CUR 412: Game Theory and its Applications,

Lecture 6

Prof. Ronaldo CARPIO

April 5, 2016

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Announcements

▸ HW #2 is due next week.

▸ I will return HW #1 next week.
▸ The midterm will be in class on April 19. It will cover Chapters 1-4
(only the sections that we’ve gone over in lectures).
▸ Midterm will be closed-book. No programmable calculators or
smartphones allowed.
▸ Previous midterms and solutions are on the course website.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Review of Last Week

▸ We introduced the concept of a mixed strategy: a probability

distribution over a player’s set of actions.
▸ Instead of players choosing an action, they choose a probability
distribution over their set of actions.
▸ Players rank outcomes based on their expected payoff, using the
probability distributions generated by all players’ mixed strategies.
▸ A mixed strategy Nash equilibrium (MSNE) is a mixed strategy
profile where no player can get a higher expected payoff by
unilaterally changing his mixed strategy.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Review of Last Week

▸ A pure strategy is a mixed strategy where one action has probability

1, and all other actions have probability 0.
▸ This is equivalent to our previous concept of strategy in games
without randomization.
▸ A game with mixed strategies can have more NE than the
corresponding game without randomization (i.e. if we only allow
pure strategies).
▸ We’ll see that NE of the non-random game are a subset of the NE
when mixed strategies are allowed.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Review of Last Week

▸ Player i’s best response function is similar as before; now it gives

the set of mixed strategies (i.e. probability distributions) that give
the highest expected payoff, conditional on the mixed strategies of
all other players α−i .
▸ In 2-player, 2-action games, we can calculate Player i’s expected
payoff to each of his actions (or equivalently, pure strategies).
Suppose Player i’s actions are {a, b}.
▸ If Player i’s expected payoff to action a, denoted Ei (a), is
strictly greater than Ei (b), then the best response is the pure
strategy (α(a) = 1, α(b) = 0).
▸ If Ei (a) = Ei (b), then any mixed strategy
(α(a) = p, α(b) = 1 − p) for 0 ≤ p ≤ 1 is a best response.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Example: BoS

B S
B 2,1 0,0
S 0,0 1,2

▸ Suppose p, q are the probabilities assigned to B by Player 1 and

Player 2, respectively.
▸ E1 (B, α2 ) = 2 ⋅ q + 0 ⋅ (1 − q) = 2q
▸ E2 (S, α2 ) = 0 ⋅ q + 1 ⋅ (1 − q) = 1 − q
▸ If 2q > 1 − q → q > 1/3, unique best response is pure strategy B
▸ If q < 1/3, unique best response is pure strategy S
▸ If q = 1/3, all mixed strategies are best responses

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
BoS: Best Response Function

⎧
⎪{0} if q < 1/3
⎪
⎪
⎪
B1 (q) = ⎨{p ∶ 0 ≤ p ≤ 1} if q = 1/3
⎪
⎪
⎪
⎪
⎩{1} if q > 1/3

▸ There are two Nash equilibria in pure strategies, and one new
equilibrium in mixed strategies.
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
The ”Colonel Blotto” game

▸ There are two players, the Attacker and the Defender.

▸ Each player has two armies.
▸ There are two locations being defended. Each player allocates his 2
armies to the 2 locations.
▸ At each locations, the Defender wins if he has at least as many
armies as the Attacker.
▸ The Defender wins the game if he wins at both locations.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
The ”Colonel Blotto” game

(0,2) (1,1) (2,0)

(0,2) -1,1 1,-1 1,-1
(1,1) 1,-1 -1,1 1,-1
(2,0) 1,-1 1,-1 -1,1

▸ Player 1 (the row player) is the Attacker.

▸ In each row and column, there is an outcome where Attacker wins,
and an outcome where Defender wins.
▸ Therefore, there is no pure strategy NE, since the loser can always
switch actions and become the winner.
▸ This can be generalized to any number of locations and number of
armies for each player.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
A Useful Characterization of Mixed Strategy NE

▸ So far, we’ve found mixed strategy NE by constructing best

response functions.
▸ For more complicated games, this is too difficult. We’ll state a
condition that must hold true at any mixed strategy NE.
▸ A player’s expected payoff to the mixed strategy α is a weighted
average of his expected payoffs to playing each action:

Ui (α) = Σai αi (ai )Ei (ai , α−i )

▸ Ei (ai , α−i ) is the expected payoff of playing action ai , when the

other players use the mixed strategies α−i .

