1  1  p2  0  (1  p1  p2 )  5  5  4p1  5p2  2.75,

from y is

p1  3  p2  4  (1  p1  p2 )  2  2  p1  2p2  2.75,

and from z is

## p1  4  p2  2  (1  p1  p2 )  0  4p1  2p2  1.5.

Hence, x is optimal. We conclude that the strategy pair a/a/b and x is a BNE.
Consider a strategy profile that has player 2 choose y. When he is type t1, player 1’s
optimal action is b as it delivers a payoff of 2 (given player 2 is choosing y) com-
pared to 1 from action a. When he is type t2, player 1’s optimal action is b, and
when he is type t3 then player 1’s optimal action is a. Thus, player 1’s optimal strat-
egy is b/b/a. Given that strategy for player 1, player 2’s expected payoff from x is

p1  5  p2  2  (1  p1  p2 )  1  1  4p1  p2  2.25,
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-23

from y is

p1  2  p2  2  (1  p1  p2 )  3  3  p1  p2  2.5,

and from z is

p1  3  p2  0  (1  p1  p2 )  2  2  p1  2p2  1.75.

## y is optimal and we conclude that b/b/a and y is a BNE.

Finally, consider a strategy profile that has player 2 choose z. When he is t1,
player 1’s optimal action is a, when he is t2 his optimal action is b, and when he is
t3 his optimal action is b, resulting in strategy a/b/b. Given that strategy for player 1,
player 2’s expected payoff from playing z is

p1  4  p2  0  (1  p1  p2 )  0  1,

## which is lower than her expected payoff from playing x:

p1  1  p2  2  (1  p1  p2 )  5  3.25.

## We conclude that there is no BNE that has player 2 using strategy z.

18. Consider the game on the bottom of page 395. Nature chooses player 1’s type which is L
with probability 1/2 and H with probability 1/2. After player 1 learns his type, player 1
chooses between actions a and b. Not knowing player 1’s type or action, player 2 chooses
between c and d. Player 3 observes player 2’s choice but does not observe player 1’s type
or action. Player 3 chooses between x and y. Find all BNE.

Nature

Type L Type H
Probability = 1
2 Probability = 1
2

1 1

a b a b

c d c d c d c d

x y x y x y x y x y x y x y x y

Player 1 1 3 4 1 2 1 3 0 2 5 2 3 1 4 3 1
Player 2 2 2 0 1 2 1 0 0 4 1 2 6 3 1 2 5
Player 3 3 2 1 2 4 1 3 4 1 0 2 5 5 4 1 2
10-24 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

ANSWER: To find some BNE, we’ll specify strategies for players 1 and 2 and
then derive player 3’s optimal strategy. Given that optimal strategy for player 3,
we’ll then derive player 2’s optimal strategy. If what is derived is the same as the
strategy originally specified for player 2 then we’ll continue. (Otherwise, the
strategies specified for players 1 and 2 are not part of a BNE.) Given the strate-
gies of players 2 and 3, we’ll then derive player 1’s optimal strategy. If what is
derived is the same as the strategy originally specified for player 1 then this
strategy profile is a BNE.
First note that if player 2 chooses c then x delivers a higher payoff for player 3
than y for any type and action of player 1. Hence, if a strategy profile has player 2
choose c then BNE requires that player 3 choose x. If player 2 chooses d then y deliv-
ers a higher payoff for player 3 than x for any type and action of player 1. Hence, if
a strategy profile has player 2 choose d then BNE requires that player 3 choose y.
Consider strategy a/a for player 1. If player 2 chooses c (and, therefore, player 3
chooses x), then her expected payoff is

p  2  (1  p)  4  3

If player 2 chooses d (and, therefore, player 3 chooses y), then her expected payoff is
7
p  1  (1  p)  6 
2
So d is optimal for player 2. Given that players 2 and 3 choose d and y, respectively,
player 1’s optimal action when she is type L is a, which gives her payoff of 1 instead
of 0; and her optimal action when she is type H is also a (giving her a payoff of
3 rather than 1 from b). Hence, a/a is indeed the optimal stratgey for player 1. One
BNE is then (a/a, d, x/y) which means that player 1 chooses a whether she is type
L or H, player 2 chooses d, and player 3 chooses x when player 2 chose c and
chooses y when player 2 chose d.
Consider strategy b/a for player 1. If player 2 chooses c (and, therefore, player 3
chooses x), then her expected payoff is

