Game Theory (Due: March 27th , before 7pm)

Take Home Exam

Instructor: Guillem Roig Name: Marı́a Sofı́a Casabianca González, Student Number: 216376

Problem 1. Drinking in Cartagena (20 Points points)

Consider a relationship between a bartender and a customer. The bartender serves aguardiente to the cus-
tomer and chooses x ∈ [0, 1], which is the proportion of aguardiente in the drink served, while 1 − x is the
proportion of water. The cost of supplying such a drink is cx, where c > 0. The customer, without knowing x,
decides on whether or not to buy the drink at the market price p. If he buys the drink his payoff is vx − p, and
the bartender’s payoff is p − cx. Assume that v > c and all payoffs are common knowledge. If the customer
does not buy the drink he gets 0 and the bartender gets −cx. Because the customer has some experience, once
the drink is bought and he tastes it, he learns the value of x, but this is only after he pays for the drink.

(a) (5 points) Find all the Sub-game perfect Nash equilibrium of this game.
In this game, we have two players, the bartender (B) and the customer (C), and considering that the customer
doesn’t know the amount of x poured into the drink, there is a information set in the moment the customer
decides either to buy or not the drink served for him, making the game a simultaneous. The bartender will
have a dominant strategy in choosing always x = 0, because this action will maximize his payoff no matter
what the customer chooses. If we evaluate the case where the customer will always buy the drink he is serve,
the bartender will choose a x = 0 maximizing his profits (in this case, p), and the customer will have incentives
to deviate and not buy the drink. For instance, if the customer will never buy the drink, the bartender will
choose, as well, a x = 0, because any deviation will cause a negative payoff.
The customer’s best response will depend on the value of x, following this distribution:
(
Not buy if x < vp
BC (x) =
Buy if x ≥ vp

Given this, the best responses for both players will look like this:

If the game is ever played, then the bartender will always choose a x of zero and under this analysis, the
only Nash Equilibrium will be the one in which the bartender plays his dominant strategy and the customer
never buys the drink.

(b) (5 points) Now assume that the customer is visiting the town of Cartagena for 10 days, and this “bar game”
will be played on each of the 10 evenings that the customer is in Cartagena. Assume that each player tries to
maximize the (nondiscounted) sum of his stage payoffs. Find all the subgame-perfect equilibria of this game.
For the “bar game” when T = 10, the game still has a unique Nash equilibrium, then I’ll will argue that
there will be no other equilibrium than the SPNE described in the previous subproblem. If the players decide

1
Marı́a Sofı́a Casabianca González – Take Home Exam 2

to play the strategies described above (subproblem A) in everyone of the 10 repetitions of the game, then they
will end up with a payoff of zero each, as was describe in the previous subproblem and this is the only SPNE
for this game.

(c) (10 points) Now assume that the customer is a local, and the players perceive the game as repeated infinitely
many times. Assume that each player tries to maximize the discounted sum of his stage payoffs, where the
discount rate is δ ∈ (0, 1). What is the range of prices p (expressed in the parameters of the problem) for which
there exists a subgame-perfect equilibrium in which every day the bartender chooses x = 1 and the customer
buys at the price p? For which values of δ (expressed in the parameters of the problem) can the price range
that you found exist?

Problem 2. Graduating from Rosario (30 points points)

A recent graduate from Rosario (player 1) has an idea for two start-up ventures, a small one and a large
one. He can trust (T) his ideas with his roommate (player 2), who is the only one who can implement these
ideas, or he can not trust (N) player 2, forget his dreams, and get a boring job, resulting in the status quo payoffs
of (0, 0) for players 1 and 2, respectively. There are two types of player 2: an honest type that always honors
trust (H) and a normal type that can choose to honor trust (H) or abuse trust (A). It is common knowledge
that a quarter of the people in the world are honest, but only player 2 knows whether he is honest or not.
The sequence of play is as follows: First, player 1 can choose T or N with respect to sharing his small idea.
A choice of N (no trust) ends the game and the payoffs are (0, 0). Following T player 2 can honor the trust,
resulting in payoffs of (1, 1), or abuse the trust and expropriate the idea, resulting in payoffs of (−1, 2). (Recall
that an honest player 2 cannot abuse trust.) If trust was offered in the first stage then first-stage payoffs are
determined (i.e., the outcome is known to both players), and player 1 can then choose whether or not to trust
player 2 with his big idea. (If trust was not offered in the first stage then the game ended.) No trust in the
second stage will result in no payoffs in addition to those obtained in the first stage. Following trust in the
second stage, if player 2 honors trust then the additional payoffs will be (7, 7), while if he abuses trust and
expropriates the idea the additional payoffs will be (−5, 11). There is no discounting between these stages.

