Mixed Strategies

Prof. John Patty

1
(Non-)Existence of Pure Strategy Nash Equilibrium

The following game has no pure strategy Nash equilibrium.

H T
H 1, −1 −1, 1
T −1, 1 1, −1
Matching Pennies

2
Defining an Equilibrium in Matching Pennies

Two ways to think about an equilibrium in this type of game:

● Allowing players to randomize (e.g., “flip a coin”), and/or

● Representing players’ uncertainty about what the other player(s) will
do.

We accomplish both through the use of mixed strategies.

3
Mixed Strategies

Fix a game γ = (N, A, v).

A mixed strategy for player i ∈ N is simply a lottery over Si.

I will write a mixed strategy for i ∈ N as σi ∈ Σi ≡ ∆(Si).

The set of all profiles of mixed strategies is written as Σ ≡ ×i∈N Σi.

4
Mixed Strategies

Probability Player i ∈ N will choose pure strategy si, given σi, is written as
σi(si) = Pr[si ∣ σi].

Then, the support of a mixed strategy σi is written as

Supp(σi) ≡ {si ∈ Si ∶ σi(si) > 0}.

“Randomization” interpretation:
Supp(σi) is the set of pure strategies that might occur if i is using σi.

“Beliefs” interpretation:
Supp(σi) is the set of pure strategies others believe i might actually play.

5
Pure Strategies are Mixed Strategies, too

A pure strategy is a mixed strategy where the pure strategy is played with
probability 1.

For example, in the matching pennies game, the pure strategy s1 = H is

identical to

σ1 ≡ (Pr[s1 = H], Pr[s1 = T ]) = (1, 0) ∈ Σ1

and the pure strategy s1 = T is identical to

σ1 ≡ (Pr[s1 = H], Pr[s1 = T ]) = (0, 1) ∈ Σ1.

6
Mixed Strategies with Continuous Strategy Set (Technical)

When Si is a continuum, σi is a cumulative distribution function over Si.

This is a very technical area of game theory.

Nash equilibria can fail to exist even if we allow for mixed strategies.

I am happy to chat about this with you if you’re interested.

“In the course,” we will consider mixed strategies only in finite games.

7
Mixed Strategies as Beliefs

A belief by player i about aj for any other player j ∈ N is a lottery over Aj .

Thus, the set of beliefs about aj is simply Σj .

So, a mixed strategy σj is a belief about how j will play, and vice-versa.

Thus, Nash equilibrium is also referred to as an “equilibrium in beliefs.”

8
Mixed Strategies: Expected Payoffs

Expected payoff from pure strategy si ∈ Si, given σ−i ∈ Σ−i, in a finite game:

EVi(si, σ−i) = ∑ σ−i(s−i)vi(si, s−i).

s−i ∈S−i

Expected payoff from mixed strategy σi ∈ Σi:

EVi(σi, σ−i) = ∑ σi(si)vi(si, σ−i).

si ∈Si

9
Best Responses with Mixed Strategies

We defined the (pure strategy) best response correspondence as

BRi ∶ S−i ⇉ Si.

Now we extend this to the following:

BRi ∶ Σ−i ⇉ Σi.

10
When Can A Mixed Strategy Be A Best Response?

If σi ∈ Σi is a best response to σ−i ∈ Σ−i,

σi ∈ BRi(σ−i),
then

EVi(σi, σ−i) = ∑ σi(si)vi(si, σ−i) ≥ max [EVi(si, σ−i)]

si ∈Si si ∈Si

This implies that

si ∈ Supp(σi) ⇒ EVi(si, σ−i) = EVi(σi, σ−i).

In words, if i ∈ N is best responding to σ−i ∈ Σ−i, then

i is indifferent between all strategies i is playing with positive probability.

11
Nash Equilibrium with Mixed Strategies

Nash equilibrium with mixed strategies is defined essentially as before:

A strategy profile σ ∗ ∈ Σ is a Nash equilibrium if, for all i ∈ N ,

EVi(σi∗, σ−i) ≥ EVi(σi′ , σ−i) for all σi′ ∈ Σi.

Equivalently, σ ∗ ∈ Σ is a Nash equilibrium if

σ ∗ ∈ BR(σ ∗).

12
Matching Pennies

Consider the matching pennies game again.

H T
H 1, −1 −1, 1
T −1, 1 1, −1
Matching Pennies

Let x ∈ [0, 1] represent the probability Player 1 (row player) plays H:

σ1 ≡ (Pr[s1 = H], Pr[s1 = T ]) = (x, 1 − x),

and y ∈ [0, 1] represent the probability Player 2 (column player) plays H:

σ2 ≡ (Pr[s2 = H], Pr[s2 = T ]) = (y, 1 − y),

13
Matching Pennies: Solving for Nash Equilibrium

Solving for a mixed strategy Nash equilibrium in matching pennies:

● Write down Player 1’s expected payoff function for H,

● Write down Player 1’s expected payoff function for T ,
● Find Player 2’s mixed strategy, y, that sets these equal,
● Do the same thing with Player 2’s expected payoffs and x.

