5th Week

Strategic Games with Mixed Strategies

 Pure and Mixed Strategy Games
— So far, we have studied the so-called pure strategy and the
strategies that are frequently encountered in real life.

— If we allow players to play on pure strategies with a certain

probability, we encounter a new concept we call mixed strategy.

— In mixed strategy a player chooses any of his/her pure strategies,

but which one is played is determined by a certain probability.

— A pure strategy is a rule that tells the player which move to choose.
Mixed strategy, on the other hand, is a rule that tells the player
with what probability to choose a move.
 Mixed Strategy Games
— Games don't always have a pure strategy balance.

— In other words, these games may not always have a saddle

point.

— In this case, the best strategies of both players do not

coincide.

— Such games that do not have pure strategy balance are called
Mixed strategy balanced games.
 Mixed Strategy Games
— This situation prevents reaching a direct solution on how the
players will behave, and the game does not have a definite
solution anyway.

— For this reason, the solution of mixed strategy games is called

decision-making under risk, not decision-making under
uncertainty.

— In mixed strategy balanced games, the frequency or

probability of which strategy the players can choose is
investigated.
 Mixed Strategy Games
— Let the profit-loss matrix be given as follows for a two-person
strategy game:

Player B
Strategies B1 B2
Player A A1 K11 K12
A2 K21 K22
 Mixed Strategy Games
— Let X1 be the probability that player A chooses strategy A1 and,
let X2 be the probability of choosing strategy A2.
— Same way; Let Y1 be the probability that player B chooses
strategy B1 and, let Y2 be the probability of choosing strategy
B2.
— Thus, the profit and loss matrix is reconstructed by including
the determined probabilities:
Player B
Frequencies - Y 1-Y
- Strategies B1 B2
Player A
X A1 K11 K12
1-X A2 K21 K22
 Mixed Strategy Games
— X1 , X2 ≥ 0 and X1 + X2 =1
ü X1 = X
ü X2 = 1-X
— Y1 , Y2 ≥ 0 and Y1 + Y2 =1
ü Y1 = Y
ü Y2 = 1-Y
Player B
Frequencies - Y 1-Y
- Strategies B1 B2
Player A
X A1 K11 K12
1-X A2 K21 K22
 Mixed Strategy Games
— If player B chooses strategy B1,
ü X*K11 + (1-X)* K21

— If player B chooses strategy B2,

ü X*K12 + (1-X)* K22

Player B
Frequencies - Y 1-Y
- Strategies B1 B2
Player A
X A1 K11 K12
1-X A2 K21 K22
 Mixed Strategy Games
— Similarly, the expected losses of player B according to the
strategies that player A will choose are as follows:
— If player A chooses strategy A1,
ü Y*K11 + (1-Y)* K12
— If player A chooses strategy A2,
ü Y*K21 + (1-Y)* K22
Player B
Frequencies - Y 1-Y
- Strategies B1 B2
Player A
X A1 K11 K12
1-X A2 K21 K22
 Mixed Strategy Games
— The best possible value for X for player A will be determined
when the expected payoffs are equal for both strategies that B
can choose.
X*K11 + (1-X)* K21 = X*K12 + (1-X)* K22

— Likewise for player B, the best possible value for Y will be

determined if the expected losses are equal for both strategies
that player A can choose.
Y*K11 + (1-Y)* K12 = Y*K21 + (1-Y)* K22
 Mixed Strategy Games
— As a result, the value of the game is obtained by using one of the
following equations, which will give the same result, according to the X
and Y values obtained from these two equations.

D = X*K11 + (1-X)* K21

D = X*K12 + (1-X)* K22

or

D = Y*K11 + (1-Y)* K12

D = Y*K21 + (1-Y)* K22

 Mixed Strategy Games (Example)
— Let's determine the strategies of the players and the value of
the game according to the profit-loss matrix given below.

Player B
Strategies B1 B2
Player A A1 1 3
A2 2 -1
 Mixed Strategy Games (Example)
— When checked, it will be seen that there is no dominant
strategy for either player in the game.
— When we try to solve the game with pure strategy, it will be
seen that there is no balance point in a game
Player B
Strategies B1 B2
Player A A1 1 3
A2 2 -1

— For this reason, this game is a mixed strategy game.

