Economics 203

Core Economics

Suggested Solutions to Problem Bank1 2

1. (PS1, Q1)

a) In the diagram, left branches mean heads and right branches mean tails.
N

1/3 1/3 1/3

1 2 1
!"#$%&'()*
!
2 2 1 1 2 2
*-. *-. *-.
" # "
/ 0 (2,-2) (-2, 2) (-2,2) 1(2,-2) (-2,2) %
(2,-2) (2,-2) (-2,2) (2,-2) /
(-2,2) (-2,2) 0
(2,-2)

# # " " # #
b) Each player
1 % has 4 information
1 % /sets,
0 so strategies
/ 0 are 4-tuples.
1 % 1 %
S1 = {H, T } × {H, T } × {H, T } × {H, T }, which contains 16 elements.
S2 = {h, t} × {h, t} × {h, t} × {h, t}, which contains 16 elements.
+23+ 3+2)+ 3+2)+ +23+ +23+ 3+2)+ 3+2)+ +23+ +23+ 3+2)+ 3+2)+ +23+
2. (PS1, Q2)

a) The extensive form is:

!"#$%&'()+
"

/ 7 0

! !
*-+ *-+
# # #
*-+ *-+
8 9 : 8 ; 8 ;

.2+ 425 .2* *26 *2* 52+ *2* +25 ,2.

1
Credit for many of the solutions is due to Doug Bernheim, and a series of previous TAs, including Bobby Pakzad-
!"#$%&'(),
Hurson, Isaac Opper, Josh Mollner, Jeff Naecker, Daniel Fragiadakis, Marissa Beck, Tomas Rodriguez Barraquer,
Ned Augenblick, Juuso Toikka, Xioachen Fen, Nageeb Ali, Peter Coles, Mark Israel, Luis Rayo, Nese Yildiz, and
others. "
2
Last updated February 13, 2015.
< #-+ #
# # #
1
! " ! " ! "
 b) S1 = {H, M, T }
S2 = {(U, U ), (U, B), (D, U ), (D, B), (S, U ), (S, B)}
c) Note that for some of the entries in the normal form, nature plays a role, and so we must
take the expected value given nature’s randomization. For example, if 1 plays H and 2
plays U U , then payoffs are 21 (3, 2) + 12 (9, 6) = (6, 4). The full normal form is here:

UU UB DU DB SU SB
H (6, 4) (6, 4) (3, 1.5) (3, 1.5) (2, 3.5) (2, 3.5)
M (1, 1) (6, 2) (1, 1) (6, 2) (1, 1) (6, 2)
T (2.5, 2) (3, 4.5) (2.5, 2) (3, 4.5) (2.5, 2) (3, 4.5)

3. (PS1, Q4)
A strategy
QN includes one action from each information set. Thus, the number of strategies is
just n=1 Mn .

4. (PS1, Q3)

a) The extensive form is:

b) S1 = {0, v/2, v}
S2 = {(a, a, a), (a, a, r), (a, r, a), (a, r, r), (r, a, a), (r, a, r), (r, r, a), (r, r, r)}
aaa aar ara arr raa rar rra rrr
0 0, v 0, v 0, v 0, v 0, 0 0, 0 0, 0 0, 0
c)
v/2 v/2, v/2 v/2, v/2 0, 0 0, 0 v/2, v/2 v/2, v/2 0, 0 0, 0
v v, 0 0, 0 v, 0 0, 0 v, 0 0, 0 v, 0 0, 0
d) No strategies are strictly dominated. One way to verify this is to note that every strategy
is a best response to some opponent’s strategy.
e) For player 1, 0 is weakly dominated by v/2 and v. For player 2, (a, a, a) and (a, a, r)
weakly dominate all other strategies.

5. (PS1, Q5)

a) For player 1, E is strictly dominated by A. For player 2, a is strictly dominated by any

mixture of b and e, or d and e. The only weakly dominated strategies are the strictly
dominated ones.
b) Iterative deletion leaves (A, C, D) for player 1 and (b, c, e) for player 2, so the game is
not dominance solvable.
c) We only need to check the strategies that survive ISD. (C, c) is the unique PSNE.

2
 d) In two player games, the set of strategies that survive iterative deletion is the same as
the set of rationalizable strategies. Thus, the answer is the same as in (b).

6. (PS1, Q6)
The question does not specify whether bids have to be non-negative; making either assumption
is fine. Let us assume that S1 = S2 = [0, ∞), but if one made the alternative assumption,
the answers should not be too different.

a) No strategies are strictly dominated. Observe that if the opponent j makes demand
xj > 100, then all demands xi yield payoff 0. Thus, no strategy xi can do strictly worse
than some other strategy for every choice of opponent strategy.
b) xi = 0 is weakly dominated by any xi ∈ (0, 100]. xi > 100 is also weakly dominated by
any xi ∈ (0, 100]. For any xi ∈ (0, 100], xi is a strict best response to 100 − xi , and thus
cannot be weakly dominated.
c) We can solve by graphing best response functions, or by inspection:
(x, 100 − x) is a NE for any x ∈ [0, 100].
We also have equilibria of the form (x1 , x2 ) where x1 , x2 ≥ 100.

7. (PS2, Q2)
Let n denote a node that is reached in equilibrium. Playing c at node n can be a best response
only if the opponent plays c at node n + 1 with positive probability. Thus, if in equilibrium c
is played at node 1, then it must be played at every node with positive probability. But this
clearly is not an equilibrium, as playing c at the last node is not a best response. Therefore
c cannot be played in equilibrium at node 1.
Note 1: It is not hard to extend this result to mixed strategies.
Note 2: It is important to distinguish between Nash equilibria and equilibria outcomes here.
There are many pure strategy NE – any profile where both players say stop at their respective
first nodes – but there is a unique equilibrium outcome where stop is played at the first node.
Note 3: In effect, we have shown a stronger result: viz., in any rationalizable strategy profile,
player 1 says stop at the root node.

8. (PS2, Q3)
Let s∗ be the unique strategy profile that survives iterated deletion of strictly dominated
strategies.
Step 1: No ρ0 6= s∗ is a NE (where ρ0 can be a mixed strategy equilibrium).
There are many ways to prove this. An indirect (but elementary) approach would recognize
that a NE is a strategy-profile that necessarily lies within the set of rationalizable strategies.
Since that set is a subset of those remaining after iterated elimination of strictly dominated
strategies, a NE could not put positive weight on a strategy deleted by ISD. Since dominance-
solvability implies that ISD leads to a unique strategy profile, s∗ is the only strategy profile
that could be a NE.
One can also supply a direct proof: let s0i denote the first pure strategy in the support of ρ0
which was deleted, and suppose this took place in round k. Then there exists a randomization
ρi over i’s remaining strategies at the start of round k that beats s0i for any s−i that remains
at the start of round k. In particular, because ρ0−i remains at the start of round k, we have
gi (ρi , ρ0−i ) > gi (s0i , ρ0−i ), so ρ0 cannot be a NE.

3
 Step 2: s∗ is a NE.
Suppose not. Then there exist i and ŝi ∈ arg maxsi gi (si , s∗−i ) such that gi (ŝi , s∗−i ) > gi (s∗i , s∗−i ).
But this implies that ŝi 6= s∗i could not have been deleted, a contradiction.
Note: we can relax the restriction that strategy sets be finite if we include an assumption
about existence of best responses:
Assumption 1 (A1) arg maxsi gi (si , s−i ) 6= ∅ for all i, s−i .
Claim Any dominance solvable game that satisfies (A1) has a unique NE which coincides
with the iterative dominance solution s∗ .

9. (PS2, Q1)
Strategy sets are S1 = S2 = S3 = [0, 1]. Let p be the price of ice cream fixed by the
regulator and πi designate payoffs for i=1,2,3. Note that in both parts of the question, the
best-response function is not always defined because of discontinuities (as in the Bertrand
game).

a) There are 3 cases to check:

Case 1: s1 = s2 = 12 . Here π1 = π2 = p2 , and any deviation yields payoff < p2 . Thus,
( 21 , 12 ) is a NE.
Case 2: s1 = s2 6= 21 . Then π1 = π2 = p2 . However, if i deviates towards 12 by a small
amount, he secures more than half of the market and a payoff > p2 . Thus, this is not a
NE.
Case 3: s1 6= s2 . This is clearly not a NE as either player can increase market share by
moving towards the other.
b) We will show that none of the 4 possibilities are Nash equilibria.
Case 1: s1 = s2 = s3 = 21 . Here π1 = π2 = π3 = p3 . By deviating to si + ε, i secures
(1−ε)p
2 > p3 . Thus, this is not a NE.
Case 2: s1 = s2 = s3 6= 21 . Again π1 = π2 = π3 = p3 . By moving ε closer to 12 , i secures
more than half of the market and a payoff > p2 . Thus, this is not a NE.
Case 3: si = sj 6= sk . Here k can clearly gain by moving closer to i and j. Thus, this is
not a NE.
Case 4: si < sj < sk . Again k can gain by moving closer to i and j. Thus, this is not a
NE.
Theses cases are exhaustive, so no pure strategy NE exists!

10. (PS2, Q4)

We begin by converting the problem to an equivalent one where strategy spaces are compact.
Claim: There exists q̄ such that P (q) < C 0 (q) for all q > q̄.
This follows because P (·) is decreasing and concave and C 0 (·) is increasing.
Now observe that any strategy q > q̄ is strictly dominated by q̄. Thus, we can throw out the
strictly dominated strategies and be sure that the set of equilibria is identical to that in the
original game. We now proceed in 4 steps:

4
a) There exists a pure strategy NE.
We verify the conditions of the existence theorem given in the notes:
- Strategy spaces are compact and convex by construction.
- Payoff function gi (q1 , q2 ...qN ) = P (Q)qi − C(qi ) is continuous.
- Payoff function is gi (q1 , q2 ...qN ) is concave (thus quasi-concave) in qi :

∂ 2 gi
= qi P 00 (Q) + 2P 0 (Q) − C 00 (qi ) < 0.
∂qi2 | {z } | {z } | {z }
<0 <0 >0

Thus, there exists a pure strategy NE.

b) Any pure strategy NE q must be symmetric.
P P
Let Q = qi , and Q−i =. j6=i qj .
i. q interior ⇒ q symmetric:
We first establish that for a particular Q−i , there exists a unique qi that is a best
response to it. qi∗ is a best response to Q−i if:

qi∗ P 0 (qi∗ + Q−i ) + P (qi∗ + Q−i ) − C 0 (qi∗ ) = 0.

Since the LHS is decreasing in qi∗ (by concavity as shown above), we are guaranteed
a unique best response. Let q ∗ (Q−i ) = arg maxθ P (θ + Q−i ) − C(θ).
Calculating the slope of the best response function, we see that
∂q ∗ P 0 (q ∗ + Q−i ) + q ∗ P 00 (q ∗ + Q−i )
0> =− 0 ∗ > −1
∂Q−i 2P (q + Q−i ) + q ∗ P 00 (q ∗ + Q−i ) − C 00 (q)
If there were an asymmetric equilibrium, then there exist i, j such that qi 6= qj . Then
q −q
qi +Q−i = qj +Q−j , and hence, Q−ii −Qj−j = −1. But qi = q ∗ (Q−i ) and qj = q ∗ (Q−j )
−i q ∗ (Q
−j )−q ∗ (Q )
since i and j are playing best responses. Therefore Q−i −Q−j = −1, which
by the Mean Value Theorem contradicts our earlier slope calculation. Hence, all
solutions must be symmetric.
ii. q not interior ⇒ q = (0, 0, ...0) (and thus symmetric):
q not interior ⇒ qi = 0 for some i. Then from i0 s FOC, we have P (Q) − C 0 (0) ≤ 0.
But for qj > 0,

qj P 0 (Q) + P (Q) − C 0 (qj ) < P (Q) − C 0 (0) < 0,

so no firm j can choose qj > 0 in equilibrium.

c) The symmetric equilibrium q is unique.
We know that the slope of the best-response function, q, is negative. Suppose there is
0
another symmetric equilibrium q 0 with WLOG, N qi = Q0 > Q = N qi .
But
N −1 0 N −1
Q0 > Q ⇒ Q0−i = Q > Q = Q−i
N N
⇒ qi0 = q ∗ (Q0−i ) < q ∗ (Q−i ) = qi
⇒ Q0 < Q,

a contradiction.

5
 d) No mixed strategy equilibrium exists.
This follows from the concavity of the objective
P function. Suppose in equilibrium i’s
opponents are mixing so that the sum qj6=i takes on N values (generalization to a
continuous distribution is straightforward), with probabilities α1 , ..., αN . Then i0 s choice
of qi solves
XN  X 
max αk qi P (qi + qj6k=i ) − C(qi ) .
qi ∈[0,q̄]
k=1
The objective is a weighted average of strictly concave functions and is thus strictly
concave. The choice set is compact. Thus, there is a unique optimum, so player i will
never mix!

11. (PS2, Q5)

The technique in this problem is to find pi (p), i’s best response to all other firms naming p,
and then to impose symmetry (finding p such that pi (p) = p).
Let πi (pi , p) be the payoff to firm i when i names pi and others name p. To calculate i’s
market share, we need to find the distance d to the consumer who is indifferent between and
i and the next firm over.
Now d is defined by
1
v − pi − td2 = v − p − t( − d)2
J
J 1
which yields d = 2t (p − pi ) + 2J .
We then have  
J 1
πi (pi , p) = (pi − c)2d = (pi − c) (p − pi ) + .
t J
i’s maximization problem yields FOC:
J 1 J
(p − pi ) + + (pi − c)(− ) = 0.
t J t
t
In the symmetric Nash equilibrium, p = pi , so the FOC reduces to p = J2
+ c. Since all firms
name the same price, market share is 1/J for each firm.
Note: it’s easy to check that when all other firms name p = Jt2 + c, a firm has no profitable
deviation to a price low enough to grab all customers between him and the next firm and
beyond.
As J grows large, pi → c and qi → 0. With many firms, i has arbitrarily close neighbors and
we approach the Bertrand outcome with no differentiation.
As t → 0, pi → c but qi → 1/J. As transportation costs approach zero, we have effectively
no product differentiation and we approach the Bertrand outcome.
12. (June 2001 Comp, Q5)
Note that the strategy set for i is Si = {Trade always, Trade iff wi , Not trade iff wi , Never
trade}. With that, we can write the normal form:
1/2 Trade always Trade iff w2 Not trade iff w2 Never trade
Trade always 4 − 9p, 4 − 9p −2p, p 4 − 7p, 4 − 10p 0, 0
Trade iff w1 p, −2p 0, 0 p, −2p 0, 0
Not trade iff w1 4 − 10p, 4 − 7p −2p, p 4 − 8p, 4 − 8p 0, 0
Never trade 0, 0 0, 0 0, 0 0, 0

6
 a) Since s−i = Never trade results in a 0 payoff always, i is indifferent among all possible
strategies. Therefore, there are no strictly dominated strategies. The set of weakly
dominated strategies for player i is given by {Not trade iff wi , Never trade}.
b) Once we delete weakly dominated strategies, the game reduces to:
1/2 Trade always Trade iff w2
Trade always 4 − 9p, 4 − 9p −2p, p
Trade iff w1 p, −2p 0, 0
From the above matrix, there are two NE:
i. (Trade always, Trade always), which gives each player a payoff of 4 − 9p. Trade
occurs in all states and the expected total surplus is 8 − 18p.
ii. (Trade iff w1 , Trade iff w2 ), which gives each player a payoff of 0. Trade never occurs
and the expected total surplus is 0.
c) From the above matrix, there is only one equilibrium: (Trade iff w1 , Trade iff w2 ), which
gives each player a payoff of 0. Trade never occurs, and total surplus is 0.
d) The socially optimal strategy profile is (Not trade iff w1 , Not trade iff w2 ). Under such
strategies, trade occurs iff the state is ω3 , so that total surplus is 8−16p. These strategies
are not an equilibrium because each player would deviate to {Trade always}, given the
opponent’s strategy. That is, the socially optimal profile involves players not trading in
a state under which they would profit from trade: it requires an “act of honesty” when
there is an incentive to “lie.”

13. (Sept 2001 Comp, Q5)

a) First, we show that (wi , fi ) where wi > 0 and fi > 0 is dominated. To do this, notice
that if we decrease both wi and fi by a little bit, we leave b unchanged, and the rest of
the utility function actually increases. Formally, consider (w̃i , f˜i ) such that w̃i = wi − ε
and f˜i = fi − ε. But then note that
1
ui (w̃i , f˜i , wj , fj ) = 2 min{w̃i + wj − f˜i − fj , 500} − (w̃i + wj ) + (f˜i + fj )
10
1 1
= 2 min{wi − ε + wj − fi + ε − fj , 500} − (wi + wj ) + ε + (fi + fj ) − ε
10 10
9
= ui (wi , fi , wj , fj ) + ε
10
> ui (wi , fi , wj , fj ).

Thus any (wi , fi ) >> 0 is dominated by (w̃i , f˜i ) such that w̃i = wi − ε and f˜i = fi − ε
for some ε > 0.
Before proceeding with the rest of the problem, it is now useful to write down and draw
the best response functions for the players. In general, this would be really hard, as
each player’s strategies are 2-dimensional, so drawing a best-response would require 4
dimensions. But the previous part tells us that in fact each player’s strategy will be
either of the form (wi , 0) or of the form (0, fi ), both of which are one-dimensional! This
allows us to draw the best responses fairly easily.
Suppose j is playing (wj , 0). What is i’s best response? He could play (wi , 0) for some
wi ≥ 0, in which case his payoffs would be as follows:

wi + wj if wi + wj ≤ 500
ui (wi , wj ) = 2 min{wi + wj , 500} − (wi + wj ) =
1000 − (wi + wj ) if wi + wj ≥ 500

7
Note that in the first case i wants to increase wi , whereas in the second case, he wants to
decrease wi . This means that for any wj , the optimal wi = 500 − wj , or as close to that
as possible. Thus for wj ≤ 500, wi = wj − 500 for a payoff of 500, whereas for wj ≥ 500,
wi = 0 for a payoff of 1000 − wj ≤ 500.
But this is not quite yet i’s best response function, because i could also play (0, fi ) for
some fi ≥ 0, in which case his payoffs would be as follows:

wj − 19

1 10 fi if wj − fi ≤ 500
ui (fi , wj ) = 2 min{wj − fi , 500} − wj + fj = 1
10 1000 − wj + 10 fi if wj − fi ≥ 500

By similar logic as above, the optimal fi = wj −500, or as close to that as possible. Thus
for wj ≤ 500, i should set fi = 0 for a payoff of wj , whereas for wj ≥ 500, he should set
9
fi = wj − 500 for a payoff of 950 − 10 wj ≥ 1000 − wj .
Now we can fully describe i’s best response function, given that j is playing (wj , 0):

(500 − wj , 0) if wj ≤ 500
BRi (wj ) =
(0, wj − 500) if wj ≥ 500

Next, suppose that j is instead playing (0, fj ). If i chooses to play (wi , 0), his payoff will
be as follows:
wi − 19

9 10 fj if wi − fj ≤ 500
ui (wi , fj ) = 2 min{wi − fj , 500} − wi + fj = 1
10 1000 − wi + 10 fj if wi − fj ≥ 500

By the logic we’ve already seen a couple times, if i is playing (wi , 0), his best option is
9
play wi = 500 + fj for a payoff of 500 − 10 fj if fj ≤ 1100. He should play wi = 1600 for
19
a payoff of 1600 − 10 fj if fj ≥ 1100.
If i instead responds by playing (0, fi ), his utility is
1 1
ui (fi , fj ) = 2 min{−fi − fj , 500} + fi + fj
10 10
19 19
=− fi − fj .
10 10
Clearly in this case i wants to minimize fi . So, if i is playing (0, fi ), his best option is
19
fi = 0 for a payoff of − 10 fj .
Now we can write down i’s best response to (0, fj ):

(500 + fj , 0) if fj ≤ 1100
BRi (fj ) =
(1600, 0) if fj ≥ 1100

We can now immediately see that the following strategies for i are not dominated, since
they are best responses:
ˆ (wi , 0) for any wi ∈ [0, 500] (best response to wj = 500 − wi ),
ˆ (wi , 0) for any wi ∈ [500, 1600] (best response to fj = wi − 500), and
ˆ (0, fi ) for any fi ∈ [0, 1100] (best response to wj = fi + 500).
The only strategies remaining to be considered are (0, fi ) for fi > 1100. The work we’ve
done for far suggests that any such strategy can be improved upon by lowering fi . And

8
 in fact, (0, 1100) dominates (0, fi ) for fi > 1100:
1 1
ui (0, fi > 1100) = 2 min{−fi + wj − fj , 500} − wj + fi + fj
10 10
1 1
= −2fi + 2wj − 2fj − wj + fi + fj
10 10
1 1
< −2 ∗ 1100 + 2wj − 2fj − wj + 1100 + fj
10 10
= ui (0, 1100)

Thus (0, fi ) for fi > 1100 is dominated.

If we draw the best response functions, we can also see that we have the following PSNE:
ˆ ((wi , 0), (wj , 0)) for any wi , wj such that wi + wj = 500, and
ˆ ((wi , 0), (0, fj )) where wi ∈ [500, 1600] and fj = wi − 500.
We can also verify that these are PSNE by inspecting the best response functions we’ve
written down. In the first NE, for example, i is best responding to j because wj ≤ 500
and wi = 500 − wj as dictated by i’s best response. The situation for j is symmetric, so
she is best responding as well.
Finally3 , it is clear that the NE strategies are exactly the strategies remaining after
IDDS, since all other strategies are dominated. More precisely, the strategies surviving
IDDS are
ˆ (wi , 0) for any wi ∈ [0, 1600] and
ˆ (0, fi ) for any fi ∈ [0, 1100]
for i = 1, 2.
b) We are going to show that there is in fact a unique strategy profile that survives IDDS,
and thus a unique PSNE. The general logic is that the the best response functions will
tell us which strategies might be dominated (those that are not a best response), and
then we will directly prove that these strategies are in fact dominated.
It is useful to note that player 1’s best-response function is the one we constructed
in the previous part. For player 2, we replace 500 with 500 + ε, so we have that her
best-response function is given by the following if 1 plays (w1 , 0):

(500 + ε − w1 , 0) if w1 ≤ 500 + ε
BR2 (w1 ) =
(0, w1 − 500 − ε) if w1 ≥ 500 + ε

If 1 instead plays (0, f1 ), then 2’s best response is as follows:


(500 + ε + f1 , 0) if f1 ≤ 1100 − ε
BR2 (f1 ) =
(1600, 0) if f1 ≥ 1100 − ε

We start by eliminating dominated strategies. For both players, this eliminates (wi , fi ) >>
0 as in the previous part. For player 1, (0, f1 > 1100) is also eliminated, also as above,
since nothing changes in the arguments in the previous part. For player 2, if we follow
the logic in the previous part, we can show that (0, f2 > 1100 − ε) is dominated. Thus
the strategies remaining are as follows:
3
Note that I am not answering the questions in the same order they are asked. Because of the relationship between
solution concepts, this is often an effective test strategy.

9
 (w1 , 0) s.t. w1 ∈ [0, 1600] (w2 , 0) s.t. w2 ∈ [0, 1600]
(0, f1 ) s.t. f1 ∈ [0, 1100] (0, f2 ) s.t. f1 ∈ [0, 1100 − ε]
Next, note that we’ve eliminated part of the support of 1’s best response function. Now,
(w1 > 1600 − ε, 0) is no longer a best response to anything. And in fact, it is dominated
by (1600 − ε, 0), as the following argument shows:
1
u1 (w1 > 1600 − ε, 0) = 2 min{ w1 + w2 − f2 , 500} − w1 − w2 + f2
|{z} | {z } 10
>1600−ε >−1100+ε
1
= 1000 − w1 − w2 + f2
10
1
< 1000 − (1600 − ε) − w2 + f2
10
= u1 (1600 − ε, 0)

So, we eliminate (w1 > 1600 − ε, 0).

But now we have eliminated part of the support of 2’s best-response function, so that
(0, f2 > 1100 − 2ε) is no longer a best-response. We can show that (0, f2 > 1100 − 2ε)
is dominated by (0, 1100 − 2ε):
1 1
u2 (0, f2 > 1100 − 2ε) = 2 min{w1 − f1 + −f , 500 + ε} − w1 + f1 + f2
| {z } |{z}2 10 10
<1600−ε <−1100+2ε
1
= 2w1 − 2f1 − 2f2 − w1 − w2 + f2
10
1
< 2w1 − 2f1 − 2(1100 − 2ε) − w1 − w2 + (1100 − 2ε)
10
= u2 (0, 1100 − 2ε)

So, we eliminate (0, f2 > 1100 − 2ε). But note that we have lopped off another chunk of
the support of 1’s best-response, so we can eliminate (w1 > 1600 − 2ε, 0). Then we can
eliminate (0, f2 > 1100 − 3ε), and so on. This continues until we hit a bound, namely
f2 = 0. At this point, we have eliminated (w1 > 500 − ε, 0) and (0, f2 > 0). Thus the
strategies remaining are as follows:
(w1 , 0) s.t. w1 ∈ [0, 500 − ε] (w2 , 0) s.t. w2 ∈ [0, 1600]
(0, f1 ) s.t. f1 ∈ [0, 1100] (0, 0)
Now we see that with w1 > 500−ε eliminated, (w2 < ε, 0) is no longer a best response for
player 2. The usual arguments show that in fact (w2 < ε, 0) is dominated by (w2 = ε, 0),
so we eliminate (w2 < ε, 0). But then (w1 > 500 − ε, 0) is dominated, so we eliminate it.
We continue this process until we hit a bound, w1 = 0. At that point, we have eliminated
(w1 > 0, 0) and (w2 < 500 + ε, 0). Thus the strategies remaining are as follows:
(0, 0) (w2 , 0) s.t. w2 ∈ [500 + ε, 1600]
(0, f1 ) s.t. f1 ∈ [0, 1100] −−
Lastly, note that we’ve now removed the part of 1’s best-response that supported playing
(0, f1 > ε). We can show that strategies of this form are dominated by (0, f1 = ε), and
so are eliminated. But then we can show that (w2 < 500 + 2ε, 0) is dominated, and

10
 so is eliminated. Continuing this process until we run into a bound (w2 = 1600), we
eliminate all (0, f1 < 1100) and all (w2 < 1600, 0). The remaining strategies are then
just as (0, 1100) for 1 and (1600, 0) for 2.
Thus ((0, 1100), (1600, 0)) is the unique NE and the unique profile that survives IDDS.
We are asked to consider the correspondence from ε to PSNE profiles. Note that for
ε > 0, the set of PSNE profiles is the singleton ((0, 1100), (1600, 0)). When ε = 0,
the set of PSNE is given in part (a). But ((0, 1100), (1600, 0)) is in this set, so the
correspondence is upper hemicontinuous.

14. (Sept 2010 Comp, Q2)

a) A strategy for a politician with quality q is a choice from the set {In, Out}, where In
signifies that he enters the election, and Out signifies that he does not.
b) i. An entrant of quality q 0 earns q by staying out, and 12 q+ 21 (q 0 +s)−k = 12 (q+q 0 +s)−k
from entering. Thus, we must have
1
q ≥ (q + q 0 + s) − k
2
or
q 0 − q + s ≤ 2k
Because this statement must be true for all q 0 ∈ [0, 1], we have

1 − q + s ≤ 2k

or
q ≥ 1 − 2k + s (1)
No further conditions must be met to have a single candidate equilibrium. We do
not have to worry about the single candidate’s incentives because he has no incentive
to deviate by not entering, given that the consequences are anarchy with a payoff
of −∞.
ii. Plainly, if (1) is satisfied for any q, it is also satisfied for q = 1. Thus, a single-
candidate equilibrium exists iff

1 ≥ 1 − 2k + s

or
k 1
≥
s 2
iii. The best single-candidate equilibrium always involves quality q = 1, and is therefore
unaffected by s and k. The worst single-candidate equilibrium is given by

q = 1 − 2k + s

Therefore, an increase in compensation, s, improves the down-side, while an increase

in the costs of running for office, k, makes the down-side worse.
c) i. An entrant of quality q 0 earns q by staying out, and N
N +1 q + 1
N +1 (q
0 + s) − k from
entering. Thus, we must have
N 1
q≥ q+ (q 0 + s) − k
N +1 N +1

11
 or
q 0 − q + s ≤ (N + 1)k
Because this statement must be true for all q 0 ∈ [0, 1], we have

1 − q + s ≤ (N + 1)k

or
q ≥ 1 − (N + 1)k + s (2)
ii. Let q be the average quality of all the candidates, and q 00 be the average quality
of the remaining candidates if a candidate of quality q 0 drops out. A candidate of
quality q 0 receives a payoff of q 00 if he drops out, and a payoff of
N − 1 00 1
q + (q 0 + s) − k
N N
if he runs. Thus, he is willing to run iff
N − 1 00 1
q + (q 0 + s) − k ≥ q 00
N N
or
q 0 − q 00 + s ≥ N k
N −1 00 1 0
Clearly, q = N q + Nq . Therefore,

N 1
q 00 = q− q0,
N −1 N −1
from which we infer that the candidate is willing to run iff
N 1
q0 − q+ q0 + s ≥ N k
N −1 N −1
or
N −1
q0 − q +
s ≥ (N − 1)k
N
That inequality plainly holds for all candidates iff it holds for qmin . So we have
N −1
qmin − q + s ≥ (N − 1)k (3)
N
iii. Plainly, there exists a set of N candidates satisfying (3) with average quality q if
and only if such a set also exists for qmin = q (so that all candidates are of the same
quality). But in that case, the condition becomes simply

k 1
≤
s N
In other words, if ks ≤ N1 , then for any q we can find a group of N candidates
satisfying (3). Now we consider (2). Clearly, (2) is satisfied for some q iff it is
satisfied for q = 1. Thus, to have an N -candidate equilibrium, we also require

1 ≥ 1 − (N + 1)k + s

12
 or
k 1
≥
s N +1
Thus, the necessary and sufficient condition for the existence of an N -candidate
equilibrium is
1 k 1
≥ ≥ (4)
N s N +1
d) Clearly, either ks ≥ 21 , in which case a single-candidate equilibrium exists, or there exists
N ≥ 2 such that N1 ≥ ks ≥ N 1+1 , in which case an N -candidate equilibrium exists. Thus
we have existence, and we see that the number of candidates rises as ks falls. When
k 1
s = N for any N , there exist equilibria with both N and N − 1 candidates; otherwise,
the number of candidates is unique.
e) From (4), it is obvious that N → ∞ as k → 0 and as s → ∞.
Next consider the effect of k → 0 on the range of quality. From the preceding analysis,
it is clear that there is always an equilibrium with q = 1. We know from (2) that
q ≥ 1 − (N + 1)k + s, and we know from (4) that N ≤ ks . Thus,
s 
q ≥ 1− +1 k+s
k
= 1−k

Thus, q → 1 as k → 0: reducing entry costs is unambiguously beneficial in the limit.

Finally consider the effect of s → ∞ on the range of quality. Again, there is always
an equilibrium with q = 1. As ks declines from N1 to N 1+1 , the lowest quality in an
N -candidate equilibrium increases. So let’s focus on bad equilibria when ks = N1 (the
worst of the worst). In that case, (2) becomes

q ≥ 1 − (N + 1)k + s
= 1−k

Thus, as s gets large with k fixed, quality does not converge to 1. Instead, the lowest
quality is a “saw tooth” function of s, and at each tooth it falls to 1 − k, no matter
how large s becomes. We conclude that very low k promotes the quality of governance,
while very high s does not.

15. (June 2013 Comp Q3)

16. (PS3, Q1)

We don’t need to consider strategies eliminated by ISD4 . Thus, we consider the reduced 3 × 3
game:
b c e
A 4, 2 1, 0 2, 4
C 0, 0 3, 3 0, 1
D 2, 4 0, 0 4, 2

First note that the unique PSNE (C, c) is a special case of a MSE.
Second, note that neither player has two pure best responses to any opponent’s pure strategy,
so there will be no pure/mixed equilibria.
4
Iterative deletion of Strictly Dominated strategies

13
 We now look for equilibria where each player mixes over all 3 strategies:

Player 1 Player 2
A λ1 b µ1
C λ2 c µ2
D (1 − λ1 − λ2 ) e (1 − µ1 − µ2 )

We pin down µ1 and µ2 by requiring that 1(!) be indifferent between A, C and D.

U1 (A) = 4µ1 + µ2 + 2(1 − µ1 − µ2 )

U1 (C) = 3µ2
U1 (D) = 2µ1 + 4(1 − µ1 − µ2 ).
1 6
Equating these three yields µ1 = 11 , µ2 = 11 .
1 6
Analogously, requiring 2’s indifference yields λ1 = 11 , λ2 = 11 .
1 6 4 1 6 4
Thus, we have the MSE ( 11 , 11 , 11 ), ( 11 , 11 , 11 ).
Now we look for mixed strategies in which at least one player plays only 2 strategies. Note
the following (using dominance against the strategies in question):
If 1 plays A and C only, 2 will only play c or e;
if 1 plays C and D only, 2 will only play b or c;
if 1 plays A and D only, 2 will only play b or e.
Similarly:
If 2 plays b and c only, 1 will only play A or C;
if 2 plays c and e only, 1 will only play C or D;
if 2 plays e and b only, 1 will only play A or D.
From this, we can rule out any 2 × 3 MSE. We also see that the only potential 2 × 2 MSE are
of the form λA + (1 − λ)D, µb + (1 − µ)e.
Using the same technique as in the 3 × 3 case, we find µ = λ = 21 .
Thus, we have a third and final MSE: ( 12 , 0, 12 ), ( 12 , 0, 21 ).

17. (PS3, Q2)

a) Is there a PSNE involving A? Then 2 must best respond and play b. But then 1 must
play B.
Is there a PSNE involving B? Then 2 must best respond and play a. But then 1 must
play A.
This rules out all PSNE.
b) At a MSNE, players must be indifferent between playing all strategies in the support of
the mixture. Routine calculation shows that for indifference to hold, player 1 must play
x−z y−m
A with probability x−z+n−v and player 2 must play a with probability y−m+u−w .

