Vous êtes sur la page 1sur 78

Chapter 8

Noncooperative Game Theory


General Equilibrium and Welfare Economics

Outline of the chapter

Introduction to Noncooperative Game Theory


Simultaneous-move games

Dynamic games

Complete information
Incomplete information
Complete information
Incomplete/imperfect information

Repeated games

Introduction

Economic agents take strategic decisions in an


environment with conflicting interests
and in environments in which information and
knowledge are diffuse.
In this case, even simple cooperation proves
difficult.
Noncooperative means: each agent decides
unilaterally, maximizing only his own utility (not
taking into account the others).
The agents may still try to reach common goal
though.
3

Introduction

Firms decisions on prices, quantities, advertising, or


capacity influence other firms decisions on these
same variables, as well as firms welfare.
Similarly, people take strategic decisions in an
environment with conflicting interests.
Game Theory allows us to formally analyze
interdependence, and predict the final outcome

Introduction

Description (ingredients) of a game:

Players
Rules
Actions/Strategies
Outcomes
Each players payoff in each outcome

Introduction

Formal definition of a game:


Set of player: I = {1, , n} with typcial element i
Strategy space S = xi Si, where Si is player is
strategy set (this is also called action space/set)
Payoff function u that maps S into Rn, where u = (u1,
, un) specifies a payoff for every player i given any
strategy profile s in S.

Introduction

We will typically assume:

Rationality
Perfect recall
(Im)Perfect Information. Imperfect information is related
with probabilities.
(In)Complete Information. Don't know if something exist or
not or something can happen.
Common knowledge. Everybody knows the description of
the game. (I know that you know everything in reiterative
ways). (This reiterations usually take 3 steps, but we are
going to consider infinitive).

Introduction

Example: matching pennies

Two players
Rules: Player 1 chooses H or T, simultaneously player 2
chooses H or T. If pennies match, player 1 gets both
pennies, otherwise, player 2 gets them.
Actions/strategies: choose H or T
Outcomes: (H,H), (H,T), (T,H), (T,T)
Payoffs: (1,-1), (-1,1), (-1,1), (1,-1)

Introduction

Depicting matching pennies graphically:

Player 1 has two strategies:

H
T

Player 2 has two strategies:

H
T

Introduction

Making use of the payoff matrix (normal form of the


game).
Simultaneous-move version of matching pennies:
Player 2

H
Player 1

H 1,-1
T

T
-1,1

-1,1 1,-1
P1 P2
10

Introduction

We will classify games into:

Looking at the type of information:

Simultaneous-move games vs. Dynamic games


Normal form games vs. Extensive form games (game tree to
analyze sequences.
Perfect information
Imperfect information

and of the knowledge each player has when


taking a decision:

Complete information
Incomplete information
11

Introduction

Definition of a strategy for player i: function from the


set of information sets to the set of
actions/strategies that determines what player i
does at each possible information set.
Notice that a strategy says what to do at each
possible information set. It is a complete contingent
plan or description of what the players do in any
possible state of the world. Even if knowing or not
the result of the event, take a strategy in advance.

12

Introduction

Pure strategies: those played with probability one


Mixed strategies: the player plays two or more
strategies with positive probability. Ex: flip a coin for
deciding.
Pure strategies are a particular case of mixed strategies

13

Introduction

The games normal form is the set of players, a set


of strategies for each player, and utility functions for
each player:

Strategies

N I, S , u
Players

I
i i 1

I
i i1
utilities

14

Introduction

and it is graphically depcited in a payoff matrix.


For the matching pennies we saw it already:
Player 2

H
Player 1

H 1,-1

-1,1

1,-1

-1,1

15

Outline of the chapter

Introduction to Noncooperative Game Theory


Simultaneous-move games

Dynamic games

Complete information
Incomplete information
Complete information
Incomplete/imperfect information

Repeated games

16

Simultaneous-move games

All players choose strategies at the same time:


players do not know what other players have
chosen when selecting their own strategies
Complete information: players know other players
relevant characteristics and the complete description
of the game
Incomplete information: players lack some
information on other players characteristics and/or
the description of the game
Example: auction
17

Simultaneous-move games

The Prisoners Dilemma


Dominant and dominated strategies

Prisoner 2

Prisoner 1

DC

DC

-2,-2

-10,-1

-1,-10 -5,-5

18

Simultaneous-move games

Here Dont Confess is strictly dominated for both


players, since it yields a lower payoff regardless of
the other players strategy
For the same reason, Confess is striclty dominant,
and we expect both players to play Confess
There is an equilibrium in dominant strategies.
Notice that we only require rationality. Players do
not need to know that the other player is rational

19

Simultaneous-move games

What if there is no dominant strategy?