 Mixed Strategy Games
— The expected payoff for player A for player B's strategy B1 is:
X + 2*(1-X)
— The expected payoff for player A for player B's strategy B2 is:
3X - (1-X)

Player B
Frequencies - Y 1-Y
- Strategies B1 B2
Player A
X A1 1 3
1-X A2 2 -1
 Mixed Strategy Games
— The expected payoff for player B for player A's strategy A1 is:
Y + 3*(1-Y)
— The expected payoff for player B for player A's strategy A2 is:
2Y - (1-Y)

Player B
Frequencies - Y 1-Y
- Strategies B1 B2
Player A
X A1 1 3
1-X A2 2 -1
 Mixed Strategy Games
— The best possible value for X for player A will be determined
when the expected payoffs are equal for both strategies that B
can choose.
X + 2*(1-X) = 3X - (1-X)
X + 2 – 2X = 3X – 1 + X
2 – X = 4X -1
5X = 3
X = 3/5
 Mixed Strategy Games
— Likewise for player B, the best possible value for Y will be
determined if the expected losses are equal for both strategies
that player A can choose.
Y + 3*(1-Y) = 2Y - (1-Y)
Y + 3 – 3Y = 2Y - 1 + Y
3 – 2Y = 3Y -1
5Y = 4
Y = 4/5
 Mixed Strategy Games
— Mixed strategy vector of player A:
X = 3/5 è X1= 3/5 and,
X2= (1-X) = 1 - 3/5 = 2/5
X = (3/5 2/5)

— Mixed strategy vector of player B:

Y = 4/5 è Y1= 4/5 and,
Y2= (1-Y) = 1 - 4/5 = 1/5
Y = (4/5 1/5)
 Mixed Strategy Games
— X = (3/5 2/5) and Y = (4/5 1/5)

— If the game is played 5 times, player A is expected to choose

strategy A1 3 times and strategy A2 2 times, while player B is
expected to choose strategy B1 4 and strategy B2 1 time.

— In other words, if the game is to be played once, player A

chooses strategy A1 with 60% probability and strategy A2 with
40% probability, while player B chooses strategy B1 with 80%
probability and strategy B2 with 20% probability.
Mixed Strategy Games
— Starting from the determined elements of the vector
X, the value of the game is obtained as follows:
D = X + 2*(1-X)
D = 3/5 + 2*(2/5) = 7/5
or
D = 3X - (1-X)
D = 3*(3/5) – (2/5) = 7/5
Mixed Strategy Games
— Likewise, starting from the determined elements of
the Y vector, the value of the game is obtained as
follows :
D = Y + 3*(1-Y)
D = 4/5 + 3*(1/5) = 7/5
or
D = 2Y - (1-Y)
D = 2*(4/5) – (1/5) = 7/5
 Pure and Mixed Strategy Games
— Since the players can determine the exact strategy in the
games with pure strategy, these games are considered as
decision making problems under uncertainty and this
uncertainty is eliminated under the assumption of rational
decision makers.
— On the other hand, in games with mixed strategy balance,
decision making problem is evaluated under the title of
decision making problems under risk, and a probability level
measurement of risk is tried to be obtained under the
assumption of players who are rational decision makers.
 6 th Week

Zero-Sum Games
 Two Person Zero-Sum Games
— There are two players (called the row player and the column
player).
— The row player must choose 1 of m strategies. Simultaneously,
the column player must choose 1 of n strategies.
— If the row player chooses his ith strategy and the column player
chooses his jth strategy, then the row player receives a reward
of aij and the column player loses an amount aij.
 Two Person Zero-Sum Games
— A two-person zero-sum game has the property that for any
choice of strategies, the sum of the rewards to the players is
zero.
— In a zero-sum game, every dollar that one player wins comes
out of the other player’s pocket, so the two players have
totally conflicting interests.
— Thus, cooperation between the two players would not occur.
— John von Neumann and Oskar Morgenstern developed a
theory of how two-person zero-sum games should be played.
— Each player chooses a strategy that will allow her/him to do
her/his best because s/he knows the strategy her opponent is
following.
 Two Person Zero-Sum Games (Example)
— For example, in the two-person zero-sum game in below table,
if the row player chose his/her 2nd strategy and the column
player chose his/her 1st strategy;
ü the row player would receive two units, and
ü the column player would lose two units.
Two Person Zero-Sum Games (Example)
Two Person Zero-Sum Games – Saddle Point (Example)
— This game has the property of satisfying the saddle point
condition:
max (row minimum) = min (column maximum)
— Any two-person zero-sum game satisfying this condition is said
to have a saddle point.
— A saddle point occurred where the row player chose row 3 and
the column player chose column 2.
Two Person Zero-Sum Games – Saddle Point (Example)
— The row player could make sure of receiving a reward of at
least five units (by choosing the optimal strategy of row 3),
and the column player could ensure that the row player would
receive a reward of at most five units (by choosing the optimal
strategy of column 2).
— If a game has a saddle point, then this game has a value of 5.
Two Person Zero-Sum Games – Saddle Point (Example)
— Many two-person zero-sum games do not have saddle points.
max (row minimum) = -1
Not equal
min (column maximum) = +1
Two Person Constant-Sum Games
— A two-person 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.