18. (PS3, Q4)

We resolve indifference for consumers facing p = v by assuming the consumer buys.5
No PSNE
Observe first that naming price p = v guarantees at least the residual demand Q − (N − 1)K
5
The problem can also be solved with any other assumption

14
 and thus payoff (Q − (N − 1)K)v > 0. Naming p > v guarantees payoff 0. Thus, we rule out
all p > v.
There are two cases to consider: everyone charges the same price, or not everyone charges
the same price.
Q
Case 1: p1 = p2 = . . . = pN = p ∈ (0, v]. Here equilibrium payoff is p N for each firm. But
Q Q
then any firm can deviate to p−ε, secure market share K > N , and earn payoff (p−ε)K > p N
for small enough ε.
Case 2: Let pL and pH be the lowest and highest prices named in equilibrium with pL < pH .
But then a firm naming pL could deviate to pH − ε, maintaining a market share of K, and
earn payoff (pH − ε)K > pL K small enough ε.
Thus, we have no PSNE.
Finding the Symmetric MSE
We look for a symmetric MSE represented by distribution function F (·) where the support is
given by [p, v] with p > 0.6
Let the (expected) equilibrium payoff be given by Π. Suppose players j 6= i are mixing using
F (·). Then for i to be mixing over [p, v], i must be indifferent over every price in this range.
Observe that by naming p ∈ [p, v], with probability F (p)N −1 , i’s price will be the highest and
he will secure residual market share Q − (N − 1)K. With probability 1 − F (p)N −1 , i’s price
will not be the highest and he will secure residual market share K. Thus, expected payoff
from naming p is given by

Π(p) = p · F (p)N −1 (Q − (N − 1)K) + (1 − F (p)N −1 )K = Π.

 
(5)

We pin down Π by noting F (v) = 1 ⇒ Π(v) = v(Q − (N − 1)K) = Π.

Now we can solve equation (5) for F (p) to find
  1
(Q − (N − 1)K)v − pK N −1
F (p) = .
(Q − (N − 1)K)p − pK
(Q−(N −1)K)
Finally, we pin down p by noting that F (p) = 0 ⇒ p = K v.
That completes our characterization of the symmetric MSE.
Comparative Statics
Suppose we let N get large, adjusting K (lower) so that the assumptions hold. Observe that
p = (Q−(NK−1)K) v, where the coefficient on v is given by the ratio of residual demand to K.
Thus, even for large N , we can have p lie anywhere in the range (0, v).
However, observe that F (p) → 1 for all p ∈ [p, v], so that the equilibrium approaches the
PSNE where all firms name p. Thus, the limiting case depends crucially on how we adjust K
as N gets large.

19. (PS3, Q5)

a)
6
The proof that this is the unique symmetric MSE is a simple extension of the 2 firm case in the notes.

15
 b) The normal form of the game is given by

r y d
aa 0, 2 10, 0 -100,-100
an 3, 2 5, 5 -50,-45
na −3, 10 5, 5 -50,-45
nn 0, 10 0, 10 0,10

c) The unique PSNE is (nn, d).

To calculate the mixed strategy Nash equilibria, observe that although no strategies are
strictly dominated, d is weakly dominated for B. We exploit this fact.
Consider NE in which A places positive probability on aa, an or na (ie any strategy
other than all weight on nn.) In such a NE, we observe that B cannot be placing any
probability on d. It follows that nn is strictly dominated by an and that na is strictly
dominated by a mixture of aa and an. Thus, we have a reduced 2x2 matrix. Solving for
the mixed NE in the standard way yields ( 53 , 25 , 0, 0), ( 58 , 38 , 0).
Now consider NE in which A places all probability on nn. That is, A uses the pure
strategy nn. Any strategy σB for B is a best response to this. Therefore, we need
restrictions on B’s strategy such that nn is a best response to σB .
Let λr , λy , λd represent the weights on r, y and d in σB , where λr = 1 − λy − λd .

πA (aa) ≤ πA (nn) ⇒ 10λy − 100λd ≤ 0

πA (an) ≤ πA (nn) ⇒ 3λr + 5λy − 50λd ≤ 0
πA (an) ≤ πA (nn) ⇒ −3λr + 5λy − 50λd ≤ 0

Observe that if the middle inequality is satisfied, then so too are the first and third. Thus,
we have NE (0, 0, 0, 1), (1−λy −λd , λy , λd ) for all λy , λd ∈ [0, 1] satisfying 3+2λy −53λd ≤
0 and λy + λd ≤ 1.

20. (PS3, Q6)

a) Observe that since it is not immediately obvious what player 1 might play, LL, L and
R are very risky, yielding very small gains if you “guess” right, and significant losses if
you guess wrong. Thus, only M seems reasonable.
b) The PSNE are (U, LL) and (D, R). The only non-pure MSE is given by ( 51 52 U +
1 1 1
52 D, 2 LL + 2 L).
To see this, observe that since no player has multiple best responses to any opponent’s
pure strategy, there are no pure/mixed NE.
Any potential 2×3 or 2×4 MSE can be ruled out because requiring indifference between
> 2 strategies for player 2 yields an insoluble system with more equations than variables.
The 2 × 2 cases must be analyzed individually.
c) M is not a part of any of the above strategies. M is, however, rationalizable. One way
to see this is that no strategies can be removed by ISD.
d) With preplay communication, players could agree on one of the NE. The self-enforcing
nature of NE would make this agreement credible. The PSNE are payoff equivalent and
pareto dominate the 2 × 2 MSE, so we would expect agreement on either of these.

21. (June 2004 Comp, Q4)

16
 a) There is no PSNE. There are three configurations to consider:
i. Both firms quote a price: optimality implies that they must quote the same price
(otherwise the payoff for the one quoting the higher price would be −k, and a
deviation to ”no price quote” would increase its payoff). Moreover, this price p∗
must be at least c + 2k (otherwise each would earn (p∗ − c)/2 − k < 0, in which
case a deviation to ”no price quote” would increase its payoff). But then either
firm could increase its profit by slightly undercutting p∗ (earning nearly p∗ − c − k >
(p∗ − c)/2 − k ≥ 0).
ii. Only one firm quotes a price. In that case, the quoted price must be v (otherwise
the quoting firm could increase its profits by deviating to a higher price). But
then the other firm, which earns zero, could obtain a strictly positive payoff (nearly
v − c − k > 0) by quoting a price slightly below v.
iii. Neither firm quotes a price. In that case, a deviation to pi = v would yield a payoff
of v − c − k > 0, instead of 0.
b) In a mixed strategy equilibrium, a player must be indifferent between all sets of strategies
selected with strictly positive probability. Here, a firm quotes no price with strictly
positive probability. When this occurs, its payoff is zero; hence the equilibrium expected
payoff is zero. When a firm names a price less than c + k or greater than v, its payoff
is strictly negative. It cannot be indifferent between this and ”no price quote” (which
yields a payoff of zero). Consequently, it can’t make both of these choices in an MSE.
The probability that the other firm quotes price lower than p is πF (p), so the firm makes
the sale with probability 1−πF (p). In that case, it earns profits p−c. Since it incurs the
cost k regardless of whether it makes the sale, its expected payoff is (1 − πF (p)) (p−c)−k.
price inthe support of F we must have (1 − πF (p)) (p−c)−k = 0.
In light of this, for any 
1 p−c−k
This implies F (p) = π p−c . Notice that this is non-negative and strictly increasing
on [c + k, v], with F (c + k) = 0. Since the support of F is the entire interval [c + k, v],
we choose π so that F (v) = 1. This implies π = v−c−k v−c . So F is flat up to c + k, strictly
increasing between c + k and v, and flat beyond v.

22. (Sept 2004 Comp, Q4)

a) Strictly positive profits. A firm not earning strictly positive profits could gain simply by
setting its price equal to v (since it would sell to the “loyal” customers).
b) No, there isn’t a PSNE for this game. We know both firms earn strictly positive profits,
which means both must quote prices in (c, v]. If the prices are different, the firm with
the lower price could gain by increasing it. If the prices are the same, either firm could
gain by slightly undercutting the other.
c) When quoting p, the expected payoff is (p − c) [K + (1 − F (p))N ]. Setting the ex-
pression
 for expected
 payoff equal to a constant, C, and solving, we get: F(p) =
K C/K−(p−c) v−p
1− N p−c . Using F (v) = 1 gives us C/K = v − c, so F (p) = 1 − K
N p−c .
Note that this equals zero when p = NNc+Kv 00 00
+K = p . Obviously, p ∈ (c, v). So F equals
zero up to p00 (which exceeds costs), and then increases continuously, reaching a value of
unity at p = v. As K 00
N → 0, p → c, and F (p) → 1 for all p > c, so we have convergence
to the standard Bertrand outcome. As K 00
N → ∞, p → v, and F (p) → 0 for all p < c, so
we have convergence to the monopoly outcome.

23. (2005 Final, Q1)

17
Let v1 = v and v2 = v 0 > v.

a) Any bid bi > vi for Bidder i is strictly dominated by a bid bi = 0. Removing these
strategies yields a game where bidding b2 > v is strictly dominated by bidding b2 = v for
Bidder 2, given the tie-breaking assumption (namely, since Bidder 2 wins in the event
of a tie, and Bidder 1 will not bid greater than v, there is no reason for Bidder 2 to do
so). All bids bi ∈ [0, v] can be rationalized by some bid of the opponent’s, bj ∈ [0, v],
and therefore, no other strategies can be removed by iterative elimination of strictly
dominated strategies.
b) There are no PSNE. To establish this, let us consider the following profiles (b1 , b2 ) ∈
[0, v]2 to see whether there is a strictly profitable deviation:
b +b
i. bi < bj : Bidder j can improve her payoffs by decreasing her bid to i 2 j while
ensuring that she still wins the auction
ii. 0 < bi = bj : Then Bidder 2 wins the object for sure, and Bidder 1 can strictly
improve her payoffs by bidding 0 instead.
iii. bi = bj = 0: Then Bidder 1 loses the object for sure and can do strictly better by
bidding any b1 ∈ (0, v).
Observe that the preceding three cases exhaust the profiles in [0, v]2 . Since all other
profiles use strategies that are not rationalizable, this establishes that there are no PSNE.
c) By (a), all b1 > v are strictly dominated and so it could not be a best response to put
positive probability on (v, ∞). Similarly, for bidder 2, b2 > v is iteratively dominated,
and so this too could not be part of a best response.
d) i. If b2 = x, then regardless of whether she wins, Bidder 2 pays x. She wins if x ≥ b1 ,
and therefore, wins the object with probability F1 (x). Her expected payoff from
bidding x is F1 (x) v 0 − x.
ii. Observe that F1 (v) = 1. Therefore, from b2 = v, Bidder 2’s expected payoff is
v 0 − v.
iii. Since Bidder 2 will mix according to F2 only if she is indifferent between all strategies
in its support,

F1 (x) v 0 − x = v 0 − v
v0 − v + x
F1 (x) =
v0
0
Observe that F1 (0) = v v−v
0 .

iv. If Bidder 1 bids x, she wins if and only if b1 > b2 . Assuming that F2 is continuous
(we will verify this later), her payoff from b1 = x is F2 (x) v − x. Since Bidder 1
mixes over the interval [0, v], and F2 (v) = 1, her expected payoff is 0.
v. From above

F2 (x) v − x = 0
x
F2 (x) =
v
Observe that F2 (0) = 0.
v2
Rv
e) E[b1 ] = 0 vx0 dx = 2v 0.
Rv x
E[b2 ] = 0 v dx = v2 .

18
 2
v
Hence, expected revenue is 2v v 0 0
0 + 2 and is decreasing in v ; since increases in v simply

serve to depress 1’s bid but do not increase 2’s bid, the expected revenue is strictly
decreasing in v 0 .

24. (2006 Final, Q3)

a) It is strictly dominated to bid bi > v. No other strategies can be iteratively eliminated.

b) There are no PSNE.
i. Case 1: 0 ≤ bi < bj ≤ v. This cannot be an equilibrium because Student j would
like to decrease his bribe slightly, thereby getting a payoff of v − bj + ε > v − bj .
ii. Case 2: 0 ≤ bi = bj < v. This cannot be an equilibrium because Student j would
like to increase his bribe slightly, thereby getting an expected payoff of v − bj − ε
instead of 12 v − bj .
iii. Case 3: bi = bj = v. This cannot be an equilibrium because Student j would like to
change his bribe to 0, thereby getting a payoff of 0 rather than 12 v − v < 0.
c) No, since this would be strictly dominated.
d) F (bi ) v − bi
e) From (d), the payoff from a bribe of v is F (v)v − v=0 because F (v) = 1 by assumption.
f) Student i must be indifferent between all of the bribes he pays with positive probability.
Therefore, from (e), F (bi )v − bi = 0 for all bi ∈ [0, v] and F (bi ) = bvi .
Rv
g) The Social Chair expects to make 2 0 vb db = v.
h) The payoff from a bribe bi < b0 is (1 − α) G (bi ) v − bi . The payoff from a bribe b̄ is
α


2 + (1 − α) v − b̄.
i) When bi = 0, i’s payoff is 0 because G(0) = 0 by assumption.
j) Again, Student i must be indifferent between all of the bribes he pays with positive
2(v−b̄)
probability. So α2 + (1 − α) v − b̄ = 0 ⇒ α = v . Also, (1 − α) G (bi ) v − bi = 0 ⇒


bi bi
G(bi ) = (1−α)v = 2b̄−v . We know G(b0 ) = 1 ⇒ b0 = 2b̄ − v.
k) The Social Chair expects to make

! Z 2b̄−v
!

2 v − b̄ 2 v − b̄ 4 v − b̄ b̄
 
b
2b̄ − v
2 1− db + b̄ = 2b̄ − v +
v 0 2b̄ − v v v v
1
4b̄2 − 4b̄v + v 2 + 4b̄v − 4b̄2 = v


=
v
Therefore, the spending cap left the total amount of bribing unchanged.

25. (June 2010 Comp, Q1)

a) The pure strategy set is the interval [0, bmax ]. A mixed strategy is a CDF on that in-
terval – specifically, a monotonically increasing uppersemicontinuous (right-continuous)
function F with F (0) ≥ 0 and F (bmax ) = 1. The mixed strategy set is the set of all such
functions.
b) There is no pure strategy equilibrium. If there is a single high bidder, the winner could
lower his bid. If there is more than one high bidder below the upper bound, then either
player would benefit from increasing his bid slightly. If there is more than one bidder at
the upper bound, each has a negative expected payoff (since the upper bound is above
5) and would benefit from changing his bid to 0.

19
 c) We need to solve for F , α, and b0 . Because F is atomless, the probability of winning,
and hence the expected payoff, when bidding 0 is 0. Because 0 is in the support of F ,
the equilibrium expected payoff is zero by the mixed strategy indifference condition. A
bid at the cap (bmax ) earns:
hα i
+ (1 − α) 10 − bmax .
2
Setting that expression equal to zero (the equilibrium expected payoff), we see that

10 − bmax
α= .
5
A bid of b0 earns:
(1 − α)10 − b0 .
Setting that expression equal to zero (the equilibrium expected payoff) and substituting
for α, we see that
b0 = 2bmax − 10.
A bid of b ∈ [0, b0 ] earns:
(1 − α)F (b)10 − b.
Setting that expression equal to zero (the equilibrium expected payoff) and substituting
for α, we see that
b
F (b) = .
2bmax − 10
To verify that the solution is a mixed strategy equilibrium, we need to check that a
player cannot gain by deviating to a bid outside the support of F . The complement of
the support of F is b ∈ (b0 , bmax ). For all such b, the expected payoff is:

(1 − α)10 − b ≤ (1 − α)10 − b0 = 0.

d) The case of no bid cap is equivalent to bmax = 10, so it is a special case of the above.
As we’ve shown, the expected payoff for each student is zero regardless of bmax . For this
game, the payoffs of all players sum to 0 in every realization (and hence in expectation),
so Bernheim’s expected payoff is also 0. This means his expected revenue must be
10 (just enough to compensate for the $10 bill) irrespective of bmax . Intuitively, the
students will always compete away all positive rents, if not by bidding close to 10, then
by matching each other with the maximum bid.

26. (Sept 2010 Comp, Q3)

a) The pure strategy set is the set [bmin , ∞] ∪ {φ}. A mixed strategy is a probability π
of playing φ, plus a CDF F (b) on [bmin , ∞] that governs choice with probability 1 −
π. Specifically, F is a monotonically increasing uppersemicontinuous (right-continuous)
function F with F (b) = 0 for b < bmin and limb→∞ F (b) = 1. The mixed strategy set is
the set of all such functions.
b) There is no pure strategy equilibrium.
Case 1: There is only one active bidder. Either that bidder bids less than $10, in which
case another student would have an incentive to bid higher, or he bids $10, in which
case he has an incentive to lower his bid.

20
 Case 2: There are at least two active bidders. If there is only one high bidder, he
must be bidding more than bmin and has the incentive and opportunity to lower his bid.
If there is more than one high bidder below $10, then any one of them would benefit
from increasing his bid slightly. If more than one bidder ties at $10, each has a negative
expected payoff, so a deviation to φ is better.
c) Assume F has an atom at bmin . Then a player’s probability of winning (and hence his
expected payoff) rises discontinuously when increasing his bid from bmin to bmin + ε, a
contradiction. Now assume F has an atom at 10. Then a player receives a negative
expected payoff from playing 10, and could improve his expected payoff by shifting that
atom to φ.
d) In light of part (3), the probability of winning when bidding 10 is unity; hence the
expected payoff when bidding 10 is 0. Because 10 is in the support of F , the equilibrium
expected payoff is zero by the mixed strategy indifference condition.
A bid at the floor earns an expected payoff of:

10π − bmin

Setting that expression equal to zero (the equilibrium expected payoff), we see that
bmin
π=
10
A bid of b earns:
10 [π + (1 − π)F (b)] − b
Setting that expression equal to zero (the equilibrium expected payoff), we see that
 
1 b
F (b) = −π
1 − π 10
Substituting for π, we obtain
b − bmin
F (b) =
10 − bmin
To verify that the solution is a mixed strategy equilibrium with the desired properties,
we need to check that F (b) is a legitimate atomless CDF on [bmin , 10]. Substituting into
the above expression for F , we see that F (bmin ) = 0 and F (10) = 1, as required; we
also see that F 0 (b) = 10−b1 min > 0, also as required. We do not have to check deviations
outside the support of F (because there are none), and we have already assured (by
construction) that a player receives the same expected payoff from φ as well as from all
b in the support of F . Thus we have a mixed strategy equilibrium.
e) The case of no bid floor is equivalent to bmin = 0, so it is a special case of the above.
As we’ve shown, the expected payoff for each student is zero regardless of bmin . For this
game, the payoffs of all players sum to 0 in every realization (and hence in expectation),
so Bernheim’s expected payoff is also 0. This means his expected revenue must be 10
(just enough to compensate for the $10 bill) irrespective of bmin . Intuitively, the students
will always compete away all positive rents with bmin = 0, and setting a higher minimum
will only cause them to compete less aggressively by opting out.

27. (2011 Final , Q1)

a) The strategy set for a student is S = {enter, not enter}.

21
 d
b) If M students enter the science fair, the payoff for each student that enters is M − K.
Students that don’t enter have a payoff of 0. In order for there to be a Nash equilibrium
in which M students enter, we need participating students to not want to drop out:
d d
− K ≥ 0 =⇒ M ≤
M K
and non-participating students to not want to enter:
d d
0≥ − K =⇒ ≤ M + 1.
M +1 K

Thus, if N < kd , then all N students enter the fair. If N ≥ kd , there are Nash equilibria
d N


in which exactly M students enter with M ≤ K ≤ M + 1. (Note that there are M
d
such equilibria.) If K is not an integer, then this M is unique. If it is an integer, then
d d
there are equilibria with M = K − 1 and M = K .
c) There are three possibilities, other than the pure-strategy equilibria described in part 2.
ˆ One student plays “enter” and the other student plays a strictly mixed strategy.
Let λ ∈ (0, 1) be the probability that the latter student enters. The student who is
strictly mixing must be indifferent between entering and not entering, which implies
d
2 − K = 0, or d = 2K. Also, the student playing a pure strategy must at least
weakly prefer entering to not entering, which implies λ d2 + (1 − λ)d − K ≥ 0. This
means λ ≤ 2(1 − K d ) = 1, so we could have any λ ∈ (0, 1). This is an equilibrium if
and only if d = 2K.
ˆ One student plays “not enter” and the other students plays a strictly mixed strategy.
Let λ ∈ (0, 1) be the probability that the latter student enters. The student who is
strictly mixing must be indifferent between entering and not entering, which implies
d − K = 0, or d = K. This contradicts the assumption that d > K, so there is no
such equilibrium.
ˆ Both students play strictly mixed strategies. If student 1 enter with probability
λ ∈ (0, 1), then student 2, who is strictly mixing, must be indifferent between
entering and not entering. This implies λ d2 + (1 − λ)d − K = 0 or λ = 2(1 − K d ).
The problem is symmetric, so both students must enter with this probability. We
need λ ∈ (0, 1). We already know 2(1 − K d ) > 0 because K < d. We also need
K
2(1 − d ) < 1, which requires d < 2K. Therefore, this is an equilibrium if and only
if d < 2K.
d) The strategy set for a student is S = {not enter} ∪ R+ .
e) Any strategy e > d−K
c is strictly dominated by not entering. There are no other strictly
dominated strategies to eliminate.
f) There are no pure-strategy Nash equilibria. Consider the following cases, which exhaust
all possibilities:
ˆ No students enter. This is not a Nash equilibra because one student could enter
with e = 0 and earn d − K > 0.
ˆ One student enters and chooses e > 0. This is not a Nash equilibrium because the
participating student could increase his payoff by deviating to e = 0.
ˆ One student enters and chooses e = 0. This is not a Nash equilibrium because a non-
participating student would find it profitable to enter and select any e ∈ (0, d−K
c ).

22
 ˆ At least two students enter and not every entrant selects the same effort level. This
is not a Nash equilibrium because the student with the lowest effort level receives a
strictly negative payoff and could instead receive a payoff of zero by not entering.
ˆ At least two students enter and select the same effort level e < d−K c . This is not
a Nash equilibrium because one of the participants could claim the whole prize by
slightly increasing their effort.
ˆ At least two students enter and select the same effort level e = d−Kc . This is not a
Nash equilibrium because the participants earn strictly negative profits and would
do better if they didn’t enter.
g) i. Not entering yields a payoff of zero. Because this mixed strategy puts positive
probability on not entering, the mixed-strategy indifference condition implies that
payoffs are zero in this equilibrium.
ii. We can’t have e > d−K c because those strategies are strictly dominated. We can’t
have e < d−Kc because then some student could deviate to entering with probabil-
ity 1 and choosing some effort e ∈ (e, d−Kc ) and receive a strictly positive payoff.
Therefore, e = d−Kc .
iii. Using the indifference condition, for any e ∈ (e, d−K
c ), payoffs are

0 = d [1 − π + πF (e)]N −1 − K − ce =⇒

K+ce

N1−1
d +π−1
F (e) =
π
N −1
d [1 − π] −K
e=
c

N1−1
iv. Claim: e = 0, which implies π = 1 − K d .
Suppose to the contrary that e > 0. Then the expected payoff from entering and
choosing some e0 ∈ [0, e) is (1 − π)N −1 d − ce0 − K > (1 − π)N −1 d − ce − K =
(1 − π + πF (e))N −1 d − ce − K. Therefore, entering with probability 1 and choosing
e0 is a profitable deviation and this is not a Nash equilibrium.
v. As N gets large, π goes to zero, so the probability of entry goes to zero.

28. (PS4, Q2)

a) Using iterative deletion, we find that 3 must play A and 2 must play L. There are 2
PSNE: (U, L, A) and (D, L, A).
b) No.
c) (U, L, A) is trembling hand perfect, but (D, L, A) is not.
To show this, we note that in this problem, ε-constrained equilibria must take a particular
form: Note that because A is a dominant strategy for 3, in any ε-constrained equilibrium,
3 must be placing no more that the required ε weight on B. Similarly, 2 must be placing
no more than ε on R. But then
π1 (U ) = (1 − ε)2 · 1 + (1 − ε)ε · 1 + (1 − ε)ε · 1 + ε2 · 0 and
π1 (D) = (1 − ε)2 · 1 + (1 − ε)ε · 0 + (1 − ε)ε · 0 + ε2 · 1
so that π1 (U ) > π1 (D). Therefore, 1 must be placing no more than the required ε weight
on D.

23
 Thus, for εn → 0, we have a sequence of ε-constrained equilibria that converges to
(U, L, A):
σ n = ((1 − εn )U + εn D, (1 − εn )L + εn R, (1 − εn )A + εn B).
Thus, (U, L, A) is trembling hand perfect.
On the other hand, since all ε-constrained equilibria take this form, it is impossible to
find a sequence of ε-constrained equilibria converging to (D, L, A), so (D, L, A) is not
trembling hand perfect.
Moral: In 3+ player games, the trembling-hand refinement can rule out more than just
NE where one strategy is weakly dominated. (In 2 player games, these two refinements
are equivalent.) Ask yourself what happened here that couldn’t in a 2 player setting.
29. (PS3, Q3)
The unique profile surviving ISD is (C, c), which is a Nash equilibrium. Therefore, it is also a
correlated equilibrium. On the other hand, it is easy to show that no pure strategy eliminated
by ISD can be a part of a correlated equilibrium. It follows that (C, c) is the unique correlated
equilibrium.
30. (PS4, Q1)
a) The NE are (B, L), (T, R) and ( 23 T + 13 B, 23 L+ 13 R). Thus, the set of payoffs is (7, 2) , (2, 7) , 14 14


3 , 3 .
b) The convex hull of the payoffs is drawn below:

(2,7)
7

5
(14/3,14/3)

2
(7,2)

0
0 1 2 3 4 5 6 7

c) Let’s look for payoffs outside the convex hull in (b) within the class of equilibria described
in the notes. (Of course, these are not the only correlated equilibria. But if we find one
outside of the convex hull, then we are done.)
L R
T λ (1 − λ)/2
B (1 − λ)/2 0

24
 The symmetry of this class makes verification easy. We only need to verify that 1 would
agree to this distribution of outcomes. If 1 is told to play B , he believes that 2 will
play L with certainty, in which case B is in fact a best response. If 1 is told to play T ,
then he believes that 2 will play L with probability λ+λ1−λ = 1+λ
2λ
and R with probability
2
1−λ
1−λ
2
λ+ 1−λ
= 1+λ . For T to in fact be a best response, we require π1 (T ) ≥ π1 (B) or
2

     
2λ 1−λ 2λ
6 +2 ≥7
1+λ 1+λ 1+λ

which reduces to λ ≤ 21 . Thus, when λ ≤ 12 , the distribution over outcomes described

above is a correlated equilibrium. Choosing λ = 21 yields the highest payoff equilibrium
in this class, with payoffs (5 14 , 5 41 ). This falls outside the convex hull of the NE.

31. (PS7, Q3)

Here a strategy is a mapping si : {1, 2, . . . , M } → {trade, not-trade}.
Mixed strategy NE: Observe that in no equilibrium will i use a strategy placing positive
probability on trading M . Suppose the contrary, that there exists an equilibrium s in which
play i places positive probability on trading M . Player i’s strategy can be a best response
only if all the other players use strategies in which 1, . . . , (M − 1) are never traded. However,
this cannot be the case in a NE because player j 6= i would do better to deviate to the strategy
“trade only when card reads 1.”
Iterating this proof rules out placing positive probability on any number 2 through M − 1.
Thus, the set of candidate strategies is reduced to si (1) = trade with probability p and
si (m) = not-trade with probability 1 for all m = 2, . . . , M .
It is easy to check that any profile where each i uses a strategy of this form is in fact a NE.
We can not use iterative deletion here. No strategy is strictly dominated. To see this, note
that if j 6= i all use the strategy “never trade,” then all strategies for i yield identical payoffs.
Correlated NE: An M is never traded in a correlated equilibrium. Suppose, towards contra-
diction, that there is a correlated equilbrium in which the correlation device assigns a positive
probability to player i using a strategy that involves trading an M . Suppose the state of the
world is such that the correlation device suggests using such strategy to player i. In order
for him to follow the suggestion in equilibrium, it must be the case that all other players
are instructed, by the correlation device, in such state of the world, to place zero probability
on any strategy that involves trading anything other than an M . But then, in this state of
the world, any j 6= i will have an incentive to deviate to the strategy that trades a 1 with
probability 1, and everything else with probability zero, a contradiction.
Repeating this argument rules out trading 2 through M in a correlated equilibrium. We
are left with the set of correlation devices that that assign probability zero to any strategy
involving trading any card greater than 1, all of which are easily verified to be correlated
equilibria.
Remark : For pedagogical purposes, we separated the 2 cases, but in fact the NE result follows
from the correlated equilibrium result.

32. (PS4, Q4)

25
 We can calculate the normal form:

aa an na nn

M −2w−2s M −2w−2s 2M −w−s M −2s 3M −w−s −2w

aa 4 , 4 4 , 4 4 , 4 M, 0

M −2s 2M −w−s M −s M −s 2M −s M −w 2M
an 4 , 4 4 , 4 4 , 4 4 ,0

−2w 3M −w−s M −w 2M −s M −w M −w 2M
na 4 , 4 4 , 4 4 , 4 4 ,0

nn 0, M 0, 2M
4 0, 2M
4 0, 0

It is easily verified that the squares marked with an × can never be NE for any M, s, and w
that satisfy the given conditions.

aa an na nn
aa × ×
an × ×
na × × × ×
nn × × ×

For the remaining squares, verifying the conditions for a NE yields:

(aa, an) and (an, aa) are NE when M ≥ 2s and M ≥ w;
(an, an) is a NE when M ∈ [s, w];
(aa, nn) and (nn, aa) are NE when M ≤ 2s.
Finally, the NE of this game correspond to Bayesian Nash equilibria of the associated Bayesian
game.

33. (PS4, Q5)

To avoid confusion with quantity, we replace the probability q in the problem with µ. Inverse
demand is given by p(Q) = a − bQ.
A BNE involves firm i choosing a quantity qci when its costs are c and a quantity qdi when
costs are d. For this to be a BNE, i’s choices of q must be optimal, taking j’s strategy as
given:
qci ∈ arg max µ · [a − b(qcj + q) − c)]q + (1 − µ) · [a − b(qdj + q) − c)]q
q

qdi ∈ arg max µ · [a − b(qcj + q) − d)]q + (1 − µ) · [a − b(qdj + q) − d)]q

q

This yields FOCS:

a − b(µqcj + (1 − µ)qdj ) − c a − b(µqcj + (1 − µ)qdj ) − d

qci = and qdi = .
2b 2b
The symmetry of the problem implies qci = qcj = qc and qdi = qdj = qd . Solving the two FOCs
simultaneously for qc and qd yields
1−µ
a− 2 (c − d) − c a − µ2 (d − c) − d
qc = and qd = .
3b 3b

26
34. (Sept 2004 Comp, Q5)
a) A BNE where the project does not go forward definitely: 2 students play n regardless
of their type, and the rest of the students do anything. Then everyone gets a payoff of
zero, and no deviation improves this.
A BNE where the project goes forward with positive probability: each student con-
tributes iff his valuation is greater than or equal to 10. Aside from resolving indifference
differently when the student’s valuation is exactly 10, there is no other equilibrium of
this type. If the probability that everyone else plays y is strictly positive, then a student
will definitely choose y if his valuation is greater than 10, and definitely choose n if it is
less than 10.
b) i. Once again, if both students choose n regardless of their valuations, this will lead
to a BNE where the project never goes forward.
ii. Claim: In every BNE, students will use a cutoff strategy: σi (vi ) = y if vi ≥ ci and
σi (vi ) = n if vi < ci for some cutoff ci .
Proof : In any equilibrium, there is some probability p−i that student −i chooses
y. Student i benefits from making a contribution iff p−i vi ≥ 10. For p−i > 0, this
means setting a cutoff ci = p10 −i
is optimal. For p−i = 0, setting ci > 45 is optimal.
ci
Let student i use cutoff ci . Then the probability that i plays y is 1− 45 ; therefore cj =
10 450 2 −45c+450 = 0.
1− i
c = 45−ci . At a symmetric equilibrium, ci = cj = c, and hence c
45
This leads to c ∈ {15, 30}. Consequently, there are two symmetric equilibria: one
where both students use 15 as a cutoff, and the other where both use 30 as a cutoff.
iii. At an asymmetric equilibrium, we would have c1 (45 − c2 ) = 450 = c2 (45 − c1 ).
This yields c1 = c2 . Therefore, there are no asymmetric equilibria.
c) i. As before, there is a BNE where the project never goes forward if all students choose
n regardless of their valuations.
ii. In every BNE, students still must use a cutoff strategy. This follows from the
argument in part (b), treating p−i as the probability that all other students choose
y. At a symmetric BNE (if it exists), all students use the same cutoff c. The
c N −1


probability that all other students choose y is 1 − 45 , so the best cutoff to
use in response is c0 = 10
N −1 . This means that any symmetric equilibrium has
(1− 45c )
c N −1


a cutoff that solves c 1 − 45 = 10. For a project to go forward with positive
probability, we must have c < 45. We now show that this is impossible if N ≥ 3.
c N −1
. Let c∗ (N ) = arg maxc∈[0,45] f (c, N ). The FOC yields


Let f (c, N ) = c 1 − 45
 c N −1 c  c N −2
1− − (N − 1) 1 − = 0.
45 45 45
Since for c ≥ 45, f (c, N ) ≤ 0, and for c < 45, f (c, N ) > 0, we can verify that
c∗ (N ) = 45
N (you can verify that this is the maximum without checking the 2
nd
df
order condition by simply checking that for c < c∗ (N ), dc > 0, and for c > c∗ (N ),
df
dc < 0).
N −1
Now, let g (N ) = f (c∗ (N ) , N ) = 45(NN−1)
N . Recognize that g (3) = 20 3 < 10.
Thus, there is no symmetric equilibrium where the project goes forward with positive
probability with 3 agents.
For N > 3, note that
dg (N − 1)N −1
= 45 (log (N − 1) − log (N )) < 0.
dN NN

27
 Therefore, for every N ≥ 3, g (N ) ≤ g (3) < 10, and hence, the unique symmetric
equilibrium is the one where the project never goes forward.

35. (PS4, Q3)

a) Each i has a dominant strategy: Name the highest xi , namely xi = 1, regardless of type.
Thus, we have a unique BNE in which each i names xi = 1 regardless of type.
b) This maximization problem is

x22 x21
   
max x1 − θ1 + x2 − θ2 ,
x1 ,x2 ∈[0,1] 2 2

which has solution x∗1 = 1 ∗

θ2 , x2 = 1
θ1 .
c) Method 1: We use the Groves mechanism described in the notes. Note that while
this implementation guarantees that truth-telling is a dominant strategy, it does not
guarantee budget balance. So we’ll have to check specifically if we can achieve budget
balance in this case.
The Groves mechanism calls for:
1
x1 (θ1A , θ2A ) = x∗1 (θ1A , θ2A ) =
θ2A
1
x2 (θ1A , θ2A ) = x∗2 (θ1A , θ2A ) = A
θ1
     
Hence Ti θ1A , θ2A = Uj x∗1 θ1A , θ2A , x∗2 θ1A , θ2A , θjA +Ki θjA = 1
− 2θ1A +Ki θjA .




