Vous êtes sur la page 1sur 11

Chapter 4

Mixed Strategy
aTwo kind of strategies:
`pure `mixed

Mixed Strategies and Mixed Strategy Equilibrium


1

aTwo kinds of equilibrium


`pure strategy `mixed strategy

aTwo games with mixed strategy equilibria:


`Matching Pennies `Market Niche
2

Matching Pennies: The payoff matrix (All payoffs in cents)


Player 2 Player 1 Heads Heads Tails

Matching Pennies: No equilibrium in pure strategies


Player 2 Player 1 Heads All Best Responses are underlined. Heads Tails

+1, -1

-1, +1

+1, -1

-1, +1

Tails

-1, +1

+1, -1

Tails

-1, +1

+1, -1

Computing Mixed Strategy Equilibria in 22 Games


aSolution criterion: each pure strategy in a mixed strategy equilibrium pays the same at equilibrium aEach pure strategy not in a mixed strategy equilibrium pays less aDetailed calculations for Matching Pennies and Market Niche aAn appealing condition on equilibria: payoff dominance
5

Matching Pennies: What about mixed strategies?


probability 1 probability x H 2 y h +1, -1 1- y t -1, +1

1- x

-1, +1

+1, -1

x, y between 0 and 1 That is, 0 x 1 and 0 y 1


6

Need to calculate player 1s expected utility from player 2s mixed strategy


probability 1 2 h H +1, -1 t -1, +1 EU1: 2y - 1 y 1- y

Need to calculate player 2s expected utility from player 1s mixed strategy


2 1 probability x H h +1, -1 t -1, +1

-1, +1

+1, -1

1 - 2y

1- x

-1, +1

+1, -1

EU1(H) = y 1 + (1- y) -1 = 2y - 1 EU1(T) = y -1 + (1- y) 1 = 1 - 2y


7

1 - 2x 2x - 1 EU2: EU2(h) = x -1 + (1- x) 1 = 1 - 2x EU2(t) = x 1 + (1- x) -1 = 2x - 1


8

In equilibrium, Player 1 is willing to randomize only when he is indifferent between H and T


EU1(H) = y 1 + (1- y) -1 = 2y - 1 EU1(T) = y -1 + (1- y) 1 = 1 - 2y

Similarly, Player 2 is willing to randomize only when she is indifferent between h and t
Player 1s Conditions: EU1(H) = EU1(T) Player 2s Conditions: EU2(h) = x -1 + (1- x) 1 = 1 - 2x EU2(t) = x 1 + (1- x) -1 = 2x - 1

In equilibrium: EU1(H) = EU1(T)


2y - 1 = 1 - 2y 4y = 2 y= 1-y=1-=

In equilibrium: EU2(h) = EU2(t)



9

1 - 2x = 2x - 1 x= and 1 - x = 1 - =
10

y=1-y=

x=1-x=

Matching Pennies: Equilibrium in mixed strategies


probability 1 probability H 2 h +1, -1 t -1, +1 EU1: 0 || T -1, +1 +1, -1 0

Mixed strategies are not intuitive: You randomize to make me indifferent.


Row randomizes to make Column indifferent. Column randomizes to make Row indifferent. Then each is playing a best response to the other.
11 12

EU2: 0 = 0 Each is playing a best response to the other!

Market Niche: The payoff matrix


Firm 2 Firm 1 Enter Enter Stay Out

Market Niche: Two pure strategy equilibria


Firm 2 Firm 1 Enter Mutual best responses form an equilibrium. Enter Stay Out

-50, -50

100, 0

-50, -50

100, 0

Stay Out

0, 100

0, 0

Stay Out

0, 100

0, 0

13

14

Market Niche: What about mixed strategies?


probability 1 probability x E 2 y e -50, -50 1- y

Need to calculate firm 1s expected utility from firm 2s mixed strategy


probability 1 2 e -50, -50 y 1- y s 100, 0 EU1: 100 - 150y

s 100, 0 E

1- x

0, 100

0, 0

0, 100

0, 0

x, y between 0 and 1 That is, 0 x 1 and 0 y 1


15

EU1(E) = y -50 + (1- y) 100 = 100 - 150y EU1(S) = y 0 + (1- y) 0 = 0


16

Need to calculate firm 2s expected utility from firm 1s mixed strategy


1 probability x 1- x 2 e -50, -50 s 100, 0

In equilibrium, Firm 1 is willing to randomize only when it is indifferent between E and S


EU1(E) = y -50 + (1- y) 100 = 100 - 150y EU1(S) = y 0 + (1- y) 0 = 0

In equilibrium: EU1(E) = EU1(S)


100 - 150y = 0 150y = 100 y = 2/3 1 - y = 1 - 2/3 = 1/3

S EU2:

0, 100 100-150x

0, 0 0

EU2(e) = x -50 + (1- x) 100 = 100 - 150x EU2(s) = x 0 + (1- x) 0 = 0


17

y = 2/3 and 1 - y = 1/3


18

Similarly, Firm 2 is willing to randomize only when it is indifferent between h and t


Firm 1s Conditions: EU1(E) = EU1(S) Firm 2s Conditions: EU2(e) = x -50 + (1- x) 100 = 100 - 150x EU2(s) = x 0 + (1- x) 0 = 0

Market Niche: Equilibrium in mixed strategies


probability 1 probability 2

2/3
e -50, -50

1/3
s 100, 0 EU1: 0 ||

2/3 1/3

In equilibrium: EU2(e) = EU2(s)


100 - 150x = 0 150x = 100

0, 100

0, 0

x = 2/3 and 1 - x = 1/3

19

EU2: 0 = 0 Each firm is playing a best response to the other!

20

Mixed Strategies and bluffing: Liars Poker


aMixed strategies as a way to be unpredictable aBluffing and mixed strategies aLiars poker, a game where bluffing pays

Liars Poker: extensive form


Call Says ace Ace 1/2 0 1/2 King 1 Says king
21

1, -1 0.5, -0.5 -1, 1 0.5, -0.5

Fold 2 Call Fold

Says ace

0, 0
22

Liars Poker: normal form


1 2 Call Say A when K Fold

Liars Poker: No pure strategy equilibrium


1 2 Call Say A when K Fold

0, 0

0.5, -0.5

0, 0

0.5, -0.5

Say K when K

0.5, -0.5

0.25, -0.25

Say K when K

0.5, -0.5

0.25, -0.25

23

24

Liars Poker: What about mixed strategies?


probability 1 probability x A when K 2 y c 0, 0 1- y f 0.5, -0.5

Each player calculates his expected utility from others mixed strategy
probability 1 probability x 2 y c 0, 0 1- y f 0.5, -0.5 EU1: 0.5 - 0.5y

A when K

1- x K when K

0.5, -0.5

0.25, -0.25

1- x K when K 0.5, -0.5 EU2:


25

0.25, -0.25 0.25 + 0.25y

x, y between 0 and 1 That is, 0 x 1 and 0 y 1

0.5x - 0.5 -0.25x - 0.25


26

In equilibrium, player 1 is willing to randomize only when he is indifferent between A and K


EU1(A) = y 0 + (1- y) 0.5 = 0.5 - 0.5y EU1(K) = y 0.5 + (1- y) 0.25 = 0.25 + 0.25y

Similarly, Player 2 is willing to randomize only when she is indifferent between c and f
Player 1s Conditions: EU1(A) = EU1(K) Player 2s Conditions: EU2(c) = x 0 + (1- x) -0.5 = 0.5x - 0.5 EU2(f) = x -0.5 + (1- x) -0.25 = -0.25x - 0.25

In equilibrium: EU1(A) = EU1(K)


0.5 - 0.5y = 0.25 + 0.25y 0.75y = 0.25 y = 1/3 1 - y = 1 - 1/3 = 2/3

In equilibrium: EU2(c) = EU2(f)



27

0.5x - 0.5 = -0.25x - 0.25 0.75x = 0.25

y = 1/3 and 1 - y = 2/3

x = 1/3 and 1 - x = 2/3

28

Liars Poker: Equilibrium in mixed strategies


probability 1 probability 2

Mixed Strategy Equilibria of Coordination Games and Coordination Problems aGames with mixed strategy equilibria which cannot be detected by the arrow diagram aThe mixed strategy equilibrium of Video System Coordination is not efficient

1/3
c 0, 0

2/3
f 0.5, -0.5 EU1: 1/3 ||

1/3

2/3

0.5, -0.5

0.25, -0.25

1/3

EU2: -1/3 = -1/3 Each player is playing a best response to the other!

29

30

Correlated Equilibrium
aMixed strategy Nash equilibria tend to have low efficiency aCorrelated equilibria
`public signal `Nash equilibrium in game that follows

Asymmetric Mixed Strategy Equilibria


aMaking a game asymmetric often makes its mixed strategy equilibrium asymmetric aAsymmetric Market Niche is an example

31

32

Asymmetrical Market Niche: The payoff matrix


Firm 2 Firm 1 Enter Enter Stay Out

Asymmetrical Market Niche: Two pure strategy equilibria


Firm 2 Firm 1 Enter Enter Stay Out

-50, -50

150, 0

-50, -50

150, 0

Stay Out

0, 100

0, 0

Stay Out

0, 100

0, 0

33

34

Asymmetrical Market Niche: What about mixed strategies?