— Of course, a two-person zero-sum game is just a two-person

constant sum game with c=0.

— A two-person constant-sum game maintains the feature that

the row and column players are in total conflict, because a unit
increase in the row player’s reward will always result in a unit
decrease in the column player’s reward.
Two Person Constant-Sum Games (Example)
— Two networks are competing 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 in table.
— Does this game have a saddle point?
— What is the value of the game to network 1?
 Two-Person Constant-Sum Games (Example)
— For example, if both networks choose a western, the matrix indicates that 35
million people will watch network 1 and 100-35=65 million people will watch
network 2.
— Thus, we have a two-person constant-sum game with c=100 (million).
 Two-Person Constant-Sum Games (Example)
— Looking at the row minimum, we find that by choosing a soap opera,
network 1 can be sure of at least max (15, 45, 14)=45 million viewers.

— Looking at the column maxima, we find that by choosing a western, network

2 can hold network 1 to at most min (45, 58,70)=45 million viewers.
 Two-Person Constant-Sum Games (Example)
max (row minimum) = min (column maximum) = 45

— According to the equation, network 1’s choosing a soap opera and network
2’s choosing a western yield a saddle point.

— Neither side will do better if it unilaterally changes strategy. (You can check
this)
 Two-Person Constant-Sum Games (Example)
— The value of the game to network 1 is 45 million viewers, and the value of
the game to network 2 is 100 - 45 = 55 million viewers.

— The optimal strategy for network 1 is to choose a soap opera, and the
optimal strategy for network 2 is to choose a western.
 7th Week

Prisoners' Dilemma: Coordination and

Anti-Coordination Games
 Two-Person Non-Constant-Sum Games
Prisoner’s Dilemma
— Two prisoners who escaped and participated in a robbery have been recaptured
and are awaiting trial for their new crime.
— Although they are both guilty, the district attorney is not sure he has enough
evidence to convict them.
— To convince them to testify against each other, the district attorney tells each
prisoner the following:
ü If only one of you confesses and testifies against your partner, the person who confesses will
go free while the person who does not confess will surely be convicted and given a 20-year
jail sentence.
ü If both of you confess, then you will both be convicted and sent to prison for 5 years.
ü Finally, if neither of you confesses, I can convict you both of a misdemeanor and you will
each get 1 year in prison.
— What should each prisoner do?
 Two-Person Non-Constant-Sum Games
Prisoner’s Dilemma
— If we assume that the prisoners cannot communicate with each other, the
strategies and rewards for each are as shown in the table below.
— The first number in each cell of this matrix is the reward (negative, because
years in prison is undesirable) to prisoner 1, and the second matrix in each
cell is the reward to prisoner 2.
— Sum of the rewards in each cell varies from a high of -2 (-1 - 1) to a low of -
20 (-20 + 0).
— Thus, this is not a constant-sum two-player game.
 Two-Person Non-Constant-Sum Games
Prisoner’s Dilemma
— As in a two-person zero-sum game, a choice of strategy by each player
(prisoner) is an equilibrium point if neither player can benefit from a
unilateral change in strategy.
— Nash equilibrium is the situation where each party makes decisions
considering what the other party will do.
— (-5, -5) is an equilibrium point, because if either prisoner changes his
strategy, then his reward decreases (from -5 to -20).
— Clearly, however, each prisoner is better off at the point (-1, -1).
 Two-Person Non-Constant-Sum Games
Prisoner’s Dilemma
— To see that the outcome (-1, -1) may not occur, observe that (-1, -1) is not an
equilibrium point, because if we are currently at the outcome (-1, -1), either
prisoner can increase his reward (from -1 to 0) by changing his strategy from
“don’t confess” to “confess” (that is, each prisoner can benefit from double-
crossing his opponent).
— This illustrates an important aspect of the Prisoner’s Dilemma type of game:
If the players are cooperating (if each prisoner chooses “don’t confess”),
then each player can gain by double-crossing his opponent (assuming his
opponent’s strategy remains unchanged).
 Two-Person Non-Constant-Sum Games
Prisoner’s Dilemma
— If both players double-cross each other, however, then both will be worse off
than if they had both chosen their cooperative strategy.