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
A Useful Characterization of Mixed Strategy NE

Ui (α) = Σai αi (ai )Ei (ai , α−i )

▸ The value of a weighted average (with positive weights) must lie

between the highest and lowest values.
▸ Suppose α∗ is a mixed strategy NE, and Player i gets expected
payoff Ei∗ in this equilibrium.
▸ This must be in between the highest and lowest expected payoffs to
actions that have a positive probability in αi∗ .
▸ Player i’s expected payoff to all strategies (mixed and pure) is at
most, Ei∗ , by definition of MSNE.
▸ The highest expected payoff to an action cannot be higher than Ei∗ ,
and the lowest cannot be lower.
▸ Therefore, all actions that have a positive probability in αi∗ must
have the same expected payoff, equal to Ei∗ .
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Characterization of Mixed Strategy NE of a Finite Game

▸ Proposition: A mixed strategy profile α∗ is a mixed strategy Nash

equilibrium if and only if, for each player i:
▸ The expected payoff (given other players’ strategies α−i ) to
every action in αi∗ with a positive probability, is the same,
equal to Ui (α∗ )
▸ The expected payoff (given other players’ strategies α−i ) to
every action in αi∗ with zero probability, is at most, equal to
the expected payoff in the first condition
▸ We can use this condition to check whether some mixed strategy
profile α is a mixed strategy NE.
▸ Check that the expected payoffs to each action in αi with positive
probability, is the same.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Example: BoS

B S
B 2,1 0,0
S 0,0 1,2

▸ In BoS, let’s check the mixed strategy profile (( 32 , 13 ), ( 31 , 23 )).

▸ Given Player 2’s mixed strategy α2 (B) = 13 , α2 (S) = 23 :
▸ E1 (B, α2 ) = 31 ⋅ 2 + 23 ⋅ 0 = 23
▸ E1 (S, α2 ) = 31 ⋅ 0 + 23 ⋅ 1 = 23
▸ Given Player 1’s mixed strategy α1 (B) = 23 , α1 (S) = 13 :
▸ E2 (B, α1 ) = 32 ⋅ 1 + 13 ⋅ 0 = 23
▸ E2 (S, α1 ) = 32 ⋅ 0 + 13 ⋅ 2 = 23
▸ Therefore, this is a mixed strategy NE.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Example 117.1
L C R
T ⋅, 2 3, 3 1, 1
M ⋅, ⋅ 0, ⋅ 2, ⋅
B ⋅, 4 5, 1 0, 7

▸ Is the strategy pair (( 34 , 0, 14 ), (0, 13 , 23 )) a MSNE?

▸ The dots indicate irrelevant payoffs (they occur with zero
probability).
▸ Given Player 2’s mixed strategy: α2 (L) = 0, α2 (C ) = 31 , α2 (R) = 32 :
▸ E1 (T , α2 ) = 0 + 13 ⋅ 3 + 23 ⋅ 1 = 53
▸ E1 (M, α2 ) = 0 + 31 ⋅ 0 + 23 ⋅ 2 = 43
▸ E1 (B, α2 ) = 0 + 31 ⋅ 5 + 23 ⋅ 0 = 53
▸ T , B occur with positive probability in α1 , and have the same
expected payoff when Player 2 plays α2 .
▸ M occurs with zero probability in α1 , and has an expected payoff
not greater than the expected payoffs to T , B.
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Example 117.1

L C R
T ⋅, 2 3, 3 1, 1
M ⋅, ⋅ 0, ⋅ 2, ⋅
B ⋅, 4 5, 1 0, 7

▸ Given Player 1’s mixed strategy: α1 (T ) = 34 , α1 (M) = 0, α1 (B) = 41 :

▸ E2 (L, α2 ) = 43 ⋅ 2 + 0 + 41 ⋅ 4 = 52
▸ E2 (C , α2 ) = 43 ⋅ 3 + 0 + 41 ⋅ 1 = 52
▸ E2 (R, α2 ) = 43 ⋅ 1 + 0 + 41 ⋅ 7 = 52
▸ C , R occur with positive probability in α2 , and have the same
expected payoff when Player 1 plays α1 .
▸ L occurs with zero probability in α2 , and has an expected payoff not
greater than the expected payoffs to C , R.
▸ Note: the fact that E2 (L, α2 ) = 52 does not imply anything about
the existence of a MSNE that has a positive probability on L.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
The ”Colonel Blotto” game

(0,2) (1,1) (2,0)

(0,2) -1,1 1,-1 1,-1
(1,1) 1,-1 -1,1 1,-1
(2,0) 1,-1 1,-1 -1,1

▸ A MSNE is where both players choose mixed strategy ( 13 , 13 , 13 ).

