Game Theory: Minimax, Maximin,

and Iterated Removal

Naima Hammoud

March 14, 2017

Last Lecture: expected value principle
Colin

A B

Rose A 2 -2 -3 3
B 0 0 3 -3
Rose’s expected
Suppose that Rose knows Colin will play ½ A + ½ B payoff if she plays
strategy A is -1/2
Rose’s Expectations for playing pure strategies
ERose (A) = 1/2 ⇥ 2 + 1/2 ⇥ ( 3) = 1/2
ERose (B) = 1/2 ⇥ 0 + 1/2 ⇥ (3) = 3/2 Rose’s expected
payoff if she plays
strategy B is 3/2
Last Lecture: expected value principle
Colin

A B

Rose A 2 -2 -3 3
B 0 0 3 -3
Suppose that Rose knows Colin will play ½ A + ½ B
Because 3/2 > -1/2
Rose’s Expectations for playing pure strategies Rose chooses to
maximize her payoff
ERose (A) = 1/2 ⇥ 2 + 1/2 ⇥ ( 3) = 1/2 by playing B. That’s
of course only if
ERose (B) = 1/2 ⇥ 0 + 1/2 ⇥ (3) = 3/2 Colin is playing
½A+½B
Last Lecture: expected value principle
Colin

A B

Rose A 2 -2 -3 3
B 0 0 3 -3

Rule of thumb: If you know your opponent is playing

a mixed strategy and will continue to play it, you should
use a strategy that maximizes your expected payoff.
 Last Lecture

• We saw that in soccer penalty-kick data collected by Ignacio Palacios-

Huerta (2003) that kickers and goal-keepers seem to be playing the
Nash equilibrium! But is that really the case?

Goalie Left Goalie Right Kicker Left Kicker Right

Nash frequency 0.42 0.58 0.38 0.62

Actual frequency 0.42 0.58 0.4 0.6

 Last Lecture

• We saw that in soccer penalty-kick data collected by Ignacio Palacios-

Huerta (2003) that kickers and goal-keepers seem to be playing the
Nash equilibrium! But is that really the case?

• The player is actually trying to maximize their own gain and minimize
the gain of the goal keeper

• It turns out that in zero-sum games, the Nash equilibrium,

maximizing your own gain, and minimizing your opponent’s gain
actually coincide.
 Zero-sum Games

zero-sum game: A zero-sum game is one in which the sum of the

individual payoffs for each outcome is zero.

Example: Matching pennies

Colin
The sum of payoffs for
Heads Tails this outcome is zero,
as is the sum of payoffs
for every other outcome.
Heads 1 -1 -1 1
Rose
Tails -1 1 1 -1
 Minimax, Maximin

zero-sum game: A zero-sum game is one in which the sum of the

individual payoffs for each outcome is zero.

Minimax strategy: minimizing one’s own maximum loss

Maximin strategy: maximize one’s own minimum gain

 Zero-sum game example

Column Column
player 2 player 2
2 3 2 3
2, 2 0, 0 1, 1 2 0 1
Row 44, 4 3, 3 2, 25 Row 44 3 25
player 1 player 1
1, 1 2, 2 2, 2 1 2 2

Since the payoffs of the column player (shown red) are just the negative of the
payoffs of the row player, we can write a matrix only showing payoffs of the
row player (on the right). Once we have that, we can find the maximin & minimax.
Maximin strategy for player 1: maximize their own minimum gain

Column
player 2
minimum gain
2A B C3
A 2 0 1 0
B44 25
Row
player 1 3
C 1 2 2
If player 1 plays the first strategy (strategy A) then their minimum gain is 0.
Maximin strategy for player 1: maximize their own minimum gain

Column
player 2
minimum gain
2A B C3
A 2 0 1 0
B44 25
Row
player 1 3 3
C 1 2 2
If player 1 plays strategy B then their minimum gain is -3.
Maximin strategy for player 1: maximize their own minimum gain

Column
player 2
minimum gain
2A B C3
A 2 0 1 0
B44 25
Row
player 1 3 3
C 1 2 2 2
If player 1 plays strategy C then their minimum gain is -2.
 Maximin strategy for player 1: maximize their own minimum gain
Minimax strategy for player 2: minimize their own maximum loss
Column
player 2
minimum gain
2 A B C 3
A 2 0 1 0
B44 3 25
Row
player 1
3
C 1 2 2 2
maximum loss 4
If player 2 plays strategy A then their maximum loss is 4 (their max loss is player 1’s max gain)
 Maximin strategy for player 1: maximize their own minimum gain
Minimax strategy for player 2: minimize their own maximum loss
Column
player 2
minimum gain
2 A B C 3
A 2 0 1 0
B44 3 25
Row
player 1
3
C 1 2 2 2
maximum loss 4 0
If player 2 plays strategy B then their maximum loss is 0 (their max loss is player 1’s max gain)
 Maximin strategy for player 1: maximize their own minimum gain
Minimax strategy for player 2: minimize their own maximum loss
Column
player 2
minimum gain
2 A B C 3
A 2 0 1 0
B44 3 25
Row
player 1
3
C 1 2 2 2
maximum loss 4 0 2
If player 2 plays strategy C then their maximum loss is 2 (their max loss is player 1’s max gain)
Maximin strategy for player 1: maximize their own minimum gain
Minimax strategy for player 2: minimize their own maximum loss
Column
player 2

