ECO316: Applied Game Theory

Lecture 4

Christopher R. Dobronyi

University of Toronto

July 12, 2018

© 2018 by Christopher R. Dobronyi

Based on materials by Martin J. Osborne
Table of contents

Mixed Strategies

Mixed Equilibrium

Best Response Functions

Expected Utility

Mixed and Pure Equilibria

Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Strategic Game with no Nash Equilibrium

Player 2
L R
T 1, 0 0, 4
Player 1
B 0, 1 2, 0

No Nash equilibrium: What happens if this game is played?

I Player 1 might choose T sometimes and B sometimes
I Player 2 might choose L sometimes and R sometimes
I How often should each player choose each action?
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Pure and Mixed Strategies

A mixed strategy for a player in a strategic game is a probability

distribution over the set of possible actions for that player

A probability distribution over a (finite) set of possible actions

{a1 , . . . , an } is a list of numbers (p1 , . . . , pn ) such that
I p1 + · · · + pn = 1
I 0 ≤ pi ≤ 1, for all i = 1, . . . , n

A pure strategy for a player in a strategic game is a mixed

strategy that assigns a probability of 1 to a single action
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Example of Mixed Strategies

Player 2
L R
T 1, 0 0, 4
Player 1
B 0, 1 2, 0

Possible mixed strategies:

I Player 1: (p, 1 − p) in which p is the probability that she
chooses T and 1 − p is the probability that she chooses B
I Player 2: (q, 1 − q) in which q is the probability that she
chooses L and 1 − q is the probability that she chooses R
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Example of Mixed Strategies

Player 2
L R
T 1, 0 0, 4
Player 1
B 0, 1 2, 0

Possible pure strategies:

I Player 1:
I (1, 0): assign a probability of 1 to T
I (0, 1): assign a probability of 1 to B
I Player 2:
I (1, 0): assign a probability of 1 to L
I (0, 1): assign a probability of 1 to R
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Mixed Equilibrium: Optimal Actions

Player 2
L(q) R(1 − q)
T 1, 0 0, 4
Player 1
B 0, 1 2, 0

Optimal action for player 1 if player 2 uses the strategy (q, 1 − q)?
I Expected payoff of T : q · 1 + (1 − q) · 0 = q
I Expected payoff of B: q · 0 + (1 − q) · 2 = 2(1 − q)
2
T B ⇔ q > 2(1 − q) ⇔ q> 3
2
T ∼B ⇔ q = 2(1 − q) ⇔ q= 3
2
BT ⇔ q < 2(1 − q) ⇔ q< 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Mixed Equilibrium: Optimal Actions

Player 2
L R
T (p) 1, 0 0, 4
Player 1
B(1 − p) 0, 1 2, 0

Optimal action for player 2 if player 1 uses the strategy (p, 1 − p)?
I Expected payoff of L: p · 0 + (1 − p) · 1 = 1 − p
I Expected payoff of R: p · 4 + (1 − p) · 0 = 4p
1
LR ⇔ 1 − p > 4p ⇔ p< 5
1
L∼R ⇔ 1 − p = 4p ⇔ p= 5
1
RL ⇔ 1 − p < 4p ⇔ p> 5
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Mixed Equilibrium: Optimal Actions

Player 2
L(q) R(1 − q)
T (p) 1, 0 0, 4
Player 1
B(1 − p) 0, 1 2, 0

Player 1: Player 2:
2 1
T B ⇔ q> 3 LR ⇔ p< 5
2 1
T ∼B ⇔ q= 3 L∼R ⇔ p= 5
2 2
BT ⇔ q< 3 RL ⇔ p> 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Mixed Equilibrium: Equilibrium

Player 2
L(q) R(1 − q)
T (p) 1, 0 0, 4
Player 1
B(1 − p) 0, 1 2, 0

Player 1: Player 2:
2 1
p = 1 optimal ⇔ q > 3 q = 1 optimal ⇔ p < 5
2 1
any p optimal ⇔ q = 3 any q optimal ⇔ p = 5
2 2
p = 0 optimal ⇔ q < 3 q = 0 optimal ⇔ p > 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Mixed Equilibrium: Equilibrium

Player 2
L(q) R(1 − q)
T (p) 1, 0 0, 4
Player 1
B(1 − p) 0, 1 2, 0

Player 1: Player 2:
2 1
p = 1 optimal ⇔ q > 3 q = 1 optimal ⇔ p < 5
2 1
any p optimal ⇔ q = 3 any q optimal ⇔ p = 5
2 2
p = 0 optimal ⇔ q < 3 q = 0 optimal ⇔ p > 3

