IE54500 – Exam 3

Dr. David Johnson

Fall 2020
1. Static Game of Complete Information
Specify the mixed-strategy Nash equilibrium of the following normal-form game (and show your work). Note
that the payoffs 𝑥, 𝑦 indicate a payoff of 𝑥 to Player A and 𝑦 to Player B.

Player B
Left Right
Player A Top 2, 1 0, 2
Bottom 1, 2 3, 0
To identify a mixed-strategy Nash equilibrium, denote 𝑝 as the probability of Player A playing Top, and 𝑞 as the
probability of Player B playing Left. Each player wishes to maximize their expected payoffs:

𝔼𝜋𝐴 = 2𝑝𝑞 + (1 − 𝑝)𝑞 + 3(1 − 𝑝)(1 − 𝑞) = −3𝑝 − 2𝑞 + 4𝑝𝑞 + 3

𝐹𝑂𝐶: − 3 + 4𝑞 ∗ = 0 ⇒ 𝑞 ∗ = 3/4
𝔼𝜋𝐵 = 𝑝𝑞 + 2𝑝(1 − 𝑞) + 2(1 − 𝑝)𝑞 = 2𝑝 − 3𝑝𝑞 + 2𝑞
𝐹𝑂𝐶: − 3𝑝∗ + 2 = 0 ⇒ 𝑝∗ = 2/3
Both of these conditions present valid probabilities, 0 ≤ 𝑝∗ , 𝑞 ∗ ≤ 1, and can hold simultaneously, so this
represents a mixed-strategy Nash equilibrium.

2. Dynamic Game of Complete Information

Suppose the following game is repeated infinitely, and that each player has an individual discount rate 𝑑1 and
𝑑2 , where 0 ≤ 𝑑𝑖 ≤ 1 for 𝑖 ∈ {1,2}. Assume that the discount rate is defined such that the present value at
time period 0 of payoff 𝜋𝑡 in time period 𝑡 is equal to 𝜋𝑡 𝑑𝑡 . Define a trigger strategy such that Player 1 chooses
Top in each stage of the game unless Player 2 chooses Right; upon seeing Right, Player 1 chooses Bottom in
every following stage, forever. Similarly, Player 2 chooses Left in each stage unless Player 1 chooses Bottom;
upon seeing Bottom, Player 2 chooses Right in every following stage, forever.

What range of discount rates for each player would result in the trigger strategies being an equilibrium?

Player 2
Left Right
Player 1 Top 5, 4 0, 7
Bottom 6, 0 2, 3
Denote the players’ discount rates as 𝛿1 and 𝛿2 . Given Player 2’s use of a trigger strategy, suppose the value
Player 1 gets from sustaining (Top, Left) indefinitely is 𝑉1, and the value from initiating a deviation is 𝑉1𝐷 .
Therefore, we can find that

𝑉1 = 4 + 4𝛿1 + 4𝛿12 + ⋯ = 4/(1 − 𝛿1 )

𝑉1𝐷 = 7 + 3𝛿1 + 3𝛿12 + ⋯ = 4 + 3/(1 − 𝛿1 )

Similarly, for Player 2, the payoff structure yields

𝑉2 = 5 + 5𝛿2 + 5𝛿22 + ⋯ = 5/(1 − 𝛿2 )

 𝑉2𝐷 = 6 + 2𝛿2 + 2𝛿22 + ⋯ = 4 + 2/(1 − 𝛿2 )

Player 1 will not pull the trigger as long as 𝑉1 ≥ 𝑉1𝐷 , so

4 3 1
≥4+ ⇒ ≥ 4 ⇒ 𝛿1 ≥ 3/4
1 − 𝛿1 1 − 𝛿1 1 − 𝛿1

Similarly, we require that 𝑉2 ≥ 𝑉2𝐷 to sustain the trigger strategy for Player 2:
5 2 3
≥4+ ⇒ ≥ 4 ⇒ 𝛿2 ≥ 1/4
1 − 𝛿2 1 − 𝛿2 1 − 𝛿2
Therefore, each player will sustain their trigger strategy if 𝛿1 ≥ 3/4 and 𝛿2 ≥ 1/4, resulting in an equilibrium.