probability 1 probability x E 2 y e -50, -50 1- y s 150, 0

Need to calculate each firms expected utility from the firms mixed strategy
probability 1 probability x 2 e -50, -50 y 1- y s 150, 0 EU1: 150 - 200y

1- x

0, 100

0, 0

1- x

S EU2:

0, 100 100 - 150x

0, 0 0

x, y between 0 and 1 That is, 0 x 1 and 0 y 1


35

36

In equilibrium, Firm 1 is willing to randomize only when it is indifferent between E and S


EU1(E) = y -50 + (1- y) 150 = 150 - 200y EU1(S) = y 0 + (1- y) 0 = 0

Similarly, Firm 2 is willing to randomize only when it is indifferent between h and t


Firm 1s Conditions: EU1(E) = EU1(S) Firm 2s Conditions: EU2(e) = x -50 + (1- x) 100 = 100 - 150x EU2(s) = x 0 + (1- x) 0 = 0

In equilibrium: EU1(E) = EU1(S)


150 - 200y = 0 200y = 150 y = 3/4 1 - y = 1- 3/4 = 1/4

In equilibrium: EU2(e) = EU2(s)



37

100 - 150x = 0 150x = 100

y = 3/4 and 1 - y = 1/4

x = 2/3 and 1 - x = 1/3

38

Asymmetrical Market Niche: Equilibrium in mixed strategies


probability 1 probability 2

Asymmetrical Market Niche: Equilibrium in mixed strategies


Although the two pure strategy equilibria (E,s) and (S,e) did not change in Asymmetrical Market Niche, the mixed strategies equilibrium did change.

3/4
e -50, -50

1/4
s 150, 0 EU1: 0 ||

2/3 1/3

0, 100

0, 0

EU2: 0 = 0 Each firm is playing a best response to the other!

39

40

Chicken
aTwo drivers race toward a cliff aStrategy choice:
`swerve `straight ahead

Chicken: The payoff matrix


player 1 player 2 drive straight ahead swerve

aMore general version of the game:


`back down `do not back down

drive straight ahead

-10, -10

1, -1

aSolution as in Market Niche Game


41

swerve

-1, 1

0, 0

42

Chicken: strategy for player 1


player 1 player 2 drive straight ahead swerve

Chicken: strategy for player 2


player 1 player 2 drive straight ahead swerve

drive straight ahead

-10, -10

1, -1

drive straight ahead

-10, -10

1, -1

swerve

-1, 1

0, 0

swerve

-1, 1

0, 0

43

44

Chicken: two pure strategy Nash equilibria


player 2 drive straight player 1 ahead drive straight ahead swerve

Chicken: The payoff matrix


player 2 probability y player 1 straight probability straight -10, -10 x 1- y swerve 1, -1

-10, -10

1, -1

1- x swerve -1, 1 0, 0

swerve

-1, 1

0, 0

x, y between 0 and 1
45

That is, 0 x 1 and 0 y 1

46

Chicken: The payoff matrix


player 2 probability y player 1 straight probability straight x -10, -10 1- y swerve 1, -1 EU1 1 11y

In equilibrium, player 1 is willing to randomize only when she is indifferent between swerve and straight

EU1(straight) = y (-10) + (1-y) 1 = 1 11y EU1(swerve) = y (-1) + (1- y) 0 = -y

In equilibrium: EU1(swerve) = EU1(straight)


1 11y = - y 1 = 10y y = 1/10 1 - y = 1 - 1/10 = 9/3

1- x

swerve

-1, 1

0, 0

-y

EU2

1 11x

-x
47

y = 1/10 and 1 - y = 9/10


48

Similarly, player 2 is willing to randomize only when he is indifferent between swerve and straight
Player 1s Conditions: EU1(swerve) = EU1(straight) Player 2s Conditions: EU2(straight) = x (-10) + (1-x) 1 = 1 11x EU2(swerve) = x (-1) + (1- x) 0 = -x

Chicken: The payoff matrix


player 2 probability 1/10 player 1 straight probability 1/10 straight -10, -10 9/10 swerve 1, -1 EU1 -0.1

In equilibrium: EU2(swerve) = EU2(straight)


1 11x = - x x = 1/10 9/10 swerve -1, 1 0, 0 -0.1

x = 1/10 and 1 - x = 9/10

EU2
49

-0.1

-0.1
50

Everyday Low Prices


aSales are mixed strategies aSears marketing campaign to do away with sales, called Everyday Low Prices aTwo types of buyers:
`informed `uninformed