1 4 2 1



5, 5 , 3, 3 is a mixed strategy Nash equilibrium
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Mixed Strategy Nash Equilibrium

A profile of mixed strategies is a mixed strategy Nash

equilibrium of a strategic game if

every player’s strategy is optimal, given the other players’ strategies,

or, equivalently, if
no player can change her strategy to increase her
expected payoff, given the other players’ strategies
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

How to Find a Mixed Strategy Nash Equilibrium

Method: We can find every mixed strategy Nash equilibrium of

a strategic game by calculating the best response function
bi (α−i ) for each player i and identifying all of the profiles of mixed
strategies α such that

αi = bi (α−i ), for all players i

Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Examples of Best Response Functions: Player 1

Player 2
L(q) R(1 − q)
T (p) 1, 0 0, 4
Player 1
B(1 − p) 0, 1 2, 0

We already found the best response functions for this game:

2
p = 1 optimal ⇔ q > 3
2
any p optimal ⇔ q = 3
2
p = 0 optimal ⇔ q < 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Examples of Best Response Functions: Player 1

Player 2
L(q) R(1 − q)
T (p) 1, 0 0, 4
Player 1
B(1 − p) 0, 1 2, 0

We already found the best response functions for this game:


{1},
 if q > 23 ,
b1 (q) = {p : 0 ≤ p ≤ 1}, if q = 23 ,

{0}, if q < 23

Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Examples of Best Response Functions: Player 2

Player 2
L(q) R(1 − q)
T (p) 1, 0 0, 4
Player 1
B(1 − p) 0, 1 2, 0

We already found the best response functions for this game:

1
q = 1 optimal ⇔ p < 5
1
any q optimal ⇔ p = 5
1
q = 0 optimal ⇔ p > 5
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Examples of Best Response Functions: Player 2

Player 2
L(q) R(1 − q)
T (p) 1, 0 0, 4
Player 1
B(1 − p) 0, 1 2, 0

We already found the best response functions for this game:


{1},
 if p < 15 ,
b2 (p) = {q : 0 ≤ q ≤ 1}, if p = 51 ,

{0}, if p > 15

Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Example of Best Response Functions: Plot

q
1

2
(p ∗ , q ∗ )
3
b1

b2

p
0 1 1
5
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Summary of Example

Player 2
L(q) R(1 − q)
T (p) 1, 0 0, 4
Player 1
B(1 − p) 0, 1 2, 0

Summary:
I No pure strategy Nash equilibria
1 4 2 1



I Unique mixed strategy Nash equilibrium at 5, 5 , 3, 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Expected Payoffs

Payoffs are not purely ordinal if players maximize expected payoffs

Example: The following functions have the same order

u(a) = 0, u(b) = 1, u(c) = 4
v (a) = 0, v (b) = 3, v (c) = 4
but represent different preferences:
1 1
2 u(a) + 2 u(c) = 2 > 1 = u(b)
1 1
2 v (a) + 2 v (c) = 2 < 3 = v (b)
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Expected Payoffs: Affine Transformations

The expected values of the payoff functions u(x) and v (x)
represent the same preferences over “lotteries” if and only if
v (x) = αu(x) + β, for some α > 0 and β

Example: The following functions represent different preferences

u(a) = 0, u(b) = 1, u(c) = 4
v (a) = 0, v (b) = 3, v (c) = 4
because
v (a) = αu(a) + β ⇒ 0 = α · 0 + β ⇒ β = 0
v (b) = αu(b) + β ⇒ 3 = α · 1 + β ⇒ α = 3
v (c) = αu(c) + β ⇒ 4 = α · 4 + β ⇒ α = 1
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Expected Payoffs: Affine Transformations

The expected values of the payoff functions u(x) and v (x)
represent the same preferences over “lotteries” if and only if
v (x) = αu(x) + β, for some α > 0 and β

Example: The following functions represent same preferences

u(a) = 0, u(b) = 1, u(c) = 4
v (a) = 1, v (b) = 3, v (c) = 9
because
v (a) = αu(a) + β ⇒ 1 = α · 0 + β ⇒ β = 1
v (b) = αu(b) + β ⇒ 3 = α · 1 + β ⇒ α = 2
v (c) = αu(c) + β ⇒ 9 = α · 4 + β ⇒ α = 2
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Back or Stravinsky: Optimal Actions