3. Static Game of Incomplete Information

Consider a Cournot duopoly operating in a market with inverse demand 𝑃(𝑄) = 𝑎 − 𝑄, where
𝑄 = 𝑞1 + 𝑞2 is the aggregate quantity produced. Both firms have total costs 𝑐𝑖 (𝑞𝑖 ) = 𝑐𝑞𝑖 for some constant
marginal cost 𝑐, but the demand is uncertain. For simplification, assume that demand is either high (𝑎 =
𝑎𝐻 ) with probability 𝜃 or low (𝑎 = 𝑎𝐿 ) with probability 1 − 𝜃. Furthermore, information is asymmetric; Firm 1
does a better job with market research, so they know whether demand is high or low, but Firm 2 does not. All
of this is common knowledge, and the two firms simultaneously choose their quantities.

What is the Bayesian Nash equilibrium of this game? In your answer, you should also state conditions related
to 𝑎𝐻 , 𝑎𝐿 , 𝜃 and 𝑐 under which all equilibrium quantities are positive.

Firm 1 knows if the demand is high or low, so their strategy space involves two choices: choosing their level of
production in the case of 𝑎 = 𝑎𝐻 , 𝑞1∗ (𝐻), and their production in the case of 𝑎 = 𝑎𝐿 , 𝑞1∗ (𝐿). They are restricted
by physical constraints that 𝑞1∗ (𝐻), 𝑞1∗ (𝐿) ≥ 0. Firm 2’s strategy space is their choice of 𝑞2∗ ≥ 0. Assuming they
act to maximize their profits, Firm 1 has profit function

𝜋1 (𝐻) = 𝑞1 ∙ (𝑎𝐻 − 𝑞1 − 𝑞2∗ − 𝑐)

𝑎𝐻 − 𝑞2∗ − 𝑐
⇒ 𝐹𝑂𝐶: 2𝑞1∗ (𝐻) = 𝑎𝐻 − 𝑞2∗ −𝑐 ⇒ 𝑞1∗ (𝐻) =
2
Similarly,
𝑎𝐿 − 𝑞2∗ − 𝑐
𝑞1∗ (𝐿) =
2
Firm 2 must maximize their expected profits because they do not know 𝑎:

𝔼𝜋2 = 𝑞2 (𝜃(𝑎𝐻 − 𝑞1∗ (𝐻)) + (1 − 𝜃)(𝑎𝐿 − 𝑞1∗ (𝐿)) − 𝑞2 − 𝑐)

⇒ 𝐹𝑂𝐶: 𝑞2∗ = 𝜃(𝑎𝐻 − 𝑞1∗ (𝐻)) + (1 − 𝜃)(𝑎𝐿 − 𝑞1∗ (𝐿)) − 𝑐

𝑎𝐻 − 𝑞2∗ − 𝑐 𝑎𝐿 − 𝑞2∗ − 𝑐 1
= [𝜃 (𝑎𝐻 − ) + (1 − 𝜃) (𝑎𝐿 − ) − 𝑐] ∙
2 2 2

𝜃𝑎𝐻 + (1 − 𝜃)𝑎𝐿 − 𝑐 𝑞2∗

⇒ 𝑞2∗ = + ⇒ 3𝑞2∗ = 𝜃𝑎𝐻 + (1 − 𝜃)𝑎𝐿 − 𝑐
4 4
𝜃𝑎𝐻 + (1 − 𝜃)𝑎𝐿 − 𝑐
⇒ 𝑞2∗ =
3
Substituting our solution for 𝑞2∗ back into our expressions for 𝑞1∗ (𝐻) and 𝑞1∗ (𝐿), we get
𝑎𝐻 − 𝑐 𝜃𝑎𝐻 + (1 − 𝜃)𝑎𝐿 − 𝑐 𝑎𝐻 − 𝑐 1 − 𝜃
𝑞1∗ (𝐻) = − = + ∙ (𝑎𝐻 − 𝑎𝐿 )
2 6 3 6
𝑎𝐿 − 𝑐 𝜃𝑎𝐻 + (1 − 𝜃)𝑎𝐿 − 𝑐 𝑎𝐿 − 𝑐 𝜃
𝑞1∗ (𝐿) = − = − ∙ (𝑎𝐻 − 𝑎𝐿 )
2 6 3 6
These are the Bayesian Nash equilibrium quantities produced, provided that each quantity is positive. 𝑞2∗ can be
written as
𝑎𝐿 − 𝑐 𝜃
𝑞2∗ = + ∙ (𝑎𝐻 − 𝑎𝐿 )
3 3
so clearly we can conclude that 𝑞1∗ (𝐻) > 𝑞1∗ (𝐿) and 𝑞2∗ > 𝑞1∗ (𝐿).