θiA j

Budget balance requires T1 (θ1A , θ2A ) + T2 (θ1A , θ2A ) =0 ∀(θ1A , θ2A ), so

1 1 1 1
A
− A + K1 (θ2A ) + K2 (θ1A ) + A − A = 0.
θ1 2θ2 θ2 2θ1

The simplest choices for K1 , K2 that satisfy this are

1 1
K1 (θ2A ) = − K2 (θ1A ) = − .
2θ2A 2θ1A
1 1
This leads to T1 θ1A , θ2A = = −T2 θ1A , θ2A .



θ1A
− θ2A
Method 2: We derive a mechanism from first principles. Consider a differentiable
direct-revelation mechanism that implements the first best outcome with budget balance.
We assume that player 2 tells the truth. Let W (θ1 , θ, θ2 ) represent the payoffs of player
1 when his true type is θ1 , his announcement is type θ, and player 2’s true type is θ2 .
I.e.,

1 θ1
W (θ1 , θ, θ2 ) = − 2 + T1 (θ, θ2 )
θ2 2θ
Since player 1 would want to maximize his payoffs, we take the FOC wrt θ:

dW θ1 dT1 (θ, θ2 )
= 3+
dθ θ dθ

28
 Since we need truth-telling to be a dominant strategy, it must be a best response for
each θ2 . Therefore, dW
dθ |θ=θ1 = 0. Making the substitution implies that

dT1 (θ1 , θ2 ) 1
=− 2
dθ θ1

Integrating up, we obtain T1 (θ1 , θ2 ) = θ11 + C1 (θ2 ). To establish budget balance, let
T1 (θ1 , θ1 ) = 0. Hence, C1 (θ1 ) = − θ11 , and therefore letting C1 (x) = − x1 assures
budget balance. Thus, using Ti (θi , θj ) = θ1i − θ1j implements the first-best allocation in
dominant strategies with budget balance.

36. (2003 Final, Q1)

Each player’s strategy set is the set of all functions s : [0, 100] → R+ . Let x denote the amount
in Lilo’s box and y denote the amount in Tim’s box. The pair of functions sL (x) = x, and
sT (y) = y constitute a Bayesian Nash equilibrium. Conditional on Tim following strategy
sT , Lilo’s problem is to choose z(x) for each x ∈ [0, 100] to solve:
Z 100  
1 z+y
max 1{z+y≤x+y} dy.
z 0 100 2

If z > x, then this expression is 0. For all z < x, this expression is increasing in z. Therefore,
Lilo would not want to deviate from her equilibrium strategy sL (x) = x. The analysis is
symmetric for Tim.

37. (2007 Final, Q1)

a) A pure strategy is a mapping from [0, 1] to R+ . An example: b(s) = s for all s ∈ [0, 1].
b) An example: b(s) = 3 for all s ∈ [0, 1] is weakly dominated by the strategy b(s) = 2 for
all s ∈ [0, 1]. It is not strictly dominated because when the other bidder bids according
to b(s) = 0 for all s ∈ [0, 1], there is no other strategy that yields a strictly higher
(expected) payoff.
c) Suppose bidder 2 bids according to b2 (s2 ) = αs2 . When observing signal s1 , bidder 1’s
R b1
expected payoff from bidding b1 is 0 α (s1 + s2 − αs2 )ds2 if b1 ≤ α (so that the upper
R1
limit of the integral does not exceed 1) and 0 (s1 + s2 − αs2 )ds2 if b1 ≥ α. Note that in
the second case the payoff does not depend on b1 .
d) Bidder 1’s maximization problem is
Z b1
α
max (s1 + s2 − αs2 )ds2 .
0≤b1 ≤α 0

Ignoring the b1 ≤ α constraint, the FOC yields b∗1 (s1 ) = α−1

α
s1 . For any s1 such that the
α
constraint is binding, any b1 (s1 ) ∈ [α, ∞) is optimal (and in this case, α−1 s1 ∈ [α, ∞)).
∗ α
Thus, b1 (s1 ) = α−1 s1 is a best response whenever α > 1. For α ≤ 1, any b1 (s1 ) ∈ [1, ∞)
is optimal.
e) Unlike in the example from the notes, the value of the item to one player is dependant
on the other player’s signal. If the other player’s bid is high, that suggests that the
value of the object is higher. Therefore, it is a best response to bid more than one’s own
valuation.

29
 α
f) Symmetry implies α−1 = α. Therefore α = 2 and a symmetric BNE is bi (si ) = 2si for
i = 1, 2.
g) From the previous part, if bidder 1’s strategy is b1 (s1 ) = αs1 with α > 1, bidder 2’s best
response is  α α
α−1 s2 if α−1 s2 ≤ α
b2 (s2 ) = α
∈ [α, ∞) if α−1 s2 ≥ α
α α
In particular, b2 (s2 ) = α−1 s2 is a best response for bidder 2. Let β = α−1 . Similarly,
we can check that given b2 (s2 ) = βs2 , bidder 1’s best response is b1 (s1 ) = αs1 .
Therefore, there exist a continuum of asymmetric equilibria of the form b1 (s1 ) = αs1
α
and b2 (s2 ) = βs2 where β = α−1 , α > 1.
α
h) In an equilibrium where b1 (s1 ) = αs1 and b2 (s2 ) = α−1 s2 with α > 1, bidder 1 wins if
1
and only if s1 ≥ α−1 s2 .
α
If α < 2, α < α−1 and therefore, bidder 1 is less aggressive than bidder 2. For any s1 ,
the expected payoff to bidder 1 is
Z (α−1)s1
α 1
π1 (s1 ) = (s1 + s2 − s2 )ds2 = (α − 1)s21 .
0 α−1 2
α
If α > 2, α > α−1 and therefore, bidder 1 is more aggressive than bidder 2. For any s1 ,
the expected payoff to bidder 1 is
( R
(α−1)s1 α
0 (s1 + s2 − α−1 s2 )ds2 = 12 (α − 1)s21 if s1 ≤ 1
α−1
π1 (s1 ) = R1 α 1 1
0 (s1 + s2 − α−1 s2 )ds2 = s1 − 2(α−1) if s1 > α−1

π1 (s1 ) is increasing in α. A larger α means bidder 1 is more aggressive in bidding. In

such an asymmetric equilibrium, the more aggressive bidder 1 is, the less aggressive
bidder 2 is. By being more aggressive, bidder 1 increases his chance of winning and
reduces the payment upon winning.

38. (2006 Final, Q1)

a) Yes there is; if every other individual discloses her grade, an individual is indifferent
between disclosing her grade and not doing so since in either case, she would receive the
same marks.
b) Yes it is. If this student believes that all other students are disclosing their grades, she
is indifferent between disclosing and not disclosing. As such, it is 1-rationalizable for
her to not disclose her grade. Since by (a), it is a NE for all students to disclose their
grades, it is rationalizable for all other students to disclose their grades. Therefore, it is
rationalizable for a student with a 10 to not disclose her grade.
c) – d) Proving that there is no such mixed strategy equilibrium suffices to establish that
there is no such PSNE in (c). Suppose by way of contradiction that there is some mixed
strategy equilibrium where an inidividual with a strictly positive grade does not disclose
her grade with positive probability. Let M > 0 be the highest grade that is not disclosed
in equilibrium with positive probability. Then all students with grades of 0 shall not
disclose their grades with probability 1 (since not disclosing their grades yields a positive
expected grade). Therefore, any student with a grade M shall receive a grade strictly
less than M if she does not disclose her grade. Therefore, she would strictly prefer to
disclose her grade, leading to a contradiction.

30
39. (2012 Fall Comp, Q3)

40. (2012 June Comp, Q4)

41. (PS5, Q3)

a)
b) This problem is a colossal pain. The technique is to use backward induction, carefully
checking all equilibrium outcomes in each subgame.
A strategy in this game is given by ((r, p, λ), (q, µ)) where r is the probability E places
on In, p and q are the probabilities placed on Fight by E and I, respectively, and λ and
µ are the probabilities placed on Here by E and I, respectively. To simplify notation,
if, for example, the entrant plays F ight with probability 1, we write ((r, F, λ), (q, µ)).
We first look at the final subgame:

H T
H 3, 1 0, 0
T 0, 0 x, 3

Three cases:
x < 0 Here the game is dominance solvable. The unique equilibrium is (H, H), with
outcome (3,1).

x = 0 (H, H) is still an equilibrium, with outcome (3,1). We also have a continuum of

equilibria characterized by (λ, T ) where λ ≤ 34 and the corresponding equilibrium
outcomes are (0, 3(1 − λ)).

x > 0 Here we have two pure strategy equilibria, as well as a mixed strategy equilibrium:
(H, H) yielding outcome (3,1), (T, T ) yielding outcome (x, 3), and (λ = 34 , µ = 3+x
3
)
3x 3
yielding outcome ( 3+x , 4 ).
We now replace the H/T subgame with each of the outcomes above (backward induc-
tion).
In all three cases, (H, H) is an equilibrium, with outcome (3,1). Replacing the final
subgame with this outcome, the F/A subgame is reduced to:

F A
F −3, −1 1, −2
A −2, −1 3, 1

This subgame is dominance solvable, with equilibrium (A, A) and outcome (3,1). Re-
placing this subgame with its outcome, we find that the Entrant will choose In. Thus,
∀x we have SPNE:
((In, A, H), (A, H)). (6)

Case x = 0: The F/A subgame becomes

F A
F −3, −1 1, −2
A −2, −1 0, 3(1 − λ)

31
This game has a unique equilibrium – a mixed strategy equilibrium given by (p =
4−3λ 1
5−3λ , q = 2 ) with outcome (−1, −1). It follows that E will play Out. Thus, when x = 0
we have SPNE    
4 − 3λ 1 3
Out, ,λ , ,T for any λ ≤ . (7)
5 − 3λ 2 4
Case x > 0, outcome = (x, 3): Now the F/A subgame becomes
F A
F −3, −1 1, −2
A −2, −1 x, 3

Since x is in the normal form, we have subcases.

i. x > 1: Here the F/A subgame is dominance solvable. The equilibrium is (A, A),
with outcome (x, 3). It follows that E will play In. Thus, when x > 1, we have
SPNE
((In, A, T ), (A, T )). (8)
ii. x = 1: Here we have a continuum of equilibria given by (p, A) where p ≤ 54 , and
corresponding outcomes (1, 3 − 5p). For any outcome, E will play In. Thus, when
x = 1, we have SPNE
4
((In, p, T ), (A, T )) for any p ≤ . (9)
5
iii. x < 1: Here we have a unique equilibrium – a mixed equilibrium given by (p =
4 1−x 3x−2
5 , q = 2−x ) with outcome ( 2−x , −1). Whether this outcome beats the payoff 0
choice of Out depends on x.
When x < 2/3 we have SPNE
   
4 1−x
Out, , T , ,T . (10)
5 2−x
When x = 2/3 we have SPNE
   
4 1−x
r, , T , ,T for any r ∈ [0, 1]. (11)
5 2−x
When x > 2/3 we have SPNE
   
4 1−x
In, , T , ,T . (12)
5 2−x
3x 3
Case x > 0, outcome = ( 3+x , 4 ): Now the F/A subgame becomes

F A
F −3, −1 1, −2
3x 3
A −2, −1 3+x , 4

Since x is in the normal form, we again have subcases.

i. x > 32 : Here the F/A subgame is dominance solvable. The equilibrium is (A, A),
3x 3
with outcome ( 3+x , 4 ). In this case, E will play In. Thus, when x > 32 we have
SPNE    
3 x
In, A, , A, . (13)
4 3+x

32
 ii. x = 32 : Here we have a continuum of equilibria, given by (p, A) for p ≤ 11
7
, with
3−11p
corresponding outcomes (1, 4 ). For every outcome, E will play In. Thus, when
x = 23 we have SPNE
   
3 x 7
In, p, , A, for any p ≤ . (14)
4 3+x 11
iii. x < 32 : Here we have a unique equilibrium – a mixed equilibrium given by (p =
7 3−2x 7x−6
11 , q = 6−x ) with outcome ( 6−x , −1). Whether this outcome beats the payoff 0
choice of Out depends on x.
When x < 6/7 we have SPNE
   
7 3 3 − 2x x
Out, , , , . (15)
11 4 6−x 3+x
When x = 6/7 we have SPNE
   
7 3 3 − 2x x
r, , , , for any r ∈ [0, 1]. (16)
11 4 6−x 3+x
When x > 6/7 we have SPNE
   
7 3 3 − 2x x
In, , , , . (17)
11 4 6−x 3+x
42. (PS5, Q2)
a) i. Si = {f or, against} for i = A, B, C.
ii.
iii. Here g(·) maps strategies to the triplet of payoffs where A’s payoff is listed first, B’s
second and C’s third.
g(f, f, f ) = (1, 1, −1) g(a, f, f ) = (1, 1, −1)
g(f, f, a) = (1, 1, −1) g(a, f, a) = (0, 0, 0)
g(f, a, f ) = (1, 1, −1) g(a, a, f ) = (0, 0, 0)
g(f, a, a) = (0, 0, 0) g(a, a, a) = (0, 0, 0)
iv. The Nash equilibria are (f, f, f ), (f, f, a) and (a, a, a). The bill passes in the first
two equilibria.
v. Observe that for A and B, voting against the bill is weakly dominated. For C,
voting for the bill is weakly dominated. Thus, only in (f, f, a) does no player use a
weakly dominated strategy.
vi. There is no pure strategy SPNE. There are several ways to see this. One is to
reduce the game by replacing subgames with their equilibrium outcomes. There are
8 subgames: the game itself, and the subgames following completion of the show/no
show game. For the subgame following sss, there are two equilibrium outcomes:
(1 − ε, 1 − ε, −1 − ε) and (−ε, −ε, −ε). For the subgame following ssn, there are also
two equilibrium outcomes: (1 − ε, 1 − ε, −1) and (−ε, −ε, 0). Every other subgame
has a unique equilibrium outcome. Thus, there are 2 × 2 = 4 reduced games to
check. None of these has a pure strategy NE. (Although we know for sure that there
exists a mixed strategy NE!)
Another, perhaps simpler method is to rule out profiles case by case. Ask, is there
an equilibrium where as the entry component of their strategies A, B and C choose:

33
 sss No. If following sss, an equilibrium where the bill passes is chosen, then C would
do better to not show up. If the no pass equilibrium is chosen in the subgame,
then A or B would have done better to not show up.
ssn No. Then A or B would do better by not showing up.
sns No. Then A would do better by not showing up.
snn No. Then C would do better by showing up.
nss No. Then B would do better by not showing up.
nsn No. Then C would do better by showing up.
nns No. Then C would do better by not showing up.
nnn No. Then A or B would do better by showing up.
b) i. SA = {f,a}
SB = {ff,fa,af,aa}
SC = {ffff,fffa,ffaf,ffaa,faff,fafa,faaf,faaa,afff,affa,afaf,afaa,aaff,aafa,aaaf,aaaa}

ii.
iii. Yes. This can be shown using backward induction, replacing each subgame with its
equilibrium outcome. One equilibrium is given by (f,ff,aaaa).
iv. No. Again, using backward induction to replace each subgame with its equilibrium
outcome, we find that the equilibrium outcome is that the bill passes.

43. (June 2004 Comp, Q5)

a) Since we are looking for SPNE where voters vote sincerely, their actions are easily spec-
ified at every information set (they vote for the alternative they strictly prefer, and in
the case of indifference, can settle indifference any which way). As such, what one needs
to determine – by backward induction – is what the agenda-setter shall propose taking
into account that voters shall vote sincerely. The SPE outcomes are characterized for
each initial status quo.
Table 1
Status Quo Alternatives majority-preferred to status quo Outcome
(0, 0, 0) (0, 0, 0), (0, 1, 1), (1, 0, 1), (1, 1, 0), (1, 1, 1) (0, 0, 0)
(0, 0, 1) (0, 0, 0), (0, 0, 1), (1, 1, 0), (1, 1, 1) (0, 0, 0)
(0, 1, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0), (1, 0, 1), (1, 1, 1) (0, 0, 0)
(1, 0, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0), (1, 0, 0), (0, 1, 1), (1, 1, 1) (0, 0, 0)
(0, 1, 1) (0, 0, 1), (0, 1, 0), (0, 1, 1) (0, 0, 1)
(1, 0, 1) (0, 0, 1), (1, 0, 0), (0, 1, 1), (1, 0, 1) (0, 0, 1)
(1, 1, 0) (0, 1, 0), (1, 0, 0), (0, 1, 1), (1, 0, 1), (1, 1, 0) (0, 1, 0)
(1, 1, 1) (0, 1, 1), (1, 0, 1), (1, 1, 0), (1, 1, 1) (0, 1, 1)
b) From part A, we know that there are only four possible outcomes on the equilibrium
path: (0, 0, 0), (0, 0, 1), (0, 1, 0), and (0, 1, 1). In round 1, each z is associated with one
of these outcomes (in the sense that, if z becomes the status quo for the second round, it
will lead to the outcome indicated in the table 1). Consequently, the two round game is
just like a one round game where the possible choices are (0, 0, 0), (0, 0, 1), (0, 1, 0), and
(0, 1, 1). For this “reduced form” one round game, we can make up a table like table

34
 Table 2
Status quo Possibilities Outcome
(0, 0, 0) (0, 0, 0), (0, 1, 1) (0, 0, 0)
.
(0, 0, 1) (0, 0, 0), (0, 0, 1) (0, 0, 0)
(0, 1, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0) (0, 0, 0)
(0, 1, 1) (0, 0, 1), (0, 1, 0), (0, 1, 1) (0, 0, 1)
Combining table 2 with table 1, we can make up the required table for the two round
game.
Table 3
Status Quo Outcome Proposal(s)
(0, 0, 0) (0, 0, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0), (1, 0, 0)
(0, 0, 1) (0, 0, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0), (1, 0, 0)
(0, 1, 0) (0, 0, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0), (1, 0, 0)
(1, 0, 0) (0, 0, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0), (1, 0, 0)
(0, 1, 1) (0, 0, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0), (1, 0, 0)
(1, 0, 1) (0, 0, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0), (1, 0, 0)
(1, 1, 0) (0, 0, 0) (0, 0, 0), (0, 0, 1), (0, 1, 0), (1, 0, 0)
(1, 1, 1) (0, 0, 1) (0, 1, 1), (1, 0, 1)
For example, consider the status quo (1, 0, 1). We know from table 1 that, if this
becomes the status quo for round 2, the outcome will be (0, 0, 1). From table 2, we see
that, in the reduced form one round game, (0, 0, 1) leads to the outcome (0, 0, 0), which
therefore appears in the “outcome” column of table 3. From table 1, we see that this
outcome is obtained if (0, 0, 0), (0, 0, 1), (0, 1, 0), or (1, 0, 0) becomes the status quo for
round 2. Consequently, the agenda setter can propose any of these alternatives in round
1.
c) From part B, we know that there are only two possible outcome: (0, 0, 0) and (0, 0, 1).
In round 1, each z is associated with one of these outcomes (in the sense that, if z
becomes the status quo for the second round, it will lead to the outcome indicated in
the table 3). Consequently, the three round game is just like a one round game where
the possible choices are (0, 0, 0) and (0, 0, 1). For this “reduced from” one round game,
we can make up a table like table 1.
Table 4
Status quo Possibilities Outcome
(0, 0, 0) (0, 0, 0) (0, 0, 0)
(0, 0, 1) (0, 0, 0), (0, 0, 1) (0, 0, 0)
Combining table 4 with table 3, we can make up the required table for the three round
game.
Table 5
Status Quo Outcome Proposal(s)
(0, 0, 0) (0, 0, 0) Anything but (1, 1, 1)
(0, 0, 1) (0, 0, 0) Anything but (1, 1, 1)
(0, 1, 0) (0, 0, 0) Anything but (1, 1, 1)
(1, 0, 0) (0, 0, 0) Anything but (1, 1, 1)
(0, 1, 1) (0, 0, 0) Anything but (1, 1, 1)
(1, 0, 1) (0, 0, 0) Anything but (1, 1, 1)
(1, 1, 0) (0, 0, 0) Anything but (1, 1, 1)
(1, 1, 1) (0, 0, 0) Anything but (1, 1, 1)
For example, consider the status quo (1, 0, 1). We know from table 3 that, if this

35
 becomes the status quo for round 2, the outcome will be (0, 0, 0). From table 4, we see
that, in the reduced form one round game, (0, 0, 0) leads to the outcome (0, 0, 0), which
therefore appears in the “outcome” column of table 5. From table 3, we see that this
outcome is obtained if anything other than (1, 1, 1) becomes the status quo for round 2.
Consequently, the agenda setter can propose any of these alternatives in round 1.
d) True. From part C, the statement is true with three rounds. Moreover, with more
than three rounds, we know the outcome will be (0, 0, 0) regardless of the status quo
inherited by the third to last round. Therefore, all play before the third-to-last round
is irrelevant, as all status quos lead to the same outcome. If the agenda setter proposes
(0, 0, 0) in the first round, all legislators are willing to vote for it, even though they prefer
(1, 1, 1).
44. (PS5, Q4)
Recall that in an extensive form game, a (pure) strategy consists of an action taken at each
information set. In this game, information sets consist of monopoly nodes and duopoly nodes.
That is, in the extensive form, each player has exactly t + 1 time t information sets - one
duopoly node (no one exited in time 0 . . . t − 1) and t monopoly nodes (opponent exited at
each time 0 . . . t − 1).
Using backward induction on the subgames that begin with monopoly nodes, actions in
monopoly nodes are easily specified:
Firm A:
t = 1 . . . 19 ⇒ stay in
t = 20 ⇒ α stay in, 1 − α out
t > 20 ⇒ out
Firm B:
t = 1 . . . 24 ⇒ stay in
t = 25 ⇒ β stay in, 1 − β out
t > 25 ⇒ out
We now must specify actions taken at duopoly nodes. Again we rely on backward induction.
In the duopoly subgame beginning at any t > 25, for each firm, any strategy specifying “out”
at time t strictly dominates any strategy specifying “stay in”. Thus, “out” is chosen by each
firm at each duopoly node with t > 25.
Consider the t = 25 duopoly subgame. For firm A, any strategy specifying “out” strictly
dominates any strategy specifying “in”, so A plays out at t = 25. Firm B mixes between
“out” and “in”.
Consider the t = 21−24 duopoly subgames. Here A’s dominant strategy is to exit immediately
regardless. It follows that B stays in and earns monopoly profits.
Consider the t = 20 duopoly subgame. Here we see that it is a dominant strategy for B to
play “in”, because in the continuation subgame, A will drop out in the next period and B
will reap 4 periods of monopoly profit. Thus, A must play “out”.
Consider the t = 11 − 19 duopoly subgames. For each t, it is a dominant strategy for B to
play “in” because (by backward induction) in the continuation subgame, A will drop out in
the next period. Thus A plays “out”.
Consider the t = 10 duopoly subgame. “in” is a dominant strategy for A because duopoly
profits are positive. “in” is also a dominant strategy for B because as before, A will drop out
in the next period.

36
 In the t = 0 − 9 subgames, duopoly profits are positive for both, so both play “in”.
Thus, the SPE take the following form:
A’s strategy:
Monopoly node: Stay in 0-19. Any mixture at 20. Out after 20.
Duopoly node: Stay in 0-10. Exit 11 or later.
B’s strategy:
Monopoly or duopoly node: Stay in 0-24. Any mixture at 25. Out 26 or later.

45. (PS6, Q3)

a) The planner solves

max [yp + y1 + y2 ]
ap ,a1 ,a2

which yields ap = a1 = a2 = 1. The resulting total family income is 11.5

b) Each player i has a dominant strategy of choosing ai = 3. This yields a unique NE of
(3,3,3). The resulting total family income is 5.5.
a2
c) Notice that by naming action ai , i creates a (combined) negative externality of 3i on
the others. By taxing i the value of this externality (a Pigouvian tax), i internalizes his
a2
effect on the others. Thus, let i’s transfer be ti (ai , aj , ak ) = − 3i . i’s problem is now
" ! #
a2i a2j a2k a2i
max ki + ai − − − −
ai 6 6 6 3

which has solution ai = 1, coinciding with the social optimum.7

d) Let’s start with the subgame where the parent faces incomes yp , y1 , y2 . The parent solves
 
1 1
max ln cp + ln c1 + ln c2 subject to cp + c1 + c2 = yp + y1 + y2 .
cp ,c1 ,c2 2 2

This is a standard Cobb-Douglas maximization problem (with prices all 1 and income
yp + y1 + y2 ). The parent will redistribute the income in shares 1 : 12 : 12 . Therefore, the
goal of all parties going into the subgame is to maximize total income in stage 1.
Therefore, in stage 1, i solves
max[yp + y1 + y2 ]
ai

which yields solution ai = 1 for all i, coinciding with the social optimum.
The SPNE is thus
Children: name ai = 1 for i = 1, 2
1
Parent: name ap = 1; facing Y = yp + y1 + y2 → redistribute in shares 1 : 2 : 12 .

46. (June 2003 Comp, Q4)

7
This tax does not balance the budget, but it is a simple matter to find one that does. One possibility is
a2 a2 2
j +ak
ti (ai , aj , ak ) = − 3
i
+ 6
.

37
 a) The normal form of the game is

A/B C D
x x
C 2,1 − 2 x, 1 − x
1−x 1+x
D 1 − x, x 2 , 2

For B, neither strategy is strictly dominant. We can identify behavior using the following
table:

x A’s dominant strategy PSNE

∈ 0, 31


D (D, D)
= 13
None (D, D)
1 2
3, 3 None None
= 23 None (C, C)
> 23 C (C, C)

If x ∈ 0, 13 , (D, D) is the unique rationalizable



profile, and therefore, the unique Nash
2

equilibrium. Similarly, if x ∈ 3 , 1 , (C, C) is the unique rationalizable profile, and
therefore the unique Nash equilibrium. If x = 13 , observe that if A plays C with positive
probability, any best response by B involves playing C with positive probability. Note
that for every strategy of B’s that puts positive probability on C, A does strictly better
by playing D than by putting any weight on C. Therefore, the unique NE is (D, D).
Similarly, if x = 23 , the unique NE is (C, C).
Now consider x ∈ 13 , 32 : the unique N E is the MSNE ((1 − x) C + xD, (3x − 1) C + (2 − 3x) D).

b) Given a deployment x, any SPNE involves the play of the corresponding NE above.
Therefore, a choice of x results in expected payoffs for A as given below:

x A’s expected payoffs

1 1−x
≤ 3 2
1 2

3x(1−x)
∈ 3, 3 2
2 x
≥ 3 2

Observe that maxx∈( 1 , 2 ) 3x(1−x)

2 = 83 < 1
2 = maxx∈[0, 1 ] 1−x
2 = maxx∈[ 2 ,1] x2 . Therefore,
3 3 3 3
in any SPNE, A sets x = 0 or x = 1.

47. (June 2003 Comp, Q5)

a) In Version #1, i receives payoffs

S (n) = −bn if i chooses Out

B (n) = c + dn if i chooses In

No player has dominant strategies; in particular, there is a PSNE where every player
chooses OU T and receives payoff S (0) and another PSNE where every player chooses IN
and receives payoff B (N − 1). It can be shown that these are the only two equilibria.
Let there be another equilibria where there exists M ∈ (0, N ) such that individuals
{1, ..., M } play IN and individuals {M + 1, ..., N } play OU T (given that payoffs are
symmetric, any other equilibrium is equivalent to some such equilibrium). For individual
M to be playing a best-response, it must be that B (M − 1) > S (M − 1). Since

38
 B (M ) > B (M − 1) > S (M − 1) > S (M ), it must be a unique best-response for M + 1
to also play IN , leading to a contradiction.
Since both equilibria exist regardless of how S (0) relates to B (N − 1), there is no
tendency for the equilibrium to necessarily be efficient.
b) In Version #2, i receives payoffs

S (i, n) = ai − bn if i chooses Out

B (i, n) = c + dn − ei if i chooses In

The game is indeed dominance solvable and involves every individual choosing IN . Let
R (i, n) = B (i, n) − S (i, n), and let the abbreviation ISD stand for iterative elimination
of strictly dominated strategies.
Lemma 1: It is strictly dominated for 1 to choose OU T .
Proof : Observe that the incremental utility of going IN for player 1 when n others
choose IN is
R (1, n) = c − e − a + (d + b) n

Observe that for all n > 0, R (1, n) > R (1, 0) = c − e − a > 0. Therefore, regardless of
the choice of others, it is a dominant strategy for 1 to choose IN .
Lemma 2: If ISD prescribes that individuals {1, ..., M } must choose IN , then ISD
prescribes that M + 1 also chooses to go IN .
Proof : Observe that R (M + 1, k) is strictly increasing in k. After ISD prescribes that
individuals {1, ..., M } choose IN , then there are at least M players who choose IN .
Therefore, for every n ≥ M ,

R (M + 1, n) ≥ R (M + 1, M )
= R (M, M − 1) + (d − e) − (a − b)
≥ R (M, M − 1) > 0

where the last inequality follows from the fact that ISD prescribed that individual M
choose IN . Therefore, ISD prescribes that individual M + 1 chooses to go IN .
Therefore, by induction, in any rationalizable strategy profile, all individuals choose to
go IN . Thus, this is what happens also in every CE and MSNE.
c) For the case of N = 3, observe that by (b), on the path of play, all 3 individuals shall
choose to go IN . We have to specify off-path behavior for a SPNE, and this involves
the computation of multiple tedious cases.

48. (2004 Final, Q2)

a) Normal Form
c n
c 2a, 2a a, 1 + a
n 1 + a, a 1, 1
Contributing yields a < 1, so not contributing is a dominant strategy. Thus, the unique
NE is (n, n).
b) Notice that punishing is costly, so no player will choose to punish his opponent in the
subgame. The SPNE involves (n, n), followed by no punishment in the second period.

39
 c) i. Now players care about equity. Inequity in monetary payoffs is not desirable, where
inequity in favor of your opponent is worse than inequity in your favor.
ii. Normal form:
c n
c 2a, 2a a − α1 , 1 + a − β2
n 1 + a − β 1 , a − α2 1, 1
Notice that (n, n) is still a NE, as deviating to c reduces payoff and decreases equity.
(c, n) and (n, c) can never be equilibria, as deviating from c to n increases payoff and
decreases inequity. However, (c, c) can be sustained provided players have enough
distaste for inequity in their favor. This condition is given by 2a > 1 + a − βi , or

βi > 1 − a for i = 1, 2.

iii. Imagine that players start the sub-game with wealth levels Y1 and Y2 .
Suppose Y1 > Y2 . There is no equilibrium in which only 1 punishes, as this reduces
payoff for 1 and decreases equity. For 2, punishing 1 dollar reduces payoff by 1, but
via the reduction in inequity, increases payoff by α2 (K − 1). This will be profitable
1
provided α2 (K − 1) > 1 or α2 > K−1 . In this case, 2 will punish until the players
have equal monetary payoff:
Y1 − Y2
Y1 − Kp = Y2 − p ⇒ p∗ = .
K −1
K 1
The resulting payoffs are π1 = π2 = K−1 Y2 − K−1 Y1 .
1
If instead we have α2 < K−1 , no player punishes. If α2 = K−1 1
, 2 picks any p ≤ p∗ .
The case Y1 < Y2 is symmetric to the case above.
If Y1 = Y2 , neither player punishes.
1
There are other equilibria. For example, if both players have αi > K−1 and Y1 > Y2 ,
then (p1 , p2 ) is a NE when
Y1 − Y2
p2 = p1 + .
K −1
1
Call i an enforcer if i has αi > K−1 .
If neither player is an enforcer, then there will be no punishment in the subgames,
and the equilibria are as in (d).
If both players are enforcers, then (n, n) can be sustained in equilibrium, but so too
can (c, c). Notice that i would never deviate to n, as that would reduce j’s stage
1 payoff, and following punishment, players will have equal payoffs, which will be
lower than j’s payoff from the first stage. (n, c) and (c, n) can never be sustained,
as deviating from c to n increases payoff, and reduces punishment.
If only 1 is an enforcer, (n, n) can be sustained and (c, n) and (n, c) cannot, for
reasons identical to the two enforcers case. (c, c) can be sustained provided 1 has
strong distaste for inequity in his favor: β1 > 1 − a, the condition from part (d).
If only 2 is an enforcer, the analysis is identical.

49. (2004 Final, Q3a)

a) i. In period 1, each player chooses to clap (C) or not clap (N ). Should both players
have clapped in period 1, each player chooses C or N in 2. Thus, we have S1 =
S2 = {CC, CN, N C, N N }.

40
 Normal form:

CC CN NC NN
d
CC 2 − 2c, d2 − 2c d − 2c, −c d − c, 0 ∗ d − c, 0 ∗
CN −c, d − 2c −c, −c d − c, 0 d − c, 0
NC 0, d − c ∗ 0, d − c 0, 0 0, 0
NN 0, d − c ∗ 0, d − c 0, 0 0, 0

ii. The PSNE are marked with a * above.

iii. We look at behavioral mixed strategies, keeping Kuhn’s theorem in mind.
There is a unique proper subgame which is reached when both players clap in period
1. Notice that in the final period, cost c is sunk for each player. Clapping costs
c and yields at worst d2 > c, so clapping is a dominant strategy. To find the first
period action, we have reduced normal form:

C N
d
C 2 − 2c, d2 − 2c d − c, 0
N 0, d − c 0, 0

This reduced game has 3 equilibria. (C, N ) and (N, C) are pure equilibria.
(λC + (1 − λ)N, λC + (1 − λ)N ) is the mixed strategy equilibrium, where λ = 2(d−c)
2c+d .
Thus, we have three SPNE, corresponding by the actions given above, combined
with (C, C) in the subgame. Corresponding payoffs are (d − c, 0), (0, d − c) and
(0, 0).
b) i. Starting in the third period, observe that clapping is a dominant strategy. Moving
back to the second period, clapping is again a dominant strategy. This is the case
since d2 > 4c, clapping guarantees a positive payoff, even if one is forced to split the
prize in the end. We now find actions in the first period by looking at the reduced
normal form:
C N
d
C 2 − 3c, d2 − 3c d − c, 0
N 0, d − c 0, 0
This reduced game has 3 equilibria. (C, N ) and (N, C) are pure equilibria.
(λC + (1 − λ)N, λC + (1 − λ)N ) is the mixed strategy equilibrium, where λ = 2d−2c
d+4c .
Thus, we have three SPNE, corresponding by the actions given above, combined
with (C, C) in periods 2 and 3. Corresponding payoffs are (d − c, 0), (0, d − c) and
(0, 0).
ii. Starting in period 3, we see that clapping is a dominant strategy. To find period
2 actions, we solve the same reduced normal form game as in (a) iv, which yielded
three possible outcomes.8 To find first period actions, we examine the reduced
normal form using each of these three second period outcomes:
8
Note that in this case, we have a sunk cost of c, so all payoffs are reduced by c. But this does not affect our
equilibria.

