Sei sulla pagina 1di 28

EC319 Economic Theory and Its Applications,

Part II: Lecture 1

Leonardo Felli

NAB.2.14

16 January 2014
Course Outline

Games of Incomplete Information

I Lecture 1: Static Bayesian Games: Examples and Definition.

I Lecture 2: Bayesian Nash Equilibrium, Auctions and


Purification of Mixed Strategy.

I Lecture 3: Revelation Principle.

I Lecture 4: Adverse Section and Price Discrimination.

I Lecture 5: Dynamic Bayesian Games: Definition and PBE.

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 2 / 28
Course Outline (contd)

I Lecture 6: Dynamic Bayesian Games: Strong PBE.

I Lecture 7: Signaling Games.

I Lecture 8: Optimal Auctions.

I Lecture 9: Double Auctions and Efficient Trade.

I Lecture 10: Principal Agent: Moral Hazard.

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 3 / 28
Admin

I My coordinates: 32L.4.02, x7525, lfelli@econ.lse.ac.uk

I PA: Katharine Buckle, 32L.1.03, k.buckle@lse.ac.uk.

I Office Hours:
I Monday 13:00-14:00, Thursday 11:00-12:30
I or by appointment (e-mail lfelli@econ.lse.ac.uk).

I Course Material: available at:


http://econ.lse.ac.uk/staff/lfelli/teaching

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 4 / 28
References:

I Robert Gibbons, A Primer in Game Theory, London:


Harvester-Wheatsheaf, 1992.

I Martin Osborne, An Introduction to Game Theory, Oxford:


Oxford University Press, 2009.

I Steve Tadelis, Game Theory: An Introduction, Princeton:


Princeton University Press, 2013.

I Bernard Salanie, The Economics of Contracts: A Primer,


Cambridge: The MIT Press, 2nd Edition, 2005.

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 5 / 28
Static Bayesian Games: Example 1.

Consider the following simple normal form game:

1\2 L R
U 1, 1 0, 0
D 0, 0 6, 6

This game has two pure strategy Nash equilibria (U, L) and (D, R)
with payoffs respectively (1, 1) and (6, 6).

There also exists one mixed strategy Nash equilibrium (6/7, 6/7)
with payoffs (6/7, 6/7).

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 6 / 28
Static Bayesian Games: Example 1 (contd)
Assume now that player 2 has some small uncertainty when
casting her choice of strategy in the game.
9
With probability she thinks she is playing Game A:
10
1\2 L R
U1 1, 1 0, 0
D1 0, 0 6, 6

1
while with probability she thinks she is playing Game B:
10
1\2 L R
U2 1, 0 0, 10
D2 0, 0 6, 6

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 7 / 28
Static Bayesian Games: Example 1 (contd)

We assume that:

I Player 2 does not discover which game she is playing until


after she has decided her strategy;

I Player 1 has perfect information and knows exactly which


game he is playing.

This is a game of incomplete information.

The purpose of our analysis is to obtain predictions on what will be


the outcome of this game.

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 8 / 28
Static Bayesian Games: Example 1 (contd)

This is a simultaneous move game hence Nash equilibrium should


be the basic tool to get a prediction.
In particular we can determine the strategy choice of player 1 as a
function of what he expects player 2 to do as best reply:

I If the game played is A then player 1s best reply if he expects


player 2 to play L is U1 , while his best reply if he expects
player 2 to play R is D1 .

I If the game played is B then player 1s best reply if he expects


player 2 to play L is U2 , while his best reply if he expects
player 2 to play R is D2 .

This is not a surprise since the payoffs to player 1 are the same in
both games.

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 9 / 28
Static Bayesian Games: Example 1 (contd)

To be able to determine player 2 best reply however strategies and


payoffs are not enough.

We need to take into account player 2s beliefs about which game


is played.

As mentioned in the description of the game player 2 believes that:

I with probability 9/10 the game played is A, and


I with probability 1/10 the game played is B.

We need to compute player 2s expected payoffs in the game


given her beliefs on which game is played.

