Ec-247
Games and Information
Instructor: Sourav Bhattacharya
sourav@[Link]
Office: K 106
Class 7: Bayesian Nash equilibrium
Weak Perfect Bayesian Nash equilibrium
A Cournot Duopoly with uncertainty
Two firms 1 and 2
Market demand function
Marginal cost of firm 1 is 20
Firm 2 may be a high cost firm or a low cost firm. Firm 2 knows its own type.
Firm 1 has a belief about firm’s 2’s probability of being low cost and maximizes its
expected payoff accordingly.
This “belief” is commonly known by both firms.
Assume that firm 1 believes that the two types of firm 2 are equally likely.
Firm 2 knows that firm 1 acts with this belief, firm 1 knows that firm 2 knows that
firm 1 has this belief, …..
A Cournot Duopoly with uncertainty
Firm 2’s problem
If then its reaction function is
If then its reaction function is
Firm 1’s problem
,
Hence, firm 1’s reaction function is
A Cournot Duopoly with uncertainty
Firm 2 has two “types”: high cost type and low cost type
The low cost type’s reaction function is
The high cost type’s reaction function is
Firm 1’s reaction function is
Solving the game:
,
Which implies that , and
Bayesian Nash equilibrium
Two players – one-sided uncertainty
Two players 1 and 2
1 takes action from the set
2 takes action from the set
There is a state of the world whose realization is drawn from a distribution over
the space Θ (Think of this space as {good, bad} or [0,1])
Player 2 has private information: he knows the realized value of .
Player 1 only knows the prior distribution it is drawn from.
Utility function of the two players are and respectively.
Types
Two players – one-sided uncertainty
Player 2 has private information: he knows the realized value of .
The realized value of is known as the “type” of player 2.
In the Cournot Duopoly with uncertainty case, and it took one of two values or .
Player 1 only knows the prior distribution the type is drawn from.
Notice that in the duopoly with cost uncertainty case, the payoff function of firm
2 alone depends on the state.
The payoff function of firm 1 does not depend on firm 2’s cost.
But firm 1 cares about the state because it affects firm 2’s action.
Strategies
Two players – one-sided uncertainty
Player 1’s strategy is simply an action .
In the Cournot Duopoly with uncertain costs, it was still simply a number
Player 2’s strategy is a specification of action in Y for each possible type in Θ
Therefore, it is a function which specifies an action in Y for each type .
For instance, in the case of Cournot Duopoly with uncertainty, strategy of firm 2
was the function Since took only two values, the full specification of the function
is two numbers and .
We wrote this as the pair in short.
Bayesian Nash equilibrium
Two players – one-sided uncertainty
The appropriate equilibrium concept for games with uncertainty is Bayesian Nash
Equilibrium (BNE).
For two player simultaneous move games with one-sided uncertainty as defined
earlier, BNE is a strategy profile such that
(1) For each type in Θ, is that value of which maximizes
(2) is that value of which maximizes , where the expected utility is taken with
respect to its distribution F of the random variable .
Sequential game
Two players – one-sided uncertainty
But we are interested in games with a sequential nature
Two players 1 and 2
There is a state of the world whose realization is drawn from a distribution over the
space Θ (Think of this space as {good, bad} or [0,1])
Player 2 has private information: he knows the realized value of .
Player 1 only knows the prior distribution it is drawn from.
Utility function of the two players are and respectively.
First, 2 takes action from the set
Then, 1 takes action from the set
[Interpretation: the action is a message from 2 (A) to 1 (DM) about the state of the
world.]
Strategies in the sequential game
Player 2’s strategy is a message/action for each possible type in Θ
Therefore, it is a function .
Player 1 observes the message , and can condition his action on it.
Therefore, player 1’s strategy is a specification of an action for each observed
message. In other words, it is a function ,
We denote by the action chosen by 1 on observing message .
Separating and pooling messages
If each type of player 2 plays a different message, then we say that the messages
are separating.
If all types play the same message, we say that it is a pooling message.
Suppose that there are only two types L and H ()
Let the strategy function be and
If , then the message “reveals” the type to player 1. This is a separating strategy.
If , then the message does not reveal the type to player 1. This is a pooling
strategy. After observing this message, all that the player 1 knows is that the state
follows distribution F.
Bayesian learning – if A plays mixed
strategy
Suppose that there are only two types L and H ()
Assume that they are equally likely.
Suppose M = {a,b}
Let the strategy function be the following:
(i)
(ii) is a mixed action: with probability 0.5 and with probability 0.5.
Now, if player 1 observes message , she knows that .
If player 1 observes message , from Bayes Rule she knows that
Weak Perfect Bayesian Nash
equilibrium
Two players – one-sided uncertainty
The appropriate equilibrium concept for sequential games with uncertainty is
Weak Perfect Bayesian Nash Equilibrium (WPBNE).
For two player sequential move games with one-sided uncertainty as defined
earlier, WPBNE is a strategy profile and a belief function such that
(1) For each type in Θ, is that value of which maximizes
(2) is that value of which maximizes , where the expectation of the random
variable is taken with respect to the belief
(3) The belief is calculated from Bayes Rule wherever possible.
