Gaming Theory - Study Notes
Gaming Theory - Study Notes
Gaming Theory - Study Notes
CHAPTER 6
Game Theory
6.1 INTRODUCTION
The theory of games (or game theory or competitive strategies) is a mathematical theory that
deals with the general features of competitive situations. This theory is helpful when two or
more individuals or organisations with conflicting objectives try to make decisions. In such
situations, a decision made by one decision-maker affects the decision made by one or more of
the remaining decision-makers and the final outcome depends upon the decision of all the
parties. Such situations often arise in the fields of business, industry, economics, sociology and
military training. This theory is applicable to a wide variety of situations such as two players
struggling to win at chess, candidates fighting an election, two enemies planning war tactics, fu-
ms struggling to maintain their market shares, launching advertisement campaigns by companies
marketing competing product, negotiations between organisations and unions, etc. These
situations differ from the ones we have discussed so far wherein nature was viewed as a
harmless opponent
The theory of games is based on the min-max principle put forward by J. von Neumann which
implies that each competitor will act so as to minimize has maximum loss (or maximize his
minimum gain) or achieve best of the worst. So far only simple competitive problems have been
analysed by this mathematical theory. The theory does not describe how a game should be
played; it describes only the procedure, and principles by which plays should be selected.
Though the theory of games was developed by von Neumann (called father of game theory) in
1928, it was only after 1944 when he and Morgenstern published their work named Theory of
Games and Economic Behaviour', that the theory received its proper attention. Since, so far the
theory has been capable of analysing very simple situations only, there has remained a wide gap
between what the theory can handle and the most actual situations in business and industry. So,
the primary contribution of game theory has been its concepts rather than its formal application,
to the solution of real problems.
2
6.2 CHARACTERISTICS OF GAMES
A competitive game has the following characteristics
(i) There are finite number of participants or competitors. If the number of participants is 2, the
game is called two-person game for number greater than two, it is called n-person game.
(ii) Each participant has available to him a list of finite number of possible courses of action. The
list may not be same for each participant.
(iii) Each participant knows all the possible choices available to others but does not know which
of them is going to be chosen by them.
(iv) A play is said to occur when each of the participants chooses one of the courses of action
available to him. The choices are assumed to be made simultaneously so that no participant
knows the choices made by others until he has decided his own.
(v) Every combination of courses of action determines an outcome which results in gains to the
participants. The gain may be positive, negative or zero. Negative gain is called a loss.
(vi) The gain of a participant depends not only on his own actions but also those of others.
(vii) The gains (payoffs) for each and every play are fixed and specified in advance and are
known to each player. Thus each player knows fully the information contained in the payoff
matrix.
(viii) The players make individual decisions without direct communication.
GAME MODELS
There are various types of game, models. They are based on the factors like the number of players
participating, the sum of gains-or losses and the number of strategies available, etc.
1. Number of persons : If a game involves only two players, it is called two-person game; if there-
are more than two players, it is named n-person game. An n-person game does not imply that
exactly n players are involved in it. Rather it means that the participants can be classified into n
mutually exclusive groups, with all members in a group having identical interests.
2. Sum of payoff's : If the sum of payoffs (gains and losses) to the players is zero, the game is
called zero-sum or constant-sum game-, otherwise non zero-sum game.
3. Number of strategies : If the number of strategies (moves or choices) is finite, the game is
called a finite game; if not, it is called infinite game.
3
6.2.1 DEFINITIONS
1. Game : It is an activity, between two or more persons, involving actions by each one of them
according to a set of rules, which results in some gain (+ve, -ve or zero) for each.
If in a game the actions are determined by skills, it is called a game of strategy, if they are
determined by chance, it is termed as a game of chance. Further a game may be finite or infinite.
A finite game has a finite number of moves and choices, while an infinite game contains an
infinite number of them.
2. Player : Each participant or competitor playing a game is called a player. Each player is
equally intelligent and rational in approach.
3. Play : A play of the game is said to occur when each player chooses one of his courses of
action.
4. Strategy : It is the predetermined rule by which a player decides his course of action from his
list of courses of actions during the game. To decide a particular strategy, the player need not
know the-other's strategy.
5. Pure strategy : It is the decision rule to always select a particular course of action. It is
usually represented by a number with which the course of action is associated.
6. Mixed strategy : It is decision, in advance of all plays, to choose a course of action for each
play in accordance with some probability distribution. Thus, a mixed strategy is a selection
among pure strategies with some fixed probabilities (proportions). The advantage of a mixed
strategy, after the pattern of the game has become evident, is that the opponents are kept
guessing as to which course of action will be adopted by a player.
Mathematically, a mixed strategy of a player with m possible courses of actions is a set X of m
non-negative numbers whose sum is unity, where each number represents the probability with
which each course of action (pure strategy) is chosen. Thus if p
i
is the probability of choosing
course i, then
) p ,....... p , p , p ( X
m 3 2 1
=
,
where
1 p ....... p p p
m 3 2 1
= + + + +
and
m ..... , 3 , 2 , 1 i 0 p
i
=
Evidently a pure strategy is a special case of mixed strategy in which all but one xi are zero. A
player may be able to choose only m pure strategies but he has an infinite number of mixed
strategies to choose from.
4
7. Optimal strategy : The strategy that puts the player in the most preferred position irrespective
of the strategy of his opponents is called an optimal strategy. Any deviation from this strategy
would reduce his payoff.
8. Zero-sum game : It is a game in which the sum of payments to all the players, after the play
of the game, is zero. In such a game, the gain of players that win is exactly equal to the loss of
players that lose e.g., two candidates fighting elections, wherein the gain of votes by one is the
loss of votes to the other.
9. Two-person zero-sum game : It is a game involving only two players, in which the gain of
one player equals the loss to the other. It is also called a rectangular game or matrix game
because the payoff matrix is rectangular in form. If there are n players and the sum of the game
is zero, it is called n-person zero-sum game. The characteristics of a two-person zero-sum game
are
(a) only two players are involved,
(b) each player has a finite number of strategies to use,
(c) each specific strategy results in a payoff,
(d) total payoff to the two players at the end of each play is zero.
10. Nonzero-sum game : Here a third party (e.g. the 'house' or a 'kitty') receives or makes some
payment. A payoff matrix for such a game is shown below. The left-hand entry in each cell is the
payoff to A, and the right-hand entry is the payoff to B
TABLE 6.1 Payoff Matrix of Player A and Player B
11. Payoff : It is the outcome of the game. Payoff (gain or game) matrix is the table showing the
amounts received by the player named at the left-hand-side after all possible plays of the game.
The payment is made by player named at the top of the table.
Let player A have m courses of action and player B have n courses of action. Then the game
can be described by a pair of matrices which can be constructed as described below.
5
(a) Row designations for each matrix are the courses of action available to player A.
(b) Column designations for each matrix are the courses of action available to player B.
(c) The cell entries are the payments to A for one matrix and to B for the other matrix. The cell
entry a
ij
is the payment to A in A's payoff matrix when A chooses the course of action i and B
chooses the course of action j.