41
 Second period actions (C, N ) → outcome (d − c, 0) →

C N
C d − 2c, −c d − c, 0
N 0, d − c 0, 0

This game is dominance solvable with unique NE (C, N ). Thus, we have

SP N E 1 = (CCC, N N C) (18)

Similarly, second period actions (N, C) → outcome (0, d − c) →

SP N E 2 = (N N C, CCC) (19)

The third outcome involved payoffs (0, 0), yielding reduced normal form:

C N
C −c, −c d − c, 0
N 0, d − c 0, 0

This reduced game has 3 equilibria. (C, N ) and (N, C) are pure equilibria.
(µC + (1 − µ)N, µC + (1 − µ)N ) is the mixed strategy equilibrium, where µ = 1 − dc .
Thus, we have three SPNE, corresponding by the actions given above, combined
with the mixture λ in period 2, and (C, C) in period 3. Corresponding payoffs are
(d − c, 0), (0, d − c) and (0, 0). To sum up,

SP N E 3 = ((C, λC, C), (N, λC, C)) (20)

SP N E 4 = ((N, λC, C), (C, λC, C)) (21)
SP N E 5 = ((µC, λC, C), (µC, λC, C)) (22)

c) i. Costs (T − 1)c are sunk. Clapping yields an additional

−c + (1 − p)d + p · p [−c + (1 − p)d + p · p [−c + (1 − p)d + p [. . .

which is a geometric series, reducing to

1
[−c + (1 − p)d] .
1 − p2
ii. We look for a stationary equilibrium in which players clap with probability p in
every period. For A to be indifferent between clapping and not clapping, we must
have
1
[−c + (1 − p)d] = 0
1 − p2 |{z}
| {z } payoff from not clapping
payoff from clapping

which yields
d−c
p∗ = .
d
Equilibrium payoff is (0, 0).

42
 iii. One asymmetric equilibrium is that 1 chooses to clap in every period and 2 chooses
not to clap in every period. This yields payoffs (d − c, 0). There are other equilibria,
for example, asymmetric equilibria with a tail in which players clap with probability
p∗ .
50. (2007 Final, Q2)
a) i. Candidate 1’s strategy set is who he will attack: S1 = {2, 3}. Player 2 has two
information sets in which he can make a decision (1 attacked 2 unsuccessfully, 1
attacked 3 unsuccessfully). In the other situations, he is not in the game or can
only attack 1 (which we will take as a set move rather than a choice). Therefore,
his strategy set is S2 = {11, 13, 31, 33}. Finally, player 3 has four information sets
in which he can make a decision (1 attacked 2 unsuccessfully then 2 attacked 1 un-
successfully, 1 attacked 2 unsuccessfully then 2 attacked 3 unsuccessfully, 1 attacked
3 unsuccessfully then 2 attacked 1 unsuccessfully, 1 attacked 3 unsuccessfully then 2
attacked 3 unsuccessfully). So, his strategy set contains 16 elements: S3 = {1, 2}4 .
Note that even though player 3 has 4 information sets, the actions and payoffs are
the same for each.
ii. Two things to note about this game: (1) whenever it is a candidate’s turn to attack,
if the number of surviving opponents is less than 2, he has no choice to make and
(2) for each i = 1, 2, 3, all subgames starting with candidate i attacking while both
opponents still surviving are the same in terms of payoff, disregarding the history.
So this game essentially consists of three (classes of) subgames, namely, for each
i = 1, 2, 3, Gi denotes the (class of) subgame starting with candidate i deciding
between attacking one of his two opponents.
First consider subgame G3 . Candidate 3’s survival probability is now 1. Given
this, he would like to attack the opponent with higher pi . Therefore, candidate 3’s
strategy is to attack 1 if p1 > p2 , and to attack 2 if p1 < p2 .
Now consider subgame G2 . If p1 > p2 , candidate 2’s survival probability is p2 (1 −
p3 ) + (1 − p2 ) if he attacks 1 and p2 + (1 − p2 ) if he attacks 3; if p1 < p2 , candidate 2’s
survival probability is p2 (1−p3 )+(1−p2 )(1−p3 ) if he attacks 1 and p2 +(1−p2 )(1−p3 )
if he attacks 3. In both cases, attacking 3 yields strictly higher survival probability
than attacking 1 for candidate 2. Therefore, candidate 2’s strategy is to attack 3.
Finally consider subgame G1 (the whole game). Note that whomever candidate 1
attacks, if his attack is not successful, we are in subgame G2 which has a unique
NE as shown above; therefore 1’s surviving probability conditional on his attack not
successful is the same whether he attacks 2 or 3. Now if his attack is successful,
attacking 2 gives him surviving probability of 1 − p3 (since now that 2 is out, 3
will for sure attack 1), and attacking 3 gives him surviving probability of 1 − p2 .
Therefore, candidate 1’s strategy is to attack 2 if p2 > p3 and to attack 3 if p2 < p3 .
To summarize, the unique SPNE is: candidate 1 attacks 2 if p2 > p3 and attacks 3 if
p2 < p3 ; if candidate 1’s attack is not successful, candidate 2 attacks 3; If neither 1
nor 2’s attack is successful, candidate 3 attacks 1 if p1 > p2 , and attacks 2 if p1 < p2 ;
in all other events, no choice needs to be made.
iii. If p3 < p2 , candidate 3 survives with probability p1 + (1 − p1 )(1 − p2 ); if p3 > p2 ,
candidate 3 survives with probability (1 − p1 )(1 − p2 ).
b) As in the previous part, we only need to consider subgames G1 , G2 and G3 .
First consider subgame G3 . Candidate 3’s survival probability is now 1. Given this, he
would like to minimize the danger posed by the survival of other opponents and therefore

43
 will not want to spend on the charity cause. So he will attack the opponent with higher
pi (just like in the previous part). Now given that p1 < p2 , he will attack 2.
Now consider subgame G2 . Since p1 < p2 , candidate 2’s survival probability is p2 (1 −
p3 ) + (1 − p2 )(1 − p3 ) if he attacks 1, p2 + (1 − p2 )(1 − p3 ) if he attacks 3, and 1 − p3 if
he spends on charity. Attacking 3 yields strictly higher survival probability. Therefore,
candidate 2’s strategy is to attack 3.
Finally consider subgame G1 (the whole game). Given the outcomes of the subgames,
if candidate 1 spends on charity his survival probability is 1. If he attacks either of the
two opponents, his survival probability is strictly less than 1. Therefore candidate 1’s
strategy is not to attack but instead to spend on the charity cause.
To summarize, the unique SPNE is: candidate 1 does not attack; if candidate 1 does
not attack or his attack is not successful, candidate 2 attacks 3; if there has been no
successful attack, candidate 3 attacks 2; in all other events, no choice needs to be made.
It differs from the SPNE in part (1) only in that candidate 1 now does not attack. This
will ensure that 2 and 3 fight between themselves and therefore ensure himself a survival
probability of 1.

51. (June 2007 Comp, Q1)

52. (Sept 2007 Comp, Q2)

a) The extensive form is straightforward. Pure strategies: each player makes a decision at
only one information set, so the pure strategy set for each is {perform,wait}, except for
the leader at t = 4, who has no choices (and who therefore is not really a player – we will
henceforth ignore him). There are eight strategy profiles of the form {s1 , s2 , s3 }, where
si is either perform or wait. All strategy profiles with s1 = perform lead to payoffs of
(4,8,8). All strategy profiles with s1 = wait and s2 = perform lead to payoffs of (6,2,6).
All strategy profiles with s1 = s2 = wait and s3 = perform lead to payoffs of (4,4,0).
Finally, the strategy profile (wait,wait,wait) leads to payoffs of (1,1,1).
b) i. Leader 3 will wait (0 < 1). Therefore, leader 2 will perform (2 > 1). Therefore,
leader 1 will wait (4 < 6). The task will be performed at t = 2.
ii. Payoffs are (6,2,6). Only performing at t = 1 makes leader 3 better off, but that
makes leader 1 worse off. Therefore, the outcome is Pareto efficient.
iii. Yes, the outcome must be Pareto efficient. The proof is simple: There can’t be an
earlier performance date that makes everyone better off; if there was, the leader at
that date would break the equilibrium by performing the task. Nor can there be a
later date that makes everyone better off, because vt − ct is assumed to be strictly
decreasing in t.
c) i. Leader 3 will wait (0 < 1). Leader 2 will wait (2 < 4). Leader 1 will wait (4 < 6).
So the task will be performed at t = 4.
ii. Payoffs are (1,1,1), compared with (6,2,6) in the subgame perfect Nash equilibrium,
so everyone is worse off when they behave naively. Clearly, the outcome is not
Pareto optimal.
iii. The task is performed later with naive players than in the subgame perfect Nash
equilibrium (t = 4 instead of t = 2). More generally, the task is performed weakly
later when agents are naive. Let Vte be the continuation utility that leader t expects
to receive in equilibrium if period t has arrived and he decides to wait. Let Vtn be
the continuation utility that a naive leader t expects to receive if period t has arrived

44
 and he decides to wait. Obviously, Vtn ≥ Vte . Therefore, if wait is the equilibrium
strategy for leader t, it must also be the naive choice for leader t. Consequently,
the team cannot perform the task sooner if all leaders are naive.

53. (Sept 2009 Comp, Q1)

a) i. The strategy set for player H is the set [0, 1] and the strategy set for player S is
the set of functions from [0, 1] to [0, 1]. A typical strategy for H is 0.3 and a typical
strategy for S is the function s(h) = h.
ii. A pure strategy Nash equilibrium is: h = 0.5 and s(h) = h.
iii. Any outcome where h and s play the same x ∈ [0, 1] can be a pure strategy Nash
equilibrium.
iv. Subgame perfection does not narrow down the set of possible outcomes.
b) i. The strategy set for player S is the set [0, 1] and the strategy set for player H is
the set of functions from [0, 1] to [0, 1]. A typical strategy for S is 0.3 and a typical
strategy for H is the function h(s) = s.
ii. A pure strategy Nash equilibrium is: s = .5 and h(s) = 1(s ≤ 21 ), where 1(.) is the
indicator function.
iii. The only two pure strategy Nash equilibrium are one where (s, h) = (.5, 1) and
another where (s, h) = (.5, 0).
iv. Subgame perfection does not narrow down the set of possible outcomes.
c) i. The strategy set for both S and H now is just the set [0, 1]. Thus, a typical pure
strategy is s = 0.4 and h = 0.2.
ii. We will rule out two possible PSNE. First, suppose there is a PSNE with (s, h) =
(x, y) for x 6= y. Then, H can deviate and play x and increase her payoff. Thus,
s, h) = (x, y) is not a PSNE. Next, suppose there is a PSNE with (s, h) = (x, x) for
some x. There exists a  ∈ [−1, 1] such that x +  ∈ [0, 1]. So H can improve his
payoff by playing x +  for some . So this cannot be an equilibrium either.
iii. Remember, the first step to a problem like this is to first write S’s payoff as a
function of s, conditional on H’s randomization. This is:
Z 1
ΠS (s|F ) = −(h − s)2 dF (h)
0

Our payoff is concave, so we can just take the first derivative and set it equal to zero
to find that s ∗ (f ). We get that s∗ = E[h].
iv. Again, we can write that:
Z 1
ΠH (h|F ) = (h − s)2 dG(s)
0

This time, however, the payoff is convex, not concave. Thus, setting the first deriva-
tive to zero gives us the minimum not the maximum. The convexity means that
H will be pushed to the corners; we can show that h ∗ (G) is 0 if E[s] < 21 , 1 if
E[s] > f rac12, and either 0 or 1 if E[s] = 21 .
v. A mixed strategy Nash equilibrium is then where H plays zero with fifty percent
chance and one with fifty percent chance, and S player 12 for sure. We can show using
the above results, that given their opponents response, both players are optimizing.
Thus, it’s a Nash equilibrium.

45
 vi. The above results show that if E[s] is anything other than 0.5, the hider has to
play only h = 0 or h = 1 and cannot randomize at all. If he doesn’t randomize,
however, then S must play h and we’re back to fully pure strategies, and we’ve
shown there cannot be a PSNE. So it must be that E[s] = 12 . But we’ve also shown
that s∗ (F ) = E[h]. So the only way that it can be 12 with H randomizing between
0 and 1 is for H to split between the two equally.
d) If H randomizes uniformly around the circle, then S gets the same payoff for every point
on the circle. Likewise, if S randomizes uniformly around the circle, then H gets the
same payoff for every point on the circle. Thus, it is a mixed strategy Nash equilibrium
for both of them to randomize uniformly around the circle.

54. (2010 Final, Q1)

a) A strategy is a mapping from information sets to actions. In this game, every action is
observed by both players, so at any time t ≥ 0 the information known is the history of
moves made by the players, call it ht = (m0 , m1 , . . . , mt−1 ) ∈ {e, w}t−1 . Player 1 only
drives on odd days and Player 2 only drives on even days. Thus, a strategy for Player
1 is a mapping s1 : {ht ∈ {e, w}t−1 |t is odd} → {e, w} and a strategy for Player 2 is a
mapping s2 : {ht ∈ {e, w}t−1 |t is even} → {e, w}.
To simplify the problem, we will express strategies a function of a time and position pair
(t, k), where k denotes the “driving” positions on the map:

M = {−(x − 1), ... − 1, 0, 1, ..., x − 1}

Note that when x ≥ 2, different histories can lead to the same position at the same
point in time. So technically, a player’s strategy could specify different actions based on
different histories. This won’t make any difference in the arguments that follow, so we
will ignore the possibility for simplicity of exposition.
b) Player 2 is never called on to act in this case, so the problem is just an optimization
problem “in solitude” for Player 1. Since v < v by assumption, the only Nash equilibrium
and the only subgame perfect Nash equilibrium of the game is given by the strategy
s∗1 (0) = w, leading to payoffs (δv, δv).
c) i. No. Under any strategy profile either x = 1 or x = −1 is reached. If the players
remain oscilating forever, their payoffs are (0, 0) and it must be the case that at
whichever of these positions is reached, Player 2’s strategy prescribes going back to
0. He could deviate and prescribe going to the nearest endpoint and get strictly
positive profits.
ii. Consider the strategies given by:

s1 (t, 0) = e ∀ odd t

s2 (t, 1) = s2 (t, −1) = e ∀ even t

Player 2 is getting his prefered outcome, so it is clear that he has no incentive to
deviate. By deviating, Player 1 can only postpone the arrival of the party at the far
east, thereby hurting himself. So he has no incentive to deviate either.
iii. For a strategy profile s∗ to be a Nash equilibrium in which the Far East is reached
(and Nash is a necessary condition for subgame perfect Nash), it must be the case
that s2 (2, −1) = e. Otherwise Player 1 could set s1 (1, 0) = w and they would

46
 reach the Far West at t = 2, which is Player 1’s best outcome. For Player 2 to
set s2 (2, −1) = e in a subgame perfect equilibrium, it must be the case that the
payoff that he gets by doing so is at least δ 2 v, which is what he would get by setting
s2 (2, −1) = w. The most that he could hope to get by setting s2 (2, −1) = e is δ 4 v
(if they went all the way to the Far East after his move). So a necessary condition
for an equilibrium of this kind is v ≤ δ 2 v. If v > δ 2 v, there is no such equilibrium.
For the case v ≤ δ 2 v, consider the strategies that we proposed in part (ii). It can
be verified that they constitute a subgame perfect Nash equilibrium.
iv. Let t∗ be the largest even t such that v ≤ δ t−2 v and consider the strategies given
by:
s1 (t, 0) = w ∀ odd t < t∗ − 1, s1 (t, 0) = e ∀ odd t ≥ t∗ − 1
s2 (t, −1) = e ∀ even t, s2 (t, 1) = w ∀ even t < t∗ , s2 (t, 1) = e ∀ even t ≥ t∗
Note that in this proposed equilibrium the Far East is reached whent t = t∗ and
∗ ∗
the payoffs are (δ t v, δ t v). Player 1 has no profitable deviation because altering his
strategy by going west (w) at t ≥ t∗ − 1 only weakly delays the arrival of the party
at the Far East, and by going east (e) at t < t∗ − 1 does not affect the outcome
given the “self-flagellating” strategy of Player 2. Player 2 has no profitable deviation
either. It is clear that he can’t possibly improve by switching s2 (t, 1) from e to w for
t ≥ t∗ , since this can only delay the arrival of the party at the Far East. Moreover,
given that position k = 1 is never reached when t < t∗ on the equilibrium path
(given Player 1’s strategy), switching s2 (t, 1) from w to e for t < t∗ has no effect.
Since on the equilibrium path, the last time that the party visits −1 is at t∗ − 2,
switching s2 (t, −1) from e to w has no effect for t > t∗ − 2. Now lets check that
Player 2 prefers to oscilate rather than to concede going to the Far West by setting
s2 (t, −1) = w at some point t ≤ t∗ − 2. It is clear that if he finds it profitable to
do this, then the sooner the better. The soonest he could break the oscilation is
at t = 2 when they first arrive at k = −1. If Player 2 were to deviate by setting
s2 (2, −1) = w, he would get δ 2 v, which is strictly less than his equilibrium payoff
∗ ∗
δ t v given that v ≤ δ t −2 v.

So we have established that there is a Nash equilibrium in which the party reaches
the Far East at time T = t∗ . No such equilibrium exists for T > t∗ and we prove
it by contradiction. Suppose that there exists equilibrium s∗ in which the Far East
(F E) is reached at even time T > t∗ (T has to be even simply because endpoints
are always reached in even periods). In any Nash equilibrium it must be true that
if position k = 1 is reached at time t on the equilibrium path, then s2 (t, 1) = e (i.e.,
Player 2 uses the first actual opportunity to get to the Far EAst). This implies that
in s∗ , the first time that k = 1 is reached is actually T − 1. But this implies that
s1 (t, 0) = w for all t < T − 1. In particular s1 (1, 0) = w. This in turn implies that
s2 (2, 1) = e (otherwise the game would end in the Far West when t = 2). However
setting s2 (2, 1) = w would yield δ 2 v to Player 2 and this is greater than δ T v because
T > t∗ and by assumption t∗ is the largest even t such that v ≤ δ t−2 v.
v. In any subgame perfect equilibrium, it must be true that s2 (t, 1) = e for all even t.
So if the Far East is reached with delay (T > 2), it must be that Player 1 is delaying
the arrival. In particular, it must be that s1 (1, 0) = w. Player 1 would be strictly
better off accelerating the process, by setting s1 (1, 0) = e and getting δ 2 v instead of

47
 δ T v. So the answer is T = 2. If the Far East is the outcome in a subgame perfect
equilibrium, then it must be reached with no delay.
d) i. See (c)(i).
ii. Consider the strategies given by:

s1 (t, 0) = w ∀ odd t

s2 (t, 1) = s2 (t, −1) = w ∀ even t

Player 1 is getting his preferred outcome so it is clear that he has no incentive to
deviate. By deviating, Player 2 can only postpone the arrival of the party at the
Far West, thereby hurting himself, so he has no incentive to deviate either.
iii. Consider the strategies given by:

s1 (t, 0) = w ∀ odd t

s2 (t, 1) = e ∀ even t, s2 (t, −1) = w ∀ even t

In any subgame in which Player 1 is called to drive, he gets his prefered outcome, so
he never has an incentive to deviate. In subgames that start at position k = 1, Player
2 is acting optimally by setting s2 (t, 1) = e. So we just need to check subgames in
which Player 2 is asked to drive at position k = −1. Given Player 1’s strategy, by
deviating in such a subgame Player 2 can only delay the arrival at the Far West, so
he is strictly better off not deviating.
iv. There are no Nash equilibria in which the party arrives at the Far West at time
T > 2. The reason is that in a Nash equilibrium that ends at the Far West, k = 1
can’t be reached on the equilibrium path, since at that point Player 2 would deviate
by choosing (e). This means s1 (t, 0) = w for all t. But this means that unless
s2 (2, −1) = w, Player 2 is not best responding, since he would just be delaying the
party’s arrival at the Far West.
v. See (d)(iv).

55. (June 2010 Comp, Q2)

a) A strategy for player t is a mapping st : Rt− → R− . That is, for any feasible (x0 , .., xt−1 ),
it assigns some xt ≤ 0.
b) i. On the equilibrium path, we obviously must have x1 = 0. Every pair (x∗ , 0) with
x∗ ≤ 0 is a Nash path. Simply let
 ∗ 
x − x0
s1 (x0 ) = min ,0 .
δ

Then player 0 earns a discounted payoff of x∗ from all x0 ∈ [x∗ , 0], and a payoff of x0
from all x0 < x∗ . Thus, player 0 has no profitable deviation and x0 = x∗ , s1 (x0 ) =
∗
min{ x −x


δ , 0} is a NE.
0

ii. The set of subgame perfect Nash paths is simply (0, 0). We know that x1 = 0
regardless of x0 . Consequently, we must have x0 = 0 as well. If x0 = 0 and
s1 (x0 ) = 0 for all x0 , then neither player has an incentive to deviate at any node, so
we have a subgame perfect Nash equilibrium.

48
 iii. Let xmin be the lower bound. For part (i), we can sustain any path (x∗ , 0) with
x∗ ≥ δxmin . For lower x0 , no conceivable punishment is severe enough to deter a
deviation to zero. For part (ii), there is no change.
c) i. We can sustain any feasible (x∗0 , . . . , x∗T −1 , 0). Clearly, we must have xT = 0 on a
Nash path or player T would deviate. To sustain any sequence with that restriction,
use the following strategies:
s0 = x∗0
 
∗ st−1 (x0 , . . . , xt−2 ) − xt−1
st (x0 , . . . , xt−1 ) = xt + min ,0
δ
Any gain from deviation is thereby completely negated in the following period.
ii. By backward induction, only (0, 0, . . . , 0) is sustainable.
d) i. Consider the following strategies:
s0 = x
 
st−1 (x0 , . . . , xt−2 ) − xt−1
st (x0 , . . . , xt−1 ) = min ,0
δ
Note that these strategies generate the path (x, 0, 0, . . . ). Moreover, starting from
any point where the current player is supposed to play xt , they generate (xt , 0, 0, ...),
and the continuation path ( xtδ−y , 0, 0, . . . ) is used to punish any deviations by the
current player to y > xt . Because the PDV of (y, xtδ−y , 0, 0, . . . ) equals the PDV of
(xt , 0, 0, . . . ), no player has an incentive to deviate at any node, so this is a subgame
perfect Nash equilibrium.
ii. Let Xt = ∞ k−t x . If the first deviation from the desired path occurs in period
P
k=t δ k
t and player t deviates by playing y 6= xt , punish the deviation by shifting to the sub-
game perfect continuation equilibrium that generates the path (min{ Xtδ−y , 0}, 0, 0, . . . ).
Because the PDV of (y, min{ Xtδ−y , 0}, 0, 0, . . . ) equals Xt for y > xt and is strictly
less than Xt for y < xt , player t has no incentive to deviate from the equilibrium
path. Because all the punishment paths are themselves subgame perfect equilibria,
the overall equilibrium is subgame perfect.
iii. With a lower bound, the only possible subgame perfect outcome will be (0, 0, . . . ).
Look at the strategies described above. If X0 < 0, then t consecutive deviations to 0
would imply that the next player would have to choose X0 δ −t , which will clearly be
below the lower bound for t sufficiently large. More generally, the problem is that,
because of positive discounting, the punishments necessary to maintain incentives
must explode without bound. Any bound limits incentives, thereby limiting what
can be supported, which further limits punishments and causes the punishments to
unravel.

56. (2012 Final, Q1)

a) i. The contractors problem is to maximize payoff by choosing effort x (or equivalently

√
by choosing µ = x). Thus, they solve the problem:

max R · µ − µ2
µ

Taking the derivative and setting it equal to zero, we quickly get that µ∗ (R) = R
2.

49
 ii. Given that the contractor will chose to set µ∗ = R2 , the DoD’s problem is to chose
R to max:
R
max B · µ∗ (R) − R · µ∗ (R) = (B − R) ·
R 2
Again, taking the derivative and setting it equal to zero, we get that R∗ = B2 .
Plugging this into the µ∗ (R) equation above, we get that µ∗ (B) = B4 .
Remember µ is the probability of success, so it cannot be more than one. According
to the equations above, if B > 4, (or more directly, if R > 2) the probability of
discovery is greater than one, which obviously cannot happen. Thus, the optimal
choices are actually capped by the constraint that the probability of discovery is
between zero and one. Thus, the full answer is that:

0 if B < 0,

∗
R (B) = B2 if B ∈ [0, 4],

2 if B > 2.


and

0
 if B < 0,
µ∗ (R) = B
if B ∈ [0, 4],
4
1 if B > 4.


b) i. The profit function for contractor 1 is now:

R
· µ2 − µ21


Π1 (µ1 |µ2 ) = µ1 · R · (1 − µ2 ) +
2
Remember that the difference between strategic complements and substitutes de-
2Π
pends on the sign of dµd1 dµ 2
, assuming that the profit is differentiable.9 Taking the
derivative in this problem we get that:

d2 Π1 R
=− <0
dµ1 dµ2 2
Now you just need to remember that this means that they are strategic substitutes.
ii. Since we’ve already written down the profit function for the first contractor, we can
just differentiate it and set that equal to zero to find the first
contractors optimal
∗ 1 R
choice. This gives us that µ1 (R, µ2 ) = 2 · R · (1 − µ2 ) + 2 · µ2 . To make this slightly
easier to work with, we can rewrite this as µ∗1 (R, µ2 ) = 21 · R − µ2 · R2 .

We can now use the fact that contractor one and contractor two are symmetric,
so by exchanging 1 and 2 in the previous equation, we also have contractor two’s
optimal choice of µ2 and a function of R and µ1 . We don’t want two separate best
response functions, but instead a Nash equilibrium. That is just the intersection of
the two, however, which we can solve for by solving finding µ∗1 (R, µ∗2 (R, µ1 )). This
gives us that:
 
∗ ∗ 1 ∗ R

1 1 R

R
µ1 (R, µ2 (R, µ1 )) = · R − µ2 · = · R− · R − µ1 · ·
2 2 2 2 2 2
9
If not, then we use the more general ideas of supermodularity/submodularity.

50
 Doing some algebra, we get that:
R · (8 − 2R) 2R
µ∗1 (R) = =
16 − R2 4+R
Remember, however, that µ1 is the probability of success, so that region which the
above equation is valid is the region where the result is when the optimal µ∗1 (R) is in
[0, 1], which only occurs when R ∈ [0, 4). Above that, µ∗1 = 1 and below that µ∗1 = 0.
These, together with a derivative showing that it is monotonically increasing in R
answers the question.
iii. The overall probability of success is just the probability that either contractor suc-
ceeds. An easier way to write this
is that it is one minus the probability that neither
2
succeeds. This is 1 − 1 − µ∗ (R) .
iv. Because of the way they pay, they pay R if at least one contractor succeeds. Thus,
all that matters is the probability listed above. We can thus solve the DoD’s problem
as:    

2
max B − R · 1 − 1 − µ∗ (R)
R

which, because of the concavity of the problem, we can just get by setting the
derivative to zero. Doing this, we get that R∗ (B) = 4+B2B
, and we can just plug this
into the previous equations to get the total probability of success.10 Even without
doing this, we know that there is a B above which success is certain. We know this
because the for high enough B, we can get R to be as large as we want. We also
know that µ∗ is strictly increasing in R until it gets to 1, which happens at a finite
number (4 to be exact). Thus, for high enough B, the optimal R will be greater
than 4 and thus µ = 1.
c) Again, this answer requires a TON of algebra. After solving for it, we find that the
answer does depend on B. At low B, the R is very low, so the difference between the
contractor’s optimal choice in 1 vs. 2 is minimal. Thus, having two contractors working
is beneficial. At high B, however, the benefit of having two is overwhelmed by the
negative effect of the strategic substitutes. To see this, note that at B = 4, we have that
the probability of success is one in part 1, but less than one (because µi < 1) in part 2.

57. (PS5, Q5)

a) Notation: a strategy for 1 in this game is given by (p1 , p2 ) where p1 , p2 and (1 − p1 − p2 )

are the probabilities 1 places on x, y, and z, respectively. A strategy for 2 is given by
q, the probability 2 places on choosing l. Beliefs for 1 will always be ( 12 , 12 ). Beliefs for
2 are (µ1 , µ2 , 1 − µ1 − µ2 ).
Observe first that y is strictly dominated by x for 1, so in any NE, and hence in any
PBE, we have p2 = 0.
We look first for equilibria where 1 places positive probability on z, so that 2’s informa-
tion set is reached with positive probability. In this case, Bayes’ law implies 2’s beliefs
must be µ1 = µ2 = 21 . Now we can pin down 2’s strategy. When 1 places positive proba-
bility on z (1−p1 > 0), π2 (l) = 10p1 +5(1−p1 ) > 10p1 +4.5(1−p1 ) = π2 (r), so 2 must play
l with probability 1 (q = 1). When 2 plays l with probability 1, π1 (x) = 2 < 3 = π1 (z),
10
I will stand by many things that I do, but algebra is certainly not one of them. If you solve for a different answer
than me, there is at least a 50% chance that you are right. Obviously, the algebra is the least important part of this
problem, both during the test and prepping for the test.

51
 so 1 must play z with probability 1 (p1 = p2 = 0). Thus, the unique PBE in which 1
places positive probability on z is:
1
1’s strategy: z 2’s strategy: l 2’s beliefs: µ1 = µ2 = (23)
2
Now we look for equilibria in which 1 plays only x. For 1 to prefer x over z, we must
have:
π1 (x) ≥ π1 (z) ⇒ 2 ≥ 21 · q · 0 + 1
2 · q · 6 + 12 (1 − q)(−10) + 12 (1 − q)(−10)
⇒ 2 ≥ 13q − 10
⇒ q ≤ 12
13 .

Then 2’s strategy gives us a restriction on 2’s beliefs.

12
If q ∈ (0, 13 ], then 2 must be indifferent between l and r. Thus, we have

π2 (l) = π2 (r) ⇒ µ1 · 10 + µ2 · 0 + (1 − µ1 − µ2 )5 = µ1 · 2 + µ2 · 7 + (1 − µ1 − µ2 )5
⇒ 8µ1 = 7µ2 .

Thus, we have PBE:

1’s strategy: x
2’s strategy: q ∈ (0, 12 13 ]
2’s beliefs: (µ1 , µ2 ) where 8µ1 = 7µ2 and µ1 + µ2 ≤ 1.
If q = 0, then 2 must (weakly) prefer r to l. Thus, we have

π2 (l) ≤ π2 (r) ⇒ µ1 · 10 + µ2 · 0 + (1 − µ1 − µ2 )5 ≤ µ1 · 2 + µ2 · 7 + (1 − µ1 − µ2 )5
⇒ 8µ1 ≤ 7µ2 .

Thus, we have PBE:

1’s strategy: x
2’s strategy: r
2’s beliefs: (µ1 , µ2 ) where 8µ1 ≤ 7µ2 and µ1 + µ2 ≤ 1.
b) First note that any PBE in which all information sets are reached with positive proba-
bility is also a sequential equilibrium, so the equilibrium given in (23) is a SE.
Now consider the equilibria where 1 plays only x. Observe that for any strictly mixed
strategy for 1, the beliefs generated for 2 must have µ1 = µ2 . It follows that the limit of
beliefs generated by any sequence of strictly mixed strategies must also have µ1 = µ2 .
Therefore, in any sequential equilibrium (SE) we must have µ1 = µ2 . Combining this
with the previous restrictions on µ1 and µ2 , the set of candidate SE is reduced to
1’s strategy: x
2’s strategy: q where q ∈ [0, 12 13 ]
2’s beliefs: µ1 = µ2 = 0.
Can we in fact find strictly mixed strategies and corresponding beliefs that converge to
the equilibrium? Yes. We just need to pick the strategies carefully, making sure the
weight 1 places on z goes to zero faster than the weight on y.
(n) (n)
Let σ1 = (1−εn −ε2n , εn , ε2n ). Let σ2 = q+εn . Observe that as n → ∞, the strategies
converge to the equilibrium strategies. Also, the beliefs converge to the equilibrium
beliefs.
1 2n
(n) (n) ε εn
µ1 = µ2 = 1 2n 21 2n 1 n = n → 0 and
2ε + 2ε + 2ε 2ε + 1

52
 1 n
(n) 2ε 1
µ3 = 1 2n 1 2n 1 n = → 1.
2ε + 2ε + 2ε
2εn+1
Thus the candidates are indeed SE.
58. (PS5, Q6)
a)
b) SA = {E, F }, SB = {GGG, GGH, GHG, GHH, HGG, HGH, HHG, HHH}
Note that player B’s strategies are in the form: (action if A chooses E and nature chooses
C, action if A chooses F , action if A chooses E and nature chooses D).
Normal form:
GGG GGH GHG GHH HGG HGH HHG HHH
E 0, 0 1.5, 1.5 0, 0 1.5, 1.5 3.5, 3.5 5, 5∗ 3.5, 3.5 5, 5
F 3, 3 3, 3 7, 7 ∗ 7, 7 ∗ 3, 3 3, 3 7, 7 ∗ 7, 7∗
c) The PSNE are marked in the normal form with a ∗.
d) Subgame perfection requires that B’s strategy specifies playing H at information sets
where A plays E. Thus, the pure strategy SPNE are (E, HGH) and (F, HHH)
e) Observe that sequential rationality implies that B play H at information sets where A
plays E. Therefore, in this problem we have that the PBE are a subset of the pure
strategy SPNE.
Let µ be the probability that B places on being at the node where nature played C,
given that he reaches the information set in which A played F .
Observe that in the candidate PBE (F, HHH), B’s information set in which A played
F is reached with positive probability. Thus, beliefs µ = .7 are determined from Bayes’
law and we have a PBE (F, HHH), µ = .7.
Consider (E, HGH). To sustain this as a PBE, B’s choice of G when facing F must be
sequentially rational. This places restrictions on µ. That is,
1
πB (G) ≥ πB (H) ⇒ 0 · µ + 10(1 − µ) ≥ 10 · µ + 0(1 − µ) ⇒ µ ≤ .
2
Thus we have PBE (E, HGH), µ ≤ 12 .
f) Observe that for any strictly mixed strategy for A, the beliefs generated for B must have
µ = .7. It follows that the limit of beliefs generated by any sequence of strictly mixed
strategies must also have µ = .7. Therefore, in any SE we must have µ = .7. Thus, the
unique SE is (F, HHH), µ = .7 (for which it is easy to show consistency).
59. (2003 Final, Q3)
a)
b) We solve for the SPNE using backwards induction. If there was a mock trial, P knows
he will earn v − C by going to trial. Thus he will reject D’s offer if it is less than v − C
and accept it if it is more than v − C. Therefore, D will offer q = v − C (because going to
trial costs him v + C > v − C) and P will accept it. (P is indifferent between accepting
and rejecting but we need him to accept for this to be an equilibrium. If he rejected
the offer, D would have an incentive to deviate to a higher q.) So, P will definitely
continue the suit if v > C and abandon the suit if v < C. He can continue or abandon
or randomize between the two when v = C. Defendants and plaintiffs expected payoffs
are E[(C − v)1v>C ] − B and E[(v − C)1v>c ] − B, respectively.