Player 2
B(q) S(1 − q)
B 2, 1 0, 0
Player 1
S 0, 0 1, 2

Optimal action for player 1 if player 2 uses the strategy (q, 1 − q)?
I Expected payoff of B: q · 2 + (1 − q) · 0 = 2q
I Expected payoff of S: q · 0 + (1 − q) · 1 = 1 − q
1
BS ⇔ 2q > 1 − q ⇔ q> 3
1
B∼S ⇔ 2q = 1 − q ⇔ q= 3
1
S B ⇔ 2q < 1 − q ⇔ q< 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Back or Stravinsky: Optimal Actions

Player 2
B S
B(p) 2, 1 0, 0
Player 1
S(1 − p) 0, 0 1, 2

Optimal action for player 2 if player 1 uses the strategy (p, 1 − p)?
I Expected payoff of B: p · 1 + (1 − p) · 0 = p
I Expected payoff of S: p · 0 + (1 − p) · 2 = 2(1 − p)
2
BS ⇔ p > 2(1 − p) ⇔ p> 3
2
B∼S ⇔ p = 2(1 − p) ⇔ p= 3
2
S B ⇔ p < 2(1 − p) ⇔ p< 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Back or Stravinsky: Optimal Actions

Player 2
B(q) S(1 − q)
B(p) 2, 1 0, 0
Player 1
S(1 − p) 0, 0 1, 2

Player 1: Player 2:
1 2
BS ⇔ q> 3 BS ⇔ p> 3
1 2
B∼S ⇔ q= 3 B∼S ⇔ p= 3
1 2
S B ⇔ q< 3 S B ⇔ p< 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Back or Stravinsky: Equilibrium

Player 2
B(q) S(1 − q)
B(p) 2, 1 0, 0
Player 1
S(1 − p) 0, 0 1, 2

Player 1: Player 2:
1 2
p = 1 optimal ⇔ q > 3 q = 1 optimal ⇔ p > 3
1 2
any p optimal ⇔ q = 3 any q optimal ⇔ p = 3
1 2
p = 0 optimal ⇔ q < 3 q = 0 optimal ⇔ p < 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Back or Stravinsky: Equilibrium

Player 2
B(q) S(1 − q)
B(p) 2, 1 0, 0
Player 1
S(1 − p) 0, 0 1, 2

Player 1: Player 2:
1 2
p = 1 optimal ⇔ q > 3 q = 1 optimal ⇔ p > 3
1 2
any p optimal ⇔ q = 3 any q optimal ⇔ p = 3
1 2
p = 0 optimal ⇔ q < 3 q = 0 optimal ⇔ p < 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Back or Stravinsky: Equilibrium

Player 2
B(q) S(1 − q)
B(p) 2, 1 0, 0
Player 1
S(1 − p) 0, 0 1, 2

Player 1: Player 2:
1 2
p = 1 optimal ⇔ q > 3 q = 1 optimal ⇔ p > 3
1 2
any p optimal ⇔ q = 3 any q optimal ⇔ p = 3
1 2
p = 0 optimal ⇔ q < 3 q = 0 optimal ⇔ p < 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Back or Stravinsky: Equilibrium

Player 2
B(q) S(1 − q)
B(p) 2, 1 0, 0
Player 1
S(1 − p) 0, 0 1, 2

Player 1: Player 2:
1 2
p = 1 optimal ⇔ q > 3 q = 1 optimal ⇔ p > 3
1 2
any p optimal ⇔ q = 3 any q optimal ⇔ p = 3
1 2
p = 0 optimal ⇔ q < 3 q = 0 optimal ⇔ p < 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Bach of Stravinsky: Plot of Best Response Functions

q
1

b2

b1
1
3

p
0 2 1
3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Bach or Stravinsky: Summary

Player 2
B(q) S(1 − q)
B(p) 2, 1 0, 0
Player 1
S(1 − p) 0, 0 1, 2

Summary:
I Two pure strategy Nash equilibria:

{(0, 1) , (0, 1)} and {(1, 0) , (1, 0)}

I Three mixed strategy Nash equilibria:

2 1 1 2



{(0, 1) , (0, 1)} , {(1, 0) , (1, 0)} and 3, 3 , 3, 3
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Existence of a Mixed Strategy Nash Equilibrium