▸ This case of the Colonel Blotto game has no other equilibrium, but
other cases may have many equilibria.
▸ In general, it is a difficult problem to find all equilibria of a game.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Exercise 117.2: Choosing Numbers

▸ Players 1 and 2 choose a positive integer from 1...K .

▸ If the players choose the same number, Player 2 gets a payoff of −1
and Player 1 gets a payoff of 1.
▸ Otherwise, both players get a payoff of 0.
▸ First, show that one MSNE is if both players choose each integer
with equal probability 1/K .

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Exercise 117.2: Choosing Numbers

▸ Player 1’s expected payoffs to his actions 1, ..., K are:

▸ E1 (1) = E1 (2) = ... = E1 (K ) = 1/K
▸ Player 2’s expected payoffs to his actions 1, ..., K are:
▸ E2 (1) = E2 (2) = ... = E2 (K ) = −1/K
▸ All actions with positive probability have the same payoff, so the
condition for a MSNE is satisfied.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Exercise 117.2: Choosing Numbers

▸ Show there is no other MSNE.

▸ Let Player 1’s mixed strategy be (p1 , ..., pK ).
▸ Let Player 2’s mixed strategy be (q1 , ..., qK ).
▸ Player 1’s expected payoffs to his actions 1, ..., K are:
▸ E1 (1) = q1 , E1 (2) = q2 , ..., E1 (K ) = qK
▸ Player 2’s expected payoffs to his actions 1, ..., K are:
▸ E2 (1) = −p1 , E2 (2) = −p2 , ..., E2 (K ) = −pK

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Exercise 117.2: Choosing Numbers

▸ Player 1’s expected payoff, given both player’s mixed strategies, is:

p1 q1 + p2 q2 + ... + pK qK

▸ Player 2’s expected payoff, given both player’s mixed strategies, is

the negative of Player 1’s expected payoff:

−p1 q1 − p2 q2 − ... − pK qK

▸ Suppose that Player 1 does not place equal probability on each

number: there exists a number i such that pi that is strictly greater
than the other p’s.
▸ Player 2’s expected payoff to playing i is −pi , so Player 2 will put
zero probability on i ∶ qi = 0.
▸ However, if qi = 0, then Player 1’s expected payoff to i is 0, and
Player 1’s best response is to put zero probability on i, a
contradiction.
Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Implications of Mixed Strategy NE

▸ A MSNE α∗ that is not a pure strategy equilibrium is never strict.

Player i is indifferent between αi∗ and the actions that have a
positive probability in αi∗ .
▸ In MSNE, Player i’s probabilities are such that they induce the
other players to become indifferent among their actions.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Existence of Equilibrium in Finite Games

▸ Does every finite game have a MSNE?

▸ A famous result, proved by Nash, shows that this is true.
▸ Proposition: Every strategic game with vNM preferences in which
each player has finitely many actions has a mixed strategy Nash
equilibrium.
▸ We won’t prove this, but if you are interested, it uses Kakutani’s
fixed point theorem.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Dominated Actions

▸ Recall: an action a1 strictly dominates a2 if it gives a higher payoff,

no matter what other players do.
▸ We can extend this to mixed strategies.
▸ Definition: In a strategic game with vNM preferences, player i’s
mixed strategy αi strictly dominates action ai′ if:

Ui (αi , a−i ) > ui (ai′ , a−i ) for every list of the other players’ actions a−i

▸ Note that the other players’ actions are pure strategies.

▸ It is possible for an action that is not strictly dominated by a pure
strategy, to be strictly dominated by a mixed strategy.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Example of Dominated Actions

L R
T 1 1
M 4 0
B 0 3

▸ T is not strictly dominated by M or B.