2A B C3 minimum gain

A 2 0 1 0 maximin
Row
player 1 B44 3 25 3
C 1 2 2 2
maximum loss 4 0 2
minimax
Take the maximum of the minimum gains, i.e. the maximum of row minima
(maximin), and the minimum of the maximum losses, i.e. the minimum of column
maxima (minimax). If they are equal, you have a saddle point.
Maximin strategy for player 1: maximize their own minimum gain
Minimax strategy for player 2: minimize their own maximum loss
Column
saddle point
player 2

2A B C3 minimum gain

A 2 0 1 0 maximin
Row
player 1 B44 3 25 3
C 1 2 2 2
maximum loss 4 0 2
minimax
If a saddle point exists, it should always be played.
Here player 1 plays A and player 2 plays B
Maximin strategy for player 1: maximize their own minimum gain
Minimax strategy for player 2: minimize their own maximum loss
Column
saddle point
player 2

2A B C3 minimum gain

A 2 0 1 0 maximin
Row
player 1 B44 3 25 3
C 1 2 2 2
maximum loss 4 0 2
minimax

A saddle point is a Nash equilibrium

 More examples
player 2 player 2

2 3 2 3
3 2 1 0 0 3 1 1 0 0
60 1 2 07 60 07
player 1 6 70 player 1 6 1 2 70
41 0 2 15 0 41 0 2 15 0
3 1 2 2 1 maximin 3 1 2 2 1
3 2 2 2 minimax 3 1 2 2
None of the row minima The highlighted entry is the
equals any of the column saddle point, and both
maxima, so no saddle points players will play it.
 Dominated strategies: iterated removal
Dominated strategy: There is some other strategy that does better than it.

• A dominated strategy will never be played, so we can remove it from the

game

• We can iterate until we get to to the dominant strategy

• This is called iterated removal of dominated strategies

 iterated removal example
Column
player 2

Left Center Right

Up 3 0 2 1 0 0
Row
player 1
Middle 1 1 1 1 5 0
Down 0 1 4 2 0 1
 Column
player 2

Left Center Right

Up 3 0 2 1 0 0
Row
player 1
Middle 1 1 1 1 5 0
Down 0 1 4 2 0 1
Column player will never play Right because it is strictly dominated by Center.
The payoffs of player 2 playing Right are (0, 0, 1), which are dominated by
(1, 1, 2) from playing Center. Therefore we can remove Right.
 Column
player 2

Left Center Right

Up 3 0 2 1 0 0
Row
player 1
Middle 1 1 1 1 5 0
Down 0 1 4 2 0 1

Row player will never play Middle because it is strictly dominated by Up.
Payoffs of Middle are (1, 1) which are dominated by (3, 2) from Up.
 Column
player 2

Left Center

Up 3 0 2 1
Row
player 1
Down 0 1 4 2

The new game matrix is now smaller.

 Column
player 2

Left Center

Up 3 0 2 1
Row
player 1
Down 0 1 4 2

Column player will never play Left because it is strictly dominated by Center.
Payoff of (0, 1) from Left versus (1, 2) from Center.
 Column
player 2

Center

Row Up 2 1
player 1
Down 4 2

Now row player is better off playing Down than Up,

because the payoff is 4 instead of 2.

So (4, 2) is a unique Nash equilibrium

“FAITH” — TELEVISION’S NEW HIT GAMESHOW
You have observed the host to be 99.98% accurate in the last
10,000 games. If he predicted that the contestant chooses
only Box #2, he rewards their faith with the million dollars.
!
Do you take both boxes or only Box #2?

Box 2
Box 1
$1 million
$1000
or nothing

4
THE MATRIX FOR NEWCOMB’S PROBLEM

HOST
Predicts that you Predicts that you
select both boxes select Box #2

You select both $1,000 $1,001,000

CONTESTANT
You select Box #2 $0 $1,000,000
TWO ARGUMENTS

Argument 1: Have faith and take Box #2

In your observations of the last 10,000 games, the host has been
shown to possess 99.98% accuracy in predicting the contestants
choice. If you select both boxes, you will almost certainly get
only $1000.
If you have faith (in the host, in your observations), and select Box
#2, you will win the million dollars.

Argument 2: Take both boxes

What does it matter what the host predicted? Either there is one
million dollars in Box #2 or there isn’t. The host’s prediction does
not change the contents of the box here and now. By opening
both boxes, you get either $1000 or $1,001,000. This is better
than $0 or $1,000,000.