.
TABLE 6.2 Player As Payoff Matrix
TABLE 6.3 Player Bs Payoff Matrix
(d) In a two-person zero-sum game, the cell entries in B's payoff matrix will be the negative of
6
the corresponding cell entries in A's payoff matrix. A is called maximizing player as he would
try to maximize the gains, while B is called minimizing player as he would try to minimize his
losses.
Thus the sum of payoff matrices for A and B is a null matrix. Here, the objective is to determine
the optimum strategies of both the players that result in optimum payoff to each, irrespective of
the strategy used by the other.
EXAMPLE 6.1 Consider the game in which two players simultaneously reveal 1, 2, or 3 fingers
each. If the sum of the revealed fingers is even, player B pays to player A the sum in dollars, if
the sum is odd, player A pays to player B the sum in dollars.
For this very simple two-person, zero-sum game, the pure strategies can be identified with the
individual activities. Furthermore, both players have the same set of pure strategies, {1, 2, 3}.
The payoff matrix is given in Table 6.4.
TABLE 6.4 payoff matrix (Example 6.1)
The term strategy refers to the decision by the Player A (or Player B) to choose a particular row
(or column). If each player uses the same strategy every time the game is played, then the game
is said to involve pure strategies. In the next section, we will look at games involving mixed
strategies, where the players try to outguess each other and vary their strategies each time the
game is played.
Pure Strategies
Consider a two-person, zero-sum game involving a payoff matrix P, in which the
Player A chooses a row and the Player B chooses a column. The entry in the chosen
row and column is the amount that the Player B must pay the Player A. The pure
strategy for each player is as follows.
1. Player A: Determine the smallest entry in each row. Choose the row containing the
largest of these smallest entries.
2. Player B. Determine the largest entry in each column. Choose the column
containing the smallest of these largest entries.
7
REMARK Be sure you see that these guidelines are only for strategies that are pure (to be used
each time the game is played). We will see that for many games, using the same strategy each
time is not the best (or optimal) strategy.
Usually, a player's strategy is to try to make a choice that will win the largest amount (or lose the
smallest amount). Such a strategy is called an optimal strategy. We will see that pure strategies
may or may not be optimal.
Examples 6.2 and 6.3 show how to find the pure strategies for various types of payoff matrices.
EXAMPLE 6.2 Finding Pure Strategies
Determine the pure strategies for the payoff matrix A.
(
(
=
1 2
7 5
A
discussed at the beginning of this section. If each player uses a pure strategy, what will the
payoff be? For this game, is the pure strategy optimal?
SOLUTION
1. For the Player A, the smallest entry in the first row is a
12
= -7 and the smallest entry in the
second row is a
22
= 1.
Since the largest of these two is 1, the Player A should choose the second row.
2. For the Player B, the largest entry in the first column is a
11
= 5 and the largest entry in the
second column is a
22
= 1.
8
Since the smallest of these two is 1, the Player B should choose the second column.
Assuming the Player A chooses the second row and the Player B chooses the second column, the
payoff for the game is p
22
= 1, which means that the Player B must pay the Player A $1 every
time the game is played.
Note that if either player uses a strategy other than the pure strategy, then that player will not do
as well. Thus, the pure strategy is the optimal strategy for both players.
REMARK Remember that entries in the payoff matrix are always listed in terms of amounts
that the Player B must pay the Player A. Thus, positive numbers are good for the Player A, and
negative numbers are good for the Player B.
EXAMPLE 6.3 Finding Pure Strategies
Determine the pure strategies for the following payoff matrix as shown in Table 6.5.
TABLE 6.5 payoff matrix (Example 6.3)
If each player uses a pure strategy, what will the payoff be?
SOLUTION
1. For the Row Player, the smallest entry in the first row is a
12
= - 6, the smallest entry in the
second row is a
22
= -1, and the smallest entry in the third row is a
32
= -7.
Since the largest of these three is -1, the Row Player should choose the second row.
2. For the Player B, the largest entry in the first column is a
21
= 5, the largest entry in the second
column is a
22
=- 1, and the largest entry in the third column is a
23
=- 6.
9
Since the smallest of these three is -1, the Player B should choose the second column.
Assuming the Player A chooses the second row and the Player B chooses the second column,
the payoff for the game is a
22
= -1, which means that the Player A must pay the Player B $1
every time the game is played.
In Examples 6.2 and 6.3, note that the payoff is both the smallest entry in its row and the largest
entry in its column. We call such an entry a saddle point.
This result tells us that if a payoff matrix has a saddle point, then each player should use a pure
strategy, choosing the row or column that contains the saddle point.
It is possible for a payoff matrix to have more than one saddle point. For instance, the payoff
matrix
(
(
(
=
9 1 10 2
3 11 5 3
P
saddle point
Consider a two-person, zero-sum game involving a payoff matrix P. If the matrix P bas an
entry that is both smallest entryin its row and the largest entry in its column, then that entry is
called a saddle point of the payoff matrix. Games that have a saddle point are called strictly
determinedand in such games, the optimal strategy for each player is, the, pure strategy.
10
has two saddle points: a
11
= 3 and a
14
= 3. Both of these entries are the least entries in their row
and the greatest entries in their column. Thus, the pure strategy for the Player A is to choose the
first row, and the pure strategy for the Player B is to choose either the first column or the fourth
column.
Many payoff matrices do not have saddle points, and are therefore not strictly determined. We
illustrate this type of matrix in Example 6.4.
EXAMPLE 6.4 A Payoff Matrix with No Saddle Point
Show that the payoff matrix given in Table 6.6
TABLE 6.6 payoff matrix (Example 6.4)
(
(
(
(
=
4 3
3 1
0 6
5 1
P
has no saddle point. If each player uses a pure strategy, what will the payoff be?
SOLUTION
To begin, we determine the pure strategy for each player.
1. For the Rowr A, the smallest row entries are a
11
=-1 (first row), a
22
= 0 (second row), a
32
= -3
(third row), and a
41
=-3 (fourth row). Since the largest of these is 0, the pure strategy for the
Player A is to choose the second row.
2. For the Player B, the largest column entries are a
21
= 6 (first column) and a
12
= 5 (second
column). Since the smallest of these is 5, the pure strategy for the Player B is to choose the
second column.
If each player uses a pure strategy, the payoff will be a
22
= 0, which means that no money is
exchanged between the two players. Note that a
22
= 0 is not a saddle point because it is not the
largest entry in the second column.
If you study the payoff matrix in Example 6.4, you can see why this game is not strictly
determined (that is, why a pure strategy might not be the best for each player). For instance,
suppose you are the Player A. Knowing that the Player B is likely to choose the second column,
you might try altering your strategy and choose the first row to get a payoff of $5. On the other
hand, the Player B knows that you might alter your strategy and choose the first row, so the
Player B might also decide to use an alternative strategy and choose the first column, which
11
would yield $1 to the Player B.
