Decision & Game Theory

Chapter 7
Two-Person Zero-Sum Games

Sonia REBAI
Tunis Business School
University of Tunis
Zero-Sum Games
✓ Zero-sum games are games for which player strategies can not increase
or decrease the available resources. The sum of the gains is constant
and therefore if the gain increases for one of the players it decreases
inevitably for the other player.

✓ The Battle-of-the-Sexes game or the Prisoner's dilemma are non-zero

sum games.
 Player 2
A B C
1 (7,-7) (4,-4) (3,-3)
Player 1 2 (9,-9) (5,-5) (2,-2)
3 (6,-6) (3,-3) (1,-1)

This payoff table corresponds to a zero-sum game and can be simplified as

follows: Player 2
A B C
1 7 4 3
Player 1 2 9 5 2
3 6 3 1
✓ A two-players constant-sum game is a two-player game in which, for
any choice of both player’s strategies, the row player’s reward and
the column player’s reward add up to a constant value C .

✓ A two-players zero-sum game is just a two-players constant-sum

game with C = 0.
Pure Strategy - Saddle Point
✓ A game is a pure strategy, if the optimal strategy for a player is the same
whatever the game of the 2nd.

✓ In a game of pure strategy, the 1st player identifies the minimum (what the
2nd player allows him) for each of his strategies. Then he selects the
maximum among the identified values. In other words the 1st player
maximizes the minimum allowed gains (offensive).

✓ However, the 2nd player minimizes the maximum loss (defensive).

 Player 2
A B C Min
1 7 4 3 3 Max
PLayer 1 2 9 5 2 2
3 6 3 1 1
Max 9 5 3

Min

Player 1 should play strategy 1 and player 2 should play C.

Since min-max = max-min, we say that the game possesses a saddle point
leading to a stable solution.
✓ A saddle point is the combination of strategies in which each player can
find the highest possible payoff assuming the best possible play by the
opponent.
✓ None of the players will benefit from changing his move in an attempt
to make advantage of the opponent strategy to improve his own
position.
✓ When there is a saddle point, both players should stick to their min-max
and max-min solutions respectively.
✓ A game does not always possess a saddle point.
✓ Every saddle point is an equilibrium point but not every equilibrium point
is a saddle point.

✓ A saddle point occurs when each player is achieving the highest possible
payoff and thus neither would benefit from changing strategies if the other
didn’t also change - which is why it is also called an equilibrium point.

✓ However, there are equilibrium points in variable sum games where the
players are not achieving the best possible payoff (so they aren’t saddle
points) but neither will benefit by changing their strategy.
✓ The optimal strategies and value for a two-players constant-sum game
may be found by the same methods used to find the optimal strategies
and value for a two-players zero-sum game.
Example
During the 8 to 9 P.M. time slot, two networks are vying for an audience of
100 million viewers. The networks must simultaneously announce the type
of show they will air in that time slot. The possible choices for each network
and the number of network 1 viewers (in millions) for each choice are shown
below. Does this game have a saddle point? What is the value of the game to
network 1? Network 2
S1 S2 S3
S1 35 15 60
Network 1 S2 45 58 50
S3 38 14 70
Two players and Zero-Sum Games: Randomized Strategies

B C Min
1 1 5 1
2 7 4 4 Max
Max 7 5

Min

✓ If player 1 chooses 1, player 2 will choose B. Player 1 will then change

his preference for strategy 2. Hence, player 2 will change for C.
✓ A game is said to have randomized or mixed strategies, if a player needs
to change strategy when the game is repeated.

✓ Whenever a game does not have a saddle point, game theory advises
each player to assign a probability distribution over his/her set of
strategies.

✓ Each player has to look for the frequency to play each of the strategies
so that the result will be independent of the policy adopted by the
other player.
Let denote by

pi = probability that player 1 will choose strategy i (i=1,2,…,m).

qj = probability that player 2 will choose strategy j (j=1,2,…,n).

Matrix 2x2 B C
1 1 5
2 7 4

ü Let P be the probability of playing strategy 1. The expected value of the

gain is independent of the game of the 2nd iff

1 * P + 7 * (1-P) = 5*P + 4*(1-P) or P = 3/7

ü Thus, player 1 must play 3/7 of the time the 1st strategy and 4/7 of the
time the 2nd strategy. His expected profit = 31/7
ü Similarly, for the 2nd player, the probability q of playing strategy B is
chosen so that the expected value of his loss is independent of the game
of the 1st player.