Players will not choose dominated strategies, so
we may just remove them from their choice sets
As we delete strategies, maybe some strategies
become dominated, and we may erase them
Notice that this requires knowing that the other
player is rational

20

Simultaneous-move games

Example: consider the following game


Player 2

Player 1

Left

Center

Right

Up

1,1

2,0

1,1

Middle

0,0

0,1

0,0

Down

2,1

1,0

2,2
21

Simultaneous-move games

There are no dominant strategies, but M is


dominated:
Player 2

Player 1

Left

Center

Right

Up

1,1

2,0

1,1

Middle

0,0

0,1

0,0

Down

2,1

1,0

2,2
22

Simultaneous-move games

And then, C becomes dominated


Player 2

Player 1

Left

Center

Right

Up

1,1

2,0

1,1

Middle

0,0

0,1

0,0

Down

2,1

1,0

2,2
23

Simultaneous-move games

Now U is dominated by D:
Player 2

Player 1

Left

Center

Right

Up

1,1

2,0

1,1

Middle

0,0

0,1

0,0

Down

2,1

1,0

2,2
24

Simultaneous-move games

And hence, player 2 will not choose L:


Player 2

Player 1

Left

Center

Right

Up

1,1

2,0

1,1

Middle

0,0

0,1

0,0

Down

2,1

1,0

2,2
25

Simultaneous-move games

However, sometimes iterated deletion can not be


started:
Player 2

Player 1

Left

Center

Right

Up

2,1

2,2

0,3

Middle

1,1

1,1

1,1

Down

0,1

0,0

2,2
26

Simultaneous-move games

Strategy si is a best response for player i to his


rivals playing strategies s-i if:

ui(si, si ) ui(s'i , si )

s'i Si

i.e. si maximizes player is utility, given that the


remaining players choose s-i

27

Simultaneous-move games

A strategy profile s=(s1,,sI) constitutes a Nash


equilibrium of game N I, Si iI1 , ui iI1 if for
every player i=1,,I:

ui(si, si ) ui(s'i , si )

s'i Si

i.e. no player has any profitable deviation,


conditional on the remaining players playing the
equilibrium strategies
28

Simultaneous-move games

The players unilaterally choose an action. This


either leads to a Nash equilibrium or not. In a Nash
equilibrium, the players dont have an incentive to
change their behavior. Otherwise, they do.
If there is only one Nash equilibrium, then it is focal
for the players to choose it. Otherwise,
miscoordination can occur.
Common knowledge of rationality is required.
A dominant strategy is always part of a Nash
equilibrium. Dominated or never best response
strategies are never part of a Nash equilibrium.
29

Simultaneous-move games
For instance, in the following game:
Player 2

Player 1

b1

b2

b3

b4

a1

0,7

2,5

7,0

0,1

a2

5,2

3,3

5,2

0,1

a3

7,0

2,5

0,7

0,1

a4

0,0

0,-2

0,0

10,-1

30

Simultaneous-move games

Since a Nash equilibrium is the intersection of the


two players best response functions, it is useful to
characterize these best responses
Identify player 1s best response to each of player
2s strategies. Do the same for player 2, and search
for intersections.

31

Simultaneous-move games
For instance, player 1s best response function
would be given by:
Player 2

Player 1

b1

b2

b3

b4

a1

0,7

2,5

7,0

0,1

a2

5,2

3,3

5,2

0,1

a3

7,0

2,5

0,7

0,1

a4

0,0

0,-2

0,0

10,-1

32

Simultaneous-move games
And player 2s:
Player 2

Player 1

b1

b2

b3

b4

a1

0,7

2,5

7,0

0,1

a2

5,2

3,3

5,2

0,1

a3

7,0

2,5

0,7

0,1

a4

0,0

0,-2

0,0

10,-1

33

Simultaneous-move games
This way, the unique Nash equilibrium is a2,b2:
Player 2

Player 1

b1

b2

b3

b4

a1

0,7

2,5

7,0

0,1

a2

5,2

3,3

5,2

0,1

a3

7,0

2,5

0,7

0,1

a4

0,0

0,-2

0,0

10,-1

34

Simultaneous-move games

Coordination Games
The Nash equilibrium need not be unique. For
instance:
Player 2

Player 1

Left

Right

Left

100,100

0,0

Right

0,0

100,100

35

Simultaneous-move games

This game has two Nash equilibria:


Player 2

Player 1

Left

Right

Left

100,100

0,0

Right

0,0

100,100

36

Simultaneous-move games

These would be the equilibria even if payoffs were:


Player 2

Player 1

Left

Right

Left

100,100

0,0

Right

0,0

1,1

37

Simultaneous-move games

Another coordination game


2
Plaza

Metro

3,

0,
3

Plaza

0,
Metro

0
1,

Coordination Games

There are two


2
NE in pure actions:
Plaza
1
when both players
3,
choose the same
action (therefore
Plaza
coordination games).
There is also one
0,
NE in mixed strategies.
Metro
Probability
distribution of two
choices

Metro

0,
3

0
1,

Calculate NE in mixed actions

Idea: make the other player indifferent between


2
choosing Plaza and
Plaza
Metro
1
Metro.
Only then is he willing
3,
0,
to play a mixed strategy.
Plaza
3
If he were not indifferent,
then hed be better off
1,
to play one pure strategy. 0,
Metro

Calculate NE in mixed actions

Idea: make the other player indifferent between


p
2
choosing Plaza and
1-p
Plaza
Metro
1
Metro.
Only then is he willing
3,
0,
to play a mixed strategy.
Plaza
3
If he were not indifferent,
then hed be better off
1,
to play one pure strategy. 0,
Metro

Calculate NE in mixed actions

Idea: make the other player indifferent between


p
2
choosing Plaza and
1-p
Plaza
Metro
1
Metro.
Only then is he willing q
3,
0,
to play a mixed strategy.
Plaza
3
If he were not indifferent,
then hed be better off
1,
to play one pure strategy. 0,
Metro
1-q

Coordination Games with Conflicting


Interests: The battle of sexes
2
Opera

Football

3,

0,
1

Opera

0,
Football

0
1,

Coordination Games with Conflicting


Interests: The battle of sexes

Player 1 is the woman and player 2 the man.


2
Opera

Football

3,

0,
1

Opera

0,
Football

0
1,

Coordination Games with Conflicting


Interests: The battle of sexes

Player 1 is the woman and player 2 the man.


2
Opera
What are the NE?
Football
1

3,

0,
1

Opera

0,
Football

0
1,

Coordination Games with Conflicting


Interests: The battle of sexes

Player 1 is the woman and player 2 the man.


2
Opera
What are the NE?
Football
1
and in mixed
3,
0,
strategies?

0,
Football

Opera

1,
0

Anti-Coordination Game: Chicken


Game
2
Swerve
1

0,

-1,
0

Swerve

+1,
Straight

Straight

+1
-10,

-1

-10

Anti-Coordination Game: Chicken


Game

Two cars are driving towards each other.


2
Swerve
1

0,

-1,
0

Swerve

+1,
Straight

Straight

+1
-10,

-1

-10

Anti-Coordination Game: Chicken


Game

Two cars are driving towards each other.


2
Who swerves?
Swerve
1

0,

-1,
0

Swerve

+1,
Straight

Straight

+1
-10,

-1

-10

Anti-Coordination Game: Chicken


Game

Two cars are driving towards each other.


2
Who swerves?
Swerve
Straight
and the loser is a 1
0,
-1,
chicken.

Swerve

+1,
Straight

+1
-10,

-1

-10

Anti-Coordination Game: Chicken


Game

Two cars are driving towards each other.


2
Who swerves?
Swerve
Straight
and the loser is a 1
0,
-1,
chicken.
Motivation: there is a Swerve
0
+1
resource that is rivalrous
but non-excludable and
+1,
-10,
sharing comes at Straight
-1
-10
a cost: negative externality

Anti-Coordination Game: Chicken


Game

NE: There are two in pure strategies and one in


2
mixed strategies.
Swerve

0,

-1,
0

Swerve

Straight

Straight

+1,

+1
-10,

-1

-10

Zero-sum games: e.g. Matching Pennies


2
Left

Right

+1,
-1

Left

Right

-1,

-1,

+1
+1,

+1

-1

Zero-sum games: e.g. Matching Pennies

Note that the sum of the two players payoffs is


2
always zero:
Left
Right
So what one player, 1
gains, the other loses
+1,
-1,
and vice versa.
Left

-1

Right

-1,

+1

+1,
+1

-1

Zero-sum games: e.g. Matching Pennies

Note that the sum of the two players payoffs is


2
always zero:
Left
Right
So what one player, 1
gains, the other loses
+1,
-1,
and vice versa.
Left
-1
+1
A resource is rivalrous
and excludable.
Right

-1,

+1,

+1

-1

Zero-sum games: e.g. Matching Pennies

What is the NE?


2
Left

Right

How many NE are there?