We will say more about this type of "mixed strategy" in the next section. For now the point is
this. If a payoff matrix has a saddle point, then the best strategy for each player is to use a pure
strategy every time the game is played. If a payoff matrix does not have a saddle point, then the
game becomes more interesting because one player may be able to outguess the other player and
increase his or her payoff.
We summarize these ideas as follows.
6.3 CREATING MODELS FOR GAMES
So far in this section, we have described games by listing a payoff matrix. In practice, most
applications of game theory involve an initial stage in which you must create the payoff matrix.
In order fora game to qualify as a two-person, zero-sum game, the game must have the following
features.
1. The game must have two players, usually called the Row Player and the Column Player.
2. Each time the game is played, the Row Player can make any of m different moves, and the
Column Player can make any of n different moves. Moreover, these moves are made
simultaneously so that neither player knows the move that the other player is going to make.
3. After each player has made a move, the rules of the game determine the payoff. This can be an
amount that the Column Player pays the Row Player (listed as a positive amount), an amount
that the Row Player pays the Column Player (listed as a negative amount), or a tie (listed as
zero).
Guidelines for Choosing a Strategy
Consider a two-person, zero-sum game involving a payoff matrix P. To determine the best
strategy for the players, use the following steps.
1. Determine the pure strategy for the Row Player.
2. Determine the pure strategy for the Column Player.
3. Determine the payoff that would result if both players used a pure strategy. If this payoff
is a saddle point of the matrix P, then the best strategy for both players is to use a pure
strategy every time the game is played. U the payoff is not a saddle point of the matrix P,
then the players should use a mixed strategy (which will be discussed in Section 6.##).
12
6.3.1 The Maxmin-Minmax Principle
This principle is used for the selection of optimal strategies by two players. Consider two player
A and B. A is a player who wishes to maximize his gain while player B wishes to minimize his
losses. Since A would like to maximize his minimum gain, we obtain for player A, the value
called Maxmin Value and the corresponding strategy is called the maxmin strategy.
On the other hand, since player B wishes to minimize his losses, a value called the Minmax
Value which is the minimum of the maximum losses is, found. The corresponding strategy is
called the minmax strategy. When these two are equal (maxmin value = minmax value), the
corresponding strategies are called optimal strategy and the game is said to have a saddle point.
The value of the game is given by the saddle point.
The selection of maxmin and minmax strategies by A and B is based upon the so-called
maxmin-minmax principle which guarantees the best of the worst results.
Saddle Point A saddle point is a position in the payoff matrix where the maximum of row
minima coincides with the minimum of column maxima. The payoff at the saddle point is called
the value of the game.
We shall denote the maxmin value by the minmax value of the game by and the value of the
game by .
Note (i) A game is said to be fair if
maxmin value=minmax value =0, i.e., if . =
(ii) A game is said to be strictly determinable if maxmin value= minmax value 0. . =
EXAMPLE 6.5 Solve the game whose pay off matrix is given by Table 6.7
TABLE 6.7 payoff matrix (Example 6.5)
13
SOLUTION
B
1
B
2
B
3
Row minima
1
-4
-1
Column maxima
A
1
A
2
A
3
1 5 1
3 4 0
1 3 1
1 5 1
Maxi (minimum) = Max (1,-4,-1)=1 Mini (maximum) = Min (1, 5, 1) = l .
ie., Maxmin value 1 = =Minmax value
Saddle point exists. The value of the game is the saddle point which is 1. The optimal strategy is
the position of the saddle point and is given by (A
1
, B
1
,).
EXAMPLE 6.6 For what value of , the game with the following matrix is strictly
determinable?
B
1
B
2
B
3
A
1
A
2
A
3
4 2
7 1
2 6
SOLUTION
B
1
B
2
B
3
Row minima
2
-7
-2
Column maxima
A
1
A
2
A
3
4 2
7 1
2 6
-1 6 2
The game is strictly determinable, if
= = Hence 2 = , 1 =
2 1
EXAMPLE 6.7 Determine which of the following two person zero sum games arc strictly
14
determinable and fair. Given the optimum strategy for each player in the case of strictly
determinable games.
(a)
B
1
B
2
A
1
A
2
4 7
2 5
(b)
B
1
B
2
A
1
A
2
3 4
1 1
SOLUTION
Maxi (minimum)= = Max (-5,-7)= -5 Mini (maximum)= =Min (-5, 2)=-5
(a)
B
1
B
2
A
1
A
2
4 7
2 5
Since 5 = , 5 = , the game is strictly determinable. There exists a saddle point = -5. Hence
the value of the game is -5. The optimal strategy is the position of the saddle point given by
(A
1
, B
1
).
(b)
B
1
B
2
A
1
A
2
3 4
1 1
Maxi (minimum) = = Max (1, - 3) = 1.
Mini (maximum) = = Min (4, 1) = 1.
Since 1 = = = , the game is strictly determinable. The value of game is 1. The optimal
strategy is (A
1
, B
2
).
15
EXAMPLE 6.8 Solve the game whose payoff matrix is given in Table 6.8.
TABLE 6.8 payoff matrix (Example 6.8)
(
(
(
(
6 2 4 3 5
6 2 0 3 4
2 2 1 2 3
3 5 0 0 2
SOLUTION
TABLE 6.9 Payoff matrix of player A with Maxmin and Minmax values
B
1
B
2
B
3
Row minima
-2
1
-4
-6
Column maxima
A
1
A
2
A
3
(
(
(
(
6 2 4 3 5
6 2 0 3 4
2 2 1 2 3
3 5 0 0 2
5 3 1 5 6
Maxi (minimum) = = Max (-2, 1, -4, -6)=1
Mini (maximum) = = Min (5, 3, 1, 5, 6) = 1.
Since 1 = = = , there exists a saddle point. the value of the game is 1. The position of the
saddle point is the optimal strategy and is given [A
2
, B
3
]
Applications to Business
Game theory has many applications other than "games." In the next two examples, we show how
game theory can be used to develop optimal strategies in business situations.
EXAMPLE 6.9 Leasing a Computer
A data processing firm rents computer time to several other companies. The firm has decided to
offer a new service which will require the purchase or lease of an additional computer system.
The management has already determined to lease (rather than buy) and has narrowed the choice
to three types: a large computer system, a medium-sized computer system, and a small computer
system. For each system, the firm's clients might have a high acceptance (meaning that 75% of
the clients would use the service), or a low acceptance (meaning that only 25% of the clients
would use the service). For each of these possibilities, the accounting department has predicted
the following profits for the firm given in Table 6.10.
16
TABLE 6.10 Payoff matrix of Example 6.9
Computer
Size
High
Acceptance
Low
Acceptance
Large $250,000 $30,000
Medium $200,000 $40,000
Small $150,000 $75,000
If you were required to decide which system to lease, what would your decision be?
SOLUTION
In practice, of course, you would probably try to predict the likelihood of either a high or low
acceptance. If there was a strong likelihood of a high acceptance for the large system, then
ordering the large system would be very tempting because it offers the greatest possible profit.