53
 c) Without the mock trial, we don’t have any proper subgames. However, for a sequential
equilibrium, players still must be optimizing at every information set given beliefs and
beliefs must be consistent. In this case, that means that beliefs are dictated by the
random variable describing nature’s choice of v. Then, since E[v] > C, P will not
abandon the suit. D will offer E[v] − C and P will accept. Payoffs are C − E[v] and
E[v] − C, respectively.
d) Given the answers to parts (c) and (d),R D would strictly prefer to go to Rmock trial if
1 1
E[(C − v)1v>C ] − B = C(1 − F (C)) − C vdF (v) − B > C − E[v] = C − 0 vdF (v) or
RC RC
0 vdF (v) > B + CF (C). If 0 vdF (v) < B + CF (C), then D will forego the mock
trial. And if equality holds, D will go to the mock trial with some probability.
e) With perfect bayesian equilibria, we do not require beliefs to be consistent. This means
a player can have any beliefs at an information set that is reached with zero probability
on the equilibrium path. We now construct a PBE that differs R C from the sequential
equilibrium constructed above. Assume B + CF (C) − 2C < 0 vdF (v) < B + CF (C)
and E[v] + 2c < 1. Then, as we know from above, in the sequential equilibrium, D would
choose to forego the mock trial. However, we can construct a PBE where D chooses to
go ahead with a mock trial. On the equilibrium path, where D chooses to do a mock
trial, everything is the same as in part (c). At the other branch, let P believe that v = 1
with probability 1. (This is an “irrational” belief, but it is not eliminated by Bayesian
updating because the information set is never reached.) Therefore, P will continue with
the suit and reject any offer less than 1 − C. Then at D’s next information set, let
D believe that v ∼ F [0, 1] and have D offer to pay an amount less than 1 − C, which
is surely rejected. Note that given D’s belief at this information set, offering 1 − C
has payoff C − 1, whereas offering anything less yields the expected payoff from trial:
−E[v] − C. And −E[v] − C > C − 1 by assumption, so D prefers to go to trial. Then at
his first information set, D chooses the mock trial. Given P ’s crazy belief, D’s expected
payoff from foregoing the mock trial is now −E[v] − C, and by our initial assumption on
the parameters, this is now less than −B + E[(C − v)1v≥C ].

60. (Sept 2006 Comp, Q2)

a) i. The strategy set for the owner is the set of all functions gO : {0, V } → R+ . The
strategy set for the buyer is the set of all functions gB : R+ → {accept, reject}.
ii. No. Suppose there is such a PBE. Then there are two possibilities:
Case 1: Separating equilibrium (gO (0) 6= gO (V )). When the value is V , the com-
pany must be sold at price p = V ; when the value is 0, the company is not sold or is
sold at price p = 0. But then, when the value is indeed 0, the owner should deviate
to p = V because gB (V ) = accept.
Case 2: Pooling equilibrium (gO (0) = gO (V )). The pooling price must be p ≥ V ,
for otherwise the owner won’t sell when the value is V . But meanwhile the pooling
price must be p ≤ πV < V , for otherwise the buyer won’t buy.
iii. Yes. The seller quotes p = 0 for value 0, and p = V for value V . The buyer believes
that the value is 0 for any p < V and V for p ≥ V and he doesn’t accept any offer.
This constitutes an PBE in which no sale is made, regardless of the value.
iv. Part (ii): In any separating equilibrium in which the good is sold when the value
is V , we must have V ≤ gO (V ) ≤ βV and gB (gO (V )) = accept. When the value
is 0, either gO (0) = 0 or gB (gO (0)) = reject. Either way, the owner would do
better deviating to gO (0) = gO (V ) because the buyer is accepting at that price.

54
 So there does not exist any separating PBE. In any pooling equilibrium, let the
pooling price be p. Then we must have p ≥ V for otherwise the owner won’t sell
when the value is V . For the buyer to buy, we must have πβV ≥ p. Together we
have πβV ≥ V , so we need πβ ≥ 1 or β ≥ π1 . In this case, we have the following
PBE for any p ∈ [V, βV ]: gO (0) = gO (V ) = p, gB (p) = accept, gB (p̂) = reject
for all p̂ 6= p, the buyer believes the value is V with probability π for p and the
value is 0 with probability 1 for all other prices.
Part (iii): Yes. The seller quotes p = 0 for value 0, and p = βV for value V . The
buyer believes that the value is 0 for any p < βV , and V for p ≥ βV and doesn’t
accept any offer. This constitutes a PBE in which no sale is made, regardless of
the value.
b) Notice that perfect Bayesian equilibria in this setting are the same as subgame perfect
Nash equilibria because every information set is a singleton.
i. The strategy set for the buyer is R+ . A strategy for the owner is a function gO :
R+ × {0, V } → {accept, reject}. The strategy set for the owner is the set of all such
functions.
ii. No. Suppose the buyer names price p. If p < V , the company is sold only when its
value is 0. If p ≥ V , it is a dominant strategy for the owner to accept the offer for
both values. But then the buyer will have a negative expected payoff and could do
better by naming p = 0.
iii. Yes. The buyer names p = 0. If the value is 0, the owner sells if and only if p > 0; if
the value is V , the owner sells if and only if p > V . This constitutes a PBE where
nothing is sold.
iv. The owner’s response function is unchanged. When the value is V , he must accept
offers p > V and reject offers p < V . When the value is 0, he must accept offers
p > 0. He can accept or reject offers that are exactly equal to the company’s value.
Therefore, the buyer will not name any price p ∈ (0, V ). If he names p ≥ V ,
both value 0 and value V will be sold, and the buyer receives an expected payoff of
πβV − p. So the condition is the same as in previous question.
When β < π1 , there exists a PBE in which no sale is ever made and there does not
exist a PBE in which a sale is made when the value is V .
When β > π1 , there exists a PBE in which a sale is made when the value is V (In
particular, the buyer names price p = V . The seller accepts offers p ≥ x when the
company’s value is x and rejects all other offers.) There does not exist a PBE in
which no sale is ever made.
When β = π1 , both types of PBE exist.
c) To avoid ambiguity, denote by q the probability that the owner has a liquidity problem.
i. The strategy set for the buyer is R+ . A strategy for the owner is a function gO :
R+ × {0, V } → {accept, reject}. His strategy set is the set of all such functions.
(When the owner has a liquidity problem he has no choice to make, so this does not
alter the owner’s strategy set.)
ii. The owner’s response function (when he is allowed to make a choice) is unchanged.
The buyer won’t name p ≥ V , as argued in the previous part. Now if he names
p < V , his expected payoff is πqV − p, which is nonnegative at any price p ≤ πqV .
So a PBE in which a sale is made when the company’s value is V is: the buyer
names price p = 0, the owner accepts offers p ≥ x when the company’s value is x

55
 and rejects all other offers. When the value is V , a sale is made only when the owner
has a liquidity problem.
iii. There is no PBE in which no sale is made because whenever the owner has a liquidity
problem a sale will be made regardless of the value of the company.

61. (2011 Final, Q2)

a) The extensive form is:

b) Professor’s strategy set: {Y ee, Y ep, Y pe, Y pp, N ee, N ep, N pe, N pp}
Student’s strategy set: {ss, sn, ns, nn}
c) See the solution to part 1.
d) There does exist such a PSNE. Y pe, nn and N ep, nn are both Nash equilibria in which
the professor plays e at some information set. In both cases, the professor cannot do
any better because, given that the student plays n at both information sets, p gives the
highest possible payoff. The student cannot do any better because on the equilibrium
path n is a best response to p and actions off the equilibrium path do not affect payoffs.
e) There does not exist such a SPNE. In each subgame, we need the professor to play p
since p strictly dominates e in each subgame.
f) An example could be any of the following strategy profiles: the professor plays Y pp with
probability α ∈ (0, 1) and N pp with probability 1 − α and the student plays nn with
probability 1.
g) A PBE must be subgame perfect. There are two subgames – after the professor chooses
Y and after the professor chooses N. In both subgames, e is strictly dominated by p. So
the professor must play p. Then the student must play n because it is the best response
to p. Therefore, given that the professor plays p at both information sets, µK = 1.
In conclusion, the set of PBE are:

(Y pp, nn), µY = 1, µN = 1

(N pp, nn), µN = 1, µY = 1
(αpp, nn), α ∈ (0, 1), µY = µN = 1
where α is the probability that the professor plays Y .

56
 h) In a sequential equilibrium, beliefs must be consistent. Consider strictly mixed strategies
where qan > 0 is the probability that the professor plays action a. Then Bayes rule says:
n qn
qK n
qp|K
p|K
µK = lim n qn + qn qn = lim n + qn =1
n→∞ qK p|K K e|K
n→∞ qp|K e|K

n → 1 and q n
because qp|k e|K → 0 (in all of the equilibria in part 7 the professor plays p at
both information sets).
Therefore, the set of SE are:

(αpp, nn), α ∈ [0, 1], µY = µN = 1

where α is the probability that the professor plays Y .

i) The new extensive form looks like:

j) In the “Y” subgame, the unique Nash equilibrium is (e, s) because e is a strictly dominant
strategy and s is the student’s best response to e. In the “N” subgame, the unique Nash
equilibrium is (p, n) because p is a strictly dominant strategy and n is the student’s best
response to p. The professor strictly prefers the payoff in the “Y” subgame to that in
the “N” subgame. Thus the unique PBE is (Y ep, sn), µY = 0, µN ≥ 12 . (These beliefs
come from the analysis in part 7 because the student’s payoffs are the same.)

62. (Sept 2011 Comp, Q3)

63. (Sept 2012 Comp, Q2)

a) Since firm B does not have an opportunity to adopt the technology, we know that the
payoff for firm A is X if they do not adopt the technology and Z − P if they do. Since
we are only interested in a subgame perfect Nash equilibrium, firm A must optimize in
the last period, which means that they must adopt the technology whenever Z − P ≥ X.
Thus, firm C will set P = Z − X.
b) Again, we start at the end and determine whether or not player B will adopt the tech-
nology. We must also condition on whether or not player A adopted the technology
(which we can because that decision is observable). Thus, if player A adopts, player B
will adopt the technology iff Y − PB ≥ W . If not, then player B adopts iff Z − PB ≥ X.

57
 Given that, in the penultimate stage, firm C offers a price of PB = Y − W in the nodes
after A adopts and PB = Z − X in the nodes after A does not. Again, to find the full
SPNE, we continue to move backward. From before, we get that B adopts regardless, so
the difference in payoffs between adopting at price PA and not adopting is the difference
between Y − PA and W .
Firm C can then chose whether to sell to both firms at Y − W or just to firm B at
Z − X. So they will either offer a high price to A and then Z − X to be or Y − W to
both firms.
c) i. The second-stage equilibria involves player A adopting the technology iff one of three
things occurs: one, PA ≤ Y − W ; two, PA ∈ (Y − W, Z − X] and PB ≥ Z − X; three,
PA ∈ (Y − W, Z − X], PB ∈ (Y − W, Z − X], and player B does not adopt. The
game is symmetric, so player B’s choices are identical. Note, the third case is the
most confusing one; it states that if both PA and PB are in the range (Y − W, Z − X]
exactly one of them randomizes in equilibrium. Thus, we need to condition not just
on what the two prices are (which in the other cases determine who adopts), but
also on what the two choices are.
ii. Firm C will never offer a price that is not either Y − W or Z − W , because they
could otherwise increase their offered price without changing either firms’ decisions.
By offering PA = PB = Y − W , they get πC = 2 ∗ (Y − W ). If they offer PA = Z − X
and PB = 100, 000, 000, 000, they can get Z − X. Thus, they get the same price
when offering simultaneously as when offering sequentially.
d) In this case player A adopts the technology iff one of the following occurs: one, PA ≤
Z − X; two, PA ≤ Y − W and PB ≤ Y − W and B adopts. In this case, we know
that firm C offers PA = PB = Z − X, which gives it the same profits as offering sales
sequentially and less than offering to only one firm.
e) Suppose that firm C offers a pure strategy (PA , PB ). Then a PBE requires firm A to
know PB , given that it was offered PA . But it does not place any restrictions on what
A believes about PB when offered any other PA .
Let’s suppose first that Y − W ≥ Z − X. Then the claim is that the following is a PBE:
firm C sets prices PA = PB = Y − W , both firm B and C knowing the other price when
offered Y − W , both B and C adopt, and any beliefs off the equilibrium path. To see
this, note that regardless of the off the equilibrium path beliefs, firm A will not adopt of
PA > Y − W . Thus, firm C cannot make any more than 2 ∗ (Y − W ), which they make
by offering Pi = Y − W , so they have no incentive to deviate. Next, given that they
know the other firm is offered PB = Y − W (and know that firm B knows PA = Y − W ),
firm A will adopt the technology at PA = Y − W . Thus, there is a PBE where both
adopt.
Next, suppose that 2 ∗ (Y − W ) ≤ Z − X. Then the following is a PBE: firm C sets
prices PA = PB = Z − X, firm A adopts and firm B does not, they both know the other
price when offered Z −W , and they both believe that the opponent’s price is huge at any
other price offer. Again, firm C cannot offer any higher price and still get one to adopt
and (given that both believe off the equilibrium path that the other will not adopt) firm
C and only get both to adopt if the price is lowered to Y − W for both. Given that
2 ∗ (Y − W ) ≤ Z − X, they will not have an incentive to do that. Thus, there is a PBE
where only one adopts.
Finally, there is no PBE where no one adopts. This is because regardless of beliefs, each
firm will adopt if P ≤ min{Z − X, Y − W }. Thus, by changing the price offered to

58
 PA = PB = min{Z − X, Y − W }, firm C can guarantee itself positive profits, instead of
the zero that occurs when neither firm A or B adopt.

64. (2013 Final Q1)

65. (2013 Sept Comp Q3)

66. (PS6, Q1)

Note that in this problem, a strategy for player 1 is a price, and a strategy for player 2 is a
function describing what price to name for each possible choice of price for 1.
We solve using backward induction. Consider the subgame where 2 faces price p1 . 2’s optimal
choice of p2 in this case is given by
If p1 <c any p2 greater than p1
If p1 =c any p2 greater than or equal to c
If p1 =c+1 p2 = c + 1
If p1 >c+1 p2 = p1 − 1
The only non-obvious component of this strategy is that when p1 > c + 1, undercutting by
a penny is better than matching p1 . Let’s verify that this is optimal. Suppose 1 names
p1 = c + k where k ≥ 2. p2 = c + k − 1 is optimal when π2 (c + k) < π2 (c + k − 1) or
 
Q(c + k) Q(c + k) k−1
· k < Q(c + k − 1) · (k − 1) ⇔ <2 .
2 Q(c + k − 1) k

In the final inequality we have LHS < 1 because Q(·) is decreasing, and the RHS ≥ 1 when
k ≥ 2, so p2 = c + k − 1 is indeed optimal.
Note that there is a multiplicity of equilibria in this p1 subgame. However, the eqa are all
outcome equivalent (for 1 and 2), so that 1 can anticipate the outcome after each choice p1 .
Given the subgame outcomes, the only choice that yields positive payoff for 1 is p1 = c + 1.
Thus, the set of SPNE is given by
1’s strategy: p1 = c + 1
2’s strategy: Any p2 (p1 ) that satisfies the conditions above.
Q(c+1)
The equilibrium outcome is that both players name price c + 1 and reap payoff 2 · 1.

67. (PS6, Q2)

We begin by analyzing the subgame where 1 locates at a and 2 locates at 1 − b.
If a = 1 − b, the game reduces to Bertrand competition and we have eqm prices p1 = p2 = c.
Both firms receive 0 profit.
If a < 1 − b, we need to find the prevailing equilibrium prices and the profits for each firm.
Profit for 1 is given by

π1 (p1 , p2 , a, b) = (p1 − c) · θ̂(p1 , p2 , a, b)

where θ̂(p1 , p2 , a, b) is the consumer indifferent between purchasing from 1 or 2. We have

1 − b + a p2 − p1
v − p1 − t(θ − a) = v − p2 − t(1 − b − θ) ⇒ θ̂(p1 , p2 , a, b) = + .
2 2t

59
 To find 1’s best response to 2 naming p2 , 1 solves the maximization problem:
  
1 − b + a p2 − p1
max (p1 − c) · + ,
p1 2 2t
t(1−b+a)+c
which yields p1 (p2 ) = 21 p2 + 2 .
Analogous reasoning for 2 gives p2 (p1 ) = 21 p1 + t(1−a+b)+c 2 . Solving these two equations
simultaneously yields the equilibrium prices
t t
p∗1 = c + (3 + a − b) p∗2 = c + (3 + b − a).
3 3
Profits are given by
t t
π1 (p∗1 , p∗2 , a, b) = (3 + a − b)2 π2 (p∗1 , p∗2 , a, b) = (3 + b − a)2 .
18 18
Now, following location choices of a and 1 − b, the firms can anticipate the profits above,
giving a reduced-form location game. But this location game has no NE! To see this, observe
that for any choice of 1−b, 1’s payoff is increasing in a on (0, 1−b). However, naming a = 1−b
yields 0 payoff for both.
Equilibria with a < 1 − b are ruled out because 1 could gain by moving to the right. But
equilibria with a = 1 − b (0 profits) are also ruled out, because either firm could earn positive
profits by deviating to one side.
Thus, the game has no (pure strategy) SPNE.
68. (PS7, Q1)
The crux of this problem is recognizing the difference between a) and b). The difference lies
in the best-response condition for firms which do not enter in equilibrium. See below.
Q
First, some N -player Cournot eqm results: Inverse demand is given by p(Q) = ab − bI ; eqm
c (a−bc)I c a N (a−bc)
quantity (per firm) is q = N +1 ; eqm price is p = b − N +1 b ; eqm profit (per firm) is
(a−bc)2 I (a−bc)I−Q−i
πc = b(N +1)2
; best response function for i is qi (Q−i ) = 2 .

a) A Nash equilibrium will be a situation in which N firms enter and (i) each one produces
the Cournot quantity, (ii) each firm earns non-negative profits (iii) taking these quantities
as given, no additional firm would earn positive profits by entering. Therefore
(a − bc)I
(i) qi = ,
N +1
(a − bc)2 I
(ii) πi = − K ≥ 0,
b(N + 1)2
(a − bc)2 I
(iii) πj = − K ≤ 0,
4b(N + 1)2
where πj represents the profits earned by a potential (N + 1)th entrant. For firms
choosing Not Enter in eqm, the optimal deviation is to enter and best respond to the
quantity choices of the others. That this deviation is not profitable is condition (iii).
Equilibrium price and consumer surplus are given by
a N (a − bc)
p = −
b N +1 b
Z N qi 2
(a − bc)2

N
CS = [p(q) − p] dq = I.
0 N +1 2b

60
 Therefore, from (iii), as I % ∞ we must have N % ∞ and

(a − bc)2
p & c, N qi % ∞, and CS/I % .
2
Curiously, qi % ∞, since otherwise firms would have losses for large N (recall p & c and
K > 0).
b) In the second stage, given N, firms will produce the same quantity as above. Conditions
(i) and (ii) are identical. However, for firms that do not enter, deviating to Enter would
yield the (N + 1) firm Cournot outcome in stage 2. Therefore, condition (iii) becomes

(a − bc)2 I
(iii) πj = − K ≤ 0.
b((N + 1) + 1)2

The same asymptotic results hold. Comparison of the conditions shows that for fixed I,
more firms will enter in (b).

69. (PS7, Q2)

Assume (as in the lecture notes) that v is large, so that in eqm, firms always serve the entire
market, and monopoly profits are always positive. (Verify that large v achieves this!) Also
assume c = 0.
Let’s begin with the second stage where given location choices, firms name price.
Case 1. 1 plant for firm 1. Here 1 is a monopolist and names v − t, the highest price that
serves the entire market.
Case 2. 1 plant for firm 1, 1 plant for firm 2 at opposite end. This is the standard Hotelling
model, yielding price p = t for each firm.
Case 3. 1 plant for firm 1, 1 plant for firm 2 at same end. Here we have Bertrand competition
and firms price at cost.
Case 4. 2 plants for firm 1 at opposite ends. Here each plant captures half the interval, and
the highest price that achieves this is p = v − 4t at each plant.
Case 5. 2 plants for firm 1, 1 for firm 2. Here we have competition and p = c at one end
and p = 2t for firm 1’s plant at the other end.
To summarize, we have (omitting symmetric cases):
1’s location 2’s location prices π1 π2
10 00 v−t v−t−F 0
t t
10 01 t, t 2 −F 2 − F
10 10 0, 0 −F −F
11 00 v − 4t (in each plant) v − 4t − 2F 0
11 10 p = 0 at one end, 2t at other t
8 − 2F −F
Consider 2’s entry/location decision, given 1’s choice. Observe that 2 will enter only if 1 has
placed a single plant and 2t ≥ F . In this case 2 places the plant opposite 1’s.
Consider 1’s location choice. If 2t ≥ F , 1 anticipates entry from 2 when he places only one
plant. He then compares π1 (10, 01) = 2t − F to π1 (11, 00) = v − 4t − 2F , and therefore chooses
to place two plants. If 2t < F , he knows 2 will not enter, and compares π1 (10, 00) = v − t − F
to π1 (11, 00) = v − 4t − 2F . Thus, when F ≤ 43 t, 1 places two plants, and when F > 34 t, he
places one.

61
 The SPNE11 is then:
Firm 1) if F ≤ 43 t, place two plants
if F > 43 t, place one plant
name Hotelling prices at pricing stage
Firm 2) if F ≤ 2t and 1 has placed only 1 plant, place plant at opposite end
else, don’t place
name Hotelling prices at pricing stage
The eqm outcome is that 1 places one or two plants (based on F ≷ 43 t), and 2 does not enter.
Notice that nothing changes if 1 is granted monopoly power. However, if 1 can only place
one plant, then 2 enters iff F ≤ 2t (assuming entry when indifferent). Observe that 2’s
entry reduces transportation costs, but doubles the fixed cost. Social surplus is then given
by v − 3t − F whenever F > 2t and by v − 12 t
− 2F otherwise. Therefore, for F ∈ [ 12 t, 43 t), the
3
policy increases social welfare, for F > 4 t reduces SW, and for F < 21 t has no effect.

70. (PS6, Q4)

Throughout, if the consumer is indifferent between buying and not buying, assume he buys.
This prevents firms from running into open-ness problems when naming prices.12

a) Let’s begin with the stage 2 subgame. If Upstart has not entered, then CTWE has a
monopoly in both markets and will extract all surplus, naming pS = RS and pJ = RJ .
If Upstart has entered, then in the stuff market, CTWE is still a monopolist and names
ps = Rs . In the junk market, we have Bertrand competition with differing costs. From
earlier analysis, we know that there is a continuum of equilibria; pJ = pJU = p is an
equilibrium for each p ∈ [cJU , cJ ], where the consumer buys from the lower cost firm.
However, we are restricting attention to equilibria where no firm prices below cost. Thus,
we must have pJU = pJ = cJ in the entry subgame. Profit in the entry subgame is given
by
πCT W E = RS − cS πU = cJ − cJU .
Returning to stage 1, we see that Upstart enters if cJ − cJU > K (and is indifferent
under equality). But this is exactly the parametric assumption we are given, so Upstart
enters.
The SPNE is then:
Upstart: Enter; If Entry, set pJU = cJ
CTWE: If No Entry, set pS = RS , pJ = RJ
If Entry, set pS = RS , pJ = cJ
b) Again we begin with the stage 2 subgame.
With no entry, CTWE will extract all surplus, naming pB = RS + RJ . With entry, we
have a variant of Bertrand competition, for which we must find the NE. We solve using
a series of claims about equilibria of the subgame with entry.
Claim 1: In any NE, the consumer makes a purchase.
If instead equilibrium prices are so high that the consumer purchase neither stuff nor
junk, either firm could do better by naming a price just above cost.
11
Trivially, there are multiple eqa in which we alter choices under indifference or alter where 1 places his single
plant.
12
An alternative would be to treat the consumer as a player in the game, automatically pinning down his behavior
under indifference.

62
 Claim 2: In any NE, the consumer is indifferent between the two firms.
Otherwise, the preferred firm could do better by raising its price by ε, thereby keeping the
customer and raising profits. Thus, we have RS +RJ −pB = RJ −pJU , so pB = pJU +RS .
Claim 3: In any NE, some firm must be pricing at cost.
By assumption, neither firm is pricing below cost. If both were pricing above cost, in
view of claim 2, one of the firms (the firm losing out when the customer is indifferent)
could do better by undercutting the other by ε, securing the customer for itself while
only marginally reducing markup.
Which firm prices at cost depends on the parametric assumptions. CTWE pricing at
cost implies Upstart prices below cost, which is not allowed:
pB = cJ + cS ⇒ pJU + RS = cJ + cS ⇒ pJU = cJ + cS − RS < cJU
where the final inequality is the given parametric assumption. If Upstart prices at cost,
then CTWE prices above cost:
pJU = cJU ⇒ pB − RS = cJU ⇒ pB = cJU + RS > cJ + cS .
Here we resolve the consumer’s indifference in favor of CTWE (otherwise, at these prices
CTWE would undercut by ε). Thus, NE prices are pJU = cJU , pB = cJU + RS . Profits
are given by
πCT W E = cJU + RS − cS − cJ πU = 0.
We now return to the stage 1. Anticipating 0 profits in the entry subgame, Upstart will
not enter. Thus, we have SPNE:
Upstart: Not Enter; If Entry, set pJU = cJU
CTWE: If No Entry, set pB = RS + RJ
If Entry, set pB = cJU + RS .
c) Now CTWE just compares profits in (a) and (b). In (a), Upstart enters and CTWE’s
profits are RS − cS . In (b), entry is deterred and profits are higher: RS + RJ − cS − cJ .
Thus, CTWE ties its products together. The complete SPNE is for CTWE to tie its
products together and then for the parties to behave in the tie/no tie subgames as
described in (a) and (b).
To compare social welfare with and without tying, we combine profits and consumer
surplus in (a) and (b).
(a) (RS − cS ) + (RJ − cJU − K)
(b) (RS + RJ − cS − cJ )
So SWa > SWb because cJU + K < cJ . Banning tying increases total welfare by shifting
production to the lower cost firm.

71. (PS7, Q4)

Assume that player 1 gets to make an offer first.
a) T odd (so that 1 makes period T offer).
T −1 T −1
π1 (T ) = 1 − δ2 + δ1 δ2 − . . . + δ1 2 δ2 2
T −1 T −3
2 2
1−δ2
δ1t δ2t δ1t δ2t
P P
= − δ2 → 1−δ1 δ2
t=0 t=0
δ2 (1−δ1 )
π2 (T ) = 1 − π1 (T ) → 1−δ1 δ2

63
 b) T even (so that 2 makes period T offer).
T T
−1
π1 (T ) = 1 − δ2 + δ1 δ2 − . . . + δ12 δ22
T −2 T −2
2 2
1−δ2
δ1t δ2t δ1t δ2t
P P
= − δ2 → 1−δ1 δ2
t=0 t=0

π2 (T ) = 1 − π1 (T ) → δ1−δ2 (1−δ1 )
1 δ2
Identical analysis applies if 2 makes the first proposal.

72. (PS7, Q5)

We solve using backward induction.

a) If T is even: in period T , player 1 will accept any offer, so player 2 will offer a split of
(0, 1) and (continuation) payoffs will be (0, 1 − c). Thus, in period T − 1, 2 will accept
any offer above 1 − c, so 1 offers a split of (c, 1 − c) and (continuation) payoffs will again
be (0, 1 − c). In period T − 2, again 1 will accept any offer, because rejecting leads to a
subgame where he gets payoff 0. Thus, 2 offers (0, 1).
This cycle continues back to time 1, so we have the following equilibrium:
1 offers (c, 1 − c) in every period, 2 offers (0, 1) in every period. 1 accepts all offers and
2 accepts all offers ≥ 1 − c.
The equilibrium outcome is that 1 offers (c, 1 − c) in the first period and 2 accepts
immediately, yielding payoffs (0, 1 − c).
If T is odd, the roles are reversed.
Here the strategies are exactly as above, and depend on the parity of T .
b) As one can tell, the equilibrium outcomes and strategies do not converge to a stationary
limit as T → ∞; in particular, all that the strategies and outcome depend upon is
T (mod2).
c) We restrict attention to stationary equilibria of the following form: at every offer node,
1 offers (x, 1 − x), 2 offers (y, 1 − y), 1 accepts any offer ≥ y, and 2 accepts any offer
≥ 1 − x.
We can do a bit more to relate x and y:
When 2 faces an offer (x, 1 − x), rejection yields payoff 1 − y − c. Thus, we must have
1 − x ≥ 1 − y − c or x ≤ y + c. When 2 faces an offer (x + ε, 1 − x − ε), she rejects and
gets 1 − y − c. Thus, we must have 1 − x − ε ≤ 1 − y − c ∀ε or x ≥ y + c − ε for every
ε > 0. Thus, we must have x = y + c.
The proposed equilibrium is now that player 1 always offers (x, 1 − x) at each offer node,
2 always offers (x − c, 1 − x + c), 1 accepts all offers ≥ x − c and 2 accepts all offers
≥ 1 − x. Note that we require x ∈ [c, 1].
We verify that this is in fact a SPE using the one-step-deviation principle.
When 1 faces offer (s, 1 − s), rejection yields (continuation) payoff x − c. Thus, accepting
when s ≥ x − c and rejecting when s < x − c is optimal.
When 2 faces offer (1 − s, s), rejection yields (continuation) payoff 1 − x + c − c = 1 − x.
Thus, accepting when s ≥ 1 − x and rejecting when s < 1 − x is optimal.
When 1 makes an offer (s, 1 − s), any offer with s < x will be snapped up by 2 yielding
payoff for 1 of s − c < x − c. Any s > x will be rejected, yielding payoff x − 2c < x − c
for 1. Thus, (x, 1 − x) is optimal.

64
 When 2 makes an offer (s, 1 − s), any offer with 1 − s < 1 − x + c will be snapped up by
1 yielding payoff for 2 of 1 − s − c < 1 − x + c − c. Any offer with 1 − s > 1 − x + c will
be rejected, yielding payoff 1 − x − c < 1 − x for 2. Thus, (x − c, 1 − x + c) is optimal.
By the one-step-deviation principle, the profile is a SPE.
There are other non-stationary equilibria, including non-trivial equilibria where agree-
ment is not reached in the first period. See Ariel Rubinstein, “Perfect Equilibrium in a
Bargaining Model”, Econometrica 97-109, 1982 for details.

73. (2004 Final, Q1)

a) In the N firm Cournot setting, we have

 2
c 1 1 1
q = p= πc =
N +1 N +1 N +1

Splitting will result in a 3 firm Cournot setting, so the board decides between taking 2/3
of total profits when N = 3 or 1/2 of total profits when N = 2. Since
 2  2
1 1
2 >
4 3

the firm prefers to split.

b) In the second stage, per (spun-off) firm profits are
 2
1
π= .
N1 + N2 + 1

Possible NE in the reduced game (ignoring additional symmetric cases) are then (1,1)
(1,2) (1,3) (2,2) (2,3) (3,3). We could solve using the normal form, or else by inspection.
We have eliminated (1,1) in part (b). (1,2) and (1,3) are eliminated, as in each case firm
1 prefers to split 2 ways rather than 1. (2,2) is ruled out as one firm would prefer to
split in 3. (2,3) is ruled out, as deviating from 2 to 3 is preferable. Finally, (3,3) is the
unique NE of the reduced game.
Thus, there is one SPNE: (3,3), followed by second period Cournot quantities.

74. (2005 Final, Q3)

a) i. In round 2, the proposer captures the entire surplus and offers 0. This offer is
accepted in equilibrium.
ii. If Vit is i0 s continuation value before the proposer is selected at t = 2, Vi2 = 31 .
iii. In round 1, if a proposer offers the other two agents their discounted continuation
value, they would accept this offer and reject anything less. Since 1 − 2δ δ
3 > 3 , the
proposer would be willing to offer each of the respondents their discounted cont.
value of 3δ , and keep 1 − 2δ3 .
iv. Vi1 = 31 (1 − 2δ
3 ) + 32 ( 3δ ) = 1
3
b) By induction on a., it can be shown that Vit = 13 for every t. Therefore, the offer that
shall be made is 3δ to each agent, and that will be accepted.
i. No, it does not.

65
 ii. The proposer would offer 3δ to one of the two other agents. She would keep 1 − 3δ
of the surplus.
iii. Vi1 = 13 1 − 3δ + 32 21 3δ = 13 .

iv. There are many possible asymmetric SPE. Here is one: at t = 1, Let 1 and 2 always
make their proposals
to one another, and let 3 make his proposal to 1.
Then V11 = 31 1 − 3δ + 23 3δ = 31 + 9δ
V21 = 31 1 − 3δ + 31 3δ = 31

V31 = 13 1 − 3δ = 31 − 9δ .

75. (June 2005 Comp, Q4)

a) The equilibrium: In stage 5, B voids the transaction iff quality is bad or p > v. In stage
3, S ∗ chooses e = 1 iff p ≥ c. In stage 2, B accepts the contract with the lowest price in
the interval [c, v]; if there is no contract with a price in this interval, both are rejected.
In stage 1, S1 and S2 set p = c. The Bertrand result holds because, at any price above
costs, one seller can profitably steal business by undercutting the other.
b) i. The normal form for the reduced subgame looks like this:
Monitor Not Monitor
e = 1 p − c, v − p − M p − c, v − p
e = 0 0, −M p, −p
Case 1: p < M . In that case, not monitoring is a dominant strategy for B, which
means the game is dominance solvable, and the outcome is (0, Not Monitor). Payoffs
are (p, −p).
Case 2: p = M . It is easy to check that (0, Not Monitor) is the only pure strategy
equilibrum, with payoffs (p, −p).
Case 3: p > M . It is easy to check that there are no pure strategy equilibria. Let
θ denote the probability that S ∗ plays e = 1, and let π denote the probability that
B monitors. Indifference for S ∗ requires

p − c = (1 − π)p

or
c
π= .
p
Indifference for B requires

θ(v − p) − M = θv − p

or
M
θ =1−
p
S ∗ ’s expected payoff in this equilibrium is plainlyp − c (since
 this is what he gets
with certainty from e = 1). B’s expected payoff is 1 − p (v −p)−M = v −p− vM
M
p
ii. In stage 2, B accepts the contract with the price yielding the greatest value of
v − p − vM
p provided this value is non-negative, and rejects both offers if these values
are both negative. In stage 1, S1 and S2 choose p to maximize v − p − vM p . The
first order condition is:
vM
=1
p2

66
 or 1
p = (vM ) 2
The second derivative is
vM
−2 < 0,
p3
so the function is strictly concave, and the first order condition guarantees a maxi-
mum. At the maximum, B’s payoff is
1
v − 2 (vM ) 2 > 0,
v
as required, where the inequality follows from the fact that M < 4. Here, the
1
Bertrand result does not hold, because p = (vM ) > M > c. A seller can’t
2
1
profitably steal business by reducing price to a level between c and (vM ) 2 because
the continuation equilibrium is less attractive to the buyer in light of the effect on
effort and monitoring.
c) In part A, the seller would set p = v. In part B, the seller would select the largest price
consistent with non-negative payoffs for B: v − p − vM p ≥ 0, which we can rewrite as
2
p − pv − M v ≥ 0. Notice that this is definitely negative for p = v, so the monopoly
price is below v. Since B’s payoffs are strictly concave, the monopoly price is the larger
of the two roots of p2 − pv + M v = 0.