Therefore, if 𝑞1∗ (𝐿) > 0, then all three quantities are positive. This would imply that
(𝑎𝐿 − 𝑐) 𝜃
> ∙ (𝑎𝐻 − 𝑎𝐿 ) ⇒ 2𝑎𝐿 − 2𝑐 > 𝜃𝑎𝐻 − 𝜃𝑎𝐿
3 6
⇒ 2𝑎𝐿 − 𝜃(𝑎𝐻 − 𝑎𝐿 ) > 2𝑐
If this condition holds, then all three quantities will be positive, satisfying the equilibrium.

4. Dynamic Game of Incomplete Information

The hiring manager at a consulting firm believes that job applicants can be classified along two dimensions:
they are either high quality or low quality in terms of their skills, and they are either Clean or Dirty in
appearance. The quality of an applicant is an innate characteristic (i.e., their type), but the hiring manager
cannot directly observe or determine it during the interview process. The applicant can choose their level of
cleanliness and use it as a signal of quality. The manager’s prior belief is that 40% of applicants are high quality,
as chosen by Nature.

Regarding the payoff structure for both players: in the extensive form game tree below, a payoff of 𝑥, 𝑦
indicates that the Applicant’s payoff is 𝑥 and the Manager’s payoff is 𝑦. Applicants always prefer to be hired
and receive a payoff of 3 if this occurs. However, being clean is a costly signal, requiring a clean applicant to
sacrifice 1 unit of reward even if they are not hired. The Manager believes that hiring a Dirty consultant looks
bad to the firm’s clients; as such, hiring a Dirty applicant reduces the Manager’s payoff. This can be overlooked
somewhat for high-quality applicants, such that the Manager prefers to Hire high-quality applicants no matter
which signal is sent. However, the Manager would only want to hire a low-quality applicant if they send a
Clean signal. Payoffs are summarized in the game tree below.

Find all of the pure-strategy equilibria in this signaling game.

For notational convenience, I’ll call Clean L or Left, and Dirty R or Right; Hire is H and Don’t Hire is D. High-
quality applicants are type 𝑡1 and low-quality applicants are type 𝑡2 . Strategies are written like 𝑃1: (𝑡1 →
𝐿, 𝑡2 → 𝑅) to mean “Player 1 (Applicant) chooses L if type 𝑡1 , R if 𝑡2 ” and, for Player 2, 𝑃2: (𝐿 → 𝐻, 𝑅 → 𝐷)
means “Player 2 (Manager) chooses H if he observes L from Player 1 and D if he observes R.”

Because Player 1 is always on the equilibrium path, we can assume he plays a pure strategy, which means we
have four cases to examine. Let’s start with potential separating equilibria:

1) If P1 plays 𝑃1: (𝑡1 → 𝐿, 𝑡2 → 𝑅), then P2’s best response is 𝑃2: (𝐿 → 𝐻, 𝑅 → 𝐷). We can find this
trivially because, when P1 plays Left, P2 updates their belief about P1’s type given L, using Bayes Rule
to find that
𝑃(𝐿|𝑡1 ) ∙ 𝑃(𝑡1 ) 1 ∙ 0.4 0.4
𝑝̃ = 𝑃(𝑡1 |𝐿) = = = =1
𝑃(𝐿) 0.4 ∙ 1 + 0.6 ∙ 0 0.4

Because P2 knows that P1 is 𝑡1 if she plays L, P2’s expected payoff calculation given L is simple: H gives
3, D gives 0, so he plays H. Similarly, the best response to R is D, because this induces belief
𝑞̃ = 𝑃(𝑡1 |𝑅) = 0.