For this problem, however, we assume that no other information is available to aid you in
making your decision. Assuming that you are the Row Player, the payoff matrix is given in
Table 6.11.
TABLE 6.11 Payoff matrix of Example 6.9
(
(
(
(
(
=
000 , 75 000 , 150
000 , 40 000 , 200
000 , 30 000 , 250
A
Since you are the Row Player, your pure strategy is to choose the third row (the small system).
REMARK Be sure you see that the decision that was made in Example 6.9 corresponds to a
pure strategy. If you take a course in decision theory, you will see that other strategies could
have been employed, and these might lead to different choices.
EXAMPLE 6 .10 : Hard-Sell or Soft-Sell?
Your company, Company A, is planning an advertising campaign fora new product. Your major
competitor, Company B, is also producing new advertisements for its product. Each company
can gear its ads toward a "hard-sell" which would emphasize the benefits of its product and the
shortcomings of the competitor's product, or gear its ads toward a "soft-sell" that would only
emphasize the benefits of its product. The market research department has predicted the
17
following market shares (for your company) for hard-sell, soft-sell combinations as in Table
6.12.
TABLE 6.12 Payoff matrix of Example 6.10
Which type of advertising would you recommend for your company? Is this "game" strictly
determined?
SOLUTION
For this "game," the payoff matrix is as follows.
(
(
(
=
45 . 0 30 . 0
55 . 0 40 . 0
P
The pure optimal strategy for Company A, the Player A, is to choose the first row, and the pure
optimal strategy for Company B, the Player B, is to choose the first column. If both companies
use a pure optimal strategy, then the payoff will be p
11
= 0.40. Moreover, since this payoff is a
saddle point, the game is strictly determined.
EXERCISES
1. Determine which of the following two-person zero sum games are strictly determinable and
fair. Give the optimum strategies for each player in the case of strictly determinable games
(a)
(
2 0
0 5
(b)
(
4 1
2 0
[Ans: Not strictly determinable] [Ans: Fair]
2. For the game with payoff matrix
(
6 4 6
2 2 1
determine the best strategies for players A and B and also the value of the game for them. Is this
18
game (1) fair, (ii) strictly determinable. [Ans: Value of game is -2. Game is not fair but strictly
determinable]
3. Determine the optimal minmax strategies for each player in the following game as in Table
6.13.
TABLE 6.13 Payoff matrix
(
(
(
3 2 0 4
8 4 6 5
7 0 2 5
[Ans: 4 = , (A
2
, B
3
) is the optimum strategy].
In Exercises 4-7, consider a two-person game in which each person simultaneously chooses an
integer: 1, 2, or 3. Construct a payoff matrix for the given game.
4. If the sum of the two numbers is odd, then the Player A pays the Player B that sum (in
dollars). If the sum of the two numbers is even, then the Player B pays the Player A $4.
5. If both players choose the same number, then no money is exchanged. Otherwise, the player
with the higher number wins $1.
6. If both players choose the same number, then no money is exchanged. If the two numbers
differ by 1, then the player with the higher number wins $2. If the two numbers differ by 2, then
the player with the higher number wins $1.
7. If the players choose the same number, no money is exchanged. If the sum of the two numbers
is even, then the player with the higher number wins $2. If the sum of the two numbers is odd,
then the player with the lower number wins $2.
8. Coin-Tossing Game Two players each have a coin. Simultaneously, each player chooses to
lay his or her coin on the table (heads up or tails up). If both players choose heads, then the
Player Bwins $2. If both players choose tails, then. the Player A wins $3. If the players choose
different sides, then the player who chooses heads wins $1. What is the pure strategy for each
player? Is this game strictly determined?
9. Two players each have a coin. Simultaneously, each player chooses to lay his or her coin on
the table (heads up or tails up). If both players choose heads, then the Player A wins $2. If both
players choose tails, then the Player A wins $3. If the coins differ, then the Player B wins $1 or
$2, depending on whether the Player A's coin is heads or tails, respectively. What is the pure
strategy for each player? Is this game strictly determined?
19
10. Two players play "Stone, Scissors, and Paper," a game in which each player simultaneously
chooses either a stone, a pair of scissors, or paper. If there is a tie, there is no payoff. A stone
"breaks" scissors and is paid $2. Paper "covers" stone and is paid $3. Scissors "cut" paper and is
paid $4. What is the pure strategy for each player? Is this game strictly determined?
11. Repeat Exercise 10 using the following payoffs. A stone "breaks" scissors and is paid $3.
Paper "covers" stone and is paid $2. Scissors "cut" paper and is paid $4.
12. Two players each have three cards numbered 2, 3, and 4. If both players choose the same
number, no money is exchanged. Otherwise, the player with the higher number wins the sum of
the two numbers. What is the pure strategy for each player? Is this game strictly determined?
13.Two players simultaneously choose an integer from 1 to 3. If they choose the same number,
the Row Player wins $1. If the two numbers differ by 1, the player with the higher number wins
$3. If the two numbers differ by 2, the player with the higher number wins $2. What is the pure
strategy for each player? Is this game strictly determined?
14. A person buys a large tract of land, which can be used for condominiums, a hotel, or a
restaurant. There is a possibility that a highway will be built near this land. The weekly profit (in
dollars) for each type of construction is asin Table 6.14.
TABLE 6.14 Payoff matrix
Highway No Highway
Condominiums -1000 6000
Hotel 8000 -3000
Restaurant 6000 5000
What is the pure strategy for this "player?" Is this game strictly determined?
15. In a presidential campaign, there are two candidates and three major issues. Each candidate
must decide to campaign as "pro" or "anti" with respect to each of the three issues. The
"payoffs" (listed in terms of Candidate A) are as follows.
16. In a presidential campaign, there are two candidates and three major issues. Each candidate
must decide to campaign as "pro" or "anti" with respect to each of the three issues. The
"payoffs" (listed in terms of Candidate A) are in Table 6.15.
20
TABLE 6.15 Payoff matrix
Candidate B
Issue 1 Issue 2 Issue 3
Candidate A
Issue 1 3 -2 -1
Issue 2 -1 2 4
Issue 3 2 3 1
What is the pure strategy for Candidate A? Is this game strictly determined?
17.There are three construction sites in town. Two companies supply snacks from trucks. Site 1
has 50% of the customers, Site 2 has 30%, and Site 3 has 20%. If both trucks go to the same site,
they split the customers equally. If they go to different sites, each gets the entire business at that
site, and they split the business at the third site. Construct the payoff matrix and determine where
the trucks should go.
18. Repeat Exercise 17 assuming that Site 1 has 35% of the customers, Site 2 has 40% of the
customers, and Site 3 has 25% of the customers.
19. Several firms are building new offices and are taking bids from the two most prominent
construction companies in the area: Company A and Company B. When bidding, the companies
choose between three strategies to win the bid: cost, time, and quality. From past history, the
market share for Company A is given in Table 6.16.