76. (Sept 2005 Comp, Q4)

a) i. The condition for improving on autarky is u0 (W ) < δu0 (0). The social optimum
involves a gift g ∗ satisfying u0 (W − g ∗ ) = δu0 (g ∗ ).
ii. Let’s define strategies recursively. Generation 1 gives g ∗ to generation 0 in period 1.
Every subsequent generation t gives g ∗ to generation t − 1 in period t if generation
t − 1 has chosen the action prescribed by its strategy, and 0 otherwise. If every
generation follows these strategies, each will give g ∗ , which conforms with the desired
social convention.
Let’s check that this is a subgame perfect Nash equilibrium. Given any history, if
generation t follows its prescribed strategy, assuming t + 1 does so as well, t will
receive a payoff of u(W − g ∗ ) + δu(g ∗ ). If generation t instead chooses some other
gift g, it will receive u(W − g) + δu(0). Since u(W − g) + δu(0) ≤ u(W ) + δu(0) ≤
u(W − g ∗ ) + δu(g ∗ ), the deviation isn’t beneficial. Accordingly, we have a subgame
perfect Nash equilibrium. This answer does not depend on the value of δ (though,
for δ small and u0 (0) finite, the optimal convention may involve no gift).
iii. All of them are sustainable, and the answer doesn’t depend on δ (except that δ
affects which conventions make every generation better off). Choose any g such
that u(W ) + δu(0) < u(W − gb) + δu(b g ). Then the argument in part a(ii) goes
through with gb replacing g ∗ .
iv. None of them are sustainable, and the answer doesn’t depend on δ (except that
δ affects which conventions make every generation worse off). Suppose on the
contrary that we have an equilibrium supporting a convention with a gift g 0 such
that u(W −g 0 )+δu(g 0 ) < u(W )+δu(0). Let generation t deviate by giving nothing.
Then t consumes W in the first period of life, and not less than 0 in the second period
of life. It’s payoff is therefore bounded below by u(W ) + δu(0), so it’s better off –
a contradiction.

67
 v. With a finite horizon, there are no gifts. This follows from backward recursion: the
last generation gives nothing to its predecessor because there are no consequences,
so the second-to-last generation gives nothing to its predecessor, and so forth.
b) No social conventions are sustainable (aside from giving nothing). Suppose on the
contrary that there’s an equilibrium supporting a convention with a gift of b0 . Assume
that generation t has received b0 in the first period of life, and that there have been
no deviations from equilibrium strategies. If t follows the convention, its payoff is
v(b0 ) + γv(F − b0 ). If t gives nothing, its payoff is v(b0 ) + γv(F ). The deviation plainly
raises t’s payoff – a contradiction.
c) i. With δ > 1, we know from part (a) that we can sustain a convention with any
g ∈ [0, W ], and b = 0. The question is whether we can sustain other conventions.
Let’s define strategies recursively. In the first half of period t, generation t − 1 gives
b0 to generation t, regardless of the history that it has observed. In the second half
of period t, generation t gives g 0 to generation t − 1 if t − 1 followed his strategy in
both the first half of period t, and in the second half of period t−1 (which is satisfied
trivially in period 1); otherwise, he gives nothing. If every generation follows these
strategies, then, within each period, the old generation gives its child b0 , and then
the young generation gives its child g 0 .
Let’s check that this is a subgame perfect Nash equilibrium. Notice that generation
t makes two decisions in a row: one in the second half of period t, and the next in
the first half of period t + 1. Suppose it has received a gift of b (possibly not the
equilibirum level) from generation t − 1 in period t. If it follows its equilibrium
strategy, its payoff will be (W −g 0 )+b+δg 0 +γ(F −b0 ) = (W +γF +b)+(δ−1)g 0 −γb0 .
If it deviates to some other gifts g 00 and b00 , its payoff will be (W + γF + b) − g 00 − γb00 .
Clearly, its most profitable deviation involves g 00 = b00 = 0, which yields a payoff of
W + γF + b. Accordingly, we have an equilibrium provided that (W + γF + b) +
(δ − 1)g 0 − γb0 ≥ W + γF + b, or equivalently (δ − 1)g 0 ≥ γb0 . To put it differently,
we can support
 all conventions defined by (g 0 , b0 ) above the straight line defined by
γ
g = δ−1 b. An increase in γ reduces the set of sustainable conventions, and an
increase in δ expands it.
In contrast to part (b), we can sustain conventions involving transfers of food. In
(b), there’s no way to discipline the elderly, so we can’t force them to transfer food.
Here, we can discipline them by withholding water.
ii. We maximize welfare by making the gifts as large as possible. Clearly, we can set
g 0 = W . This is good in its own nright,  and  relaxes
o the constraint on b0 as much as
possible. We then set b0 = min F, δ−1 γ W . For the optimal convention, the
level of water transferred does not depend on δ or γ. The level of food transferred
rises with δ and falls with γ.

77. (Sept 2005 Comp, Q4)

a) In the case of indifference, we shall assume that the consumer buys the car from Company
A. At an equilibrium, it must be that vA −pA = vB −pB ; if vA −pA > vB −pB , then firm
A can increase its price without losing the consumer. Similarly, if vB − pB > vA − pA ,
firm B can increase its price without losing the consumer. If equality holds, then
pA = vA − vB + pB . It also must be that pB = c; if pB > c, then Firm B could lower
its price infinitesimally, draw away the consumer and make a strictly positive profit.

68
 Therefore, the equilibrium outcome is pB = c, pA = c + vA − vB , and the consumer buys
the product from Firm A.
b) Answers below:
i. Given a value of d, Firm A’s profits are f (d) − d.
ii. Firm A will choose d∗ such that f 0 (d∗ ) = 1. In the case where f (d) = d1/2 , d∗ = 41 .
c) Answers below:
i. Firm i’s profits = −di + max {f (di ) − f (dj ) , 0}
ii. There does not exist any such equilibrium. If firms choose the same level of adver-
tising, then in the Nash equilibrium of the 2nd stage game, both firms shall price
at marginal cost. Since each firm makes a loss of their advertising budget, Firm i
would be better off switching to di = 0.
iii. Consider the strategy profile where di = 0, and dj = d∗ from (b) (ii), and in the
sub-game, pi (di , dj ) = c + max {f (di ) − f (dj ) , 0}. This gives 0 utility to Firm i
and f (d∗ ) − d∗ to Firm j. Observe that conditional on dj = d∗ , Firm i receives a
strictly negative payoff from deviating and setting di ∈ (0, d∗ ]. Let us now consider
deviations where Firm i sets di > d∗ . Observe that her payoff from doing so
is f (di ) − di − f (d∗ ). Since d∗ is the unique maximizer of f (d) − d, note that
f (di ) − di < f (d∗ ) − d∗ , and therefore, f (di ) − di − f (d∗ ) < −d∗ < 0, since by
assumption, d∗ > 0. Therefore, this deviation also yields a strictly negative payoff.
Thus, this is an SPE.

78. (June 2006 Comp, Q1)

a) i. If p < v + s, buying is a dominant strategy, so all consumers will buy. If p > v + sI,
not buing is a dominant strategy, so no consumer will buy.
ii. For any p ∈ [v + s, v + sI], there is a continuation equilibrium in which everyone
buys (since the benefit of buying is then v + sI − p ≥ 0), and one in which no one
buys (since the benefit of buying is then v + s ≤ 0).
iii. For any p∗ ∈ [v + s, v + sI], one can construct a subgame perfect Nash equilibrium
in which the monopolist chooses p∗ – simply use the “everyone buys” continuation
equilibrium for p ≤ p∗ , and the “no one buys” continuation equilibrium for p > p∗ .
Since we have p∗ ≥ v + s > c, this price is optimal for the monopolist.
b) i. A strategy is a mapping from {w1 , ..., wM } into {BUY, NOT BUY}. An example:
choose NOT BUY regardless of the consumer’s realized value.
ii. Given the strategies of other consumers, let NE be the expected number of them
who will buy. If the consumer buys with value wk , then wk + s(NE + 1) − p ≥ 0.
But wr + s(NE + 1) − p > wk + s(NE + 1) − p ≥ 0, which means she definitely buys
with value wr for r > k.
iii. Suppose consumers buy iff vi ≥ wk . When a given consumer decides to buy, she
expects the total number of buyers to be Λk + 1. If vi ≥ wk , she therefore expects
buying to generate a payoff of vi + s(Λk + 1) − p ≥ wk + s(Λk + 1) − p = Wk (p) ≥ 0.
If vi < wk , she expects buying to generate a payoff of vi + s(Λk + 1) − p ≤ wk−1 +
s(Λk + 1) − p < wk−1 + s(Λk−1 + 1) − p = Wk−1 (p) ≤ 0. So all actions are optimal.
iv. If Wk (p) ≥ 0, then one of two things must be true: either (a) Wr (p) ≥ 0 and
Wr−1 (p) ≤ 0 for some r ∈ {2, ..., k}, or (b) W1 (p) > 0. In case (a), the conclusion
follows from part iii. In case (b), there is an equilibrium in which consumers buy
irrespective of type.

69
 v. If WM (p) ≥ 0, exsitence follows from step iv. If WM (p) < 0, there is an equilibrium
in which no consumer ever buys, irrespective of type.
vi. Suppose that, with p, there is an equilibrium in which consumers buy iff vi ≥ wk .
Then we must have Wk (p) ≥ 0. But then Wk (p0 ) > Wk (p) ≥ 0. By step iv, there
exists an equilibrium with p0 where consumers buy iff vi ≥ wr for some r ≤ k.
vii. For c and c0 with c > c0 , let p and p0 be the monopolist’s optimal prices. We know
that (p − c)x(p) ≥ (p0 − c)x(p0 ), and (p0 − c0 )x(p0 ) ≥ (p − c0 )x(p). Adding these
two inequalities, cancelling terms, and rearranging yields (c − c0 )(x(p0 ) − x(p)) ≥ 0.
With c > c0 , we must therefore have x(p0 ) − x(p) ≥ 0. With x non-increasing, this
requires p ≥ p0 , as required.

79. (June 2009 Comp, Q1)

a) i. The consumer who is indifferent between buying from the two firms will have θ that
solves:
pB − pA + tB
v − pA − tA θ = v − pB − tB (1 − θ) =⇒ θ =
tA + tB
Sales for firm A are equal to the θ above. Sales for firm B are 1 − θ.
ii. The payoff for firm A is πA = (pA − c)θ = (pA −c)(p B −pA +tB )
tA +tB . The payoff for firm B
is πB = (pB − c)(1 − θ) = (pB −c)(p A −pB +tA )
tA +tB .
iii. Firm i’s best response to pj is the solution to the first order condition (because
payoffs are strictly concave in one’s own price):

∂πi pj + tj + c − 2pi 1
= = 0 =⇒ pi = (pj + tj + c)
∂pi ti + tj 2

iv. Nash equilibrium prices are where the best response functions cross:
1 1
pA = (pB + tB + c) and pB = (pA + tA + c) =⇒
2 2
 
1 1
pA = (pA + tA + c) + tB + c =⇒
2 2
1 1
pA = (tA + 2tB + 3c) and pB = (2tA + tB + 3c)
3 3
v. Plugging in equilibrium prices from (iv), Nash equilibrium payoffs are:

(ti + 2tj )2
πi =
9(ti + tj )

(p −c)(p −p +t )
b) Firm i chooses pi and ti to maximize πi = i ti +t j
j
i j
. Profits are decreasing in
ti , so ti = 0 is optimal for every possible pj , tj . As shown above, the optimal price is
pi = 21 (pj + tj + c). Thus, the equilibrium is pA = pB = c and tA = tB = 0.
c) i. A strategy for firm i is choice ti ∈ {0, t} and a function fi : {0, t}2 → R+ that
determines a price for every possible pair (tA , tB ). Thus the strategy set is {0, t} ×
R4+ .

70
 ii. Subgame perfection means that prices in the second stage must be determined as in
part (a). Note that firm i’s profits are increasing in ti for all values of tj .
∂πi (ti + 2tj )(ti )
= >0
∂ti 9(ti + tj )
Therefore, it is a dominant strategy to play ti = t and the equilibrium is tA = tB = t
and prices defined as in part (a)(iv).
d) i. A strategy for firm i is choice ti ∈ [0, t] and a function fi : [0, t]2 → R+ that
determines a price for every possible pair (tA , tB ). Let A denote the set of such
functions fi . Then a firm’s strategy set is [0, t] × A.
ii. The same reasoning from part (c) applies, so the equilibrium is tA = tB = t and
prices defined as in part (a)(iv).
e) When prices and transportation costs are chosen simultaneously, the firms compete
away all the profits. But when transportation costs are chosen first, firms can commit
to differentiating their products and thereby earn positive profits.
80. (2011 June Comp, Q5)
a) A pure strategy consists of a 5-tuple (a1 , ..., a5 ), where a1 denotes the player’s choice in
the first period, a2 denotes the second-period choice conditional on (C, C) being chosen
in the first period (where I adopt the convention that the first element in such a pair
represents the player’s own first-period choice while the second represents the choice
of the other player), a3 denotes the second-period choice conditional on (C, F ) being
chosen in the first period, a4 denotes the second-period choice conditional on (F, C)
being chosen in the first period, and a5 denotes the second-period choice conditional on
(F, F ) being chosen in the first period. Thus, the strategy set for each player is {C, F }5 ,
which contains 32 elements. The strategy profile set therefore contains 322 = 1024
elements.
b) A strategy is strictly dominated if a player follows his intended first-period choice
with C regardless of what the other player chooses. Such strategies have the forms
(C, C, C, a4 , a5 ) and (F, a2 , a3 , C, C). There are 8 such strategies.
It is easy to check that no other strategy is strictly dominated. Note in particular
that (C, F, a3 , a4 , a5 ) is a best response to (C, C, F, a04 , a05 ), (C, a2 , F, a4 , a5 ) is a best
response to (C, a02 , a03 , C, F ), (F, a2 , a3 , F, a5 ) is a best response to (C, F, F, a04 , a05 ), and
(F, a2 , a3 , a4 , F ) is a best response to (F, a02 , a03 , F, F ).
c) A strategy is weakly dominated if a player follows his intended first-period choice with C
for any choice of his opponent. In addition to the 8 strictly dominated strategies, weakly
dominated strategies have the forms (C, F, C, a4 , a5 ), (C, C, F, a4 , a5 ), (F, a2 , a3 , C, F ),
and (F, a2 , a3 , F, C). There are 16 strategies with those forms, making 24 weakly dom-
inated strategies in all.
No other strategy is weakly dominated. (C, F, F, a4 , a5 ) is a best response to (C, C, F, a04 , a05 ).
Only strategies of the form (C, F, a3 , a4 , a5 ) yield an equal payoff in response to (C, C, F, a04 , a05 ),
and all of those yield the same or lower payoffs in response to any other strategy. Simi-
larly, (F, a2 , a3 , F, F ) is a best response to (F, a02 , a03 , F, F ). Only strategies of the form
(F, a2 , a3 , a4 , F ) yield an equal payoff in response to (F, a02 , a03 , F, F ), and all of those
yield the same of lower payoffs in response to any other strategy.
d) In the second period, on the equilibrium path, the players must be making mutual best
responses, which means the outcome must be (F, F ). It follows that the equilibrium

71
 outcome must also be (F, F ) in the first period (otherwise a deviation to F in both
periods would yield a strict improvement). It then follows that each player’s strategy
must call for F in response to (C, F ) in the first period (otherwise the other player would
deviate to C in the first period). That leave four strategies of the form (F, a2 , a3 , F, F ).
Any such pair are mutual best responses. Thus, there are 16 Nash equilibria, all of
which yield (F, F ) in both periods on the equilibrium path.
d) In a two-player game, the set of THP equilibria is equivalent to the set of Nash equilibria
in which no player chooses a weakly dominated strategy. Combining steps c) and d),
we see that all 16 Nash equilibria are THP.
e) By backward induction, the unique SPNE involves both players choosing (F, F, F, F, F ).
f) It is easily checked that (i) (F, F ) is the unique equilibrium following (C, C), (ii) (C, F )
is the unique equilibrium following (F, C), (iii) (F, C) is the unique equilibrium following
(C, F ), and (iv) (C, C) is the unique equilibrium following (F, F ) (because the tax exceeds
the gain from deviating from C to F ). Therefore, all equilibrium strategies have the
form (a1 , F, F, C, C), and the first period game reduces to
C F
C X, X 3, 3
F 3, 3 X, X
When X > 3, we can have (a1 , a01 ) = (C, C) or (F, F ). For X < 3, we can have
(a1 , a01 ) = (C, F ) or (F, C). For X = 3, we can have all four possibilities.

81. (PS8, Q1)

If s∗ is a Nash equilibrium of the stage game, then the strategies “each player i plays s∗i in
each period” form a SPNE of the repeated game.
To see that the above observation is correct, note that with these strategies, the future play
of i’s opponents is independent of how he plays today. So his optimal action today is to
maximize his current period payoff, i.e., to play a static best response to s∗−i . By definition
of s∗ being a NE of the stage game, s∗i is a static best response to s∗−i .

82. (PS8, Q2)

a) The stage game is dominance solvable with PSNE (C, c). Because the stage game has a
unique NE, the only equilibrium of the repeated game is to play the static equilibrium
(C, c) in every period. Since the stage game is dominance-solvable, (C, c) would be
played at the final stage regardless of the prior history. That would imply that play
at the final period is independent of play at the penultimate period; then, each agent’s
optimal action at the penultimate period is to maximize the current period payoff and
play (C, c). By induction, this implies that the unique NE is to play (C, c) in every
period.
b) To sustain the cooperative choice (B, b), it must be that no player wishes to deviate by
choosing c (the optimal deviation):

1 δ
9 ≥ 11 + 5.
|1 −
{zδ } | 1−δ }
{z
eqm payoff opt. dev. payoff
So we require δ ≥ 31 .

72
 This is a SPNE by construction. We can verify this using the one-step deviation principle.
We see that no player wishes to deviate if everyone has cooperated (shown above) and
certainly no player wishes to deviate in a subgame where Nash reversion is prescribed.
Thus, we have a SPNE.

83. (PS8, Q3)

We will describe one of many possible SPNE in which (N F, N F ) is chosen on the equilibrium
path. Consider the following strategies where the “if” refers to play in the previous round:
Period 1 Play (N F, N F )
Period t > 1 if (N F, N F ) (N F, N F )
if (N F, F ) (F, F )
if (F, N F ) (F, F )
if (F, F ) (N F, N F )
To check that this is an equilibrium, we must ensure that no player wishes to deviate when
(F, F ) is prescribed or when (N F, N F ) is prescribed.
When (N F, N F ) is prescribed, 1’s incentive constraint is:

3 δ2 · 3
≥ 4+δ·1+ =⇒
1−δ 1−δ
3 ≥ (4 + δ) (1 − δ) + 3δ 2 =⇒
1
δ ≥
2

2’s incentive constraint is:

3 ≥ 4 (1 − δ) + 3δ 2

which is satisfied when δ ≥ 21 .

When (F, F ) is prescribed, 1 has no incentive to deviate. 2’s incentive constraint is:

3δ 3δ 2
0+ ≥ 1+0·δ+ =⇒
1−δ 1−δ
3δ ≥ (1 − δ) + 3δ 2 =⇒
1
δ ≥
3

So for δ ≥ 12 , (N F, N F ) can be sustained using the strategies specified above.

84. (PS8, Q4)

85. (PS8, Q5)

To sustain the monopoly outcome, we require
πm δ
≥ πd + πc
1−δ 1−δ
where π m represents (per firm) monopoly profits in the N -player Cournot game, π c represents
Cournot profits, and π d represents i’s profits from best responding to the N − 1 opponents
all naming q m , the collusive monopoly quantity.

73
 (a−c)2 1
Solving the N -player model yields π m = 4bN , which is just N times single firm monopoly
(a−c)2
profits, and πc = (N +1)2 b
. To find πd, we find i’s optimal deviation:

max [(p(qi + (N − 1)q m ) − c)qi ]

qi

N +1 a−c N +1 2 (a−c)2
and profit π d =


which yields qi = N 4b N 16b . Thus, our “no deviate” condition
becomes 2
1 (a − c)2 (a − c)2 (a − c)2

N +1 δ
≥ +
1 − δ 4bN N 16b 1 − δ (N + 1)2 b
which reduces to
(N + 1)2
δ≥ .
(N + 1)2 + 4N
Observe that as N → ∞, δ → 1. Hence with more firms, it becomes harder to sustain
collusion. This matches common intuition, although that intuition is largely built on detection
and enforcement problems which are not present in this model. In this model as N → ∞,
2
we have π m → 0 and π c → 0, while interestingly, π d → (a−c)
16b . Hence, for large N the payoff
from deviating to π d dominates the system and unless δ is very close to 1, prevents collusion.
Question: Does this reasoning hold for Cournot settings with general demand function p(Q)?

86. (PS8, Q6)

a) Observe that profits are decreasing in q on [q m , q c ], so that our problem becomes

min q subject to q sustainable via Nash reversion.

q∈[q m ,q c ]

The condition that q is sustainable in equilibrium via Nash reversion is

π(q) δ
≥ π d (q) + πc (24)
1−δ 1−δ
where π(q) are (per firm) single stage profits from naming q, π d (q) are 1’s profits from
best responding to 2 naming q, and π c are Cournot profits. Routine analysis yields

(a − c − qb)2 (a − c)2
π(q) = (a − 2bq − c)q π d (q) = πc = .
4b 9b
Substituting into (24) yields

(a − 2bq − c)q (a − c − qb)2 δ (a − c)2

≥ + . (25)
1−δ 4b 1 − δ 9b
To simplify the algebra, we use a 1-1 change of variables q = a−c
br .
Substituting into (25) yields

r−1 2 1
     
1 r−2 δ 1
2
≥ + ⇔
1−δ r r 4 1−δ 9

(9 − 5δ)r2 − (54 − 18δ)r + (81 − 9δ) ≤ 0. (26)

The problem is now
max r subject to (26).
r∈[3,4]

74
 Note that the condition r ∈ [3, 4] corresponds to the condition q ∈ [q m , q c ].
Since (9 − 5δ) > 1, the LHS of (26) is a convex quadratic, so all values of r between the
9−δ
two roots will satisfy the inequality. The roots are {3, 3 9−5δ }, where since δ ∈ (0, 1), the
second root is the greater of the two. Thus, our solution is
 
9−δ
r = min 3 ,4 .
9 − 5δ
9−δ 9 9 9 9−δ
Notice that 3 9−5δ = 4 ⇒ δ = 17 , so that for δ ≥ 17 , r = 4 and for δ < 17 , r = 3 9−5δ .
Finally, substituting r back into q, and then into π(q), maximum sustainable profit is
(a − c)2 9
when δ≥ ,
8b 17

(a − c)2 (9 − 5δ)(9 − 7δ) 9

when δ< .
9b(9 − δ)2 17
b) Assume consumers split business equally when firms name the same price.
Suppose we wish to sustain some price p ∈ (c, pm ]. Let corresponding (per firm) profits
be given by π(p) > 0. The condition that p is sustainable in equilibrium via Nash
reversion is
π(p)
≥ 2π(p)
1−δ
which reduces to δ ≥ 21 . The condition is independent of p, so that either all p ∈ (c, pm ]
can be sustained, or no p ∈ (c, pm ] can.
Thus, maximum sustainable profits are
1
πm when δ≥ ,
2
1
0 when δ< .
2
87. (Sept 2003 Comp, Q4)
a) Both firms naming c1 and when indifferent assuming the consumer buys from firm 1, is
a NE for the stage game. Infinite repetition of this stage NE is a SPNE for the repeated
game, in which both firms earn zero profits. No equilibrium can yield lower payoff to
either firm, as 0 is the minmax payoff to each firm.
Below assume the consumers resolve indifference by randomizing equally between the
two firms at any p ∈ (c2 , v], and buying soley from firm 1 for all other price. No price
above v can be sustained because the consumers will not buy. To sustain p = v, consider
reversion to the static NE described above.
1 1 1 1
For firm 1, this requires 1−δ 2 (v−c1 ) ≥ v−c1 , and, for firm 2, this requires 1−δ 2 (v−c2 ) ≥
1
v−c2 . Both yield δ ≥ 2 , the same condition as in the standard Bertrand case. Therefore,
the set of sustainable prices is [c1 , v] if δ ≥ 21 ; [c1 , c2 ] otherwise.
b) For δ ≥ 21 , the set of sustainable prices is [c1 , v] since we can already sustain all prices
without conglomerate markets. For δ < 21 , to sustain a price p, require 1−δ 1
(p − c1 ) ≥
2 −c1 )
(p − c1 ) + (p − c2 ). Solving yields p ≤ c2 + δ(c1−2δ ≡ p∗ . Note that firms will not charge
more than v, as this would lead to 0 profits. Therefore, the set of sustainable prices
for δ < 12 is [c1 , min{p∗ , v}]. Since p∗ > c2 as long as δ > 0, it is possible to sustain
∗
some price above c2 for all δ > 0. dp dδ > 0, so the highest sustainable price is strictly
v−c2
increasing in δ until δ = 2v−c 1 −c2
≤ 21 (when p∗ = v) and then constant at v for δ > 12 .

75
88. (June 2001 Comp, Q4)
A strategy for player P is a first period choice m1 ∈ {1, 2}, and a second period choice
m2 (m1 , x1 ) ∈ {1, 2} for each possible first period history (m1 , x1 ) ∈ {1, 2} × [0, 1]. Thus,
her strategy set is given by: SP = {1, 2} × {m2 such that m2 : {1, 2} × [0, 1] → {1, 2}}. A
strategy for manager i ∈ {1, 2} is a first period choice x1 ∈ [0, 1] (if elected in the first
period), and a second period choice x2 (m1 , x1 ) ∈ [0, 1] (if elected in the second period)
for each possible first period history (m1 , x1 ) . Thus, her strategy set is given by: Si =
[0, 1] × {x2 such that x2 : {1, 2} × [0, 1] → [0, 1]} .

a) i. Notice that in any (subgame perfect Nash) equilibrium x2 = 1. Thus, P ’s payoff will
be given by (1 − x1 ) , and each manager will receive either zero or one in the second
period. Moreover, the equilibrium x1 must satisfy:

x1 + δ ≥ 1,

where the left hand side is the most m1 can obtain when selecting x1 , and the right
hand side is the least m1 can obtain by deviating in the first period. Therefore, P ’s
payoff is bounded above by δ.
On the other hand, the strategy profile: m1 = 1, m2 (m1 , x1 ) = m1 iff x1 ≤ 1 − δ;
x1 = 1 − δ (for both managers), and x2 = 1 (for both managers) constitutes an
equilibrium that gives payoff δ to P. Therefore, from the previous paragraph, δ is
the highest possible payoff that P can obtain in equilibrium.
ii. Notice that the strategy profile: m1 = 1, m2 (·) = 1; and x1 = x2 = 1 (for both
managers) constitutes an equilibrium that gives a payoff of zero to P. Since P ’s
payoff is bounded below by zero, zero is the lowest possible equilibrium payoff for
P.
iii. The set of equilibrium payoffs for P is convex. To see this notice that any payoff π ∈
[0, δ] (i.e., between the highest and lowest possible equilibrium payoffs) is sustained
by the following equilibrium profile: m1 = 1, m2 (m1 , x1 ) = m1 iff x1 ≤ 1 − π;
x1 = 1 − π (for both managers), and x2 = 1 (for both managers).
b) There does exist an equilibrium such that, on the path of play, xt < 1 for all t. Fix
α ∈ (0, 1), and consider the following profile:

m1 = 1,

 mt−1 if xt−1 ≤ 1 − α and mt−1 ∈ {1, 2}
mt = {1, 2}mt−1 if xt−1 > 1 − α and mt−1 ∈ {1, 2}
1 if mt−1 ∈
/ {1, 2},


1 − α for i ∈ {1, 2}
xt =
1 for i ∈
/ {1, 2}.

Notice that the above profile constitutes an equilibrium for all α ≤ δ. Finally, provided
N ≥ 2 (and α ≤ δ) the above profile constitutes an equilibrium with the desired property.

89. (2005 Final, Q2)

90. (2007 Final, Q3)

76
 a) To sustain any positive profit π (i.e., any price above c), the incentive constraints (i.e.,
no profitable deviation) are
1 π
≥ (1 + δ + ... + δ ki −1 )π + 0, ∀i = 1, 2
1−δ 2

For all values of π, this gives δ ki ≥ 12 for i = 1, 2.

1
Therefore, the highest symmetric price sustainable is pm if δ ki ≥ 2 for both i = 1, 2 and
c otherwise.
∗ ∗
b) Let k ∗ be the nonnegative integer such that δ k ≥ 21 and δ k +1 < 21 . So the conditions
in the above question to sustain pm are ki ≤ k ∗ for i = 1, 2.
In any SPNE, if ki > k ∗ for some i, the best response for −i is to choose φ; if any firm
chooses φ, the best response for the other firm is to choose φ. Therefore we have a SPNE
where both firms choose φ at the beginning, and both price at c in all periods regardless
of history.
Any other SPNE must involve ki ≤ k ∗ for both i = 1, 2. And since c0 < 0, this implies
1 πm
that any other SPNE must involve k1 = k2 = k ∗ . Such a SPNE exists if c(k ∗ ) ≤ 1−δ 2 .
To summarize:
We have a SPNE where both firms choose φ at the beginning, and both price at c in all
periods regardless of history;
1 πm
If c(k ∗ ) ≤ 1−δ ∗
2 , we have a second SPNE where both firms choose k at the beginning,
both price at pm if no deviation is detected, both price at c if a deviation is detected.

91. (2007 Final, Q4)

a) Suppose firm 1 plays a with probability α and firm 2 plays a with probability β. The
minmax payoff for firm 1 is:
π 3
min max (αβ + (1 − α)(1 − β)) + π(α(1 − β) + β(1 − α)) = π
0≤β≤1 0≤α≤1 2 4

Similarly, the minmax payoff for firm 2 is:

π π
min max (αβ + (1 − α)(1 − β)) =
0≤α≤1 0≤β≤1 2 4

b) We know each must at least obtain their minmax payoff in any NE of the repeated game.
But note that the stage game is a constant-sum game, so if any firm obtains a payoff
strictly higher than his minmax payoff, it must be the case that the other firm obtains
a payoff strictly below his minmax payoff. Therefore, the highest payoff each firm can
obtain in any NE is his minmax payoff.

92. (June 2007 Comp, Q2)

a) The unique PSNE is (C,c).

b) Consider player 1. If the proposed equilibrium strategies are (A,a), cooperation will
give a payoff of 8 for the rest of time, while deviating will give him a payoff of 9 today,
but 2 after that (for the rest of time). Therefore, we need:

1 δ 1
8≥9+ (2) ⇒ δ ≥
1−δ 1−δ 7

77
 Similarly, for player 2, we need:
1 δ 1
8≥9+ (4) ⇒ δ ≥
1−δ 1−δ 5
(Therefore, if both have the same discount factor, we require δ ≥ 15 )
c) Remember that the most we can punish a player that deviates is by giving them their
minmax payoff for the rest of time. This punishment provides a lower bound on the
possible discount factors needed to achieve (A,a) in any Nash Equilibrium. Player 1’s
minmax payoff is 2 and player 2’s minmax payoff is 4 (the same as the payoffs from
(C,c)). Therefore, we cannot sustain any NE (or SPNE, which is a subset of NE) at a
lower discount rate than the one we found in (b).
93. (June 2007 Comp, Q3)
(a−c) 2
a) Note that the profits to one firm if both play a−c
4b are 8b . The profits to one firm if
a−c
both firms play 2b are 0. The profits from the optimal deviation if the opponent plays
a−c 9 (a−c)2
the 4b are 64 b . The profits from the optimal deviation if the opponent plays the
a−c 1 (a−c)2
2bare 16 b .
We have 1 constraint for the cooperation regime:
1 (a − c)2 9 (a − c)2 δ k+1 (a − c)2
≥ + (δ + ... + δ k )(0) +
1 − δ 8b 64 b 1 − δ 8b
Then, we have k constraints given that we are in the ith period of the punishment regime
(note that deviation leads back to being punished for k periods):
δ k+1−i (a − c)2 1 (a − c)2 δ k+1 (a − c)2
(1 + δ + ... + δ k−i )(0) + ≥ + (δ + ... + δ k )(0) +
1−δ 8b 16 b 1 − δ 8b
b) Note that if the constraint for the first period of the punishment (i = 1) holds, the
constraints for the rest of the periods of punishment will hold as well. Therefore, we
just need to satisfy 2 constraints. Doing the math leads to two equations:
1 (a − c)2 9 (a − c)2 δ k+1 (a − c)2
≥ + (δ + ... + δ k )(0) +
1 − δ 8b 64 b 1 − δ 8b
δ k (a − c)2 1 (a − c)2 δ k+1 (a − c)2
(1 + δ + ... + δ k−1 )(0) + ≥ + (δ + ... + δ k )(0) +
1 − δ 8b 16 b 1 − δ 8b
Note that as k rises, the first equation becomes easier to satisfy (because the punishment
is harsher) while the second becomes harder to satisfy (because players want to deviate
from this harsher punishment more). However, for any k, if δ satisfies the first equation,
it will satisfy the second equation. Focusing on the second equation, we see that k = 1
leads to the lowest δ (δ ≥ 12 ) so a quicker punishment is best for the players.
94. (June 2006 Comp, Q2)
a) i. Let T be the total net contributions of all other players. For a liberal, payoffs
depend on choice as follows: −(T + b − c)2 − b2 . Maximizing over b, we obtain the
first order condition −2(T + b − c) − 2b = 0. Consequently, b = max {(c − T ) /2, 0}.
Similarly, for a conservative candidate, the payoff is −(T − g + c)2 − g 2 . Maximizing
over g, we obtain the first order condition 2(T − g + c) − 2g = 0. Consequently,
g = max {(c + T ) /2, 0}.