Player 1 always prefers to be hired, though, so she has incentive to deviate and play L if she is type 𝑡2 ,
given Player 2’s declared strategy. Therefore, this is not an equilibrium.
2) If Player 1 plays 𝑃1: (𝑡1 → 𝑅, 𝑡2 → 𝐿), Player 2’s best response is found, similarly as above, to be
𝑃2: (𝐿 → 𝐻, 𝑅 → 𝐻). However, even though Player 1 is always hired, she still has incentive to deviate
from L to R if she is of type 𝑡2 . Consequently, this is also not an equilibrium.
3) If Player 1 plays 𝑃1: (𝑡1 → 𝐿, 𝑡2 → 𝐿), then the manager has no new information to update his beliefs:
𝑃(𝐿|𝑡1 ) ∙ 𝑃(𝑡1 ) 1 ∙ 0.4
𝑝̃ = 𝑃(𝑡1 |𝐿) = = = 0.4
𝑃(𝐿) 1
Given this strategy from the applicant, Player 2 has an expected payoff of Hiring of
𝔼𝜋2 (𝐻|𝐿) = 3 ∙ 0.4 + 1 ∙ 0.6 = 1.2 + 0.6 = 1.8
Clearly, the expected payoff from D is 0, as always, so the best response to L is H. Player 1 would have
incentive to deviate to R if she would still be hired, because Clean is costly. Thus, in an equilibrium,
Player 2’s best response to R must be D, Don’t Hire. However, his beliefs must be consistent with this
strategy for it to be a credible threat in order to discourage Player 1 from deviating. With 𝑞̃ = 𝑃(𝑡1 |𝑅),
 𝔼𝜋2 (𝐻|𝑅) = 2𝑞̃ − 1(1 − 𝑞̃) = 2𝑞̃ + 𝑞̃ − 1 = 3𝑞̃ − 1
𝔼𝜋2 (𝐷|𝑅) = 0𝑞̃ + 0(1 − 𝑞̃) = 0

We require that 𝔼𝜋2 (𝐻|𝑅) ≤ 𝔼𝜋2 (𝐷|𝑅), so 0 ≥ 3𝑞̃ − 1 ⇒ 𝑞̃ ≤ 1/3.

Therefore, an equilibrium exists for player strategies 𝑃1: (𝑡1 → 𝐿, 𝑡2 → 𝐿) and 𝑃2: (𝐿 → 𝐻, 𝑅 → 𝐷)
when Player 2 has beliefs 𝑝̃ = 0.4 and 𝑞̃ ≤ 1/3.
4) If Player 1 plays 𝑃1: (𝑡1 → 𝑅, 𝑡2 → 𝑅), then Player 2 has beliefs 𝑞̃ = 0.4 by Bayes Rule, and thus expected
payoffs to H and D, given R, of
𝔼𝜋2 (𝐻|𝑅) = 2 ∙ 0.4 − 1 ∙ 0.6 = 0.2
𝔼𝜋2 (𝐷|𝑅) = 0
Therefore, under this strategy from the applicant, the manager’s best response is still to hire, even though
the applicant is Dirty. Player 1 has no incentive to deviate from R in this case, no matter what Player 2
threatens to do when faced with L: if Player 2 would not hire upon seeing L, it is in Player 1’s best interest to
stick to R because they prefer to be hired. If Player 2 would hire upon seeing L, then Player 1 would still
prefer not to incur the cost of cleanliness. Thus, two other equilibria exist: for 𝑃1: (𝑡1 → 𝑅, 𝑡2 → 𝑅), and
either of 𝑃2: (𝐿 → 𝐻, 𝑅 → 𝐻) or 𝑃2: (𝐿 → 𝐷, 𝑅 → 𝐻), provided that 𝑞̃ = 0.4 (and for any beliefs about 𝑝̃).

Note that these two equilibria are both dependent upon the manager’s (Player 2) beliefs about the
prevalence of high-quality applicants and the relative payoff of high versus low quality, because their
incentives hinge upon whether or not the expected payoff of hiring the Dirty (R) applicant is positive or
negative.

5. Extra Credit
Choose a number, from 0 to 5, describing how many points of extra credit you would like for putting up with
all of this COVID-19 stuff.

Thanks to all of you for your contributions to the class and interactions with me this semester. I’ve really
enjoyed teaching the class and hope that you’ve been able to do your best despite all of the many challenges
we’ve faced along the way.