▸ But, it is strictly dominated by α1 (T ) = 0, α1 (M) = 21 , α1 (B) = 12 .
▸ If Player 2 plays L: expected payoff is 12 ⋅ 4 = 2
▸ If Player 2 plays R: expected payoff is 12 ⋅ 3 = 1.5

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Strictly Dominated Actions in Mixed Strategy NE

▸ In a pure strategy Nash equilibrium, no player uses a strictly

dominated action.
▸ Extend to mixed strategies: In a mixed strategy Nash equilibrium,
no player will place a positive probability on a strictly dominated
action.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Weak Domination

▸ Definition: A mixed strategy αi weakly dominates action ai′ if

Ui (αi , a−i ) ≥ ui (ai′ , a−i ) for every list a−i of the other players’ actions

Ui (αi , a−i ) > ui (ai′ , a−i ) for some a−i of the other players’ actions
▸ We say action ai′ is weakly dominated.
▸ As before, a weakly dominated action may be played with positive
probability in a mixed strategy NE.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Pure Equilibria when Randomization is Allowed

▸ The pure version of Bos had 2 NE. These were also equilibria in the
game where mixing was allowed, which had 3 NE. This is true in
general.
▸ Suppose G is a strategic game without randomization. Preferences
are represented by the payoff function ui .
▸ Suppose G ′ is a strategic game with randomization, that has the
same players and actions as G , and preferences are represented by
expected values of ui .
▸ If a∗ is a NE of G , then the mixed strategy profile in which each
player assigns probability 1 to action ai∗ is also a mixed strategy NE
of G ′ .
▸ If α∗ is a mixed strategy NE of G ′ in which all players assign
probability 1 to an action ai , then the action profile (a1 ...an ) is a
Nash equilibrium of G .

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Illustration: Reporting a Crime

▸ Suppose a crime is observed by a group of n people.

▸ Each person would like the police to be informed, but prefers that
someone else makes the phone call.
▸ Specifically, each person gains value v from the police being
informed, but pays cost c if she calls.
▸ Game with vNM preferences:
▸ Players: n people
▸ Actions: Each player chooses to {Call, Don′ tCall}
▸ Preferences: Each player i has expected value preferences over
a payoff function that gives 0 if no one calls, v − c if player i
calls, and v if someone other than player i calls.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Nash Equilibrium in Pure Strategies

▸ Nash equilibrium in pure strategies: exactly one player will call.

▸ Case 1: only Player i calls. Let’s check this is a Nash equilibrium.
▸ Player i gets payoff v − c, everyone else gets payoff v .
▸ Player i can switch to Don’t call, but payoff will be lowered
from v − c to 0.
▸ All other players can switch to Call, but payoff will be lowered
from v to v − c.
▸ Case 2: All players choose Don’t call.
▸ Any player can switch to Call, payoff increases from 0 to v − c.
▸ Case 3: Two or more players call. Suppose players i, j call.
▸ Either player can switch to Don’t call, increase payoff from
v − c to v .

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Symmetric Mixed Strategy Nash Equilibrium

▸ There is no symmetric (i.e. all players choose the same action) NE

in pure strategies.
▸ However, there is one in mixed strategies. Suppose that all players
place probability p on Call, 1 − p on Don’t call.
▸ Equilibrium condition: EV to all actions with positive probability are
equal, given other players’ mixed strategies α−i .
▸ EV to Call, given α−i ∶ v − c
▸ EV to Don’t call, given α−i ∶

Prob(nobody else calls) ⋅ 0 + Prob(at least 1 person calls) ⋅ v = v − c

(1 − Prob(nobody else calls)) ⋅ v = v − c

c
→ = Prob(nobody else calls)
v ´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶
(1−p)n−1

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Symmetric Mixed Strategy Nash Equilibrium

c c 1
= (1 − p)n−1 → p = 1 − ( ) n−1
v v

▸ What happens as n increases?

▸ p decreases, so the equilibrium probability that any given player will
call goes down to zero as n → ∞.
▸ The probability that no one calls is:
c
(1 − p)n = (1 − p)n−1 ⋅ (1 − p) = (1 − p)
v
▸ So as n increases, the equilibrium probability that at least one
person calls also goes down to zero as n → ∞.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6
Next week

▸ Please finish Chapter 4 and read Chapter 5.1-5.2.

▸ HW #2 is due next week.
▸ I will return HW #1 next week.
▸ The midterm will be in class on April 19. It will cover Chapters 1-4
(only the sections that we’ve gone over in lectures).
▸ Midterm will be closed-book. No programmable calculators or
smartphones allowed.
▸ Previous midterms and solutions are on the course website.

Prof. Ronaldo CARPIO CUR 412: Game Theory and its Applications, Lecture 6