(a) (5 points) Draw the whole game tree and write down the pure strategies of each player for the whole game.
How many pure strategies does each player have?

Player 1 {TTTTT, TTTTN, TTTNT, TTTNN, TTNTT, TTNTN, TTNNT, TTNNN, TNTTT, TNTTN,
TNTNT, TNTNN, TNNTT, TNNTN, TNNNT,TNNNN, NTTTT, NTTTN, NTTNT, NTTNN, NT-
NTT, NTNTN, NTNNT, NTNNN, NNTTT, NNTTN, NNTNT, NNTNN, NNNTT, NNNTN, NNNNT,
NNNNN}
Player 2 {}

(b) (5 points) Using sequential rationality argue that following an abuse of trust in the first stage, there will
be no trust in the second stage, effectively ending the game. Write down the reduced game tree that is implied
by this argument, and the resulting pure strategies for each of the two players.
Marı́a Sofı́a Casabianca González – Take Home Exam 3

(c) (5 points) Can there be a pooling perfect Bayesian equilibrium in which the normal type in the first period
honors trust for sure? Can there be a separating perfect Bayesian equilibrium in which the normal type in the
first period abuses trust for sure?

(d) (15 points) Find the perfect Bayesian equilibrium of this game. Is there a unique one (Note that this could
be semi-separating or semi-pooling)?

Problem 3. A University Choice (20 points points)

A prospective student, player 1, is deciding whether to go to Barcelona University (B) or Sabadell Univer-
sity (S). Both are universities that accept only the very best students. The difference is that Sabadell offers
a more holistic and comfortable lifestyle, while Barcelona requires harder work and imposes more “studying”
costs on its students. For this reason a player who goes to Barcelona learns to become more self sufficient and
becomes more productive in the workforce. The cost of learning and level of final productivity depend on the
type of player 1, who can be either excellent (E) or just really good (G). Player 1 knows his type, but all others
in society know only that a proportion p of young adults are type E. The cost of learning and the level of
productivity from each choice are given as follows:

Once player 1 finishes university, he is hired by a firm (player 2), which can place him in one of two jobs:
low-tech (L) or high-tech (H). The wage for the L job is wL = 2 and that for the H job is wH = 6. The payoffs
to player 1 are the wages less the cost of education. The firm’s profits depend on both the job assignment
and the type of employee. If the employee is assigned to an H job, the net profits to the firm are equal to the
productivity of the employed player 1 less the wage he is paid. If the assignment is to an L job, the net profits
are half the productivity of the employed player 1 less the wage he is paid.

(a) (5 points) Draw this game in extensive form.

(b) (15 points) Assume that p = 1=2, find all the pure-strategy Bayesian Nash equilibria (Do not consider
semi-separating or semi-pooling equilibria).

Problem 4. Traffic Game (15 points points)

Marı́a Sofı́a Casabianca González – Take Home Exam 4

Two cars are coming together into an intersection. A driver has to decide whether it lets the other person
go first or it either goes trough. If one goes and the other waits, the one that goes gets a utility of 1, and the one
that waits gets 0. If they both wait, each one gets a utility of -1. If they both go, the worst happens because
they crash into each other giving them a utility of -10.

