Perfect Information Cournot Competition Outcome

Under symmetric costs Cournot competition, recall that

q1c = q2c = 13 [a − c]. Aggregate quantity in Cournot competition is
Q c = 32 [a − c]
We also know that in the limit (when number of firms are very large), the
Cournot result coincides with perfect competition.

Saha Debdatta (SAU) May-June, 2020 1 / 28

Games with Imperfect Information

Solution concepts:
Bayes Nash Equilibrium for static games
Perfect Bayesian Nash Equilibrium for dynamic games
An important concept here is the concept of beliefs. These are necessary
for players who do not know all the aspects of the game. They must form
beliefs about the part of information that they do not know, else we
cannot solve the game. We mostly work with Bayesian belief structures.

Saha Debdatta (SAU) May-June, 2020 2 / 28

Bayes Nash Equilibrium: Role of Beliefs

A Bayesian Nash equilibrium (BNE) is defined as a strategy profile that

maximizes the expected payoff for each player given their beliefs and given
the strategies played by the other players.
Belief: The belief of a player in a given information set determines what
node in that information set he believes the game has reached. The belief
may be a probability distribution over the nodes in the information set,
and is typically a probability distribution over the possible types of the
other players. Formally, a belief system is an assignment of probabilities to
every node in the game such that the sum of probabilities in any
information set is 1.
Note beliefs are as important as strategies in these games. Ex-ante beliefs
are specified for static games and are updated in dynamic games, following
Bayes’ law.

Saha Debdatta (SAU) May-June, 2020 3 / 28

Application of Bayes Nash Equilibrium: Cournot
Competition

Consider Cournot competition in quantities with imperfect information.

Suppose firm 1 has cost c, which is common knowledge. However, firm 2’s
costs is privately known by it. Suppose that these costs are either ch > c
with probability θ and cl < c with probability (1 − θ). Suppose, as in the
case of Cournot competition with perfect information, both firms face a
linear inverse demand: P(Q) = a − Q, where Q = q1 + q2 . How will this
game be solved, when both players choose quantity simultaneously?

Saha Debdatta (SAU) May-June, 2020 4 / 28

Solution of Cournot Competition with Asymmetric
Information
For firm 2, cost is either ch or cl , given that cost is c for firm 1. Hence,
each type of firm 2 (where type is identified by cost) maximizes output,
given the choice of firm 1. Now for firm 1, profit maximization involves
taking an expectation of profits over the two possible types of the
opponent to decide its own best response. The unique Nash equilibrium of
the game is driven by this expectation formed by firm 1, given choices of
either type of firm 2.
Simultaneous Optimization
Let q2I (ch ; c) and q2I (ch ; c) be the optimal quantity choice by firm 2, given
the choice of q1I by firm 1. This results from the following maximization
for all i = h, l:
Maxq 2 [a − q1I − q2 − ci ]q2 . For firm 1, the maximization is:
Maxq 1 [θ{a − q1 − q2I (ch ; c) − c}q1 + (1 − θ){a − q1 − q2I (cl ; c) − c}q1 ]

Saha Debdatta (SAU) May-June, 2020 5 / 28

First Order Conditions

FOCs:
a − q1I − ch
q2I (ch ; c) =
2
a − q1I − cl
q2I (cl ; c) =
2
θ(a − q2I (ch ; c) − c) + (1 − θ)(a − q2I (cl ; c) − c)
q1I =
2
Show that q2I (cl ) > q2I (ch ) and that q2I (ch ) > q2I , where q2I is the output of
firm 2 in the standard Cournot competition.

Saha Debdatta (SAU) May-June, 2020 6 / 28

Comparing with Complete Information Cournot
Competition

Note that, in the equilibrium with imperfect information:

a − 2ch + c 1−θ
q2I (ch ; c) = + (ch − cl ) (1)
3 6
a − 2cl + c θ
q2I (cl ; c) = − (ch − cl ) (2)
3 6
a − 2c + θch + (1 − θ)cl
q1I = (3)
3
In the complete information case, when firm 2’s cost is ch , we find that
q2∗ = a−2c3h +c < q2I (ch ; c). When firm 2’s cost is known to be cl , we find
q2∗ = a−2c3 l +c > q2I (ch ; c).
In either case, for firm 1 (armed with complete information), equilibrium
output is q1∗ = a−2c+c 3
i
∀i = l, h.

Saha Debdatta (SAU) May-June, 2020 7 / 28

Dynamic Games of Incomplete Information

Appropriate solution concept: Perfect Bayesian Nash Equilibrium (PBNE)

Four requirements:
* Requirement 1 At each information set, the player with the move
must have a belief about which node in the information set has been
reached by the play of the game. For a non-singleton information set,
a belief is a probability distribution over the nodes in the information
set.
* Requirement 2 Given their beliefs, players’ strategies must be
sequentially rational.
What is the effect of these requirements?