TABLE 6.16 Payoff matrix
Company B
Cost Time Quality
Company A
Cost 50% 40% 30%
Time 55% 55% 45%
Quality 50% 60% 60%
What is the pure strategy for Company A? Is this game strictly determined?
20.Repeat Exercise 19 using the following assumptions. Company A's quality has fallen. As a
result, if Company A chooses quality as its strategy, its market share will be 10% less than
before.
21
6.4 GAMES WITHOUT SADDLE POINTS(MIXED STRATEGIES)
6.4.1 Expected Value of a Game
In Section 6.3, we looked at several examples of games whose payoff matrices have saddle
points. As we have already mentioned, such games are strictly determined and both players
should use a pure strategy every time the game is played.
For payoff matrices that do not have saddle points, the choice of a best strategy is more difficult.
To see why, let's consider the following simple payoff matrix.
(
=
1 3
2 4
A
The pure strategy for the Row Player would be to choose the first row, and the pure strategy for
the Column Player would be to choose the second column. These two choices would result in a
payoff of a
12
= -2, which is not a saddle point because it is not the largest entry in the second
column.
Note that for this payoff matrix, if the Column Player chooses the second column each time the
game is played, then the Row Player would soon catch on, and switch to the second row. On the
other hand, if the Row Player started choosing the second row, then a smart Column Player
would start choosing the first column. The idea is simply this: if a payoff matrix does not have a
saddle point, then each player should use a mixed strategy.
For instance, suppose that the probability that the Row Player will choose the first row is 0.40,
and the probability that the Column Player will choose the first column is 0.25. Over the long
run, what would the payoff for this game be? To determine this, we find the expected value of
the game as follows.
Payoff is a
11
= 4: Probability = p
1
q
1
= (0.4)(0.25) = 0.10
Payoff is a
12
=2: Probability = p
I
q
2
= (0.4)(0.75) = 0.30
Payoff is a
21
= -3: Probability = p
2
q
1
= (0.6)(0.25) = 0.15
Payoff is a
22
= 1: Probability = p
2
q
2
= (0.6)(0.75) = 0.45
Thus, the expected value for this game is
E= p
1
q
1
a
11
+ p
I
q
2
a
I2
+p
2
q
I
a
21
+p
2
q
2
a
22
= (0.4)(0.25)(4) + (0.4)(0.75)(-2) + (0.6)(0.25)(-3) + (0.6)(0.75)(1)
= -0.2.
In other words, if this game is played many times, then the average payoff per game would be
22
-0.2. In matrix notation, we can write the formula for the expected value as
| |
(
=
2
1
22 21
12 11
2 1
q
q
a a
a a
p p E
In this definition, the matrix
| |
m 3 2 1
p ... p p p P =
is called the strategy for the Player A because it gives the probabilities that the Player A will
make each of the m possible "row-moves." Similarly, the matrix
(
(
(
(
(
(
(
=
n
3
2
1
q
....
...
q
q
q
Q
is called the strategy for the Player B because it gives the probabilities that the Player B will
make each of the n possible "column-moves."
Note that the entries in a "strategy matrix" must be nonnegative because they represent
probabilities. Moreover,
1 q .... q q q and 1 p .... p p p
n 3 2 1 m 3 2 1
= + + + + = + + + +
Expected Value of a Game
Consider a two-person, zero-sum game with a payoff matrix P. If the probability that the Row
Player will choose the ith row is ri, and the probability that the Column Player will choose the
jth column is cp then the expected value of the game is
| |
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(
n
3
2
1
mn mj 3 m 2 m 1 m
in ij 3 i 2 i 1 i
n 3 j 3 33 32 31
n 2 j 2 23 22 21
n 1 j 1 13 12 11
m 3 2 1
q
....
...
q
q
q
a ...... a ...... a a a
a ...... a ..... a a a
... .... ... .... .... ...
a ....... a ...... a a a
a ...... a .... a a a
a ...... a ... a a a
p ... p p p
23
A strategy is called a mixed strategy if at least two entries in the strategy matrix are not zero. If
one of the entries is 1 (and the other entries are zero), then the strategy is called a pure strategy.
For instance,
0.00
P = [0.25 0.10 0.65] and
/
] 40 . 0 30 . 0 30 . 0 00 . 0 [ Q =
are mixed strategies, whereas
P = [0 1 0] and
/
] 1 0 0 0 [ Q =
are pure strategies.
EXAMPLE 6.11 Finding the Expected Value of a Game
A two-person, zero-sum game has a payoff matrix of
(
=
1 2
3 5
A
Find the expected value of the game if the players use the following mixed strategies.
(a)
(
=
50 . 0
50 . 0
Q ] 50 . 0 50 . 0 [ P
(b)
(
=
75 . 0
25 . 0
Q ] 60 . 0 40 . 0 [ P
(c)
(
=
60 . 0
40 . 0
Q ] 75 . 0 25 . 0 [ P
SOLUTION
(a) For these strategies, the expected value is
25 . 0
50 . 0
50 . 0
] 01 . 1 50 . 1 [
50 . 0
50 . 0
1 2
3 5
] 50 . 0 50 . 0 [ PAQ E
=
(
=
(
= =
(b) For these strategies, the expected value is
25 . 0
75 . 0
25 . 0
] 60 . 0 80 . 0 [
75 . 0
25 . 0
1 2
3 5
] 60 . 0 40 . 0 [ PAQ E
=
(
=
(
= =
(c) For these strategies, the expected value is
24
10 . 0
60 . 0
40 . 0
] 00 . 0 25 . 0 [
60 . 0
40 . 0
1 2
3 5
] 75 . 0 25 . 0 [ PAQ E
=
(
=
(
= =
Note in Example 6.11 that the expected value of the game can change, depending on the
strategies that are used by the Player A and the Player B.
EXAMPLE 6.12 Comparing Expected Values for Mixed Strategies A two-person, zero-sum
game has a payoff matrix of
(
(
(
=
1 2 4
1 4 2
2 1 3
A
The Player A knows that the strategy used by the Player B will be
(
(
(
=
50 . 0
20 . 0
30 . 0
Q
Which of the following strategies would be better for the Row Player?
(a) P = [0.30 0.40 0.30] (b) P = [0.20 0.60 0.20]
SOLUTION
(a) For the strategy P= [0.30 0.40 0.30], the expected value of the game is
| | 02 . 2
50 . 0
20 . 0
30 . 0
1 2 4
1 4 2
2 1 3
30 . 0 40 . 0 30 . 0 PAQ =
(
(
(
(
(
(
=
(b) For the strategy R [0.20 0.60 0.20], the expected value of the game is
| | 98 . 1
50 . 0
20 . 0
30 . 0
1 2 4
1 4 2
2 1 3
20 . 0 60 . 0 20 . 0 PAQ =
(
(
(
(
(
(
=
Since larger payoffs are better for the Row Player, it would be better for the Row Player to use
the first strategy.
