Microeconomics II - problem set 2

M.M

December 2022

Exercise 1

Player 2
a b c
A 1, 1 7, 0 0, 0
Player 1 B 0, 7 5, 5 0, 0
C 0, 0 0, 0 4, 4

It is possible to define two different pure Nash equilibria strategies in the stage game above (thus,
considering the game as if it was played just once). The best responses for each player are high-
lighted and the Nashes are (A,a) and (C,c).
The exercise requires proving the existence of a SPNE beginning with an action profile (B,b).
I thus consider the following strategies:
1) if both players play (B,b) in the first period, then (C,c) is played in the second period.
2) if instead (B,b) is not played in period one, then (A,a) is played in period 2.
This basically means that one player can use the ”bad Nash” to punish (credibly) the other player
if she deviates from (B,b) in the first stage. It is the existence of this ”bad Nash” with which to
threaten the opponent, that makes possible the existence of a SPNE beginning with (B,b).
This is a SPNE equilibrium (regardless of the fact that the action of the first period is not a Nash
considering a single-stage game). Indeed, let me analyze the incentives to deviate. Playing (B,b)
and (C,c) leads to a cumulative payoff for both players of 9. This is higher than playing (C,c)
two times (cumulative payoff of 8) and also higher than the payoff that a player who deviates will
obtain after the punishment (indeed, in that case, the deviating player will get a cumulative gain
of 7). Thus, no player has the incentive to deviate, and the above-described strategy combination
is SPNE.

1
Exercise 2

Answer 1

Player 2
a b c
A 1, 1 7, 0 0, 2
Player 1 B 0, 7 5, 5 0, 0
C 2, 0 0, 0 4, 4

This game is particularly similar to the one in exercise 1, but with the important difference of
having just one Nash equilibrium in pure strategies. Namely: (C,c). Differently from exercise 1,
there does not exist here a ”bad” Nash that can be used as a credible threat to incentivize players
not to deviate and coordinate, for instance, to reach (B,b) at the first stage. Let me go through an
example to observe why (B,b) cannot be an action profile played in a SPNE.
Say that player 1 plays B in the first stage, and player 2 plays b. Then, they play (B,b) in the
second stage and (C,c) in the third. The total payoff is 14 for both players. Let me consider
the case where player 2 deviates and plays a in the first stage, then she gets a payoff of 7. After
that, both players will play pure Nash (thus C,c). As a consequence, player 2 at the end gets a
total cumulative payoff of 15, which is higher than 14. So, action (B,b) cannot be the first action
of a SPNE since there exists an incentive to deviate for players that cannot be curbed by a bad Nash.

Answer 2
This game has a unique SPNE, which is the repetition of the Nash of the stage game 3 times. This
follows from the theorem stating that if the stage game G has a unique Nash equilibrium, then
the finitely repeated game G(T) (G(3) in this case) with T < ∞ has a unique SPNE, which is the
repetition of the Nash of the stage game T times. Indeed, upon deviation, there exists NO ”bad”
Nash that can be used to punish the deviator, as instead happened in exercise 1. Thus, the only
subgame perfect Nash equilibrium of the game is reached if both players play (C,c) in every period
from 1 to 3.

2
Exercise 3

Answer 1.

Player 2
C D
Player 1 C 6, 6 0, 8
D 0, 2 4, 0

There is no pure strategy Nash equilibrium, this can be observed considering the best responses of
both players. I thus now aim at finding the mixed strategy Nash equilibria of the game. Let me
define p as the probability that player 1 plays C and (1 - p) the probability that player 1 plays D.
Similarly, q the probability that player 2 plays C and (1 - q) the probability that player 2 plays D.
It follows that:
E1 (u|C) = 6q + 0(1 − q) = 6q
E1 (u|D) = 0q + 4(1 − q) = 4 − 4q
4
q = 10 = 25
And analogously,
E2 (u|C) = 6p + 2(1 − p) = 4p + 2
E2 (u|D) = 8p + 0(1 − p) = 8p
p = 42 = 12
Thus, the mixed strategy Nash equilibrium involves player 1 playing C with probability p = 12 and
D with the same probability; while player 2 plays C with q = 25 and D with probability equal to 35 .
More formally, the mixed Nash is: {( 21 C, 12 D)( 52 C, 35 D)}

Answer 2.
Analogously with respect to the previous exercise, this game has one unique Nash equilibrium (with
the only difference that in this case, the equilibrium is in mixed strategy). Following the theorem
mentioned in exercise 2, I can thus conclude that the SPNE is such that each player plays her Nash
mixed equilibrium at each stage game. This leads to a SPNE since each of the 15 times the game
is played, the Nash equilibrium is played as well. Finally, it is important to recall that the above
reasoning is valid as long as there exists one unique Nash equilibrium and the number of times the
game is played is finite.