Saha Debdatta (SAU) May-June, 2020 8 / 28

Effect of Requirements 1 & 2

Saha Debdatta (SAU) May-June, 2020 9 / 28

What Requirements 1 and 2 Do?

Two pure strategy Nash equilibria: (L, L’) and (R, R’) without
Requirements 1 & 2. With these requirements, we can rule out equilibria
like (R,R’). Suppose p and 1 − p are the beliefs of player 2 that player 1
has played L and M respectively in the information set in figure 3 are
reached. Now, the expected payoff to player 2 from playing R’ is
p.0 + (1 − p).1 = (1 − p). The expected payoff from playing L’ is
p.1 + (1 − p).2 = (2 − p). Since (2 − p) > (1 − p) for any value of p,
playing L’ strictly payoff dominates R’s for a rational player 2. Note that
for a given equilibrium in an extensive form game, an information set is
on-the-equilibrium path if it will be reached with a positive probability if
the game is played following equilibrium strategies. An information set is
off-the-equilibrium path if it is certain not to be reached if the game is
played following equilibrium strategies.

Saha Debdatta (SAU) May-June, 2020 10 / 28

Other Requirements for PBNE

* Requirement 3 At information sets on-the-equilibrium path, beliefs are

determined by Bayes’ rule and the players’ strategies.
* Requirement 4 At information sets off-the-equilibrium path, beliefs are
determined by Bayes’ rule and the players’ strategies where possible.
PBNE
A Perfect Bayesian Nash Equilibrium (PBNE) consists of strategies and
beliefs that satisfy Requirements 1-4.

Saha Debdatta (SAU) May-June, 2020 11 / 28

Role of Requirements 3 and 4

Refer to figure 4.1.4 from Gibbons and class discussion. Compare

equilibria (D, L, R’) and (A, L, L’).

Saha Debdatta (SAU) May-June, 2020 12 / 28

Signaling Games: Application of PBNE

Signaling games are a special type of dynamic game of incomplete

information, where there are two kinds of players (a Sender (S) and a
receiver (R)). We consider the discrete version of this game, where the
sender (S) has one of two types T = {t1 , t2 } (privately known by S and
not known by R), S sends a message from the message space
M = {m1 , m2 } and the receiver chooses among actions A = {a1 , a2 }.

Timing of the Game

First, Nature chooses the type of the sender S, who receives a private
signal about her/his type. The receiver R only knows the type-space and
the ex-ante probability distribution over the type-space. Upon knowing
her/his type, S sends R a message (to maximize expected utility). Upon
receiving the message, R updates his/her beliefs about the type
conditional on the message sent by S and selects an action to maximize
expected utility. Now, both S and R realize the final payoffs.

Saha Debdatta (SAU) May-June, 2020 13 / 28

Extensive Form Representation of a Signaling Game

Figure: Signaling Game Extensive Form

Saha Debdatta (SAU) May-June, 2020 14 / 28

Strategies in the Signaling Game: Sender

Pooling and Separating Strategies of the Sender

1 Sender Strategy 1 (Pooling): Play m1 if nature draws t1 and play m1
if nature draws t2 .
2 Sender Strategy 2 (Separating): Play m1 if nature draws t1 and play
m2 if nature draws t2 .
3 Sender Strategy 3 (Separating): Play m2 if nature draws t1 and play
m1 if nature draws t2 .
4 Sender Strategy 4 (Pooling): Play m2 if nature draws t1 and play
Play m2 if nature draws t2 .

Saha Debdatta (SAU) May-June, 2020 15 / 28

Receiver’s Pure Strategies

1 Receiver Strategy 1: Play a1 if the Sender chooses m1 and play a1 if

the Sender chooses m2 .
2 Receiver Strategy 2: Play a1 if the Sender chooses m1 and play a2 if
the Sender chooses m2 .
3 Receiver Strategy 3: Play a2 if the Sender chooses m1 and play a1 if
the Sender chooses m2 .
4 Receiver Strategy 4: Play a2 if the Sender chooses m1 and play a2 if
the Sender chooses m2 .

