Chapter 3 (Extension): Game Theory

Reference:
• Chapter 1 and 2, Prisner, E. Game theory through examples. Mathematical
Assocation of America, 2014.
• Chapter 14, Hillier and Lieberman, Introduction to Operations Research ,
McGraw Hill, 2010, 9th Edition. ISBN 978-0-07-337629-5.
 Contents
1. Introduction to game theory
a. What is a game?
b. Game classification
2. Simultaneous games
a. Bimatrix description
b. Which option to choose
c. Dominated move
d. Best response
e. Nash equilibria
3. Solving Nash Games
 1. Introduction
1.1. What’s a game?
• Game in game theory refers mathematical game, which has been
applied to economics, politics, business, and other various areas.
• Ingredients:
- Rules
- Outcomes and payoffs: Mathematical game has many possible
outcomes, each producing payoffs for players;
- Uncertainty of the outcome: Since a game contains some random
elements, or has more than one player, the outcome of the game
cannot be predicted;
- No cheating

¨ Rational behavior is usually assumed for all players in a game. That is,
players have preferences and beliefs about the world, and try to
optimize their individual payoffs.
 1.2. Game Classification (1 of 2)
Games can be categorized according to several criteria:
¨ How many players are there in the game? Usually there should be
more than one player.
¨ Is play simultaneous or sequential?
- Simultaneous game: Each player has only one move (action), and all moves
are made simultaneously.
- Sequential game: Each player moves at a time, and players may have to
move several times.
¨ Do players have perfect information? Every player knows other players’
moves when about to move
¨ Is the game zero-sum? A player can have a positive payoff only if
another has a negative payoff.
 Game classification (2 of 2)
Games can be categorized according to several criteria:
¨ Is the game cooperative or non-cooperative? Even if players negotiate,
the question is whether the results of the negotiation can be enforced
- Cooperative game: The results of the negotiations can be put in a contract
and be enforced.
- Non-cooperative game: A player can move differently from what was
promised in the negotiation. Then the communication is called “cheap
talk.”
¨ Within the scope of this chapter extension, we’ll focus on several main
types of simultaneous games
2. SIMULTANEOUS GAMES
 2.1. Normal Form – Bi-matrix Description
¨ Assumption:
- Each player only has finite options.
- Players simultaneously make their moves, determine the outcome of the
game, and receive their payoffs
How many outcomes are possible?
players
Player has possible moves, for
 There are possible outcomes.
¨ Symmetric game: All players have the same options, and if the two
players interchange their moves, the payoffs also interchange.
 2.2. Which Option to Choose ?
¨ Some people always expect the worst, they want to
maximize their payoffs. So, for any of their options,
they would concentrate on their smallest possible
payoff, and choose the option with highest value. This
value is called the maximin value or security level
¨ This move is called a maximin move (strategy), since it
maximizes the minimum possible payoff.
¨ To choose the maximin move, the player doesn’t have to
know the payoffs of other players
 Example 1 – two players (1 of 3)
Two companies share a market, If both don’t advertise, they get $5,000,000
each. If both advertise, both lower their gain to $3,000,000. If A advertises, but B
doesn’t, A gets $6,000,000 and B only $2,000,000, and conversely if B
advertises and A doesn’t. What is the payoff pattern of two players?
 Two-person Zero-Sum Games
¨ A game with two players, where a gain of one player equals
a loss to the other
¨ Example: Consider a coin-matching game in which each of
two players A and B selects a head (H) or a tail (T). If the
outcomes match (i.e. H and H, or T and T) player A wins
$1.00 from player B. Otherwise, A loses $1 to B. In this
game each player has two strategies (H or T), which yeild
the following 2x2 game matrix expressed in terms of the
payoff to A
Player A
H T
Player B H 1 -1
T -1 1
 Example 2 – two players zero-sum

Assume we are play Rock – Scissors – Paper for one dollar. Develop the
payoff matrix

In zero-sum game, the worst case of either A or B is to lose the game. In

other words, there is no way to implement the maximin move in this game.
 Example 3 – three of more players (1 of 2)
Three companies consider purchasing masks for their employees from a
manufacturer during pandemic. If there are only one company deciding to
purchase, the price is $10,000 per shipment. However, if the manufacturer
see any increase in demand, they would raise their price by $2,000 per
shipment for one more client. Develop a bimatrix if all three companies
decide at the same time.