In Example 6.12, can you think of a strategy for the Player A that is better than either of the two
given? Can you think of a strategy for the Player B that is better than the one given? The
25
following theorem, called the Fundamental Theorem of Game Theory tells us that there is
always a best (or optimal) strategy for the Player A and a best (or optimal) strategy for the
Player B.
V AQ P AQ P and V AQ P PAQ
* * * * * *
= =
To see why P* and Q* are called optimal, remember that large payoffs are good for the Player
A. Thus, the statement
* * *
AQ P PAQ
means that if the Player A chooses any strategy P other than the optimal strategy P*, then the
payoff will be less than or equal to the payoff for the optimal strategy. Similarly, since small
payoffs are good for the column player, the statement
* * *
AQ P AQ P
means that if the Column Player chooses any strategy Q other than the optimal strategy Q*, then
the payoff will be greater than or equal to the payoff for the optimal strategy.
6.4.2 Finding Optimal Mixed Strategies
Note that the Fundamental Theorem of Game Theory tells us of the existence of the strategies P*
and Q*, but does not tell us how to find these optimal strategies. Of course, if the game is strictly
determined, then we know that the optimal strategy is to use a pure strategy.
Finding optimal mixed strategies fora game can be a tedious process. In Section 6.7##, we will
show how linear programming can be used to find optimal mixed strategies for an m x n payoff
matrix. For a 2 x 2 payoff matrix, however, the problem is straightforward, as seen in the
following result.
Fundamental Theorem of Game Theory
Consider a two-person, zero-sum game with a payoff matrix A. There exist optimal
strategies P* and Q* and a unique payoff P*AQ* = V such that for any other strategies P
and Q,
V AQ P AQ P and V AQ P PAQ
* * * * * *
= =
The number V is called the value of the game. H the value of the game is zero, then the
game is called fair.
26
EXAMPLE 6.13 Finding Optimal Mixed Strategies
In Example 6.11, we found the expected value for three different pairs of strategies for the
payoff matrix
(
=
1 2
3 5
A
Find the optimal strategies for this payoff matrix. What is the value of the game? Is the game
fair?
SOLUTION
First, you should check to see that this payoff matrix has no saddle points. Next, applying the
formula for the optimal row strategy produces
11
3
) 2 ( ) 3 ( 1 5
) 2 ( 1
) a a ( ) a a (
a a
p
21 12 22 11
21 22
1
=
+
=
+ +
=
which implies that
.
11
8
p 1 p
1 2
= =
Similarly, applying the formula for the optimal column strategy produces
11
4
) 2 ( ) 3 ( 1 5
) 3 ( 1
) a a ( ) a a (
a a
q
21 12 22 11
12 22
1
=
+
=
+ +
=
Optimal Mixed Strategy for 2 x 2 Payoff Matrix
Consider a 2 x 2 payoff matrix
21 12 22 11
22 21
12 11
a a a a ,
a a
a a
A +
(
=
that has no saddle points. The optimal strategy for the Player A is to play the first tow with a
probability of
) a a ( ) a a (
a a
p
21 12 22 11
21 22
1
+ +
= and the second row with a probability of . p 1 p
1 2
= The
optimal strategy for the Player B is to play the first column with a probability of
) a a ( ) a a (
a a
q
21 12 22 11
12 22
1
+ +
= , and the second column with a probability of
1 2
q 1 q =
27
which implies that
.
11
7
q 1 q
1 2
= =
Thus, the optimal strategies for the payoff matrix are
(
=
11
8
11
3
P
*
and
(
(
(
=
11
7
11
4
Q
*
Using these strategies, we find the value of the payoff matrix to be
. 09 . 0
11
1
11
7
11
4
1 2
3 5
11
8
11
3
AQ P V
* *
= =
(
(
(
= =
3
Because the value of the game is not zero, we conclude that the game is not fair. (It favors the
Player B.)
EXAMPLE 6.14 Finding Optimal Mixed Strategies
Two people play a card-matching game as follows. Both players have two cards, a 2 and a 3.
Simultaneously, each player chooses one of the cards and lays it face up on the table. The
payoffs for the game are as in Table 6.17.
TABLE 6.17 payoff Matrix (Example 6.14)
Player A Player B Payoff
2 2 Player 2 pays Player 1 $4
2 3 Player 1 pays Player 2 $12
3 2 Player 1 pays Player 2 $3
3 3 Player 2 pays Player 1 $9
What are the optimal strategies for each player? Is this game fair?
SOLUTION
The payoff matrix for this game is
(
=
9 3
12 4
A
Moreover, this payoff matrix has no saddle point, so the game is not strictly determined (which
28
means that pure strategies are not appropriate). The optimal strategy for the Row Player is given
by
7
3
28
12
) 3 ( ) 12 ( 9 4
) 3 ( 9
) a a ( ) a a (
a a
p
21 12 22 11
21 22
1
= =
+
=
+ +
= and
.
7
4
p 1 p
1 2
= =
The optimal strategy for the Column Player is given by
4
3
28
21
) 3 ( ) 12 ( 9 4
) 12 ( 9
) a a ( ) a a (
a a
q
21 12 22 11
12 22
1
= =
+
=
+ +
= and
.
4
1
q 1 q
1 2
= =
Thus, the optimal strategies for the two players are
(
=
7
4
7
3
P
*
and
(
(
(
=
4
1
4
3
Q
*
Using these strategies, we find the value of the payoff matrix to be
0
4
1
4
3
9 3
12 4
7
4
7
3
AQ P
* *
=
(
(
(
=
Because the value of the game is zero, we conclude that the game is fair.
EXAMPLE 6.15 Solve the following game and determine the value of the game
(
4 4
4 4
Solution It is clear that the pay off matrix does not possess any saddle point. The players will
use mixed strategies.
The optimum mixed strategy for the players are
Where
2
1
16
8
) 4 4 ( ) 4 4 (
) 4 ( 4
) a a ( ) a a (
a a
p
21 12 22 11
21 22
1
= =
+
=
+ +
= , and
2
1
p 1 p
1 2
= =
29
2
1
16
8
) 4 4 ( ) 4 4 (
) 4 ( 4
) a a ( ) a a (
a a
q
21 12 22 11
12 22
1
= =
+
=
+ +
= and
2
1
q 1 q
1 2
= =
The value of the game is = V 0
) 4 4 ( ) 4 4 (
) 4 ( ) 4 ( 4 4
=
+
The optimum mixed strategies
(
=
2
1
2
1
P
*
and
(
(
(
=
2
1
2
1
Q
*
The value of the game is 0 V = .
EXAMPLE 6.16 In a game of matching coins with two players suppose A wins one unit of
value when there are two heads, wins nothing when there are two tails and losses 1/2 unit of
value when there are one head and one tail Determine the payoff matrix, the best strategies for
each player and the value of game to A.