+1,
-1

Left

Right

-1,

-1,

+1
+1,

+1

-1

Zero-sum games: e.g. Matching Pennies

There is no intersection of the best response


functions, and thus, there is no Nash equilibrium in
pure strategies
However, both players choosing H and T with
probability 0.5 is a Nash equilibrium in mixed
strategies

57

Simultaneous-move games

We conclude that sometimes the Nash equilibrium


in pure strategies does not exist; such as in the
simultaneous-move version of the matching pennies
game, in which there is only a Nash equilibrium in
mixed strategies
This is true in general for any zero sum game

58

On the Number of NE

Did you notice how many NE we found for the


different games we analyzed?

On the Number of NE

Did you notice how many NE we found for the


different games we analyzed?

The number of NE is always odd.


Intuition (picture): best-reply correspondences
always cut an odd number of times.
They cannot cut an even number of times.

Outline of the chapter

Introduction to Noncooperative Game Theory


Simultaneous-move games

Dynamic games

Complete information
Incomplete information
Complete information
Incomplete/imperfect information

Repeated games

61

Dynamic games

Example: sequential matching pennies

Two players
Rules: Player 1 chooses H or T, then player 2 chooses H
or T knowing player 1s choice. If pennies match, player 1
gets both pennies, otherwise, player 2 gets them
Actions/strategies: choose H or T
Outcomes: (H,H), (H,T), (T,H), (T,T)
Payoffs: (1,-1), (-1,1), (-1,1), (1,-1)

Two important concepts: nodes and information sets

62

Dynamic games

The games extensive form is:


P1
H

P2
H

1 1

2 1

P2
T

1

1

1

1

1

1

63

Dynamic games

Strategies: matching pennies, dynamic version:

Player 1 has two strategies:

H
T

Player 2 has four strategies:

H if H, H if T
H if H, T if T
T if H, H if T
T if H, T if T

64

Dynamic games

Recall the normal form or Simultaneous-move


version of matching pennies:
Player 2

H
Player 1

H 1,-1

-1,1

1,-1

-1,1

65

Dynamic games

If both players choose simultaneously, the games


extensive form is:
P1
H

P2
H

1 1

2 1

P2
T

1

1

1

1

1

1

66

Dynamic games

Entry Game

Two players: entrant and incumbent


The entrant chooses to enter or not. The incumbent fights
or accommodates.
Outcomes: (Out, Fight), (Out, A.), (Enter, Fight), (Enter, A.)
Payoffs: (0,2), (0,2), (-3,-1), (2,1)

67

Dynamic games

In extensive form:
Entrant
Out

E 0

I 2

In
Incumbent
Fight

Accommodate

2

1

68

Dynamic games

There are two Nash equilibria:

(Out, Fight if in)


(In, Accommodate if in)

However, (Out, Fight if in) tells the incumbent to


choose a suboptimal action if the entrant enters
The principle of sequential rationality requires
players to choose optimal actions at every point
in the game tree (even those that are not
reached in equilibrium)

69

Dynamic games

Applying the principle of sequential rationality,


players may anticipate what other players will
do, and choose accordingly
The entrant anticipates that the incumbent will
accommodate if entry occurs, and thus, it
chooses to enter
The principle of sequential rationality allows us
to analyze these games by means of backward
induction
70

Dynamic games

Now consider this game:


Entrant
Out

In
Incumbent

E 0

I 2

Entrant
F

3

1

3

1

71

Dynamic games

If entry occurs, the entrant will play Accommodate


Entrant
Out

E 0

I 2

In
Incumbent
A

Entrant
A
1

2

3

1

72

Dynamic games

And the incumbent will also play Accommodate


Entrant
Out

E 0

I 2

In
Incumbent
A
Entrant
A

3

1

73

Dynamic games

Hence, it is as if the entrant played this game:


Entrant
Out

E 0

I 2

In

3

1

and, of course, it will choose to enter.

74

Dynamic games

Thus, the entrant enters, anticipating that both


the incumbent and the entrant will play
Accommodate
Hence, (In, Accommodate; Accommodate) is a
Nash equilibrium. However, (Out,
Accommodate; Fight) is also a Nash equilibrium
The difference is that only the former satisfies
the principle of sequential rationality

75

Dynamic games: some definitions

A subgame is a subset of the game that begins


at a singleton information set. It is a game within
the game
A strategy profile s1 (.),..., s I (.) is a Subgame
Perfect (Nash) Equilibrium if it induces a Nash
Equilibrium in every possible subgame.
A SPE is always a NE. It adds the requirement
of satisfying the principle of sequential rationality

76

Dynamic games

Recall the game:


Entrant
Out

In
Incumbent

E 0

I 2

Entrant
F

3

1

77

Dynamic games

In the post-entry subgame, the unique Nash


Equilibrium is (Accommodate, Accommodate)
Any SPE must specify both players accommodating
if entry occurs
Hence, the unique SPE is (In, Accommodate;
Accommodate)

78

Vous aimerez peut-être aussi