(a) (5 points) Find all the Nash equilibria of the game and draw the Convex Hull.
- Normal Form Game in Matrix Notation:
1/2 G W
G -10, -10 1, 0
W 0, 1 -1, -1

This game has two Nash Equilibrium in pure strategies. N E = {(G, W ), (W, G)}
- Mixed Strategies Equilibrium
Players will play Go with a associated probability p and will Wait with probability (1 − p)

1/2 pG (1 − p) W
qG -10, -10 1, 0
(1 − q) W 0, 1 -1, -1

Player 1 will be indifferent between playing G and W when the expected utility is equal between the
strategies, given:
p(−10) + (1 − p)(1) = p(0) + (1 − p)(−1)
12p = 2
1
p=
6
Since the game is symmetric we have the same p for player 2. Then, the Nash Equilibrium in Mixed
Strategies would be: N EM S = {(p, q)k( 16 , 16 )}. Given this, the expected utility by playing the mixed
strategy, for each player, would be:
   
1 1 5 5 1 5 5
E{Ui } = (−10) + (1) + (0) + (−1) = −
6 6 6 6 6 6 6
The expected utility of the mixed strategy is inefficient for clear reasons.
- Convex Hull
Marı́a Sofı́a Casabianca González – Take Home Exam 5

(b) (10 points) Find a Correlated equilibrium in which both players get strictly larger payoffs than the mixed
strategy equilibrium. What type of mechanism are you thinking about? (A reasonable mechanism will have
extra points)
Mechanism proposed: (Public Mechanism)
The government of the city in which the intersection is located wants to help their citizens, and to do so
generates a new traffic light that chooses which of the two cars can go and which has to wait. It will be located
where both cars can see it, and will randomize with equal probability which car can move forward. If the light
shows a 1, player 1 crosses first, while player 2 waits (Nash Equilibrium in pure strategies (G, W )), if it shows
a 2, then player 2 can cross the intersection first and player 1 has to wait (Nash Equilibrium in pure strategies
(W, G)).
The mechanism proposed will be an equilibrium if no player has incentives to deviate, by communicating in
this way players obtain larger utilities than the mixed strategy equilibrium.
   
1 1 1 1 5 5
× (1, 0) + × (0, 1) = , > − ,−
2 2 2 2 6 6

Then, the Convex Hull would be:

Problem 5. Multiple Choice (15 points points)

Which one of the following options is correct? (Please give justification, wrong answers will give negative
points).

(a) (3 points) Consider the following incomplete information game where µ is the believe that players are in
the decision node x3 and (1 − µ) in the decision node x4 .
Marı́a Sofı́a Casabianca González – Take Home Exam 6

a) The assessment {(A, D, F ), µ = 1} is a weakly Perfect Bayesian Equilibrium.

b) The assessment {(B, D, F ), µ < 1} is a weakly Perfect Bayesian Equilibrium.
c) The assessment {(A, D, F ), µ = 12 } is a weakly Perfect Bayesian Equilibrium.
d) None of the previous is true.

Answer: d
A profile of assessments, one for each player, will be called an equilibrium if it satisfies a reasonable consis-
tency requirement. The weakest consistency requirement usually discussed imposes Sequential Rationality and
Bayesian updating. The connection between beliefs and optimal action for Player 3 is illustrated as follows:

Then, it is optimal for Player 3 to play F if µ < 13 , to play either F or E if µ = 13 and to play E if µ > 13 .
With this in mind, if we evaluate the assessments proposed in this exercise we have that:
Assessment {(A, D, F ), µ = 1} is an inconsistent belief because if µ > 31 , optimality requires Player 3 to
choose E instead of F due to sequential rationality.
Assessment {(B, D, F ), µ < 1} is an inconsistent belief because, even if the information set nodes are
off the equilibrium path and consistency with Bayesian updating itself places no restriction whatever on
beliefs. Optimality, though, it still restricts beliefs off the equilibrium path to those for which the actions
taken are optimal. Which means that for player 3 to choose to play F, he needs a µ that is always greater
than 31 , and this is not necessarily the case for this assesment.
Assessment {(A, D, F ), µ = 12 } is an inconsistent belief because if µ > 13 , optimality requires Player 3 to
choose E instead of F due to sequential rationality.
Due to the previous explanation, the only possible correct answer is d.

(b) (3 points) A stage game that has multiplicity of equilibrium and is played a finite number of times:

a) The only sub-game perfect equilibrium is one where a Nash equilibrium is played in every stage.
b) There are strategies that are sub-game perfect that sustain no Nash equilibrium for some stages of the
game.
c) There are strategies that are sub-game perfect strategies that sustain no Nash equilibrium in every stage
game.
d) There are strategies that are sub-game perfect that sustain no Nash equilibrium in the last stage.

Answer: b
I am going to argue that option b is the only one that could meet the game that was proposed, explaining
why the other options cannot meet the conditions imposed in the statement.