A buy A not buy

C buy C not buy C buy C not buy

B buy X,X,X X,X,X B buy X,X,X X,X,X

B not buy X,X,X X,X,X B not buy X,X,X X,X,X
 Example 3 – three of more players (2 of 2)

A buy A not buy

C buy C not buy C buy C not buy

B buy X,X,X X,X,X B buy X,X,X X,X,X

B not buy X,X,X X,X,X B not buy X,X,X X,X,X

Does the result change if we consider the risk cost of not protect employees from the
padamic? Assuming that it would cost each company $100,000 if they let their employees
expose to the disease. Wearing mask would reduce the exposure rate by 35%.
 2.3. Dominated moves (1 of 2)
¨ A move, , for A strictly dominates another move , if always results in a
higher payoff for A than
- A rational player would never play a move that is strictly dominated by
another one (i.e. weak move).
- A player’s maximin move is never strictly dominated by any of her other
moves  The advice given by the maximin mechanism and dominated
move mechanism are typically the same.
¨ If all players are assumed to be rational, and experienced enough, they
reduce the game by eliminating all strictly dominated moves.
- The process of repeatedly reducing the game, as well as its result, a
game that can’t be reduced any further is denoted by IESD – iterated
elimination of strictly dominated moves.
 Dominated moves (2 of 2)
¨ A’s move weakly dominates another move, if it yields at least the same
payoff for A in all cases generated by combinations of moves of the
other players, and in at least one case an even better payoff
- Iterated elimination of weakly dominated move (IEWD) is not something
that should be performed automatically.
 Example 4 – two bars (1 of 2)
Each of two bars charges its own price of a beer, either $2, $4, or $5. The
cost of obtaining and serving the beer can be neglected. It is expected that
6,000 beers per month are drunk in a bar by tourists, who choose one of
two bars randomly, and 4,000 beers per month by natives who go to the
bar with lowest price,and split evenly in case both bars offer the same
price. What prices would the bars select?
 Solution
¨ Consider an instance: If bar A charges $2 and bar B charges $4, then all natives
will choose bar A. Therefore, bar A will serve 4000 beers to the natives, and
3000 beers to tourists, serving 7000 beers in total, making 7000*2 = 14000
dollars. Bar B will only serve 3000 beers to tourists, making 3000 *4 = 12000
dollars.
¨ The payoff matrix, with values in thousands of dollars is in the first matrix
¨ For each bar, move “4” strictly dominates move “2”, therefore we could eliminate
both moves “2” to get the reduced game to the second matrix.

¨ Now, but not before the elimination, move “4” strictly dominates move “5”.
Therefore, we eliminate these moves for both players as well and arrive at a
game with only one option, “4”, for each player, and a payoff of $ 20000 for each.
Therefore both players will choose $4 as the price of the beer.
 2.4. Best response
¨ A’s best response to B’s move M is the move that yields the highest
payoff for A, given B’s move M
- There may be several best responses to a given move
- To find A’s best response to B’s move M, we don’t even have to know B’s
payoffs.
¨ You’ll find the best responses for A’s move by looking at the rows of the
bimatrix one by one and selecting in each row the cell where the
second entry is maximum. The label of the corresponding column is the
best response to the move corresponding to that row.
- The same rule applies to find B’s best response with the columns.

What is best response to advertising/ NOT advertising in Example 1?

How about Example 4?
 Example 1 revisit – best response

B advertises B doesn’t
A advertises 3, 3 6, 2
A doesn’t 2, 6 5, 5

• The best response for A’s move when A advertises is that B advertises.
- The best response for A’s move when A doesn’t advertise is that B
advertises.
• Likewise, the best response for B’s move when B advertises is that A
advertises
- The best response for B’s move when B doesn’t advertise is that A
advertises.
 Example 4 revisit – best response

$2 $4 $5
$2 10, 10 14, 12 14, 15
$4 12, 14 20, 20 28, 15
$5 15, 14 15, 28 25, 25

• The best response for the price of “$2” is a price of “$5”

• The best response for the price of “$4” is a price of “$4”
• The best response for the price of “$5” is a price of “$4”
 Example 5 – best response
Find the best response in an example of a simultaneous three-person game
where each player has two options: A has the move and ; B has and ; C
has and . Assume the payoffs are:

0, 2.1, 0 -1, 1.1, 0.1 0.1, 1.1, 1 1.1, 0.1, -0.9

1, 0, -1 0, 1, 1.1 -0.9, 1, 0 0.1, 2, 0.1
 Best Response for Three-players
¨ Because the highest second entry in the first column is 2.1, it is
underlined. The highest second entry in the second column is
1.1, in the third column (first column of the second matrix) 1.1,
and in the fourth column 2, so they are underlined. For C best
responses, the highest third entry in the first row of the first
matrix is 0.1. The highest third entry in the second row of the
first matrix is 1.1. For the second matrix, the highest third entry
in the first row is 1, and in the second row 0.1. For A, the
highest first entry of upper-left cells in the two matrices is 0.1,
the highest first entry of upper-right cells is 1.1, and we get 1
respectively 0.1 for the lower-left respectively lower-right cells.
 2.5. Nash equilibria
¨ An outcome is called a pure Nash equilibrium provided nobody can
gain a higher payoff by deviating from the move, when all the other
players stick to their choices.
- In a pure Nash equilibrium, each move involved is the best response to
the other moves.
- A higher payoff is possible if only two of more players change their moves.
¨ Nash equilibria are self-enforcing agreements.
¨ In the symmetric Two-bar example, (4, 4) is the unique pure Nash
equilibrium.
 Example 6 – prisoners’ dilemma (1 of 2)