Answer 3.
The game is now played an infinite amount of times. In order to verify whether it is possible for a
SPNE to feature (C,c) played at each round I need to compare the payoffs for both players while
sticking to collaboration, and the one obtained following a different strategy. In particular, I will

3
consider a grim-trigger strategy that, after one deviation by one of the two players, leads to the
Nash equilibrium played forever on (in this case Nash in mixed strategy).
Notice that player 1 does not have any incentive to deviate from the cooperation indeed, upon
deviation, she will get a lower payoff. Thus, I will just consider the threshold delta over which
player 2 will prefer to deviate from (C,C) rather than sticking to it.
The normalized payoff of player 2, if she does not deviate, is:

P∞ 6
(1 − δ) t=1 δ t−1 6 = (1 − δ) (1−δ) =6

The normalized payoff of player 2, if she does deviate, is:

P∞
(1 − δ)[8 + t=2 δ t−1 X] = (1 − δ)8 + 4δ = 8 − 4δ with X = 4

Where X is the payoff after the deviation, where the two players will play Nash. This means
that both players will play the mixed Nash equilibrium. In particular player 2 will get the following
expected payoff (X above).

2 1
5 (6 2 + 2 21 ) + 35 (8 12 + 0 12 ) = 20
5 =4=X

I thus impose 8 − 4δ > 6. If this is true player two will cheat. Solving the equation, the re-
sult is δ < 12 . So, if the delta is higher than 0.5, player two is not cheating but cooperating with
player 1. This is the case in the example where δ = 0.7. So, by adopting the grim-trigger strategy,
there exists a sub-game perfect Nash equilibrium in pure strategy where (C,C) is played in every
round.

Exercise 4.

Answer 1.
The normal form of the game features the following elements.
Players: I = {F, B}
Player F strategies are {A, T }
Player B strategies are {A, T }
The payoffs for the related strategies are shown in the matrix below:

4
 Player B
A T
v v
Player F A 2 − c 1 , 2 − c2 v, 0
v v
T 0, v 2, 2

Answer 2.
The pure Nash equilibria in the game depend on the different values that v (which is strictly pos-
itive) and c (strictly positive as well) take. In particular, I can distinguish 9 different cases listed
below with their related Nash equilibria.

v v
Case 1: 2 − c1 > 0, 2 − c2 > 0

Player B
A T
v
Player F A 2 − c 1 , v2 − c2 v, 0
v v
T 0, v 2, 2

Pure Nash equilibrium: (A,A)

Player B
A T
v v
Player F A 2 − c 1 , 2 − c2 v, 0
v v
T 0, v 2, 2

Pure Nash equilibrium: (A,T)

v v
Case 3: 2 − c1 > 0, 2 − c2 = 0

Player B
A T
v
Player F A 2 − c 1 , v2 − c2 v, 0
v v
T 0, v 2, 2

Pure Nash equilibrium: (A,A), (A,T)

5
 Player B
A T
v v
Player F A 2 − c 1 , 2 − c2 v, 0
v v
T 0, v 2, 2

Pure Nash equilibrium: (A,T), (T,A)

Player B
A T
v v
Player F A 2 − c 1 , 2 − c2 v, 0
v v
T 0, v 2, 2

Pure Nash equilibrium: (T,A)

Player B
A T
v
Player F A 2 − c 1 , v2 − c2 v, 0
v v
T 0, v 2, 2

Pure Nash equilibrium (A,T), (T,A):

Player B
A T
v
Player F A 2 − c 1 , v2 − c2 v, 0
v v
T 0, v 2, 2

Pure Nash equilibrium: (A,A), (A,T), (T,A)

Player B
A T
v
Player F A 2 − c 1 , v2 − c2 v, 0
v v
T 0, v 2, 2

Pure Nash equilibrium: (A,A), (T,A)

6
 v v
Case 9: 2 − c1 = 0, 2 − c2 < 0

Player B
A T
v v
Player F A 2 − c 1 , 2 − c2 v, 0
v v
T 0, v 2, 2

Pure Nash equilibrium: (A,T), (T,A)

Answer 3.
By imposing v2 − c1 > 0 and v2 − c2 > 0 the game becomes a dominant solvable one. Indeed, for
both players, the dominant strategy is A, which is also the strategy (as highlighted in case 1 above)
that leads to the unique Nash of the game (A,A). Indeed, for both players, A is the best strategy
among the two available regardless of what the other player does. However, this Nash equilibrium
is not efficient, since the payoff at the Nash for both players is smaller than what they could have
reached if they had been able to cooperate over the strategy (T,T).

Answer 4.
I need to check whether strategy (T,T) can represent an equilibrium of an infinitely repeated game.
In order to do that, I will compute the δ over which it is convenient to cooperate over (T,T).
Firstly, I compute the normalized payoff for the player when it is cooperating with the other, reach-
ing the payoff of v2 . This is equal to:

P∞ v
v
πcooperation = (1 − δ) t=1 δ t = (1 − δ) (1−δ)
2
= 2

Then, I consider the normalized payoff of a player that cheats. That is equal to v when it cheats,
and the Nash equilibrium payoff from then on, which is equal to v2 − ci where i depends on the
player we consider. The normalized payoff for the cheater is thus:

P∞
πcheater = (1 − δ)[v + t=2 δ t−1 ( v2 − ci )] = (1 − δ)v + δ( v2 − ci )

I now compare the delta at which deviating (thus, cheating) become convenient for the player.
This means imposing the payoff of the cheater higher than the payoff of the collaboration.