p  2  (1  p)  4  3

If player 2 chooses d (and, therefore, player 3 chooses y), then her expected payoff is

p  0  (1  p)  6  3

Player 2 is then indifferent. If she chooses d (and player 3 chooses y), then player
1’s optimal strategy is a/a, and we don’t have a BNE because we originally speci-
fied player 1’s strategy to be b/a. If instead she chooses c (and player 3 chooses x),
then player 1’s optimal strategy is b/a. This gives us a second BNE of (b/a, c, x/y).
Consider strategy a/b for player 1. If player 2 chooses c (and player 3 chooses x),
then her expected payoff is
5
p  2  (1  p)  3 
2
If player 2 chooses d (and player 3 chooses y), then her payoff is

p  1  (1  p)  5  3

So d is optimal. Given that players 2 and 3 choose d and y, player 1’s optimal strat-
egy is a/a. Thus, there is no BNE involving player 1 using strategy a/b.
Finally, consider strategy b/b for player 1. If player 2 chooses c (and player 3
chooses x), then her expected payoff is

5
p  2  (1  p)  3 
2
If player 2 chooses d (and player 3 chooses y), then her expected payoff is

5
p  0  (1  p)  5 
2
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-25

So player 2 is indifferent. If she chooses d (and player 3 chooses y), then player 1’s
optimal strategy is a/a, and we don’t have a BNE. If instead she chooses c (and
player 3 chooses x), then player 1’s optimal strategy is b/a, and we still don’t have
a BNE. Hence, there is no BNE that has player 1 choose strategy b/b.

19. Dan and Kristin had their first date last night and each is deciding whether to call the
other. Each wants to call only if they both are interested. Dan is then of two types: he likes
Kristin (which Nature chooses with probability d) or he does not (with probability 1  d).
Similarly, Kristin is of two types: she likes Dan (with probability k) or does not (probabil-
ity 1  k). Assume 0 k 1 and 0 d 1. To simplify matters (without any loss of
generality), assume that Dan (Kristin) has a choice of whether or not to call only when
he (she) likes Kristin (Dan). If a person does not like the other then he or she has no
decision. The Bayesian game is shown below. (As a side note, which you can ignore if
you don’t understand, d is the probability that Dan likes Kristin conditional on Kristin
liking Dan, and k is the probability that Kristin likes Dan conditional on Dan liking
Kristin. Got it? If not, press the IGNORE button.)

Nature

## Dan likes Dan dislikes

Kristin likes Kristin dislikes
Prob = dk Prob = (1ⴚd)(1ⴚk)
Dan dislikes
Dan likes
Kristin likes
Kristin dislikes
Prob = (1ⴚd)k
Prob = d(1ⴚk)
Dan
0
0
Do
Call not 5 0
call
1 0
Kristin
Do Do Do
Call not Call not Call not
call call call
Dan 4 4 4 0 1 0
Kristin 3 3 3 0 2 0

a. When is it a BNE for Dan to call (when he likes Kristin) but Kristin not to call (when
she likes Dan)?

ANSWER: Given that Dan will call if he is interested, Kristin’s strategy is opti-
mal because she only wants to speak with Dan when they both like each other,
in which case Dan will call anyway and she gets a payoff of 3 regardless of
whether she calls or not. If you don’t feel comfortable with this reasoning, let’s
do the math. When Kristin likes Dan and given that Dan will call if he likes
Kristin, Kristin’s expected payoff from calling is d  3  (1  d)(2) while the
expected payoff from not calling is d  3  (1  d)  0. Hence, not calling is
optimal for her.
On the other hand, given that Kristin will not call, the expected payoff Dan gets
from calling is k  4  (1  k)  (5), and the payoff from not calling is 0. His
strategy is optimal if and only if

5
k  4  (1  k)  (5)  0 1 4k  5  5k  0 1 k 
9

## b. When is it a BNE for Kristin to call but not Dan?

10-26 SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION

ANSWER: Similar as in (a), given that Kristin will call if she is interested, Dan’s
strategy is optimal because he only wants to speak with her if both are interested.
On the other hand, given that Dan will not call, the expected payoff to Kristin if
she calls is d  3  (1  d)  (2), and the payoff of not calling is 0. So her strat-
egy is optimal if and only if
2
d  3  (1  d)  (2)  0 1 3d  2  2d  0 1 d 
5

## c. When is it a BNE for both to call?

ANSWER: Never because if the other person is expected to call when interested
then one is better off not calling and avoiding the unpleasant outcome of calling
when the other person is not interested. For the math, suppose that Kristin
will call if she likes Dan. If Dan likes Kristin then his expected payoff from calling
is k  4  (1  k)  (5) and the expected payoff from not calling is
k  4  (1  k)  0. Hence, it is optimal for him not to call. Conversely, it is also
the case that not calling is optimal for Kristin given that Dan is going to call. There
is then no BNE in which both call.