1*q + 5*(1-q) = 7*q + 4*(1-q) or q = 1/7

ü Thus, player 2 must play 1/7 of the time the strategy B and 6/7 of the

time the strategy C. His expected loss is

1 * 1/7 + 5 * 6/7 = 31/7

 Player 2
Matrix 2xm A B C D
1 16 4 18 6
Player 1
2 16 22 12 18

In this case we can use a graphical solution procedure

ü Let P be the probability of playing strategy 1. The expected payoff of
player 1 depends on the choice of the second player.
𝜋" = 𝜋 (p, A) = 16
𝜋$ = 𝜋 (p, B) = P * 4 + 22 * (1-P) = 22-18p
𝜋% = 𝜋 (p, C) = 18 * p + 12 * (1-P) = 6p + 12
𝜋& = 𝜋 (p, D) = P * 6 + 18 * (1-P) = 18-12p
Remember that player 2 wants to minimize the expected payoff for player 1.
ü Player 2 can minimize the expected payoff for player 1 by choosing a
strategy corresponding to the bottom line in the figure.

ü Player 1 wants to maximize this minimum expected payoff. Consequently,

player 1 should select the value of p where the bottom line peaks. Then
the Max-min is determined by

𝜋% = 𝜋& Hence p* = 1/3.

ü Thus player 1 must play 1/3 of the time the 1st strategy and 2/3 of the
time the 2nd strategy. Game value = 14
ü The 2nd player will then play strategies C or D. If we denote by q the
probability of playing strategy C, then

18*q + 6*(1-q) = 12*q + 18*(1-q)

q = 2/3

ü Thus player 2 must play 2/3 of the time strategy C and 1/3 of the time
strategy D. His expected loss = 14
Matrix nxm

ü Let xj = the probability that Player 1 will use strategy j, (j =1,2,…,m).

ü yj = the probability that Player 2 will use strategy j, (j=1,2,…,n).

ü Player 1 will choose x1, x2,…, xm so that his expected payoff is maximized.

ü On the other hand, Player 2 will choose y1, y2,…, yn so that his expected
payoff is minimized.
Let pij = the payoff under strategies i, j respectively for player 1 and 2
(i=1,2,…,m) and (j=1,2,…,n)

Player 2
1 2 .. .. n
1 P11 P12 .. .. P1n
2 P21 P22 .. .. P2n
Player 1 : : : :
: : : :
m Pm1 Pm2 .. .. Pmn
Player 1 will solve the following LP
Note that the jth
Max v constraint implies that the
s.t. p11x1+ p21x2+ …+ pm1xm ≥ v expected reward against
p12x1+ p22x2+ … + pm2xm ≥ v
column j must at least
.
equal v; otherwise, the
.
. column player could hold
p1nx1+ p2nx2+ … + pmnxm ≥ v the row player’s expected
x1+x2+ …+xm= 1 reward below v by
x1, x2, …, xm ≥0 choosing column j.
Player 2 will solve the following LP (dual of the player 1’s LP
Min w The ith constraint in the column
s.t. p11y1+ p12y2+ … +p1nyn ≤ w player’s LP implies that if the
p21y1+ p22y2+ …+ p2nyn ≤ w
row player chooses row i, then
.
the column player’s expected
.
. losses cannot exceed w ;
pm1y1+ pm2y2+ … +pmnyn ≤ w otherwise, the row player could
y1+y2+ …+yn=1 obtain an expected reward that
y1, y2, …yn ≥ 0 exceeded w by choosing row i .
Example
A B
1 14 6
2 2 12

Let denote by
² X1 and X2 the probabilities of playing respectively Strategies 1 and 2.

² Y1 and Y2 the probabilities of playing respectively Strategies A and B.

1st player
Max V
14X1+2X2 ≥ V
6X1+12X2 ≥ V
X1+X2=1
X1, X2 ≥ 0
2nd player
Min W
14Y1+6Y2 ≤ w
2Y1+12Y2 ≤ w
Y1+Y2 = 1
Y1, Y2 ≥ 0