• Adam and Bob have robbed the bank and been arrested. They
are interrogated separately. Adam & Bob have an option to
confess (move C) or to remain silent (move S). The police has
little evidence, but if both remain silent, they’ll be sentenced to
one year on a minor charge. Therefore, the police proposes a
deal: if one confess while the other remains silent, the one
confessing goes free, while the other is sentenced for three
years. However, if both talk, both will be sentenced to two years.
Find the Nash equilibria, If each player’s payoff is 3 minus the
number of years served in jail.
 Solution

¨ We get the following payoff bimatrix:

¨ The pair (C, C)—both confessing forms the unique Nash

equilibrium of this game
 Example 7 – battle of sexes (1 of 2)
A couple, Adam and Beth, decides independently whether to go to a
soccer game, or to the movie in the evening. Each person likes to
do something together with the other, but the man prefers soccer,
and the woman prefers movie. Assume that no previous
communication is possible, no cell phones

To simplify the game, we assume that the total payoff for each
player is the sum of the payoffs (in terms of satisfaction) of being at
the preferred place, which gives a satisfaction of c satisfaction units,
and being together with the partner, giving d satisfaction units. We
have two variants, depending on whether
c or d is larger, the low or high love variants. The payoff here is
satisfaction instead of money
 Example 7 – battle of sexes (2 of 2)
• Here are the payoff bimatrices for both variants, where Adam chooses the rows
and Beth chooses the columns.

• Using best response, there’re 2 Nash equilibria in this version: (soccer, soccer)
and (movie, movie)
• Games with more than one pure Nash equilibrium are sometimes called
coordination games, since if pre-game negotiations are allowed, the players have
to agree on one of them.
• The question is: which Nash equilibrium is the best?
- By concentrating on Pareto-optimal Nash equilibria.
- A Nash equilibrium is Pareto-optimal if it is not strictly dominated by any other
Nash equilibrium.
3. Solving Nash Games
 3.1. Zero-Sum Game Solution
¨ Solution: minimax solution (there is a unique minimax solution)
¨ Example: Payoffs to player A
Player B
Strategy 1 Strategy 2 Minrow
Player A Strategy 1 10 14 10
Strategy 2 7 12 7
MaxCol 10 14
Player A selects strategy 1, Player B selects strategy 1
¨ Note: The optimal solution to such a game may require each player to play a
pure strategy (i.e. a choice of only one strategy/option/action by each player)
strictly determined game when the maximum of the of a row is equal
to the minimum of the of the columns. The value 10 is called a
saddle point.
¨ Minimax may not lead to the best possible outcome if the opponent does not use
a minimax solution
 3.2. Mixed Strategies
¨When the minimax equilibrium is .
Player B’s Player A’s strategies Minrow
strategies
C1 C2 C3 C4

U1 +20 +15 +12 +35 +12

U2 +25 +14 +8 +10 +8
U3 +40 +2 +19 +5 +2
U4 -5 +4 +11 0 -5
Maxcol +40 +15 +19 +35
 Mixed Strategies
¨ Mixed strategies: each player, instead of selecting a pure
strategy only, may play all his strategies according to a
predetermined set of probabilities, x1,x2,…xm and y1, y2,…yn.
Then
m n

x
i 1
i   y j  1; x i , y j  0; i , j
j 1

• Player I select xi that will x1 x2 x3 x4

maximize the minimum y1 +20 +15 +12 +35
expected value of rows.
y2 +25 +14 +8 +10
Player II select yj that will
minimize the maximum y3 +40 +2 +19 +5
expected value of columns y4 -5 +4 +11 0
 Mixed Strategies

¨ We have the relationship:

Minimax expected payoff  Maximin expected payoff.
When xi and yj correspond to the optimal solution, the “=” holds.
¨ How to determine optimal value of xi* and yj* ?
 3.3. Linear Programming Solution
¨ Player A’s optimal probabilities, x1,x2,…xm can be determined
by solving the following LP
max z  v
  m m
 
max min   ai1 xi ,...,  ain xi  m
xi
  i 1 i 1  s.t. v   aij xi  0, j  1,2,.., n
i 1
s.t. x1  x2  ...  xm  1  x1  x2  ...  xm  1
xi  0, i  1,2,..., m xi  0, i  1,2,..., m
¨ Player B’s optimal probabilities y1, y2,…yn can be determined
by solving the following LP

min𝑧=𝑤
  n n 
min max   a1 j y j ,...,  amj y j 
yj
  j 1 j 1 
s.t. y1  y2  ...  yn  1 
y j  0, j  1,2,..., n
 Example

B1 B2 B3 Minrow
A1 0 -2 2 -2
A2 5 4 -3 -3
A3 2 3 -4 -4
Maxcol 5 4 2
 Solution

B1 B2 B3 Minrow
A1 0 -2 2 -2
A2 5 4 -3 -3
A3 2 3 -4 -4
Maxcol 5 4 2
 Solution

max z  v
m
s.t. v   aij xi  0, j  1,2,.., n
i 1

x1  x2  ...  xm  1
xi  0, i  1,2,..., m
 Solution

min𝑧=𝑤
Optimal Solution