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 10 / 28
Static Bayesian Games: Example 1 (contd)

If player 2 expects player 1 to play U1 or U2 her expected payoffs


are:    
9 1 9
2 (U1 , U2 ; L) = 1 +0 =
10 10 10
   
9 1
2 (U1 , U2 ; R) = 0 + 10 =1
10 10
The best reply is therefore R.

If player 2 expects player 1 to play D1 or D2 her expected payoffs


are:    
9 1
2 (D1 , D2 , L) = 0 +0 =0
10 10
   
9 1
2 (D1 , D2 , R) = 6 +6 =6
10 10
The best reply is therefore R.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 11 / 28
Static Bayesian Games: Example 1 (contd)
If player 2 expects player 1 to play U1 or D2 her expected payoffs
are:    
9 1 9
2 (U1 , D2 ; L) = 1 +0 =
10 10 10
   
9 1 6
2 (U1 , D2 , R) = 0 +6 =
10 10 10
The best reply is therefore L.
Finally if player 2 expects player 1 to play D1 or U2 her expected
payoffs are:
   
9 1
2 (D1 , U2 ; L) = 0 +0 =0
10 10
   
9 1 64
2 (D1 , U2 , R) = 6 + 10 =
10 10 10
The best reply is therefore R.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 12 / 28
Static Bayesian Games: Example 1 (contd)

Recall that player 1s best reply is:

I if player 2 chooses L to play U1 and U2 ,

I if player 2 chooses R to play D1 and D2 .

The unique Bayesian Nash equilibrium of the game is therefore:

I player 2 chooses R and

I player 1 chooses D1 if the game is A and D2 if the game is B.

All other strategies yield a profitable deviation by at least one of


the players given player 2s beliefs on which game is played.

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 13 / 28
Static Bayesian Games: Cournot Duopoly

Consider now the following Cournot Duopoly game with imperfect


information.

Firm 1s cost function is:

c(q1 ) = c q1 .

Firm 2 can be a high cost firm:

cH (q2 ) = cH q2 where cH > c;

or it can be a low cost firm:

cL (q2 ) = cL q2 where cL < c;

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 14 / 28
Static Bayesian Games: Cournot Duopoly (contd)

We assume that:

Firm 2 knows whether its variable cost is high cH or low cL ,

Firm 1 does not know the variable cost of firm 2 but believes that:

1
I with probability the cost is high cH ,
2

1
I with probability the cost is low cL .
2

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 15 / 28
Static Bayesian Games: Cournot Duopoly (contd)

The inverse demand function faced by both firms is:

P(q1 + q2 ) = a (q1 + q2 )
where cH < a.

Notice that firm 1 will choose a unique quantity.

Notice that firm 2 will choose a different quantity depending on


whether its cost is high or low.

We label these quantities: q2H or q2L .

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 16 / 28
Static Bayesian Games: Cournot Duopoly (contd)

Depending on the value of the cost firm 2s profit functions are:


h i
H2 (q ,
1 2q H
) = q2
H
a (q 1 + q H
2 ) cH

or h i
L2 (q1 , q2L ) = q2L a (q1 + q2L ) cL

while firm 1s expected profit function is:


1 h i
1 (q1 , q2H , q2L ) = q1 a (q1 + q2H ) c +
2
1 h i
+ q1 a (q1 + q2L ) c
2

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 17 / 28
Static Bayesian Games: Cournot Duopoly (contd)
For any given quantity chosen by firm 1, q1 , firm 2s best reply is,
depending on the value of the cost:
h i
max q2H a (q1 + q2H ) cH
q2H R+

or
1
q2H = (a q1 cH ) ,
2

and h i
max q2L a (q1 + q2L ) cL
q2L R+

or
1
q2L = (a q1 cL ) .
2

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 18 / 28
Static Bayesian Games: Cournot Duopoly (contd)

For any given quantity chosen by the two types of firm 2 q2H and
q2L , firm 1s best reply is:

1 h i 1 h i
max q1 a (q1 + q2H ) c + q1 a (q1 + q2L ) c
q1 R+ 2 2

or

 
1 1  H
 1 
L
q1 = a q2 c + a q2 c .
2 2 2

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 19 / 28
Static Bayesian Games: Cournot Duopoly (contd)

In other words the best reply of the two types of firm 2 and of firm
1 are:

1
q2H = (a q1 cH ) ,
2

1
q2L = (a q1 cL ) ,
2

 
1 1   1 
q1 = a q2H c + a q2L c .
2 2 2

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 20 / 28
Static Bayesian Games: Cournot Duopoly (contd)

The solution to this system of equations is:

(a 2c + 12 cH + 12 cL )
q1 =
3

and
(a 2cH + c) 1
q2H = + (cH cL )
3 12

and
(a 2cL + c) 1
q2L = (cH cL )
3 12

where q2L > q2H .

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 21 / 28
Static Bayesian Games: Cournot Duopoly (contd)

This is the unique Bayesian Nash equilibrium of this Cournot game


with asymmetric information.

Useful question is what is the difference with the Cournot model


with perfect information.

Consider the same model in which everything is the same except


that firm 1 knows the costs of firm 2.

In this case firm 1 will choose two different quantities depending


on whether it is competing with the high cost firm 2, q1H , or the
low cost firm 2, q1L .

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 22 / 28
Static Bayesian Games: Cournot Duopoly (contd)
These quantities are the solution to the following two problems:
h i
max q1H a (q1H + q2H ) c
q1H R+

with best reply:


1 
q1H = a q2H c ,
2

and h i
max q1L a (q1L + q2L ) cL
q1L R+

or
1 
q1L = a q2L c .
2

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 23 / 28
Static Bayesian Games: Cournot Duopoly (contd)

Firm 2s best reply are instead:

a q1H cH a q1L cL
 
H
q2 = , q2L = .
3 3

The Nash equilibria of this two Perfect Information Cournot games


are then:

(a 2c + cH ) (a 2cH + c)
q1H = q2H =
3 3

and

(a 2c + cL ) (a 2cL + c)
q1L = q2L =
3 3

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 24 / 28
Static Bayesian Games: Cournot Duopoly (contd)

Compare now the quantities chosen by firm 2 in the presence of


complete vs. asymmetric information:

(a 2cH + c) 1
q2H = + (cH cL ) > q2H
3 12

and
(a 2cL + c) 1
q2L = (cH cL ) < q2L
3 12

while for firm 1:


1 H 1 L
q1 = q + q1
2 1 2
or
q1H > q1 > q1L

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 25 / 28
Static Bayesian Game: Definition
A game of incomplete information is defined by:
I The set of players N, in the Cournot model
N = {1, 2}.

I The set of states , in the Cournot model


= {cH , cL }

I The action space for every player Ai , in the Cournot model:


Ai = R+ .

I The set of types for every player Ti , in the Cournot model:


T1 = {c}, T2 = {cH , cL }.

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 26 / 28
Static Bayesian Game: Definition (contd)
I The believes of every player on the types of the opponent: i
a conditional probability distribution on Ti , in the Cournot
model:
1 1
1 (cH ) = , 1 (cL ) = , 2 (c) = 1.
2 2

I The payoff function for every player contingent on the players


type ui , in the Cournot model:
h i
u2 (q1 , q2H | cH ) = q2H a (q1 + q2H ) cH
h i
u2 (q1 , q2L | cL ) = q2L a (q1 + q2L ) cL
1 h i
u1 (q1 , q2H , q2L ) = q1 a (q1 + q2H ) c +
2
1 h i
+ q1 a (q1 + q2L ) c
2
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 27 / 28
Static Bayesian Game: Definition (contd)
A strategy for each player is then an action choice for any type of
the player:
si (ti ) Ai .

In the Cournot model:


I player 1s strategy: q1 :
I player 2s strategy: q2 (cH ) = q2H , q2 (cL ) = q2L .

Therefore the following is a Game of Incomplete Information:

= {N, , Ai , Ti , i , ui }

Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 16 January 2014 28 / 28

Potrebbero piacerti anche