78
 ii. Notice that we can rewrite the first order condition for liberal i as bi = c − A,
where A represents aggregate (net) contributions, including i’s. Accordingly, bi =
max {c − A, 0}, which is the same for all i. A similar argument applies for conser-
vatives.
iii. Suppose that, in equilibrium each liberal gives b∗ and each conservative gives g ∗ .
Then, for liberals, T = (N − 1)b∗ + M g ∗ . For b∗ to be a best response, we must
have
c − (N − 1)b∗ + M g ∗
b∗ =
2
which we can rewrite as
c − (N + 1)b∗ + M g ∗ = 0 (27)
For conservatives, T = N b∗ + (M − 1)g ∗ . For g ∗ to be a best response, we must
have
c + N b∗ − (M − 1)g ∗
g∗ =
2
which we can rewrite as
c + N b∗ − (M + 1)g ∗ = 0 (28)
Solving (28) and (27) yields

(2M + 1)c
b∗ = ≥0
M +N +1
and
(2N + 1)c
g∗ = ≥0
M +N +1
(so the non-negativity constraints are satisfied). This leads to the policy
 
N −M
x∗ = c
M +N +1
N
iv. With M = k, we have !
1
∗ 2+ M
b = 1 c
1+k+ M
and !
1
∗ 2k + M
g = 1 c
1+k+ M
This leads to the policy !
k−1
x∗ = 1 c
1+k+ M
The overall per capita contribution is
!
4k 1
∗ k ∗ 1 1+k +M
t = b + g∗ = 1 c
1+k 1+k 1+k +M

The effect of population size (changing M ) depends on k. When k = 1 (the two

groups are of equal sizes), population has no effect either on per capita contributions
or on the outcome. When k > 1 (liberals are more numerous), an increase in

79
 population size causes each liberal to give less and each conservative to give more.
The resulting policy becomes more conservative. When k < 1, the opposite occurs.
4k
It is easy to check that 1+k ≤ 1 + k for k > 0, with equality only when k = 1, from
∗
which it follows that t declines with M (except when k = 1, in which case it is
2c 2kc
constant). Limiting contributions are 1+k for liberals, 1+k for conservatives, and
4kc (k−1)c
(1+k)2
; the limiting policy is 1+k .
v. The result does not hold. Fixing the total contributions of the other group, a
group’s per capita contributions would indeed go to zero as population increases,
just as in a standard public goods problem. However, in equilibrium, the total
contributions of the other group rises. Competition between the groups keeps the
marginal benefit of contributions for the typical individual, and hence per capita
contributions, bounded away from zero.
vi. The equilibrium is inefficient because it involves offsetting contributions by liberals
and conservatives. For the case of M = N , the equilibrium is x = 0. The most
efficient way to achieve this involves zero contributions by everyone.
b) i. The efficient outcome involves no contributions and x = 0. Every player receives
a payoff of −c2 . The best response to this, for either a liberal or a conservative,
is to contribute c/2. In that case, the associated payoff is −c2 /2. In the static
Nash equilibrium, the policy is x = 0, and each individual contributes c, so everyone
receives a payoff of −2c2 . Consequently, an equilibrium prevails as long as

c2 c2 δ2c2
− ≥− −
1−δ 2 1−δ
We can rewrite this as
1
δ≥
3
ii. Population size is irrelevant. It doesn’t affect the static equilibrium, the efficient
outcome, or the gains associated with deviating from the efficient outcome.

95. (Sept 2006 Comp, Q1)

a) Rewrite utility as U (e, y − t(y − g) − g − f ) = U (e, y(1 − t) + tg − e) since e = f + g. Now

note that this expression is maximized when f = 0 and e = g. Therefore the problem is:

max U (e, y(1 − t) − e(1 − t)).

e

U1
The FOC yields U2 = 1 − t.
U1 c
In the case when U (e, c) = ec, U2 = e = 1 − t. Together with the budget constraint
(1−t)y
c = (y − e)(1 − t), we get c = 2 and e = y2 .
b) i. A strategy for the school is e ∈ R+ . The school’s strategy set is R+ . A strategy for
a family is G : R+ → R+ . Each family’s strategy set is the set of all such functions.
ii. Let e∗ be the level set in the first period and let g ∗ be the level of contribution
of other families (which we assume to be  the
same because

 we are looking for a
symmetric equilibrium). Then f = e∗ − NN−1 g ∗ + N1 g .
     
∗ ∗ ∗ N −1 ∗ 1
U (e , (1−t)(y−g)−f ) = U e , (1 − t)y − (1 − t)g − e + g + g
N N

80
       
∗ ∗ N −1 ∗ 1
= U e , (1 − t) y − e + g +g − (1 − t)
N N
We therefore have two cases to consider:
Case (i): N1 < 1 − t. Giving nothing is a dominant strategy, so we know g ∗ = 0.
Case (ii): N1 > 1 − t. Giving e∗ is a dominant strategy, so we know g ∗ = e∗ .
Now consider the first stage in case (i). Maximizing U (e, y(1 − t) − e) with respect
y(1−t)
to e yields FOC U U2 = 1. For the specific case U = ec, we have c =
1 ∗
2 and
e∗ = y(1−t)
2 .
In case (ii), the first stage is identical to part (a).
iii. In case (ii), the cooperative and non-cooperative solutions are the same. In case (i),
they are not the same. In the parametric case that we solved for explicitly, there is
generally less education in the non-cooperative solution, though the level of private
consumption is the same.
c) i. In case (ii), the non-cooperative solution is identical to the cooperative solution in
the stage game, so the cooperative solution is trivially sustainable for all values of
y (1−t)y
δ. So below
  we focus on case (i). The stage payoff from cooperation is ( 2 )( 2 ) =
2 2
(1 − t) y4 , and the stage payoff from the static Nash equilibrium is (1 − t)2 y4 .
The best deviation from cooperative outcome is to set g = 0, which gives stage
2 2
payoff of ( y2 )[(1 − t)y − 2Ny
] = (1−t)y
4 + y4 ((1 − t) − N1 ). Therefore, the best gain
2
from deviation is y4 ((1 − t) − N1 ).
The incentive constraint for no profitable deviation is then:

y2 y2
    
1 δ
(1 − t) − ≤ (1 − t) (1 − (1 − t))
4 N 1−δ 4
1
1−t− N
Solving for δ, we get δ ≥ 1
1−t2 − N
.
1
ii. As N → ∞, the threshold converges to 1+t , which is decreasing in t. Therefore a
higher tax rate reduces the minimum δ at which cooperation is sustainable.
d) i. The strategy set for each family is the same as before. For the school, the strategy
set is R+ × F where F consists of functions f : RN +
+1
→ R+ .
ii. We need to solve for the education level that will be set in the third stage. Suppose
that in the
second stage every family except one sets g ∗ , and one sets g, so f =
e − NN−1 g ∗ + N1 g .



The (per family) payoff for all but the N th family is:
     
∗ N −1 ∗ 1
e (y − g ) (1 − t) − e + g + g .
N N

And the payoff for the N th family is:

     
N −1 ∗ 1
e (y − g)(1 − t) − e + g + g .
N N

The school’s objective is to maximize with respect to e:

     
∗ N −1 ∗ 1
(N − 1)e (y − g )(1 − t) − e + g + g
N N

81
      
N −1 ∗ 1
+e (y − g)(1 − t) − e + g + g
N N
The FOC gives (N −1) (y − g ∗ )(1 − t) − e + NN−1 g ∗ + N1 g + (y − g)(1 − t) − e + N −1
g∗ + 1

 

N Ng
−(N − 1)e − e = 0, which yields e = y(1−t) 2
1
+ 2N [(N − 1)g ∗ t + gt].
y(1−t) ∗
For large N , this converges to e∗ (g ∗ ) = 2 + g2 t , which is invariant in g and
thus each family will take it as fixed when deciding on g in the second stage. The
consumption for the family who sets g (when every one else sets g ∗ ) is
 
∗ ∗ N −1 1
c = (y − g)(1 − t) − e (g ) + g ∗ + g.
N N

This converges to c = (y − g)(1 − t) − e∗ (g ∗ ) + g ∗ as N → ∞, so the optimal choice

of g is to set g = 0.
Therefore, any subgame perfect symmetric equilibrium involves g ∗ = 0.
iii. This solution is the same as in case (i) in the previous part where the school’s
promise was binding. There is no difference because in either case, the school can
simply set the appropriate tuition level to achieve the desired education level.

96. (2009 Final, Q1)

People like their friends to hear the secret and receive α for every friend that hears (including
themselves). People do not like their enemies to hear and receive β for every enemy that
hears.
In what follows, I will use the notation “T” for a playing telling, and “NT” for a player
not telling. When writing equilibrium strategies for multiple players, I will list player 1’s
strategy, then player 2’s strategy, etc.

a) {T,NT} is player 1’s strategy set. T is the dominant strategy. Therefore, the only
SPNE is {T}.
b) Using the same logic as in (a), 2 must tell 3 in the final subgame. Therefore, since
β = 2 and α = 1, it is better for 1 to not tell 2. So, the only SPNE is {NT,T} (note
that player 3 has no decision here). This is also the unique NE.
c) Using the same logic as in (a), 3 must tell 4 in the final subgame. Using the same logic
as in (b), 2 must not tell 3. Therefore, it better for 1 to tell 2. So, the unique SPNE
is {T,NT,T}. There are 3 NE: {T,NT,T},{NT,T,NT},{NT,T,T}. In the last 2 NE,
player 2 makes an uncredible “threat” (not really a threat because it hurts him) to tell
3, even though that would not happen in a SPNE. 3 can choose either strategy as he
will never play it in equilibrium and it does not affect the others’ decisions.
d) Using the logic above, we know that in a SPNE with n players, player n-1 must tell
player n. But, then player n-2 will not tell player n-1. Then, player n-3 will tell player
n-2. Then, player n-4 will not tell player n-3 and so on. Therefore, the SPNE is
if n is even: {T,NT,T,NT,...,T,NT,T}
if n is odd: {NT,T,NT,T,...,T,NT,T}
e) Here is the general logic to solve this problem:
Person n − 1 will definitely tell person n.
Then, person n − 2 will tell person n − 1 as long as α > β.
If the above is true, person n − 3 will tell person n − 2 as long as α > 2β.

82
 If the above is true, person n − 4 will tell person n − 3 as long as α > 3β.
...
If the above is true, person n − i − 1 will tell person n − i as long as α > iβ.
This suggests the following end to this chain. We find the lowest i such that α < iβ.
Call this i∗ . Then, everyone from person n − i∗ to n − 1 will tell. Person n − i∗ − 1
will not tell.
Now, consider person n − i∗ − 2. He is like person n − 1 above and will always tell
person n − i∗ − 1.
Then, person n − i∗ − 3 will tell person n − i∗ − 2 as long as α > β.
If the above is true, person n − i∗ − 4 will tell person n − i∗ − 3 as long as α > 2β.
If the above is true, person n − i∗ − 5 will tell person n − i∗ − 4 as long as α > 3β.
...
Etc.
So, this suggests the following answer. The strategies will take the following form with
n players.
Players n − i∗ to n − 1 will tell.
Player n − i∗ − 1 will not tell.
Players n − 2i∗ − 1 to n − i∗ − 2 will tell.
Player n − 2i∗ − 2 will not tell.
etc. (continuing in groups of i∗ players telling and then one player not telling)
So, they will look something like: {...T, ..., T, N T, T, ..., T, N T, T...T }
f) i. We have a situation in which we have 1 SPNE (and NE) outcome of the stage game.
Therefore, there is no way to get any other strategies played in a finitely repeated
game.
Intuitively, this is true because of backward induction. In the final stage game, 2
will definitely tell 3 if he is told. Therefore, 1 will never tell 2 in the final stage
game. But, then, in the second-to-last stage game, 1 cannot provide any incentive
(in the form of reward and punishment) for 2 not to tell 3, as 2 knows that 1 will
never tell him the secret in the final subgame. But, then, 1 will never tell 2 in the
second to last subgame. But, this logic continues backward through all of the stage
games.
ii. Without any monitoring, 1 can only condition his strategy on *his* previous strate-
gies, not player 2’s strategies. If this is the case, there is no way for 1 to provide any
punishment based on player 2’s actions, so we cannot get him to tell 2 the secret in
any SPNE.
iii. Here, we might be able to have 1 tell 2 the secret. Note that Nash reversion is the
harshest punishment possible. Consider the following strategies.
Player 1:
If has ever received a signal that player 2 has told, play NT.
If not, play T.
Player 2:
If 1 has ever played NT in the past or if 1 has discovered that 2 has told, play T.
If not, play NT.
The question is whether these strategies form a SPNE. Consider player 1. Given
that player 2 follows this strategy, player 1’s strategy is a best response regardless

83
 of p and δ. Now, consider player 2. Consider a one-period deviation to T. (Note
that we only need to check the one stage deviations here because the situation is
the same every period.) If he tells, he gets 3 today (1 from each player 1, 2, and 3
knowing the secret). Tomorrow, with probability p, 1 will have seen that he told
and revert to NT. Then, player 2 will only get a payoff of 1 (from player 1 knowing)
in every future period. With probability 1 − p, he will get 2 in every future period.
If he does not tell, he also gets 2 in every future period. So, we need that:
δ 2δ 2
3+p + (1 − p) ≤
1−δ 1−δ 1−δ
1
Solving this gives us δ ≥ 1+p .
This makes sense – as p → 0, we need serious patience (δ → 1) to get cooperation.
As p → 1, we need the patience (δ → 21 ) required with full monitoring to get
cooperation.

97. (2011 Final, Q3)

a) The pure-strategy NE are (S, F ) with payoffs (1, 4) and (F, S) with payoffs (4, 1). There
is a symmetric mixed-strategy equilibrium in which both countries play S with proba-
bility 12 . The associated payoffs are (2, 2).
b) The highest symmetric payoff must place the most probability possible on the strategy
profile (S, S) and no probability on (F, F ). Because the game is symmetric, the easiest
way to satisfy both countries’ incentive constraints while maximizing the probability on
(S, S) is to put equal probability on (S, F ) and (F, S). Let λ be the probability placed
on (S, F ) and (F, S). This means probability 1 − 2λ is placed on (S, S). If either country
is told to play F , it believes the other country will play S fo r sure. F is a best response
to S, so this is incentive compatible. Then we just need to check that a country wants
to play S when told to play S. This is true if:
     
1 − 2λ λ 1 − 2λ
3 +1 ≥4
1−λ 1−λ 1−λ

This holds if λ ≥ 13 . For the highest symmetric payoff, we want the lowest possible λ.
When λ = 31 , the payoffs from the correlated equilibrium are the maximum symmetric
payoffs: ( 38 , 83 ).
c) No. In fact, there is no NE in which both countries play S in the first period, regardless
of the value of δ. In the second period, the countries must play a NE of the stage game
(on the equilibrium path). From (1), there are only 3 choices. (S, F ) or (F, S) give
one of the players their minmax payoff. There is no punishment worse than that, so
the player receiving the minmax payoff on the equilibrium path in the second period
would definitely want to deviate from (S, S) in the first period. So the only possibility
to sustain (S, S) in the first period is to have the second period be the mixed-strategy
equilibrium (with payoffs of 2), conditional on history (S, S) from the first period. The
worst punishments we can prescribe off the equilibrium path are minmax payoffs (1 for
both players in this game), which are generated by the two pure-strategy equilibria.
Would either country want to deviate? They are happy playing S in the first period if
and only if
3 + 2δ ≥ 4 + δ ⇐⇒ δ ≥ 1.
Clearly this is never true, so there is no such NE when δ < 1.

84
d) Define history h∗ by h∗ (1) = ∅, h∗ (2) = (S, S), h∗ (3) = ((S, S), (S, F )). Define actions
a∗ by a∗ (1) = (S, S), a∗ (2) = (S, F ), and a∗ (3) = (F, S). The following strategies form
a SPNE that generates h∗ on the equilibrium path.
Strategy for country i:
If h(t) = h∗ (t), play a∗i (t).
If h(t) 6= h∗ (t), find the first lone deviator and play the stage game NE that gives that
deviator his minmax payoff.
If h(t) 6= h∗ (t) and no lone deviator, play a∗i (t).
First we need to verify that these strategies form a NE. In periods 2 and 3, the prescribed
actions on the equilibrium path are NE of the stage game. Therefore, there is definitely
no profitable deviation in period 3. In period 2, no deviation can increase the payoff in
the current period and any deviation leads only to minmax payoffs in period 3, which
must be (weakly) worse than payoffs from (F, S). So there is no profitable deviation in
period 2. Then, we just need to check that neither country wants to deviate in period
1. Country A doesn’t want to deviate as long as
1
3 + δ + 4δ 2 ≥ 4 + δ + δ 2 ⇐⇒ δ ≥ √
3
Country B doesn’t want to deviate as long as
1
3 + 4δ + δ 2 ≥ 4 + δ + δ 2 ⇐⇒ δ ≥
3
Thus, we have a NE when δ ≥ √13 . Is it subgame perfect? Yes, because every continu-
ation game off the equilibrium path is Nash reversion. Thus, there is a SPNE in which
both countries surrender in the first period as long as δ ≥ √13 .
e) Yes, the folk theorem tells us that there is some δ < 1 such that it is possible to sustain
any feasible payoffs that are strictly greater than minmax payoffs when the horizon is
infinite. Both countries have a minmax payoff of 1. Since the payoff from cooperation
is strictly greater than 1, we will be able to sustain it in a SPNE.
To figure out the lowest possible δ such that cooperation can be sustained, we need to use
the harshest possible punishments – minmax payoffs. In this case, Nash reversion (of the
right kind) generates minmax payoffs, and so it provides the most severe punishments.
Therefore, we restrict attention to the following SPNE with symmetric strategies:
In period 1, play S (
F if h(t) contains an F and the other country played F first
In period t > 1, play
S otherwise
For any subgame off the equilibrium path, the strategies prescribe playing a NE ((S, F )
if Country A is being punished and (F, S) if Country B is being punished) of the stage
game in every period, regardless of the history. This is a NE of the subgame because
deviations cannot effect the future actions of the other country and deviations also cannot
improve current period payoffs.
We need to check that playing S on the equilibrium path is optimal, given the strategies.
3
Cooperation yields a payoff of 1−δ . Each country’s best deviation yields a payoff of 4
today and 1 for the rest of time. This deviation is not profitable if and only if
3 δ 1
≥4+ ⇐⇒ δ ≥ .
1−δ 1−δ 3

85
 1
Therefore, the lowest delta such that cooperation is sustainable is δ = 3 and the SPNE
strategies that generate this cooperation are those given above.

98. (June 2012 Comp, Q3)

99. (2013 Final, Q1)

100. (PS9, Q1)

a) If λ = 1, B is never chosen by nature, so we can effectively wipe out that branch. The
game reduces to the centipede game and unravels, with players choosing stop at each
node. If λ = 0, we wipe out the A branch, and players choose continue at every node.13
From this point on, we restrict attention to λ ∈ (0, 1).
There are 2 proper subgames, rooted at 1’s final decision nodes. Subgame perfection
requires that player 1 of type A will choose s and type B will choose c. We are left with
solving the NE of the reduced game, which has normal form:

s c
ss λ, 1 λ, 1
sc 1, 4 − 3λ 3 − 2λ, 6 − 5λ
cs 0, 1 + 3λ 3λ, 1 + 2λ
cc 1 − λ, 4 3, 6 − 3λ

where 1’s strategy gives his action when A and when B. Since λ ∈ (0, 1), strategies ss
and cs are strictly dominated. This should also be clear from the extensive form, as type
B would never play stop. Thus, we have reduced normal game

s c
sc 1, 4 − 3λ 3 − 2λ, 6 − 5λ
cc 1 − λ, 4 3, 6 − 3λ

When λ ∈ (0, 32 ), the game is dominance solvable, with unique NE (cc, c). Thus, we have
 
2
(cc, c) λ ∈ 0, . (29)
3

When λ = 32 , the unique class of NE is

 
2 2
cc, s with prob p ≤ λ= . (30)
3 3

When λ ∈ ( 32 , 1), there is a unique NE, which is a strictly mixed NE:

   
3λ − 2 2 − 2λ 2 1 2
sc + cc, s + c λ∈ ,1 . (31)
λ λ 3 3 3

b) Notice that in each of the SPNE, the information set with 2 elements is reached with
positive probability. Hence, all of the SPNE are also SE, where the beliefs for 2 are
calculated from the strategies via Bayes’ Law.
13
Note that if λ ∈ {0, 1}, another school of thought is to apply equilibrium analysis to the full extensive form,
resulting in additional, peculiar equilibria. However, these results are not meaningful and are ignored.

86
 Finally, notice that as λ → 1, player 1 of type A will play stop with high probability.
However, should 2 get a chance to play, he still mixes between c and s – no unravelling!
This may seem surprising because ex-ante, 2 assumes he is facing a type A player who
would play stop at the last node. But since in equilibrium, type A players immediately
choose stop with such high probability, if the game ever reaches 2’s decision node, 2’s
ex-post probability on A is significantly lower – low enough to justify playing either s or
c!

101. (PS9, Q2)

Technical note: The PBE concept does not have as much bite as we would like in this problem
(in particular, regarding beliefs following the rejection of offers not made in equilibrium.)
However, under the sequential equilibrium concept, for some choices of λ there are no pure
strategy SE, only mixed. Thus, we pick the lesser of two evils and use the PBE concept,
solving for all equilibria, including those involving unreasonable beliefs.

a) A strategy for the seller is a first period offer O1 and a second period contingent offer
O2 (O1 ) for each possible O1 rejected in period 1. A strategy for a buyer of type vi is
a plan b1 (O1 , vi ) to accept or reject each possible offer O1 ∈ R+ , and a second period
plan b2 (O1 , O2 , vi ) to accept or reject each O2 ∈ R+ when O1 was the (rejected) first
period offer. We also need beliefs for the seller after his first period offer is rejected. Let
µ(O1 ) = probability seller assigns to buyer having type vH .
Thus, our equilibria take the form:

[O1 , O2 (O1 )], [b1 (O1 , vi ), b2 (O1 , O2 , vi ), µ(O1 )].

Claim 1. 
accept if O2 ≤ vi
b2 (O1 , O2 , vi ) =
reject otherwise
In period 2, the buyer is sequentially rational and accepts only offers ≤ his value.
Claim 2. O2 (O1 ) ∈ {vL , vH }
Given buyer’s period 2 behavior, sequential rationality ⇒ seller names
(
vH if µ(O1 ) ≥ vvHL
O2 (O1 ) =
vL if µ(O1 ) < vvHL

Claim 3. 
A if O 1 ≤ vL
b1 (O1 , vL ) =
R if O1 > vL
Type vL buyer, should he reject, anticipates price vL or vH in period 2, and thus 0 profit.
Thus, he takes what he can get in period 1.
Claim 4. Define v ∗ ≡ vH (1 − δ) + vL δ.

 A if O1 ≤ v∗
A if O1 ∈ (v ∗ , vH ) and O2 (O1 ) = vH


b1 (O1 , vH ) =

 R if O1 ∈ (v ∗ , vH ) and O2 (O1 ) = vL
R if O1 > vH .


Type vH buyer, should he reject, anticipates price vL or vH in period 2, and thus, profit
at best δ(vH − vL ). Thus, vH accepts any offer with vH − O1 ≥ δ(vH − vL ) ⇔ O1 ≤ v ∗ .

87
 vH clearly rejects O1 > vH . But if O1 ∈ (v ∗ , vH ), vH ’s choice depends on the seller’s
strategy. If the seller will offer vH next period, it is worth accepting O1 now. Otherwise,
if the seller will offer vL , it is worth it to wait.
Thus, we have the buyer’s strategy fully characterized in terms of the seller’s strategy.
Now let’s work on the seller’s strategy.
Claim 5. In equilibrium, the seller never offers O1 < vL .
Such an offer will be accepted by all parties, so he does better to offer vL instead.
Claim 6. In equilibrium, the seller never offers O1 ∈ (vL , v ∗ ).
Such an offer will be accepted by all the high types and rejected by the low types.
Sequentially rationality would require naming vL in period 2. But then the seller does
better to name v ∗ in the first period, followed by vL in the second period.
Claim 7. In equilibrium, the seller never offers O1 ∈ (v ∗ , vH ].
If type vH accepts this offer in equilibrium, then upon rejection, the seller believes he is
facing type vL and names vL in the next period. But this contradicts claim 4. If type
vH rejects this offer in equilibrium, by claim 4, the seller must offer vL upon rejection.
But then he does better to name vL in the first period!
Thus, we have O1 = vL , OL = v ∗ , or OL > vH . Let’s look for equilibria in which...
Case 1: O1 = vL . In such an equilibrium, both vH and vL types will accept in the
first period, yielding profit vL for the seller. We must rule out profitable deviation to (i)
O1 < vL (ii) O1 ∈ (vL , v ∗ ] (iii) O1 ∈ (v ∗ , vH ] and (iv) O1 > vH .
O1 < vL will be accepted by both types and yields lower profit that vL .
O1 ∈ (vL , v ∗ ] will be accepted by vH types, by claim 4, and rejected by vL ’s. Since
such an O1 is off the equilibrium path, beliefs µ(O1 ) are unrestricted by Bayes’ law.14
Therefore, we require

vL ≥ λO1 + δ(1 − λ) max[vL , µ(O1 ) · vH ] ∀O1 ∈ (vL , v ∗ ]. (32)

(32) should be thought of as a restriction on off-the-equilibrium-path beliefs, which in

turn pin down the seller’s off-the-equilibrium-path actions O2 (O1 ). Note that by choosing
O1 = v ∗ , (32) yields
vL
vL ≥ λv ∗ + δ(1 − λ)vL ⇒ λ≤ .
vH

λv ∗ + δ(1 − λ)vL is the minimax15 payoff from deviating to O1 ∈ (vL , v ∗ ] and thus,
λ ≤ vvHL is a necessary condition for this case.
O1 ∈ (v ∗ , vH ] will be rejected by vL types. Should type vH accept O1 , by claim 4, we
must have O2 (O1 ) = vH . Should vH reject O1 , by claim 4, we must have O2 (O1 ) = vL .
The seller will never deviate to an O1 that will be rejected by vH types, as that would
yield profit δvL < vL . He will prefer not to name any O1 ∈ (v ∗ , vH ] that will be accepted
by vH ’s when

vL ≥ λO1 + δ(1 − λ)vH ∀O1 ∈ (v ∗ , vH ] with O2 (O1 ) = vH . (33)

14
Notice that this allows for unnatural concepts, such as the seller believing he can offer O1 , which vH types would
all accept, but also believing that if it is rejected, he will face type vH with certainty. Such beliefs are ruled out by
the sequential equilibrium concept.
15
That is, the smallest payoff the optimal deviation in the class could yield.

88
 Notice that we could certainly have b(O1 , vH ) = reject, O2 (O1 ) = vL ∀O1 ∈ (v ∗ , vH ], so
that the minimax payoff from deviating is δvL .
O1 > vH will be rejected by all buyers. Beliefs O2 (O1 ) are unrestricted. Thus, we require

vL ≥ δ max[vL , µ(O1 ) · vH ] ∀O1 > vH (34)

Here the minimax payoff from deviating is δvL .

Therefore, provided λ ≤ vvHL , there exist eqa of the form
Buyer’s strategy: Satisfies claims 1,3,4 and condition (33)
Seller’s strategy: O1 = vL , O2 (vL ) = vL
off-eqm path beliefs and actions satisfy claim 2, and (32), (33), (34).
∗
Case 2: O1 = v . In this case, vH types accept and vL types reject. Therefore, we must
have O2 (v ∗ ) = vL so that seller profit is π ∗ ≡ λv ∗ + δ(1 − λ)vL . For this to be an eqm,
our no-profitable deviation conditions are given by (32), (33), (34), where on the LHS of
each inequality, vL is replaced by π ∗ . Additionally, we require that vL is not a profitable
deviation:
vL
π ∗ ≥ vL ⇔ λ ≥ .
vH
vL
Therefore, provided λ ≥ vH , there exist eqa of the form
Buyer’s strategy: Satisfies claims 1,3,4 and condition (33)
Seller’s strategy: O1 = v ∗ , O2 (vL ) = vL
off-eqm path beliefs and actions satisfy claim 2, and (32), (33), (34).
Case 3: O1 = O1 > vH . In this case, the first period offer will be rejected. Note that
we cannot have O2 (O1 ) = vL , or else the seller would do better by naming vL in the
first period. Hence, O2 (O1 ) = vH , yielding profit π ≡ δλvH . For this to be an eqm, our
no-profitable-deviation conditions are given by (32), (33) and (34), where on the LHS of
each inequality, vL is replaced by π. The condition that v ∗ is not a profitable deviation
is
δvL
π ≥ π∗ ⇔ λ ≥ . (35)
(2δ − 1)vH
That vL is not a profitable deviation:
vL
π ≥ vL ⇔ λ≥ (36)
δvH
which follows from (35).
δvL
Therefore, provided λ ≥ (2δ−1)vH , there exist eqa of the form
Buyer’s strategy: Satisfies claims 1,3,4 and condition (33)
Seller’s strategy: O1 = O1 , O2 (O1 ) = vH
off-eqm path beliefs and actions satisfy claim 2, and (32), (33), (34).
This completes our characterization of PBE in this problem.
b) Technical note: Here we use the sequential equilibrium concept. This greatly simplifies
the solution. In particular, it ensures reasonable beliefs following rejection of offers not
made in equilibrium.
For this problem, we use the assumption that 2vL < vH . (The solution is more interesting
in this case.)
The forms of strategies and beliefs are the same as in part (a).
We first work with the buyer’s strategy:

89
 Claim 1. Identical to claim 1 in part (a).

accept if O2 ≤ vi
b2 (O1 , O2 , vi ) = (37)
reject otherwise

Claim 2. O2 (O1 ) ∈ [vL , vH ]

Follows from buyer’s period 2 behavior and sequential rationality of seller.
Claim 3. In period 1, buyers use cutoffs.
In particular, the component b1 (O1 , vi ) of i’s strategy can be represented as v ∗ (O1 ), so
that
accept if vi ≥ v ∗ (O1 )

b1 (O1 , vi ) =
reject if vi < v ∗ (O1 )
Let O2 (O1 ) be the period 2 component of the seller’s eqm strategy. Consider buyer with
type vi facing O1 . If O1 ≤ vL , the buyer must accept, since by claim 2, he anticipates
an offer ≥ vL in the next period. O1 > vH must obviously be rejected. Consider
O1 ∈ (vL , vH ]. If O1 ≤ O2 (O1 ), there is no point in waiting for the higher offer in the
second period, so he buys if vi ≥ O1 . If O1 > O2 (O1 ), he prefers O1 over O2 (O1 ) when
vi − O1 ≥ δ(vi − O2 (O1 )), or vi ≥ O1 −δO1−δ
2 (O1 )
≡ ṽ(O1 ). Observe that ṽ(O1 ) > O1 , so that
any buyer who prefers O1 to O2 (O1 ) will in fact accept O1 .
Hence: 

 vL if O1 ≤ vL
O1 if O1 ∈ (vL , vH ] O2 (O1 ) ≥ O1

v ∗ (O1 ) = (38)

 min[ṽ(O 1 ), vH ] if O1 ∈ (vL , vH ] O2 (O1 ) < O1 .
vH if O1 > vH .


Claims 1 and 3 give the buyer’s strategy in terms of the seller’s strategy.
Now let’s work on the seller’s offer O2 (O1 ).
Consider O1 > vL . Note that not all buyers will accept O1 . Hence, we can pin down
the seller’s beliefs µ(O1 ) by identifying the types that reject O1 , using Bayes’ Law.16
Since buyers use a cutoff strategy, following O1 , the seller must have uniform beliefs
∼ U [vL , v ∗ (O1 )]. Given such beliefs, the seller solves

max O2 · (v ∗ (O1 ) − O2 )
O2 ∈[vL ,v ∗ (O1 )]

which has solution

v ∗ (O1 )
 
O2 (O1 ) = max vL , . (39)
2
Notice that this rules out O2 (O1 ) ≥ O1 , thereby simplifying (38).
Now (38) and (39) constitute a (somewhat complicated) system involving v ∗ (O1 ) and
O2 (O1 ). Solving simultaneously yields

O2 (O1 ) v ∗ (O1 )
O1 −δvL
When O1 ∈ (vL , (2 − δ)vL ] vL 1−δ
(40)
O1 2O1
O1 ∈ (2 − δ)vL , 2−δ

When 2 vH 2−δ 2−δ
2−δ vH
When O1 > 2 vH 2 vH
16
Even for O1 off the eqm path! This is where the sequential equilibrium concept is essential.

90
 Here we have used 2vL < vH . Otherwise, there would be no middle case. Note that the
middle case corresponds to an interior cutoff for the buyer and interior O2 for the seller.
For O1 ≤ vL , we have that all buyers will accept, so v ∗ (O1 ) = vL . However, beliefs
µ(O1 ) are then unrestricted, even under the sequential equilibrium concept.
Finally, since we have pinned down the outcome following each O1 , we identify the
equilibrium choice of O1 as the O1 that maximizes the seller’s profit.
Profit is then given by

when O1 ≤ vL



 O1



∗ (O ) ∗

O1 vHv−v + δvL v v(O 1 )−vL
 1


H −vL H −vL
when O1 ∈ (vL , (2 − δ)vL ]
π(O1 ) = (41)
 vH −v ∗ (O1 ) O1 v ∗ (O1 )−O2 (O1 ) 2−δ

O1 vH −vL + δ 2−δ when O1 ∈ (2 − δ)vL , 2 vH




 vH −vL


δ · v2H · vvHH−v
/2

when O1 > 2−δ
2 vH


L

Assume additionally that vH > 4v
L . It can then be verified that the optimal O1 lies in
the range O1 ∈ (2 − δ)vL , 2−δ
2 vH .
17

Solving the profit maximization problem yields

vH (2 − δ)2
O1 = .
2(4 − 3δ)
Combined with (37) and (40), this characterizes our sequential equilibrium.