Everyday Low Pricing: The payoff matrix


Retailer 2 Retailer 1 Normal price np 7500, 7500 Sale price sp 7500, 8500

NP

aA mixed strategy equilibrium tells how often to run sales


51

SP

8500, 7500

5500, 5500

52

Everyday Low Pricing: Two pure strategy equilibria


Retailer 2 Retailer 1 np 7500, 7500 sp

Everyday Low pricing: What about mixed strategies?


probability 1 probability 2 y np 1- y sp

NP

7500, 8500

NP 7500, 7500 7500, 8500

1- x SP 8500, 7500 5500, 5500

SP 8500, 7500 5500, 5500 x, y between 0 and 1

53

That is, 0 x 1 and 0 y 1

54

Each retailer calculates its expected utility from others mixed strategy
probability 1 probability x 2 y np 1- y sp EU1 7500

In equilibrium, Retailer 1 is willing to randomize only when it is indifferent between NP and SP


EU1(NP) = y 7500 + (1- y) 7500 = 7500 EU1(SP) = y 8500 + (1- y) 5500 = 3000y + 5500

In equilibrium: EU1(NP) = EU1(SP)


7500 = 3000y + 5500 3000y = 2000 y = 2/3 1 - y = 1 - 2/3 = 1/3

NP 7500, 7500 7500, 8500

1- x

SP 8500, 7500 5500, 5500 3000y + 5500 EU2: 7500 3000x + 5500
55

y = 2/3 and 1 - y = 1/3


56

Similarly, Retailer 2 is willing to randomize only when it is indifferent between c and f


Retailer 1s Conditions: EU1(SP) = EU1(NP) Retailer 2s Conditions: EU2(np) = x 7500 + (1- x) 7500 = 7500 EU2(sp) = x 8500 + (1- x) 5500 = 3000x + 5500

Everyday Low Pricing: Equilibrium in mixed strategies


probability

2/3
1 probability 2 np

1/3
sp EU1: 7500 ||

2/3 1/3

NP 7500, 7500 7500, 8500

In equilibrium: EU2(np) = EU2(sp)


7500 = 3000x + 5500 3000x = 200

SP 8500, 7500 5500, 5500

7500

x = 2/3 and 1 - x = 1/3

57

EU2: 7500 = 7500 Each player is playing a best response to the other!

58

Mixed strategies are not intuitive: You randomize to make me indifferent.


R E M I N D E R

Appendix: Bluffing in 1-card Stud Poker


aA version of poker with 3 kinds of cards (ace, king, and queen), 1-card hands, and players who see their cards aFor some ratios of the ante to the bet, 1card stud poker has a unique equilibrium which is in mixed strategies aEquilibrium play in poker usually calls for some bluffing aThe solution of poker has all players 60 breaking even

Row randomizes to make Column indifferent. Column randomizes to make Row indifferent. Then each is playing a best response to the other.
59

One-card Stud Poker Payoff matrix, player 1


Player 2 I: Bet AKQ II: Bet AK III: Bet AQ IV: Bet A Player 1 (4a-2b)/9, (a-2b)/9, I: Bet AKQ 0, 0 3a/9, -3a/9 (2b-4a)/9 (2b-a)/9 II: Bet AK (2b-a)/9, (a-2b)/9 -3a/9, 3a/9 (2b-4a)/9, (4a-2b)/9 0, 0 -(a+b)/9, (a+b)/9 (b-2a)/9, (2a-b)/9 (a+b)/9, -(a+b)/9 0, 0 (b-2a)/9, (2a-b)/9 (2a-b)/9, (b-2a)/9 (2a-b)/9, (b-2a)/9 0, 0
61

One-card Stud Poker. Payoff matrix, player 1, a=$1, b=$1


Player 2 I: Bet AKQ II: Bet AK III: Bet AQ IV: Bet A Player 1 I: Bet AKQ 0, 0 -1/9, 1/9 3/9, -3/9 2/9, -2/9

II: Bet AK

1/9, -1/9

0, 0

2/9, -2/9

1/9, -1/9 1/9, -1/9

III: Bet AQ

III: Bet AQ

-3/9, 3/9

-2/9, 2/9

0, 0

IV: Bet A

IV: Bet A

-2/9, 2/9

-1/9, 1/9

-1/9, 1/9

0, 0
62

One-card Stud Poker. Payoff matrix, player 1, a=$1 b=$2


Player 2 I: Bet AKQ II: Bet AK III: Bet AQ IV: Bet A Player 1 I: Bet AKQ 0, 0 -3/9, 3/9 3/9, -3/9 0, 0

II: Bet AK

3/9, -3/9

0, 0

3/9, -3/9

0, 0

III: Bet AQ

-3/9, 3/9 0, 0

-3/9, 3/9

0, 0 0, 0

0, 0

IV: Bet A

0, 0

0, 0
63