This approach works with any 2x2 game.

It is the basis of how you solve any finite game.

But it gets computationally much more complicated (i.e., takes longer).

14
Matching Pennies: Solving for Nash Equilibrium

First Player:
EV1(H, σ2) = EV1((1, 0), y),
= 1 ⋅ y + (−1) ⋅ (1 − y) ,
® ® ± ´¹¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¶
v1 (H,H) Pr[s2 =H] v1 (H,T ) Pr[s2 =T ]

= 2y − 1,

EV1(T, σ2) = EV1((0, 1), y),

= −1 ⋅ y + 1 ⋅ (1 − y) ,
¯ ® ® ´¹¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¶
v1 (T,H) Pr[s2 =H] v1 (T,T ) Pr[s =T ]
2

= 1 − 2y.
2y ∗ − 1 = 1 − 2y ∗ ⇒ y ∗ = 1/2
15
Matching Pennies: Solving for Nash Equilibrium

Second Player:
EV2(H, σ1) = EV2((1, 0), x),
= −1 ⋅ x + 1 ⋅ (1 − x) ,
¯ ® ® ´¹¹ ¹ ¹ ¹ ¸¹¹ ¹ ¹ ¹ ¶
v2 (H,H) Pr[s1 =H] v2 (T,H) Pr[s =T ]
1

= 1 − 2x,

EV2(T, σ1) = EV2((0, 1), x),

= 1 ⋅ x + (−1) ⋅ (1 − x) ,
® ® ± ´¹¹ ¹ ¹ ¹ ¸¹¹ ¹ ¹ ¹ ¶
v2 (H,T ) Pr[s1 =H] v2 (T,T ) Pr[s1 =T ]

= 2x − 1.
1 − 2x∗ = 2x∗ − 1 ⇒ x∗ = 1/2
16
Matching Pennies: Solving for Nash Equilibrium

Thus, we have the intuitive conclusion:

σ ∗ = (σ1∗ , σ2∗ ) = (( 1/2, 1/2), ( 1/2, 1/2))

is a mixed strategy Nash equilibrium of matching pennies.

Indeed, it is the unique Nash equilibrium of the game.

17
Matching Pennies: Solving for Nash Equilibrium with a Picture

1
BR2(x)

Nash
Equilibrium
y=Pr[s2=H]

BR1(y)

0
0 x=Pr[s1=H] 1

18
Mixed Strategy Example: Pure Coordination

L R
L 1, 1 0, 0
R 0, 0 1, 1
Pure Coordination Game

2 Pure Strategy Nash Equilibria: (L, L) and (R, R).

. . . and 1 more in Mixed Strategies: (( 1/2, 1/2), ( 1/2, 1/2)).

19
Mixed Strategy Example: Pure Coordination as a Picture

1
BR2(x)

Nash
Equilibria
y=Pr[s2=L]

BR1(y)

Nash
Equilibrium
0
0 x=Pr[s1=L] 1

20
Mixed Strategy Example: Battle of the Sexes

L R
L 2, 1 0, 0
R 0, 0 1, 2
Asymmetric Coordination Game

2 Pure Strategy Nash Equilibria: (L, L) and (R, R).

1 in Mixed Strategies:

((Pr[s1 = L], Pr[s1 = R]), (Pr[s2 = L], Pr[s2 = R])) = (( 2/3, 1/3), ( 1/3, 2/3)).

21
Mixed Strategy Example: Asymmetric Coordination as a Picture

1
BR2(x)
y=Pr[s2=L]

Nash
Equilibria

BR1(y)

Nash
Equilibrium
0
0 x=Pr[s1=L] 1

22
Mixed Strategy Example: Stag Hunt

S H
S 5, 5 0, 4
H 4, 0 3, 3
Risky Coordination Game

2 Pure Strategy Nash Equilibria: (S, S) and (H, H).

1 in Mixed Strategies:

((Pr[s1 = S], Pr[s1 = H]), (Pr[s2 = S], Pr[s2 = H])) = (( 3/4, 1/4), ( 3/4, 1/4)).

23
Mixed Strategy Example: Risky Coordination as a Picture

1
BR2(x) Nash
Equilibrium

BR1(y)
y=Pr[s2=L]

Nash
Equilibria

0
0 x=Pr[s1=L] 1

24
An Aside about Mixed Strategies

Pro: Nash equilibria exist for all finite games (Nash, 1950)

Con: Amplifies the multiplicity problem in some games

25
Mixed Strategies in General

Mixed strategies are difficult to think about/conceive.

And finding all mixed strategy equilibria in a finite game is complicated.

The point of them is theoretical, rather than practical.

We’ll see this again many times.

(E.g., See “semi-separating equilibrium” in the next couple of weeks...)

“Mixing,” also known as endogenous uncertainty, is one of game theory’s

most important powers for understanding social behavior.
26