(c) (3 points) The min-max payoffs for any player i:

vi = argmina−i ∈A−i [argmaxai ∈Ai Ui (ai , a−i )]

Marı́a Sofı́a Casabianca González – Take Home Exam 7

a) Establishes the minimum payoff a player can secure from itself in a finite repeated game.
b) Establishes the maximum payoff a player can secure from itself in an infinite repeated game.

c) Establishes the maximum payoff a player can secure from itself in a finite repeated game.
d) Establishes the minimum payoff a player can secure from itself in an infinite repeated game.

Answer: d
The minmax payoff serves as a lower bound on equilibrium payoffs in repeated games since it is the payoff
that player i can secure by himself. It is also an essential part to fulfill the Folk Theorem in which we look for
possible payoffs u that can be supported in time by a SPNE and are strictly above the lower bound.

(d) (3 points) Comparing the Nash Equilibrium versus the Correlated equilibrium:

a) The Nash’s Theorem implies there is always a correlated equilibrium.

b) The correlated equilibria are harder to find than Nash equilibria as we can no apply linear programming.
c) It is unusual for correlated equilibria to achieve better solutions for both players than Nash equilibria.
d) The set of Correlated equilibria is always smaller than the set of Nash equilibria.

Answer: a
Since a correlated equilibrium is a probability distribution over outcomes, that is, a joint distribution over
players’ strategies and a Nash equilibrium is just a special case of correlated equilibrium, in which the joint
distribution of strategies is the product of the corresponding marginals. Since, the Nash Theorem says that
in any n-player normal-form game with finite strategy sets Si for all players has a Nash equilibrium in mixed
strategies.Therefore if there is a Nash Equilibrium in mixed strategies and it is a special case of a correlated
equilibrium, then the Nash Theorem implies that there is always a correlated equilibrium.

(e) (3 points) Consider the following normal form game:

a) The strategies {(U, L), (D, R)} is a Trembling-hand perfect equilibrium.

b) The strategies {(U, L)} is a Trembling-hand perfect equilibrium.

c) The strategies {(D, R)} is a Trembling-hand perfect equilibrium.
d) This game has no Trembling-hand perfect equilibrium.

Answer: b
Trembling-hand proof for (U, L)

(1-) 
1/2
L R
(1- )
1, 1 2, 0
U

0, 2 2, 2
D
Marı́a Sofı́a Casabianca González – Take Home Exam 8

• If player 2 makes a mistake (plays R) with probability :

– U1 (U, σ2 ) = (1 − )(1) + 2 = 1 − 
– U1 (D, σ2 ) = (1 − )(0) + 2 = 2
Since U1 (U, σ2 ) > U1 (D, σ2 ), then player 1 will not have incentives to deviate for all  > 0
• If player 1 makes a mistake (plays D) with probability :
– U2 (σ1 , L) = (1 − )(1) + 2 = 1 − 
– U2 (σ1 , R) = (1 − )(0) + 2 = 2
Since U2 (σ1 , L) > U2 (σ1 , R), then player 2 will not have incentives to deviate for all  > 0

Since neither player has incentives to deviate, this equilibrium ((U, L)) survives and is a Trembling-hand
perfect equilibrium.
Trembling-hand proof for (D, R)

 (1 - )
1/2
L R

1, 1 2, 0
U
(1 - )
0, 2 2, 2
D

• If player 2 makes a mistake (plays L) with probability :

– U1 (U, σ2 ) = (1) + (1 − )(2) = 2 − 
– U1 (D, σ2 ) = (0) + (1 − )(2) = 2 − 2
Since U1 (U, σ2 ) > U1 (D, σ2 ), then player 1 will have incentives to deviate for all  > 0
• If player 1 makes a mistake (plays U) with probability :
– U2 (σ1 , L) = (1) + (1 − )(2) = 2 − 
– U2 (σ1 , R) = (0) + (1 − )(2) = 2 − 2
Since U2 (σ1 , L) > U2 (σ1 , R), then player 2 will have incentives to deviate for all  > 0

Since at least one of the players has incentives to deviate, this equilibrium ((D, R)) can’t survive and there
for is not a Trembling-hand perfect equilibrium.
As was shown above, only (U, L) is an equilibrium.