Saha Debdatta (SAU) May-June, 2020 16 / 28

Signaling PBNE Requirements

Requirement 1 is trivial when applied to the Sender. Applying

Requirement 1 to the Receiver yields:
Signaling Requirement 1 (SR1) After observing any message mj from M,
the Receiver must have a belief about which types could have sent this
message. Denote this beliefPby the probability distribution µ(ti |mj ), where
µ(ti |mj ) ≥ 0 ∀ti ∈ T and ti ∈T µ(ti |mj ) = 1.
Signaling Requirement 2 for the Receiver is:
Signaling Requirement 2R (SR2R) For each mj ∈ M, the Receiver’s action
a∗ (mj ) must maximize the Receiver’s expected utility, given the belief
µ(ti |mj ) about which types could have sent mj . That is, a∗ (mj ) solves
maxak ∈A µ(ti |mj )UR (ti , mj , ak ).

Saha Debdatta (SAU) May-June, 2020 17 / 28

Signaling Requirements for PBNE

Requirement 2 also applies to the Sender, but the Sender has complete
information (and hence a trivial belief) and also moves only at the
beginning of the game, so Requirement 2 is simply that the Sender’s
strategy be optimal given the Receiver’s strategy:
Signaling Requirement 2S (SR2S) For each ti ∈ T , the Sender’s message
m∗ (ti ) must maximize the Sender’s utility, given the Receiver’s strategy
a∗ (mj ). That is, m∗ (ti ) solves maxmj ∈M Us (ti , mj , a∗ (mj )).
For messages on the equilibrium path, applying Requirement 3 to the
Receiver’s beliefs yields:
Signaling Requirement 3 (SR3) For each mj ∈ inM, if there exists ti ∈ T
such that m∗ (ti ) = mj , then the Receiver’s belief at the information set
corresponding to mj must follow from Bayes’ rule and the Sender’s
strategy: µ(ti |mj ) = P p(ti p(t
)
i)
ti ∈T

Saha Debdatta (SAU) May-June, 2020 18 / 28

Application

Saha Debdatta (SAU) May-June, 2020 19 / 28

Solution
Consider each candidate equilibrium (pooling on R, pooling on L, two
separating equilibria) and follow the signaling requirements to check if
they can be sustained as PBNE.
Case I. Pooling on R (Both t1 and t2 send the same message m2 = R)
SR1. Specify the belief of the Receiver after observing the message (R, R)
Sender’s strategy is (R, R). There exists beliefs (µ2 , 1 − µ2 )
on-the-equilibrium path and (µ1 , 1 − µ1 ) off-the-equilibrium path.
SR3. Formation of the belief of (µ2 , 1 − µ2 ) by Bayes’ Law and the
Sender’s strategy

p(t1 ) L(m2 = R|t1 )q(t1 )

µ2 (t1 |m2 = R) = P =
ti ∈T p(ti ) L(m2 = R|t1 )q(t1 ) + L(m2 = R|t2 )q(t2 )
1 × 0.5
= = 0.5[= q(t1 )]
1 × 0.5 + 1 × 0.5
Signals are non-informative in this case, as both types pool on the
message R.
Saha Debdatta (SAU) May-June, 2020 20 / 28
Supporting Pooling on R as PBNE
SR2R. Given the belief (µ2 = 0.5 = 1 − µ2 =0.5), the Receiver must
maximize its expected utility/payoff
Comparing payoffs from actions u and d respectively for the Receiver, we
get:

E πR (u) = µ2 1 + (1 − µ2 )0 = 0.5
E πR (d) = µ2 0 + (1 − µ2 )2 = 1
⇒ E πR (d) > E πR (u)

Receiver’s best response following R and belief µ2 = 0.5 is to play d.

SR2S. The Sender should not regret the choice of R, R given Receiver acts
on the belief µ2 = 0.5 and plays d to this message
With (R, R) and Receiver responding with d, the two types of the Sender
gets πSd (t1 ) = 3 and πSd (t2 ) = 2. We note below the payoff to either type
of Sender due a deviation from this strategy for all possible actions of the
Receiver.
Saha Debdatta (SAU) May-June, 2020 21 / 28
 t1 deviates
→
to
L t2 deviates
→
to
L
u u
πS (t1 ) = 1 u
πS (t2 ) = 0
d πSd (t1 ) = 2 πSd (t2 ) = 3
Table: Table 1. Deviation payoffs to the Sender of Different Types

It is easy to see that t1 has no profitable deviation. By deviation, she/he is

likely to get either 1 or 2, whereas with the pooling message R, she/he gets
3. However, type t2 might be tempted to choose an off-the-equilibrium
path action only if the possibility of getting 3 (see table 1 above) is high
enough. Hence, to support pooling on R as a PBNE, it must be that the
Receiver has off-the-equilibrium path beliefs such that playing d following
a deviation by the Sender to L is not worthwhile. This needs:

E πRdev (u) ≥ E πRdev (d)

⇒ µ1 2 + (1 − µ1 )0 ≥ µ1 0 + (1 − µ1 )1
⇒ µ1 ≥ 1/3

Saha Debdatta (SAU) May-June, 2020 22 / 28

PBNE={(R, R); (d, u); (µ2 = 0.5, µ1 ≥ 1/3)}. Hence, pooling on R is a
PBNE supported by off-the-equilibrium path belief µ1 ≥ 1/3.