I There exists at least one mixed strategy Nash equilibrium

in both of the examples that we have considered so far
I Same is true for any game with finitely many actions
I Main result of John Nash’s Ph.D. thesis:

Every strategic game in which every player has finitely many

actions has a mixed strategy Nash equilibrium
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Another Example: Optimal Actions

Player 2
L(q) R(1 − q)
T 0, 1 0, 2
Player 1
B 2, 2 0, 1

Optimal action for player 1 if player 2 uses the strategy (q, 1 − q)?
I Expected payoff of T : q · 0 + (1 − q) · 0 = 0
I Expected payoff of B: q · 2 + (1 − q) · 0 = 2q

T B ⇔ 0 > 2q ⇔ q<0
T ∼B ⇔ 0 = 2q ⇔ q=0
BT ⇔ 0 < 2q ⇔ q>0
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Another Example: Optimal Actions

Player 2
L R
T (p) 0, 1 0, 2
Player 1
B(1 − p) 2, 2 0, 1

Optimal action for player 2 if player 1 uses the strategy (p, 1 − p)?
I Expected payoff of L: p · 1 + (1 − p) · 2 = 2 − p
I Expected payoff of R: p · 2 + (1 − p) · 1 = 1 + p
1
LR ⇔ 2−p >1+p ⇔ p< 2
1
L∼R ⇔ 2−p =1+p ⇔ p= 2
1
RL ⇔ 2−p <1+p ⇔ p> 2
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Another Example: Optimal Actions

Player 2
L(q) R(1 − q)
T (p) 0, 1 0, 2
Player 1
B(1 − p) 2, 2 0, 1

Player 1: Player 2:
1
T B ⇔ q<0 LR ⇔ p< 2
1
T ∼B ⇔ q=0 L∼R ⇔ p= 2
1
BT ⇔ q>0 RL ⇔ p> 2
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Another Example: Equilibrium

Player 2
L(q) R(1 − q)
T (p) 0, 1 0, 2
Player 1
B(1 − p) 2, 2 0, 1

Player 1: Player 2:
1
p = 1 optimal ⇔ q < 0 q = 1 optimal ⇔ p < 2
1
any p optimal ⇔ q = 0 any q optimal ⇔ p = 2
1
p = 0 optimal ⇔ q > 0 q = 0 optimal ⇔ p > 2
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Another Example: Equilibrium

Player 2
L(q) R(1 − q)
T (p) 0, 1 0, 2
Player 1
B(1 − p) 2, 2 0, 1

Player 1: Player 2:
1
p = 1 optimal ⇔ q < 0 q = 1 optimal ⇔ p < 2
1
any p optimal ⇔ q = 0 any q optimal ⇔ p = 2
1
p = 0 optimal ⇔ q > 0 q = 0 optimal ⇔ p > 2
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Another Example: Equilibrium

Player 2
L(q) R(1 − q)
T (p) 0, 1 0, 2
Player 1
B(1 − p) 2, 2 0, 1

Player 1: Player 2:
1
p = 1 optimal ⇔ q < 0 q = 1 optimal ⇔ p < 2
1
any p optimal ⇔ q = 0 any q optimal ⇔ p = 2
1
p = 0 optimal ⇔ q > 0 q = 0 optimal ⇔ p > 2
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Another Example: Equilibrium

Player 2
L(q) R(1 − q)
T (p) 0, 1 0, 2
Player 1
B(1 − p) 2, 2 0, 1

Player 1: Player 2:
1
p = 1 optimal ⇔ q < 0 q = 1 optimal ⇔ p < 2
1
any p optimal ⇔ q = 0 any q optimal ⇔ p = 2
1
p = 0 optimal ⇔ q > 0 q = 0 optimal ⇔ p > 2
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Another Example: Plot of Best Response Functions

q
1

b1 b2

p
0 1 1
2
Mixed Strategies Mixed Equilibrium Best Response Functions Expected Utility Mixed and Pure Equilibria

Another Example: Summary

Player 2
L(q) R(1 − q)
T (p) 0, 1 0, 2
Player 1
B(1 − p) 2, 2 0, 1

Summary:
I Two pure strategy Nash equilibria:

{(0, 1) , (1, 0)} and {(1, 0) , (0, 1)}

I Infinite mixed strategy Nash equilibria:

1

{(0, 1) , (1, 0)} and (p, 1 − p) , (0, 1) : 2 ≤p≤1