(1 − δ)v + δ( v2 − ci ) > v
2

v
δ< v+2ci

7
So, if we consider a grim trigger strategy when the player can collaborate forever, but if it cheats
it will be punished with the Nash equilibrium forever from there on, the strategy (T,T) implying
v
collaboration is sustainable only if δ > v+2ci
. However, since c1 and c2 may be different, the most
correct formulation needs to be specified as below, since the more demanding condition is deter-
mined by the lower ci . Thus, to make the (T,T) a Nash equilibrium that is played at every round,
the following condition must hold:

v
δ >
v+2c1 if c1 ≤ c2
δ > v
v+2c2 if c2 < c1

Exercise 5

Answer.
I will consider two competitive settings for the two firms competing on the market: Cournot compe-
tition (simultaneously setting quantities) and Bertrand competition (simultaneously setting prices).
I will proceed by computing the profits for firms that collude setting a specific price or quantity,
and compare them with the profits obtained cheating (thus, not colluding).

COURNOT COMPETITION
First I consider the cases when the two firms collude which implyes that they will produce the
monopoly quantity and will split in half quantity produced and profits (the two firms have the
exact same cost function). The monopoly quantity can be obtained equalizing marginal costs to
marginal revenues.

α − 2βqM = c
which leads to
qM = α−c2β that is the total monopoly quantity
qc = q2M = α−c
4β is instead the collusion quantity for each firm

Now I compute the profits for each firm if they decide to both collude forever producing the
monopoly quantity and splitting monopoly profits. The normalized payoff is:

P∞
(1 − δ) t=0 δ π2M = πM
2

πM
where 2 is equal to

8
 (α−c)2
πcollusion = {[α − β( α−c α−c α−c 1
2β )]( 2β ) − c( 2β )} 2 = 8β

This is the profit of one firm obtained when both firms collude.

Now I switch to the case where one firm decide to cheat. In such case, the cheater will obtain
higher profits for one period, then it will be punished by the other firm which will play Nash equi-
librium (here I consider a grim-trigger strategy, where the two firms will play Nash forever after
the cheat by one firm).
I first compute the ”optimal cheating quantity” that can be obtained as a best response for the
firms that wants to cheat given the collusion quantity produced by the firm that is not cheating.
The firms’1 best response is (the two firms are symmetric, thus firm 2 will behave analogously)

πi = [α − β(q1 + q2 )]q1 − cq1

α−q2 β−c
the best response for the firms are q1 = 2β

So if one firm cheats it will produce the ”best response quantity” given a production by the other
firm of q2M so that

3(α−c)
qc = (α − β α−c 1
4β − c) 2β = 8β

which is the ”optimal cheating quantity”, the ones that maximizes the profits for the cheating
firm, given what the other firm has done.
Now I compute the profits when the firm is cheating which are

9(α−c)2
πcheat = [α − β( 3(α−c)
8β + α−c 3(α−c)
4β ] 8β − 3(α−c)
8β = 64β

Finally, I compute the Nash equilibrium quantity that is produced by both players if one of them
colludes. That is found by simultaneously solving the system of the two (symmetric) best responses
of both players. The result is:

α−c
qN ash = 3β

The related profits for each firm are:

(α−c)2
πpunishment = [α − β(2 α−c α−c α−c
3β ] 3β − c 3β = 9β

9
So, I can now compute the normalized profits for a firm that cheats and is then punished with
the Nash equilibrium (using πcheat and πpunishment ) and compare it with the profits of colluding
(using πcolluding ).

(α−c)2 2 P∞ 2

8β > (1 − δ)[ 9(α−c)

9
solving for δ I get δ > 17 which represents the value over which colluding is more convenient
than cheating.

BERTRAND COMPETITION:
Bertrand competition requires less burdensome computations. I now compute the normalized profit
when each firm colludes:

P∞
(1 − δ) t=0 δ t π2M = (1 − δ) = πM
2

Then I compute the normalized profit of one firm that cheats. In bertrand competition the firm
that cheats will take the monopoly profits when cheating, getting 0 profits from the period after
the cheating one (if I consider again a grim-trigger strategy). Indeed, in Bertrand competition, the
Nash equilibrium leads to prices equal to marginal cost that means that both firms have 0 profits.

P∞
(1 − δ)[πM + t=1 δ t 0] = (1 − δ)πM = (1 − δ)πM

The firms collude (setting monopoly prices and equally splitting profits) if the normalized prof-
its of colluding are higher than the normalized profits of cheating. This depends again on the value
of the delta.

πM
2 > (1 − δ)πM

1
So firms collude if δ > 2 since cooperating would be more convenient than cheating above the
threshold level of 12 .

Conclusion
9
If I compare the two values of δ found in the two different cases, I can see that 17 > 12 which means
that it is easier to collaborate (collude) in the Bertrand case rather than in the Cournot case.

10