SOLUTION The payoff matrix for the player A is given by
H T
H
T
(
(
(
0
2
1
2
1
1
Let this be
(
22 21
12 11
a a
a a
and The optimum mixed strategies
The optimum mixed strategy for the players are
30
(
=
2 1
2 1
A
p p
A A
S and
(
=
2 1
2 1
B
q q
B B
S
Where
4
1
) 2 / 1 2 / 1 ( ) 0 1 (
) 2 / 1 ( 0
) a a ( ) a a (
a a
p
21 12 22 11
21 22
1
=
+
=
+ +
= , 1 p p
2 1
= +
4
3
p 1 p
1 2
= =
4
1
) 2 / 1 2 / 1 ( ) 0 1 (
) 2 / 1 ( 0
) a a ( ) a a (
a a
q
21 12 22 11
12 22
1
=
+
=
+ +
=
, 1 q q
2 1
= + ,
4
3
q 1 q
1 2
= =
The value of the game is = V
8
1
) 2 / 1 2 / 1 ( ) 0 1 (
) 2 / 1 ( ) 2 / 1 ( 0 1
=
+
The optimum mixed strategies
(
=
4
3
4
1
P
*
and
(
(
(
=
4
3
4
1
Q
*
The value of the game is
8
1
V =
EXERCISES
1. For a game with the following payoff matrix, determine the optimal strategy and the value of
the game.
(i)
(
0 3
3 6
[Ans.
(
=
2
1
2
1
P
*
and
(
(
(
=
4
3
4
1
Q
*
The value of the game is
4
3
V = ]
(ii)
(
1 4
5 2
31
[Ans.
(
=
4
3
4
1
P
*
and
(
(
(
=
3
1
3
2
Q
*
The value of the game is 3 V = ]
2. Two players A and B match coins. If the coins match, then A wins two units of value. If coins
do not match, then B wins two units of value. Determine the optimum strategies for the players
and the value of the game.
(
=
2
1
2
1
P
*
and
(
(
(
=
2
1
2
1
Q
*
The value of the game is 0 V = .
In Exercises 19-24, find the value of the given payoff matrix, and determine whether the game is
fair. If it is not fair, which player does it favor?
19.
(
=
6 5
4 3
A 20.
(
=
1 0
1 2
A
21.
(
=
4 3
1 2
A 22.
(
=
3 3
1 4
A
23.
(
=
9 8
10 16
A 24.
(
=
7 12
8 8
A
25. Card Game Two players have two cards each: a 3 and a 4. They each choose one of the
cards. The payoffs are as in Table 6.18.
TABLE 6.18 payoff Matrix
Player 1 Player 2 Payoff
3 3 Player 2 pays Player 1 $10
3 4 Player 1 pays Player 2 $20
4 3 Player 1 pays Player 2 $10
4 4 Player 2 pays Player 1 $15
What is the optimal strategy for each player? What is the value of the game? Is the game fair?
26. Card Game Two players have two cards each: a face card and a 10. They each choose one of
the cards. The payoffs are as in Table 6.19.
32
TABLE 6.19 payoff Matrix
Player 1 Player 2 Payoff
face card face card Player 2 pays Player 1 $5
face card ten Player 1 pays Player 2 $8
ten face card Player 1 pays Player 2 $5
ten ten Player 2 pays Player 1 $10
What is the optimal strategy for each player? What is the value of the game? Is the game fair?
27. Card Game Two players have one card each. Simultaneously, they each lay their card on the
table: either face-up or face-down. If both choose the same side, then Player 1 wins $10. If
Player I's card is face-up and Player 2's card is face-down, then Player 2 wins $5. If Player I's
card is face-down and Player 2's card is face-up, then Player 2 wins $10. What is the optimal
strategy for each player? What is the value of the game? Is the game fair?
28. Guessing Game Two people play the following guessing game. One player thinks of a
number: either 1 or 2. The second player tries to guess the number. If the guess is correct, the
second player wins $5. If the guess is incorrect, the first player wins $8. What is the optimal
strategy for each player? What is the value of the game? Is the game fair?
29. Career Choice You are offered a promotion to manage a new branch office. You assign
values to the possible outcomes of accepting or not accepting the promotion as in Table 6.20.
TABLE 6.20 payoff Matrix
Branch
Succeeds
Branch
Fails
Accept Promotion 10 4
Decline Promotion 2 9
Based on these payoffs, should you accept the promotion?
30. Political Campaign A politician is debating whether to visit a certain city during the
campaign. The politician assigns the following numerical values to the possible outcomes.
Visit and win city's support: 7
Visit and lose city's support: -5
Do not visit and win city's support: 9
Do not visit and lose city's support: -6
If you were this person, would you visit the city?
33
31. Concert Attendance Assume that a concert is being held and you are debating whether or not
to go. Assign numerical values to the following possible outcomes. Based on your assigned
values, should you attend the concert?
You attend and the concert is good.
You attend and the concert is bad.
You do not attend and the concert is good.
You do not attend and the concert is bad.
32. Committee Chair Choice Assume that you are asked to be chairperson of a committee. If you
do not accept the position, your employer will ask one of your coworkers to chair the committee.
Assign numerical values to the following possible outcomes. Based on your assigned values,
should you accept the assignment?
You chair the committee and the outcome is good. You chair the committee and the outcome is
bad. Your coworker chairs the committee and the outcome is good.
Your coworker chairs the committee and the outcome is bad.
33. Computer Game Use the computer program given in Appendix A to play the following game
10 times. Two people have two cards, a 2 and a 3. Simultaneously, each player chooses one of
the cards and lays it face up on the table. The payoffs for the game are as in Table 6.21.
TABLE 6.21 payoff Matrix
Player 1 Player 2 Payoff
2 2 Player 2 pays Player 1 $4
2 3 Player 1 pays Player 2 $12
3 2 Player 1 pays Player 2 $3
3 3 Player 2 pays Player 1 $9
You are Player 1, the Row Player. How much money did you win? Compare your results with
the expected value for playing this game ten times.
6.5 DOMINANCE PROPERTY
In some games, it is possible to reduce the size of the payoff matrix by eliminating redundant
rows (or columns). If a game has such redundant rows (or columns), those rows or columns are
dominated by some other rows (or columns), respectively. Such property is known as dominance
property.
34
EXAMPLE 6.17
Players A and B play a game in which each player has three coins (20p, 25p and 50p). Each of
them selects a coin without the knowledge of the other person. If the sum of the values of the
coins is an even number, A wins B's coin. If that sum is an odd number, B wins A's coin.
(a) Develop a payoff matrix with respect to Player A.
(b) Find the optimal strategies for the players.
SOLUTION
The payoff matrix with respect to Player A is shown in Table 6.22.
Dominance property for columns
(a) In the payoff matrix of Player A, if all the entries in a column (X) are lesser than to the
corresponding entries of another column (Y), then column Y is said to be strictly dominated by
column X. Under such situation, the column Y of the payoff matrix can be deleted.
(b) In the payoff matrix of Player A, if all the entries in a column (X) are lesser than or equal to
the corresponding entries of another column (Y), then column Y is is said to be weekly
dominated by column X. Under such situation, the column Y of the payoff matrix can be
deleted.