102. (PS9, Q3)

103. (PS9, Q4)
104. (Sept 2003 Comp, Q5)
a) No. For type H, dw dw 
 
de U = 2e. For type L, de U = 1. Notice that the type H indifference

curve can be either shallower or steeper than the type L indifference curve depending
on whether e ≶ 12 .
b) In equilibrium, wi = πi . For separation, we need

πH − e2 ≥ πL

and
πL ≥ πH − e
(where e denotes the education level for type H). The first condition tells us that
e2 ≤ πH − πL , or e ≤(πH − πL )1/2 . The second tells us that e ≥ πH − πL . Thus, we can
construct a separating equilibrium by having type H workers choose any e satisfying

πH − πL ≤ e ≤ (πH − πL )1/2
17
Check that when vH > 4vL we have
arg max π(O1 ) = (2 − δ)vL .
O1 ∈[vL ,(2−δ)vL ]

91
 When πH − πL < 1, this set is non-empty. If πH − πL > 1, the set is empty, and so
there are no separating equilibria.
c) Separation now requires
πH − e2 − d ≥ πL (ICHL )
and
πL ≥ πH − e − d (ICLH )
(where e and d denote levels for type H). Then the two constraints become

πH − πL − d ≤ e ≤ (πH − πL − d)1/2

So we can choose any d such πH − πL − d ∈ [0, 1], and then choose e to satisfy the
previous condition.
Clearly, for any d, we want to make e as small as possible, so e = πH − πL − d. The
payoff for type H is then

πH − e2 + (e − πH + πL ) = πL + e − e2

Let us maximize this over e. The first order condition is 1 − 2e = 0, or e = 12 . So the

best equilibrium for type H involves e = 21 and d = πH − πL − 12 .
From the preceding, the equilibrium payoff for type H is πL + 14 . This is invariant
with respect to πH . The intuition for this is that H’s maximal utility is constrained by
ICLH , and therefore is determined by πL rather than πH.
105. (2004 Final, Q3b)
We assume that d >> c.18 In this game, a strategy for player i is a sequence of actions. The
action in time t = 1, 2, 3... is i’s decision clap/not clap should i find that both players have
chosen to clap in every period up to and including t − 1.
Possible actions for A and B in period t can be described by the following matrix:
N Mix C
N I II III
Mix IV V VI
C V II V III IX
We narrow down the set of equilibria by the process of elimination.

a) Claim: There is no sequential equilibrium where in each period, players clap with
probability ∈ (0, 1).
Proof: Suppose the contrary. Then all information sets are reached with positive prob-
ability, so beliefs are determined by Bayes’ Law. Let µt be the probability A places on
player B being crazy at the beginning of period t. Notice that µ1 = λ.
Let ptA and ptB be the probability placed on C in t by A and B respectively. Since
players strictly mix in each period, the expected (continuation) payoff in each period
from clapping is 0. For B to be indifferent between C and N in period t, we require

−c + (1 − ptA )d + ptA · 0 = 0
|{z}
| {z }
payoff from clapping payoff from not clapping

18
In particular, we require d > 2c. Notice that if d ∈ (c, 2c), there is an equilibrium where actions at t = 1 are
(C, N ) and (N, C) thereafter. If d < c the unique equilibrium is to never clap.

92
 which yields ptA = d−c
d , independent of t. For A to be indifferent between C and N in
period t, we require

−c + (1 − µt )(1 − ptB )d = 0
|{z}
| {z }
payoff from clapping payoff from not clapping

c 1
which yields ptB = 1 − d 1−µt .
We can now calculate µt+1 from µt using Bayes’ Law as follows:

µt µt d
µt+1 = = = µt .
µt + (1 − µt )ptB

µt + (1 − µt ) 1 − c 1 d−c
d 1−µt

Thus, such an equilibrium would require beliefs that increase geometrically, which is
impossible because we must have µ ∈ [0, 1].
Therefore, no such strictly mixed equilibrium exists.
b) Claim: In no period will B choose the pure action “stop clapping”(N ).
Proof: Suppose the contrary – that there exists an equilibrium in which B plays “stop
clapping” in t. Then in period t + 1, A must have beliefs µt+1 = 1.19 Hence, in period
t+1, A must choose N . But then it is profitable for B to deviate to C in period t, because
C yields d − 2c > 0. Hence, the proposed strategy profile cannot be an equilibrium.
c) Claim: In any equilibrium, B wins the prize with certainty.
We prove this using a series of further claims.
Claim: In no equilibrium do both players have strictly positive payoff.
Proof: If there were such an equilibrium, it could not involve placing probability on N
in t = 1. Hence, we have (C, C) in period 1. But then continuation payoffs at t = 2
must be positive for each player, so we have (C, C) in period 2. Induction pins down
strategies as (C, C) in each period, which is clearly not an equilibrium.
From this claim, it follows that in no period can actions be (C, C).
Claim: If in period t actions are (N, C), then in period t − 1, actions must be (N, C).
Proof: Given period t + 1 actions, in period t, C yields a minimum of d − 2c > 0 for B,
so B must choose C. For A, C yields −c, so A must choose N .
Hence, if in any period, actions are given by (N, C), then so too are period 1 actions,
so that B wins the prize with certainty. Henceforth, we restrict attention to equilibria
where in no period do we have actions (N, C).
Claim: Period t actions are never (N, mix).
Proof: This is obvious, as otherwise B could deviate in t by playing C.
We have now narrowed down the set of possible actions in each period t:
N Mix C
N × × ×
Mix × V VI
C × V III ×
In any equilibrium where actions in each period are given by V , V I or V III, every
information set is reached with positive probability. Hence, the sequence µt is (weakly)
increasing in t. By an argument similar to (a), there can only be finitely many periods
19
Here we have invoked the sequential equilibrium concept.

93
 where the actions are given by either V or V III. Thus, we are left with an equilibrium
with a tail in which actions are given by V I. But this equilibrium is not sequentially
rational because at any point in the tail, given that B will always clap, A would rather
choose N than mix.
Therefore, upon eliminating all equilibria that include (N, C) at t = 1, we have none
remaining. It follows that any equilibrium must involve (N, C) in period 1, which proves
the claim.

106. (June 2005 Comp, Q5)

a) For S, a strategy is an element of {D, N D}3 (it specifies D or N D for each of the three
values). For B, a strategy is an element of R4+ (it specifies a quantity for each possible
disclosure, and for nondisclosure).
b) B chooses q to solve maxq≥0 E[(v−1)q−q 2 ], which is the same thing as maxq≥0 [E(v) − 1] q − q 2 .


The solution is q = E(v)−1

2 . This is also the payoff for S.
c) For any completely mixed strategy profile, the posterior upon observing disclosure of
v 0 attributes all probability to quality v 0 . So any limiting belief also attributes all
probability to quality v 0 . The same is true for a PBE because the information set is a
singleton. The only possible belief is to put probability 1 on v 0 .
d) In any sequential equilibrium, v2 and v3 are definitely disclosed, while v1 may or may
not be disclosed (it’s irrelevant). Here’s why.
Claim #1: S will choose the action associated with the highest expected value of v,
according to the buyer’s beliefs. This follows from part (b).
Claim #2: In any equilibrium where either v2 or v3 is not disclosed with certainty, v1
is definitely not disclosed. Consider any sequential equilibrium where either v2 or v3
is not disclosed with certainty. Following non-disclosure, the buyer’s inference must be
E[v | N D] > v1 (since some weight is placed on either v2 or v3 , or both). Following
disclosure, the buyer’s inference must be v1 (either because v1 is disclosed with positive
probability, or by part (c)). So non-disclosure of v1 is strictly better (by claim #1).
Claim #3: In any sequential equilibrium, v3 is disclosed with probability 1. Suppose
not. By claim #2, v1 is not disclosed, so E[v | N D] < v3 . Suppose the seller learns
that quality is v3 . Following disclosure, the buyer’s inference is v3 (either because
v3 is disclosed with positive probability, or by part (c)); following non-disclosure, the
buyer’s inference is E[v | N D] < v3 . So disclosure is strictly better (by claim #1), a
contradiction.
Claim #4: In any sequential equilibrium, v2 is disclosed with probability 1. Suppose not.
Then, by claim #2, v1 is not disclosed, and, by claim #3, v3 is disclosed, which means
E[v | N D] < v2 . Suppose the seller learns that quality is v2 . Following disclosure, the
buyer’s inference is v2 (either because v2 is disclosed with positive probability, or by part
(c)); following non-disclosure, the buyer’s inference is E[v | N D] < v2 . So disclosure is
strictly better (by claim #1), a contradiction.
e) Claim #1 from part (d) still applies. Moreover, since v2 and v3 are not disclosed with
certainty (since S doesn’t necessarily observe them), we have a stronger version of claim
#2: v1 is definitely not disclosed in any sequential equilibrium. Claim #3 also remains
valid (in fact, one no longer even needs claim #2 to prove it). So the question is whether
the seller will disclose v2 . We will identify conditions for which there is an equilibrium in
which v2 is disclosed, and for which there is an equilibrium in which v2 is not disclosed.

94
 First, consider a sequential equilibrium in which v2 = 4 is disclosed. Then, applying
Bayes rule,
1
3 1
Pr [v1 | N D] = 1 1 =
3 + 2 3 (1 − λ)
1 + 2(1 − λ)
1
3 (1 − λ) (1 − λ)
Pr [v2 | N D] = Pr[v3 | N D] = 1 1 =
3 + 2 3 (1 − λ)
1 + 2(1 − λ)
and
2 + 4(1 − λ) + 9(1 − λ) 15 − 13λ
E[v | N D] = =
1 + 2(1 − λ) 3 − 2λ
From disclosing v2 = 4,S receives a payoff of 12 (4 − 1). From not disclosing v2 = 4,
S receives a payoff of 21 15−13λ
3−2λ − 1 . Accordingly, disclosure is optimal provided that
4 ≥ 15−13λ 3
3−2λ , which requires λ ≥ 5 .
Next consider a sequential equilbrium in which v2 = 4 is not disclosed. Then, applying
Bayes rule,
1
1
Pr [v1 | N D] = Pr[v2 | N D] = 1 13 =
2 3 + 3 (1 − λ) 3−λ
1
3 (1 − λ) 1−λ
Pr[v3 | N D] = =
1
2 3 + 13 (1 − λ) 3−λ
and
2 + 4 + 9(1 − λ) 15 − 9λ
E[v | N D] = =
3−λ 3−λ
1
 a payoff of 2 (4 − 1). From not disclosing v2 = 4, S
From disclosing v2 =4, S receives
receives a payoff of 21 15−9λ
3−λ − 1 . Accordingly, non-disclosure is optimal provided that
15−9λ
4≤ 3−λ , which requires λ ≤ 53 .
We conclude that S discloses v2 = 4 when λ ≥ 35 , and doesn’t disclose it when λ ≤ 35 .

107. (Sept 2007 Comp, Q1)

a) Yes: along an indifference curve, dw

de = c0i (e), which by assumption is greater for i = L
than for i = H.
b) eL = 0; from the non-imitation conditions, eH = c−1 −1
L (πH − πL ), and eH = cH (πH − πL ).
c) eP ∈ [0, e∗ ] where e∗ = c−1L [λ(πH − πL )] (from the type L non-imitation condition),
where λ is the fraction of type H workers.
d) Equilibrium dominance (the intuitive criterion) isolates the most efficient separating
equilibrium.
e) i. For the separating equilibria, the conditions change to

cL (eH ) + teH = πH − πL

and
cH (eH ) + teH = πH − πL
For the pooling equilibrium, the formula for e∗ changes to

cL (e∗ ) + te∗ = λ(πH − πL )

95
 ii. Using the equation that defines eH ,
deH eH
=−
dt t + c0L (eH )
Therefore,  0
c (e ) − c0L (eH )

dUH
= eH H 0H <0
dt cL (eH ) + t
Moreover, dUdt = 0.
L

f) i. None of the conditions change from part (e).

ii. The effect on utility is the same as the effect from part (e-ii), plus the effect of the
change in the lump-sum transfer. For the lump-sum transfer, we have a per capita
revenue (and hence per capita lump-sum payment) of R = λeH t. Therefore,
 0 
dR cL (eH )
= λeH >0
dt c0L (eH ) + t
Therefore, dU dR
dt = dt > 0. As for the type H workers, we add the lump-sum effect
L

to the effect derived in part (e-ii):

 0
cH (eH ) − (1 − λ)c0L (eH )

dUH
= eH
dt c0L (eH ) + t
For λ sufficiently close to unity, this is strictly positive. So if the fraction of type
H workers is sufficiently high, everyone will be better off with a higher tax.
108. (2006 Final, Q2)
a) Given a price p1 , only buyers with type v ≥ p1 would accept such an offer. The probabil-
ity of sale is 1−p1 , and therefore, the expected revenue, π(p1 ) = p1 (1−p1 ), is maximized
at p1 = 21 .
b) i. If p1 < p2 , then for all v, v − p2 < v − p1 , and therefore, the buyer would never
be willing to buy the good in period 2. Therefore, the probability of sale in the 1st
period is 1 − p1 , and the probability of sale in the 2nd period is 0.
−δp2
ii. Observe that for indifference to hold, ṽ − p1 = δ(ṽ − p2 ), or ṽ = p11−δ . Note that
ṽ ≥ p1 .
Assume ṽ ≤ 1. Given prices p1 and p2 , and valuation v, let U (v, p1 , p2 ) = (v − p1 ) −
δ(v − p2 ) = v(1 − δ) − p1 + δp2 represent the incremental utility from buying in
period 1 versus buying in period 2. Observe that U (v, p1 , p2 ) is strictly increasing
in v; since indifference holds for type ṽ, then a buyer with valuation v ∈ (ṽ, 1] would
strictly prefer to buy in period 1 over period 2. Moreover, since v > ṽ ≥ p1 , such a
buyer prefers buying in period 1 to not buying at all. By similar reasoning, a buyer
with valuation v ∈ (p2 , ṽ) strictly prefers buying in the 2nd period to buying in the
1st and to not buying at all.
If instead ṽ ≥ 1, by the same reasoning, all types v ∈ (p2 , 1) strictly prefer buying
in period 2 to buying in period 1 and to not buying at all.
iii. The expected revenue is

  − p1 )
 p1 (1    if p1 < p2
p1 −δp2 p1 −p2
R(p1 , p2 ) = p1 1 − 1−δ + δp2 1−δ if p2 ≤ p1 ≤ 1 − δ + δp2 (42)

δp2 (1 − p2 ) if 1 − δ + δp2 < p1


96
 h 
p1 +δ−1
iv. For any given p1 , observe that R(p1 , p1 ) = p1 (1−p1 ). Note that for p2 ∈ δ , p1 ,
d 2δ(p1 −p2 )
dp2 R(p1 , p2 ) = > 0 and therefore, R(p1 , p2 ) < R(p1 , p1 ). Thus, for any opti-
1−δ h   
mal solution, it must be that p2 6∈ p1 +δ−1
δ , p1 . Now consider any p 2 ∈ 0, p1 +δ−1
δ .
Observe that R(p1 , p2 ) < R(p2 , p2 ), and therefore, p2 does not lie in this range. This
leaves us with p2 ≥ p1 , and so R(p1 , p2 ) = p1 (1−p1 ). We know from earlier solutions
that the optimal p1 = 21 . The buyer therefore buys in the first period if v > p1 .
v. The payoffs are the same.
c) After the monopolist observes that no purchase was made at t = 1, she will deduce that
v is distributed U [0, 21 ]. Therefore, it would be optimal for her to deviate and set p2 = 41 .
d) We shall now proceed to find a sequential equilibrium of this game.
∗
i. p2 = v2 .
∗
∗
ii. It must be that v ∗ − p1 = δ v ∗ − v2 = δv2 , or p1 = v ∗ 2−δ


2 .
iii. We will assume that v ∗ ≤ 1, and will later verify this is true for the optimal v ∗ .

R(v ∗ ) = p1 (1 − v ∗ ) + δp2 (v ∗ − p2 )
   ∗ 2
∗ 2−δ ∗ v
= v (1 − v ) + δ
2 2
iv. The FOC with respect to v ∗ is:
δv ∗
 
2−δ
(1 − 2v ∗ ) + = 0 =⇒
2 2
2−δ
v∗ = <1
4 − 3δ
(2−δ)2
This leads to an expected revenue of 4(4−3δ) .
2
To see how this compares to 14 , note that (2−δ)
< 1 since δ < 1. Therefore, the
4−3δ
revenue here is strictly less than in parts (a) and (b)(v).
109. (2009 Final, Q2)
For the following answers, I will list player 1’s strategy as (x, y), where x is his choice given
that he is type 1 and y is his choice given that he is type 2. I will list player 2’s strategy
as (w, z), where w is his strategy given that player 1 chose L and z is his strategy given that
player 1 chose R. The belief of being at the top node of the information set reached after
player 1 plays L will be listed as γ. The belief of being at the top node of the information
set reached after player 1 plays R will be listed as α.

a) Consider an equilbrium in which player 1 plays (R,L). This implies that γ = 0 and
α = 1. Given these beliefs, player 2 must choose (u,u). Now, we check if player 1
wants to deviate. Type 1 prefers to play R because it yields 2 while playing L yields
1. Type 2 prefers to play L as it yields 2 while R yields 1 . Therefore, we have a PBE:
((R,L),(u,u),γ = 0, α = 1).
Consider an equilbrium in which player 1 plays (L,R). This implies that γ = 1 and
α = 0. Given these beliefs, player 2 must choose (u,d). Now, we check if player 1 wants
to deviate. Type 2 would deviate: he prefers to play L rather than R because he gets
2 instead of 1. Therefore, this cannot be a PBE.

97
 b) Consider an equilbrium in which both types pool on L. This implies that γ = .5. Given
these beliefs (or any beliefs), player 2 must choose (u,-). Also, player 2 must play d
when R has been played, because otherwise type 1 would want to deviate. This is
sequentially rational for player 2 given that 3α ≤ 2(1 − α) => α ≤ 25 . Note that we
are free to choose these beliefs however we like because this information set is off the
equilibrium path. Therefore, we have a set of PBE: ((L,L),(u,d),γ = .5, α ≤ 25 ).
Is this a SE? Consider ((L,L),(u,d),γ = .5, α = 14 ). In order to justify this as a SE, we
have to create a sequence of purely mixed behavior strategies for player 1 that approach
(L,L), such that the beliefs elicited by these strategies approach α = 14 . Consider the
sequence in which type 1 plays L with probability 1 − εn and R with probability εn
and type 2 plays L with probability 1 − 3εn and R with probability 3εn , with εn → 0
as n → ∞. These strategies clearly converge to (L,L). The beliefs elicited by these
strategies are αn = 3εnε+ε
n
n
= 14 which clearly converges to 14 . Therefore, we have a SE.
c) Note that there are no subgames here except the original game, so the set of NE will
equal the set of SPNE. Consider the strategy profile ((R,R),(d,u)). I claim this is a
NE and, therefore, a SPNE. Consider player 1. Given that player 2 is choosing (d,u),
(R,R) is a best response for player 1 as both of his types prefer R. For player 2, choosing
d given that player 1 chooses L does not affect his payoff because this move is off the
equilibrium path. For player 2, choosing u given that player 1 chooses R is the unique
best response because 21 (3) + 12 (0) > 21 (0) + 12 (2). Therefore, we have a subgame perfect
NE.
The key point of this problem is that this is NOT a PBE. Why? In order to justify
the equilibrium above, it is crucial that player 2 plays d given that player 1 plays L
(otherwise, if 2 played u, player 1’s type 2 would deviate to L). However, there are no
beliefs at the information set reached after L that can justify this move. Sequential
rationality given ANY beliefs here will cause player 2 to play u at this information set.
Therefore, this is not a PBE.
d) Consider a situation in which type 2 plays L and type 1 strictly mixes. It must be that
player 2 plays (u,u) regardless of level of mixing. However, then type 1 would prefer to
play R with probability 1 instead of a strict mixture. So, we do not have a PBE.
Consider a situation in which type 2 plays R and type 1 strictly mixes. Player 2 must
play (u,-), regardless of his belief γ. But then type 2 would do strictly better to deviate
to playing L. So, this is not a PBE.
e) Yes, consider the pooling equilibrium ((L,L),(u,d),γ = .5, α ≤ 25 ). Player 2 justifies
playing d after R by believing that type 2 is more likely to deviate to R. However, type
2’s equilibrium payoff from playing L is 2. The most he can make by deviating to R is 1.
Clearly, he has no incentive to deviate. However, type 1 does have incentive to deviate
given that the reciever chooses u. Therefore, the intuitive criterion requires that α = 1
when R is off the equilibrium path, eliminating this pooling equilibrium.
f) Not much.
For part (a), the beliefs will remain the same regardless of µ because the types separate
in equilibrium. Therefore, the previous answer will hold.
For part (b), changing µ will change the belief γ = .5 to γ = µ, but this will not change
the best response for player 2 given that he sees L. Therefore, the only change is that
γ = µ.

110. (June 2009 Comp, Q2)

98
a) The president makes a decision based on the advice of the Economist (he has to de-
cide whether to choose “E” or “N” when the economist says “G” or “B”). So, his
strategy set is {EE,NE,EN,NN}. If the Economist is knowledgeable, he must choose
what to say if he learns the program is good or bad. If the Economist is ignorant,
he must choose what to say (he never learns the state of the world). So, he has
three information sets with two choices for each – therefore, he has eight strategies:
{GGG,BGG,GBG,GGB,BBG,GBB,BGB,BBB}.
1
b) i. The president will see the message “g” with probability 16 (only when the state of
the world is good and the Economist is knowledgeable). He believes that the state
of the world is good with probability 1 given he sees g. Given this, he enacts the
program.
ii. The president will see the message “b” with probability 15 3 1 3
16 . There is a 4 · 4 = 16
chance that the Economist is ignorant and the program is good (leading to a message
3
of b). Therefore, there is a 15//1616
3
= 15 = 15 chance that the program is good given
that he receives message b. Therefore, he will not enact the program.
iii. If the outcome is 1 (the president enacts and succeeded), the president must have
seen a message of g. He must believe that the Economist is knowledgeable with
probability 1. If the outcome is 0 (the president does not enact), the president must
have seen a message of b. He must believe that the Economist is knowledgeable
3
with probability 15//16
16
3
= 15 = 51 . We never see an outcome of -1 on the equilibrium
path, so we are free to set any beliefs in [0,1].
iv. Consider the ignorant Economist. If he chooses to say b, he will get a payoff of 51 .
If he chooses to say g, the president will enact the plan. Therefore, he will get a
payoff of 1 with probability 14 (when the program is good) and he will get a payoff
of µ(−1) with probability 34 (when the program is bad). There is no µ(−1) ∈ [0, 1]
such that he will want to say b. Therefore, we cannot have an equilibrium.
c) i. The president will see the message g with probability 81 (only when the state of the
world is good and the Economist is knowledgeable). He believes that the state of
the world is good with probability 1 given that he sees g. Given this, he enacts the
program.
ii. The president will see the message b with probability 78 . There is a 12 · 14 = 81 chance
that the Economist is ignorant and the program is good (leading to a message of
1
b). Therefore, there is a 7//88 = 17 chance that the program is good. Therefore, he
will not enact the program.
iii. If the outcome is 1 (the president enacts and succeeds), the president must have
seen a message of g. He must believe that the Economist is knowledgeable with
probability 1. If the outcome is 0 (the president does not enact), the president must
have seen a message of b. He must believe that the Economist is knowledgeable
3
with probability 7//88 = 37 . We never see an outcome of -1, so we are free to set any
beliefs in [0,1].
iv. Let’s set beliefs given an outcome of -1 at 0. Consider the ignorant Economist. If
he chooses to say b, he will get a payoff of 37 . If he chooses to say g, he will get a
payoff of 1 with probability 14 (when the program is good) and he will get a payoff of
0 with probability 34 (when the program is bad). Therefore, he will want to say b.
It is easy to show that the knowledgable Economist will want to give good advice.
Therefore, we have an equilibrium.

99
 Interpretation: If there are very few good Economists and only the good ones make
radical suggestions, it becomes very tempting for the bad Economists to make the
same suggestions to appear good, even if it sometimes exposes them as being bad.
As the number of good Economists rises, the bad Economists are happy to be
perceived as a “near-average” Economists by making conservative suggestions.

111. (Sept 2009 Comp, Q2)

a) Player A has just one information set. Player B has 4 information sets, 2 corresponding
to the selfish disposition and 2 corresponding to the egalitarian disposition. Within each
disposition, one of the information sets corresponds to player 1 undertaking the project
and the other information set corresponds to player 1 not undertaking the project. The
strategy set for player A is SA = {U, N U }, where U denotes “undertake” and N U
denotes “not undertake.” The strategy set for player B is SB = {(tu1 , tu2 , tnu1 , tnu2 ) :
tu1 , tu2 , tnu1 , tnu2 ≥ 0}, where (tu1 , tu2 , tnu1 , tnu2 ) is a four-tuple of transfer, specifying
the transfer that the player makes in each of his four information sets.
In any subgame perfect equilibrium, the selfish Player B makes no transfers to Player
A. The egalitarian Player B transfers 60 to A in case the project has been undertaken
and 40 otherwise. Player A gets an expected payoff of (1 − π)60 from undertaking the
project and 10π +(1−π)50 from not undertaking it. He therefore undertakes the project
if and only if π ≤ 21 .
b) i. A strategy for Player A is a function f : R+ → {u, nu} were f (g) represents A’s
decision after receiving a gift of size g. Player A’s strategy set is the set of all such
functions.
A strategy for Player B is of the form (g, t), where:
g : {e, s} → R+ represents the gift that he offers A in case he is selfish (s) and in
case he is egalitarian (e);
t : {e, s} × {u, nu} × R+ → R+ , where for example t(e, nu, g) represents the transfer
that B makes to A in the case that the the project was not undertaken, B is
egalitarian and has made a gift of size g to A.
Player B’s strategy set is the set of all such pairs of functions (g, t). (Note: It is
also correct to split B into two types and specify a strategy for each type.)
ii. The egalitarian B is willing to make a gift of size g ≤ ge∗ = 70. The selfish B is
willing to make a gift of size g ≤ gs∗ = 30.
iii. Note that in any separating equilibrium, the selfish type cannot make any gift. If
such a separating equilibrium existed then upon revelation A would not undertake
the project and therefore the selfish B would rather deviate to making a gift of
0. Given part (ii), in any separating equilibrium we must have 30 ≤ g ∗ ≤ 70. If
g ∗ < 30 then the selfish type would find it profitable to deviate to making a gift,
and if g ∗ > 70 then the egalitarian type would find it profitable to deviate to not
making a gift. We resolve the indifference of the selfish type at g ∗ = 30 by having
him not make a gift and the indifference of the egalitarian type at g ∗ = 70 by having
him make a gift. It can be verified that A only finds it worthwhile to undertake
the project when g ∗ ≤ 50 (and finds it strictly worthwhile when g ∗ < 50). There
can be no separating equilibrium in which the project is not undertaken, since this
would require g ∗ > 50, and therefore a final payoff to the egaliatrian B of less than
40. So the egalitarian B would deviate to making no gift and getting a payoff of
40. To summarize, denoting the belief of Player A that Player B is egalitarian after

100
 receiving a gift of size g by µ(g), we have the follwing separating equilibria for any
g ∗ ∈ [30, 50]:

Player A’s strategy:

f (g ∗ ) = u
f (g) = nu ∀g < 30
f (g) ∈ {u, nu} ∀g ∈ [30, 50], g 6= g ∗
f (g) = n ∀g > 50

Player B’s strategy:

g(s) = 0
g(e) = g ∗
t(s, x, g) = 0 ∀g, x ∈ {u, nu}
t(e, u, g) = 60 − g ∀g ≤ 60
t(e, u, g) = 0 ∀g > 60
t(e, nu, g) = 40 − g ∀g ≤ 40
t(e, nu, g) = 0 ∀g > 40

Player A’s beliefs:

µ(0) = 0
µ(g ∗ ) = 1
µ(g) ≤ 21 ∀g ∈ (0, 30)
µ(g) ≤ 12 if f (g) = nu and g ∈ [30, 40], g 6= g ∗
µ(g) ≥ 12 if f (g) = u and g ∈ [30, 40], g 6= g ∗
10
µ(g) ≤ 60−g if f (g) = nu and g ∈ (40, 50], g 6= g ∗
10
µ(g) ≥ 60−g if f (g) = u and g ∈ (40, 50], g 6= g ∗
µ(g) ∈ [0, 1] ∀g > 50

iv. Since the project is undertaken in all separating equilibria when B reveals himself as
e, and the final payoffs are always the same, they are all equally efficient. Efficiency
in this context just depends on whether the project is undertaken or not.
v. In any pooling equilibrium in which the project is not undertaken, gifts have to be
0. Such an equilibrium exists if and only if π ≥ 12 , as in part (a). If π ≤ 12 , we
can have pooling equilibria in which both types make a gift of size g ∗ ∈ [0, 30]. On
the equilibrium path, Player A believes that Player B is egalitarian with probability
µ(g ∗ ) = π. Off the equilibrium path, we must have µ(g) ≤ 21 for all g < g ∗ so that
after seeing a gift smaller than g ∗ , Player A chooses not to undertake the project in
a sequentially rational fashion, and the selfish Player B has no incentive to deviate
to a lower gift. Beliefs µ(g) for g > g ∗ can be anything.

112. (2010 Final, Q2)

a) First consider T = +∞. Minmax payoffs are -5. The set of feasible individually rational
payoffs is given by the following constraints: min{u1 , u2 } ≥ −5, max{u1 + 3u2 , 3u1 +
u2 } ≤ 20. That is the sustainable payoff set. For T finite, the static equilibrium is
unique, so the only subgame perfect outcome involves repetitions of (D, D), leading to
an average payoff vector (-5,-5).

101
b) i. D is a dominant strategy for type A and C is a dominant strategy for type B, so
that is what they must play in any Bayesian equilibrum.
ii. Because all types have dominant choices in the second (last) round, those are what
they must play regardless of beliefs. Thus, in the first round the players know that
their current choices do not affect future outcomes, so they can examine their first
round incentives in isolation. Both types have dominant choices in the first round,
so the only possible solution is: type A always plays D while type B always plays
C.
iii. A simple induction argument (based on the answer in part (ii)) implies that type A
always plays D while type B always plays C.
c) i. Obviously type A must play its dominant strategy, D. Consider player i of type
B. Regardless of what player j chooses, the difference between player i of type B’s
payoff from choosing C and his payoff from choosing D is -5 if j is type A, and +5 if
j is type B. Therefore, C is a dominant strategy for player i of type B if µ < 12 , D
is a dominant strategy for player i of type B if µ > 21 , and both choices are equally
good (regardless of what j does) when µ = 21 . It follows that condition Y is µ ≤ 12
and condition Z is µ ≥ 12 . There are no other (pure strategy) Bayesian equilibria
except when µ = 12 , in which case there are also equilibria in which player i of type
B chooses C while player j of type B chooses D.
ii. A. For type A, the payoff is 5 + µ(−5) + (1 − µ)10 = 15 − 15µ. For type B, the
payoff is 5 + µ(−10) + (1 − µ)5 = 10 − 15µ.
B. Let πi (ai , aj ) denote the probability i assigns to j being type A when the first
period choices are (ai , aj ). We will choose πi (ai , C) = µ and πi (ai , D) = 1.
These strategies always prescribe D for player i of type A in period 2, which
is obviously justified (for any beliefs about j’s type) because D is a dominant
choice in period 2 for that type. The strategies prescribe C for player i of type
B in period 2 when aj = C in the the first period, which is plainly optimal
in light of part (c)(i), given than condition Y holds and πi (ai , C) = µ. The
strategies prescribe D for player i of type B in period 2 for all other first period
outcomes (aj = D), which is plainly optimal given that πi (ai , D) = 1, since D is
then a dominant choice for that player type. Note finally that Bayes rule is only
applicable on the equilibria path. In this case, first period choices are (C, C) and
beliefs are derivable from the strategy profile – the posterior probability that j
is type A is indeed equal to the prior, µ.
C. Whether or not a type deviates in the first period, she will clearly be better
off playing D in the second period. Thus the best deviation must be to D in
the first period followed by D in the second. Given the strategies, the resulting
payoff is 10 − 5 = 5.
D. Because condition Y is satisfied, type B’s most attractive two-period deviation
is to D in the first period followed by C in the second period. (Technically, C
in the first period followed by D in the second period is more profitable, but
condition Y already rules out this possibility.) Given j’s strategy, the resulting
payoff type B is µ[10 − 10] + (1 − µ)[0 + 0] = 0.
E. For type A, we need 15 − 15µ ≥ 5, or µ ≤ 32 . For type B, we need 10 − 15µ ≥ 0,
or µ ≤ 23 . Both of these requirements are weaker than condition Y . Therefore,
we have a weak perfect Bayesian equilibrium iff µ ≤ 21 .
F. This is a sequential equilibrium. Let εN j (k) denotes a sequence of probabilities

102
 (indexed by k) with which player j of type N chooses D in the first round. By
constructing these perturbations such that εB A
j (k)/εj (k) goes to zero as k (and
the εNj (k)’s) go to zero, we assure that the limiting posterior upon observing
j choose D places all weight on type A for player j. On the other hand, since
1−εA B
j (k) and 1−εj (k) both converge to unity, the limiting posterior probability
that j is type A upon observing j choose C is µ. These are the beliefs that πi
assigns.
iii. A. For type A, the payoff is 5(T − 1) + µ(−5) + (1 − µ)10 = 5T + 5 − 15µ. For
type B, the payoff is 5(T − 1) + µ(−10) + (1 − µ)5 = 5T − 15µ.
B. For any such history ht , let πi (ht ) = 1. In that case, choosing D is a dominant
choice for both types in the current round, and according to the strategies fu-
ture outcomes do not depend on current choices, so the prescribed choices are
justified.
C. No information is revealed (because both types have made the same choices),
so the belief must equal the posterior: πi (h∗t ) = µ, where h∗t is the history for
which no one has played D.
D. If A deviates, her opponent will always play D, so the best deviation for A is
to choose D in some period t0 with T > t0 ≥ t and all subsequent periods. Her
resulting payoff will be 5(t0 − 1) + 10 + (T − t0 )(−5) = 10t0 − 5T + 5. Notice that
this in increasing in t0 ; therefore, her most attractive deviation (among those
considered here) is to switch to D in round T − 1, in which case the resulting
payoff is 5T − 5.
E. Once B deviates to D, her opponent will always play D, but given that condition
Y holds, that πi (h∗t ) = µ, and that no further information will be revealed
concerning her opponent’s type, it will be in her interests subsequently to always
play C. Therefore, the best possible deviation by B in period t0 with T > t0 ≥ t
yields a payoff of 5(t0 − 1) + µ[10 − 10(T − t0 )] + (1 − µ)0. Notice that this is
increasing in t0 ; therefore, the most attractive deviation among those considered
here is in t0 = T − 1. The associated payoff is 5T − 10.
F. For type A, we need 5T + 5 − 15µ ≥ 5T − 5, or µ ≤ 23 . For type B, we
need 5T − 15µ ≥ 5T − 10, or µ ≤ 23 . Both of these requirements are weaker
than condition Y . Also, condition Y directly rules out profitable deviations in
round T on the equilibrium path. Therefore, we have a weak perfect Bayesian
equilibrium iff µ ≤ 12 .
G. Here we have assumed that πi (ht ) = 1 if i alone has deviated. To see that this
is inconsistent with the strategies, focus on T = 2 and consider the argument
in part 3(b)(vi). When i alone has deviated, the posterior belief about j’ type
derived from any strictly mixed strategy close to the equilibrium strategy must
be that j is type A with probability close to µ. In that case, type A must choose
C in the second period (given that condition Y holds), which is contrary to the
strategies used here (but consistent with what we did in part (b)).

113. (Comp June 2011, Q6)

114. (Comp September 2011, Q4)

115. (2012 Final, Q2)

116. (2013 June Comp Q4)

103
117. (2013 Sept Comp Q4)

104