## d. When is it a BNE for neither to call?

5
ANSWER: By the analysis in (a), when k  9, if Kristin chooses not to call, then it
5
is optimal for Dan to call. When k  if Kristin chooses not to call, then it is optimal
9
,
2
for Dan to not call as well. Similarly by the analysis in (b), when d  5 , if Dan
2
chooses not to call then it is optimal for Kristin to call. When d  5 , if Dan chooses
5 2
not to call then it is optimal for Kristin to not call as well. Hence, k  9 and d  5
are necessary and sufficient for there to exist a BNE in which neither person calls.

## e. When is it a BNE for them to randomize? Find the mixed-strategy BNE.

ANSWER: Let x denote the probability that Dan calls and y denote the probability
that Kristin calls. Dan is indifferent between calling and not calling if

## Call: k  4  (1  k)  (5)  k  y  4: Do not call

9k  5
y
4k

9k  5
Since we are looking for a BNE in which they randomize, we want 0 y 1. 4k 1
9k  5 5
when k 1 and 4k  0 when k  9 . Kristin is indifferent between calling and
not calling if

## Call: d  3  (1  d)  (2)  d  x  3: Do not call

5d  2
x
3d
5d  2 5d  2 2
We want 0 x 1. 3d
1 when d 1 and 3d
 0 when d  5 . In sum, if
5 2 5d  2
k  and d  5 then it is a BNE for Dan to call with probability 3d when he is
9
9k  5
interested and for Kristin to call with probability 4k when she is interested.

20. Consider a market with two firms that are competing in quantities (as modeled in the
Appendix to Chapter 6). Inverse market demand is P  1  q1  q2 where qi is the quan-
tity of firm i. Each firm has a constant marginal cost of producing. Firm 1’s marginal
cost is zero so its profit is (1  q1  q2)q1. Firm 2’s marginal cost is c so its profit is
(1  q1  q2)q2  cq2. c is private information to firm 2 and equals 0 with probability 1/2
1
and equals 1/4 with probability 2 . Firm 2 is then fully informed, while firm 1 does not
know firm 2’s marginal cost. Nature chooses the value for c and then firms simultane-
ously choose quantities. Find the BNE.
SOLUTIONS MANUAL CHAPTER 10: I KNOW SOMETHING YOU DON’T KNOW: GAMES WITH PRIVATE INFORMATION 10-27

ANSWER: Let us begin examining optimal behavior for firm 2 who is the
informed player. Firm 2’s strategy consists of a quantity for the low cost senario
and a quantity for the high cost senario: (qL2 , qH2 ). When firm 2 has low cost, its
profit is (1  q1  qL2 )qL2 . Differentiating it with respect to qL2 , firm 2’s best
response function is derived (when its cost is low):

1  q1
qL2 
2

## When firm 2 has high cost, its profit is (1  q1  qH H H

2 )q2  q2 /4, which implies its
best response function is

3/4  q1
qH
2 
2

Next we look at firm 1, whose strategy is a single quantity because firm 1 does not
know whether firm 2 has low or high cost. Its expected profit is

1 1
(1  q1  qL2 )q1  (1  q1  qH
2 )q1.
2 2

Using the first-order condition, we can derive firm 1’s best response function:

1  1/2 qH L
2  1/2 q2
q1 
2

The BNE is solved by finding the values for the three quantities that satisfy the
three best response functions (firm 1’s, firm 2’s when it is low cost, and firm 2’s
when it is high cost). Plugging the best responses of firm 2 into that of firm 1’s best
3 3
response function, we have q1  8 . Plugging q1  8 into the two best response
5 3
functions of firm 2, we have qL2  16 and qH
2  16 .

## Menu de pied de page

### Obtenez nos applications gratuites

Droits d'auteur © 2022 Scribd Inc.

## Menu de pied de page

### Obtenez nos applications gratuites

Droits d'auteur © 2022 Scribd Inc.