Saha Debdatta (SAU) May-June, 2020 23 / 28

Pooling on L
Follow all the steps as we have mentioned for the pooling equilibrium on
R. We go over this more precisely now.
Following SR1 and SR3,
p(t1 ) L(m1 = L|t1 )q(t1 )
µ1 (t1 |m1 = L) = P =
ti ∈T p(ti ) L(m1 = L|t1 )q(t1 ) + L(m1 = L|t2 )q(t2 )
1 × 0.5
= = 0.5
1 × 0.5 + 1 × 0.5
1 − µ1 = 0.5
Applying SR2R, compare payoffs from actions u and d respectively for the
Receiver (after observing the pooling message L), we get:
E πR (u) = µ2 2 + (1 − µ2 )0 = 1
E πR (d) = µ2 0 + (1 − µ2 )1 = 0.5
⇒ E πR (u) > E πR (d)
Receiver’s best response following R and belief µ2 = 0.5 is to play u.
Saha Debdatta (SAU) May-June, 2020 24 / 28
SR2S: Does any Sender have an incentive to deviate?
With (L, L) and Receiver responding with u, the two types of the Sender
gets πSd (t1 ) = 1 and πSd (t2 ) = 0.
If t2 deviates to R, then irrespective of the action of the Receiver, it gets a
higher payoff. If t1 deviates to R, its payoff goes up only if Receiver plays
d upon deviation. However, as t2 will deviate for all feasible values of µ2 ,
there is no pooling possibility on the message L. In other words, there
exists no feasible off-the-equilibrium path belief which will support (L, L)
as a pooling PBNE.

Saha Debdatta (SAU) May-June, 2020 25 / 28

Separating Strategy (L, R) as PBNE?
t1 sends message m1 = L and t2 sends message m2 = R. There is no
action off-the-equilibrium path here. Applying SR1. and SR3., we get:
p(t1 ) L(m1 = L|t1 )q(t1 )
µ1 (t1 |m1 = L) = P =
ti ∈T p(ti ) L(m1 = L|t1 )q(t1 ) + L(m1 = L|t2 )q(t2 )
1 × 0.5
= =1
1 × 0.5 + 0 × 0.5
1 − µ1 = 0
By the same logic, 1 − µ2 = 1 and µ2 = 0.
Applying SR2R., for t1 playing L, Receiver’s best response is u, whereas
for t2 playing R, Receiver’s best response is d.
Applying SR2S, check whether any Sender type has an incentive to
deviate. Under (L, R), t1 gets 1 and t2 gets 2. Now, there is no freedom
to choose any off-the-equilibrium path beliefs. However, if t2 deviates and
sends the message L, it will get a payoff of 3 (as Receiver will play d),
which is higher than 1. Therefore, t2 will deviate.
(L, R) cannot be supported as a separating equilibrium.
Saha Debdatta (SAU) May-June, 2020 26 / 28
Separating Strategy (R, L) as PBNE?
t1 sends message m2 = R and t2 sends message m1 = L. There is no
action off-the-equilibrium path here. Applying SR1. and SR3., we get:
p(t1 ) L(m2 = R|t1 )q(t1 )
µ2 (t1 |m2 = R) = P =
ti ∈T p(ti ) L(m2 = R|t1 )q(t1 ) + L(m2 = R|t2 )q(t2 )
1 × 0.5
= =1
1 × 0.5 + 0 × 0.5
1 − µ2 = 0
By the same logic, 1 − µ1 = 1 and µ1 = 0.
Applying SR2R., for t1 playing R, Receiver’s best response is u, whereas
for t2 playing R, Receiver’s best response is d.
Applying SR2S, check whether any Sender type has an incentive to
deviate. Under (R, L), t1 gets 0 and t2 gets 3. Now, there is no freedom
to choose any off-the-equilibrium path. However, if t1 deviates and sends
the message L, it will get a higher payoff of 1 (Receiver will play u).
Therefore, t1 will deviate.
(R, L) cannot be supported as a separating equilibrium.
Saha Debdatta (SAU) May-June, 2020 27 / 28
Unique PBNE is Pooling on R

The unique PBNE of this application is the pooling equilibrium on R,

there is no separating PBNE. Work out the solution for game given earlier
and check that there is one pooling and one separating equilibria in the
game (check result against Gibbons, worked out example).
We now apply this to the Spence model of education as a signaling
contract.

Saha Debdatta (SAU) May-June, 2020 28 / 28