— The Prisoner’s Dilemma game is of interest because it explains why two

adversaries often fail to cooperate with each other.
Coordination Games
— In a two-player game, suppose the set of moves of
both players is {large, small}.
— The main purpose of the game is to match the
selected objects in the game: “If both players choose
large or small objects, their benefits are greater than
their other choices (one choosing large and the other
small).”
— However, both players choosing large objects will
benefit more than both players choosing small objects.
Large Small
Large 2, 2 -1, -1
Small -1, -1 1, 1
 Coordination Games (continued) Large Small
Large 2, 2 -1, -1
— This game has two Nash equilibrium: Small -1, -1 1, 1
(large, large) and (small, small)
— But (large, large) strategy profile produces a better
result for both players than (small, small) strategy
profile.
— If there was a pre-game communication between the
players, it would be possible to say that this balance
situation would easily emerge.
— But what happens if there is no communication
between the players?
Coordination Games (continued)
— If care is taken in such games, players can reach higher
levels of well-being if they agree on better strategy
profiles for them (act in coordination).

— That is why such games are called coordination

games.
Large Small
Large 2, 2 -1, -1
Small -1, -1 1, 1
Coordination Games (Example-2)
— In the dangerous coordination games, although there
are the same Nash equilibriums, there are differences
in the non-equilibrium utility levels.
— This difference may cause the balance in the game to
be formed in the Nash equilibrium, which gives less
benefit.
Large Small
Large 2, 2 -1000, -1
Small -1, -1 1, 1
 Coordination Games (Example-2 Continued)
— If 1st player has concerns that 2nd player is not rational or thinks s/he is
distracted, then 1st player will be reluctant to play his grand strategy.
— Because if 2nd player plays on small strategy somehow, 1st player can face
a damage level as high as -1000.
— For this reason, 1st player can not afford to take risks and choose the
small strategy, so that s/he can face a much lower loss, such as -1 at
most.
— Therefore, the expected result may not be (large, large) as in the
previous example.
Large Small
Large 2, 2 -1000, -1
Small -1, -1 1, 1
 Coordination Games (Example-3)
— In this coordination game, let's examine the risk behaviors of the players:
— There are two pure strategy Nah equilibriums in this game: (U,l) and (D,r)
— Both pure strategy Nash equilibriums make the game easier to predict.
When we compare these equilibriums, it is seen that (D,r) is a more
reliable result.
— Because every player is guaranteed to get a 7 by following this profile.
— However, playing on the (U,l) strategy profile may result in zero benefit if
the opponent does not comply with it.
— So unless 1st player is sure, strategy D is most suitable for him/her.
l r
U 9, 9 0, 8
D 8, 0 7, 7
 Coordination Games (Example-3 Continued)
— The same claim applies to the 2nd player as well.
— 2nd Player will not play the l move unless s/he is very sure that 1st player
has played the U move.
— This raises 1st player's suspicion that 2nd player will play move l, thus
making it tempting for 1st player to play D.
— 2nd player also realizes that 1st player will think so, so s/he acts cautiously
and chooses move r.
— Thus, the (D,r) strategy profile becomes the equilibrium behavior of the
game.
— That is, the (D,r) strategy profile is a more reasonable behavior,
l r
U 9, 9 0, 8
D 8, 0 7, 7
 Anti-coordination Games (Chicken Game)
— It is a classic game of game theory.
— It can be used to describe a situation in which 2 drivers driving
towards a single lane road are both driving their car quickly to
enter the single lane road alone.
— In this game, each player asks the other to withdraw from the
race.
— However, s/he will not prefer to give way, as it would be
dishonorable to withdraw before the other player withdraws.
— So no one would want to be a coward. (chicken gets its name
from here)
— If both are not give way, they will crash because one lane will not
be enough.
 Anti-coordination Games (continued)
Give Way Don’t Give Way
Give Way 0, 0 -1, 1
Don’t Give Way 1, -1 -10, -10

— There are 2 Nash equilibrium in the game.

— If 1st player thinks that 2nd player is going to play the
“give way”, the best move for her/him is ”don’t give
way".
— If 2nd player thinks that 1st player will play the “don’t
give way” move, will the 2nd player stay on the “give
way” move?
Anti-coordination Games (continued)
Give Way Don’t Give Way
Give Way 0, 0 -1, 1
Don’t Give Way 1, -1 -10, -10

— If 2nd player thinks that 1st player will play the “don’t give
way” move, will the 2nd player stay on the “give way”
move?

— Yes, it remains, so it is a stable state.

— So, one of the Nash equilibrium is (don’t give way, give
way).
 Anti-coordination Games (continued)
— The basic pattern of behavior that emerges from this game is
not to insist (give way) when one insists (don’t give way), but
to persist when the other does not.

— In this sense, the chicken game is an anti-coordination game.

— It is optimal for each player to follow the opposite strategies.

(However, following the same strategies in coordination games
gives better results)