Dominance property for rows
(a) In the payoff matrix of Player A, if all the entries in a row (X) are greater than to the
corresponding entries of another row (Y), then row Y is said to be strictly dominated by row
X. Under such situation, row Y of the payoff matrix can be deleted.
(b) In the payoff matrix of Player A, if all the entries in a row (X) are greater than or equal to
the corresponding entries of another row (Y), then row Y is said to be weekly dominated by
row X. Under such situation, row Y of the payoff matrix can be deleted.
35
The maxmin value (-20) is not equal to the minmax value (20). Hence, the game has no saddle
point. As a result, the game has mixed strategies.
Check for dominance property. Row III is weekly dominated by row I and hence row III is to be
deleted. The resultant matrix after deleting row III is shown in Table 6.23.
TABLE 6.23 Payoff Matrix After Deleting Row III
IN Table 6.23 the column III is weekly dominated by the column I and hence, column III is to be
deleted. The resultant matrix after deleting column III is shown in Table 6.24. Now payoff
matrix is reduced to 2x2 matrix and appply usual method to find optimal mixed strategies with
respect to paalayer A and player B.
TABLE 6.24 Payoff Matrix After Deleting Column III
The optimum mixed strategy for the players are
Where
9
4
90
50
) 25 20 ( ) 25 20 (
) 25 ( 25
) a a ( ) a a (
a a
p
21 12 22 11
21 22
1
= =
+
=
+ +
= , and
9
5
p 1 p
1 2
= =
2
1
90
45
) 25 20 ( ) 25 20 (
) 20 ( 25
) a a ( ) a a (
a a
q
21 12 22 11
12 22
1
= =
+
=
+ +
= and
36
2
1
q 1 q
1 2
= =
The value of the game is = V 0
) 25 20 ( ) 25 20 (
) 20 ( ) 25 ( 25 20
=
+
The optimum mixed strategies
(
=
9
5
9
4
P
*
and
(
(
(
=
2
1
2
1
Q
*
The value of the game is 0 V = .
EXAMPLE 6.18 Consider the 4 x 4 game as shown in Table 6.25 which represents the payoff
matrix of the Player A. Solve it optimally.
TABLE 6.25 Payoff Matrix for Example 6.18
(
(
(
(
(
(
(
(
4 3 4 3
1 4 2 4
4 2 4 3
0 4 2 3
SOLUTION The check for saddle point is done as shown in Table 6.26. Since, the maxmin
value (3) is not equal to the minmax value (4), the game has no saddle point. This means that the
game has mixed strategies for both the players.
TABLE 6.26 Payoff Matrix with Check for Saddle Point
Check for dominance property. In Table 6.26, row 3 dominates row 1 and the corresponding
reduced payoff matrix is shown in Table 6.27. In Table 6.27, column 3 dominates column I and
37
the corresponding reduced payoff matrix is shown in Table 6.28. In Table 6.28, row 4 dominates
row 2 and the corresponding reduced payoff matrix is shown in Table 6.29. In Table 6.29,
column 4 dominates column 2 and the corresponding reduced payoff matrix (2 x 2 matrix) is
shown in Table 6.30.
TABLE 6.27 Payoff Matrix after Deleting Row 1
Player B
1 2 3 4
2 3 4 2 4
Player A 3 4 2 4 1
4 3 4 3 4
TABLE 6.28 Payoff Matrix after Deleting Column I
Palyer B
2 3 4
2 4 2 4
Player A 3 2 4 1
4 4 3 4
TABLE 6.29 Payoff Matrix after Deleting Row 2
Palyer B
2 3 4
Player A 3 2 4 1
4 4 3 4
TABLE 6.30 Payoff Matrix after Deleting Column 2
Player B
3 4
Player A 3 4 1
4 3 4
The optimum mixed strategy for the players are
38
Where
4
1
) 1 3 ( ) 4 4 (
) 3 ( 4
) a a ( ) a a (
a a
p
21 12 22 11
21 22
3
=
+ +
=
+ +
= , and
4
3
p 1 p
3 4
= =
4
3
) 1 3 ( ) 4 4 (
1 4
) a a ( ) a a (
a a
q
21 12 22 11
12 22
3
=
+ +
=
+ +
= and
4
1
q 1 q
3 4
= =
The value of the game is = V
4
13
) 1 3 ( ) 4 4 (
1 3 4 4
=
+ +
The optimum mixed strategies
(
=
4
3
4
1
0 0 P
*
and
(
(
(
(
(
(
(
(
=
4 / 1
4 / 3
0
0
Q
*
The value of the game is
4
13
V = .
EXAMPLE 6.19
A company is currently involved in negotiations with its union on the upcoming wage contract.
Positive signs in Table 6.31 represent wage increase while negative sign represents wage
reduction. What are the optimal strategies for the company as well as the union? What is the
game value ?
TABLE 6.31 Conditional Costs to the Company (Rs. in lakhs)
Union Strategies
U
1
U
2
U
3
U
4
C
1
Company C
2
strategiesC
3
C
4
0.25 0.27 0.35 -0.02
0.20 0.16 0.08 0.08
0.14 0.12 0.15 0.13
0.30 0.14 0.19 0.00
SOLUTION
39
Since in a game matrix, player to its left is a maximizing player and me one at the top is a
minimizing player, Table 6.31 is transposed and rewritten as Table 6.32 since the companys
interest is to minimize the wage increase while union's interest is to get the maximum wage
increase.
TABLE 6.32 Payoff Matrix of Unions
Company strategies
C
1
C
2
C
3
C
4
U
1
Union U
2
Strategies U
3
U
4
0.25 0.20 0.14 0.30
0.27 0.16 0.12 0.14
0.35 0.08 0.15 0.19
-0.02 0.08 0.13 0.00
In Table 6.32, row U
4
is dominated by row U
1
as well as U
3
. It is, therefore, deleted to give
Table 6.33.
TABLE 6.33 Payoff Matrix after Deleting Row 4
Company strategies
C
1
C
2
C
3
C
4
U
1
Union U
2
Strategies U
3
0.25 0.20 0.14 0.30
0.27 0.16 0.12 0.14
0.35 0.08 0.15 0.19
In Table 6.33, column C
1
is dominated by C
2
as well as C
3
, while C
4
is dominated by C3.
Deleting columns C
1
and C4 we get
TABLE 6.34 Payoff Matrix after Deleting Columns 1 and 4.
Company strategies
U
1
Union U
2
Strategies U
3
C
2
C
3
0.20 0.14
0.16 0.12
0.08 0.15
In Table 6.34, row U
2
is dominated by U
1
and is, therefore, deleting U
2
we get,
TABLE 6.35 Payoff Matrix after Deleting Row 2
Company strategies
40
Union U
1
Strategies U
3
C
2
C
3
0.20 0.14
0.08 0.15
The optimum mixed strategy for the players are
Where
13
7
13 . 0
07 . 0
) 14 . 0 08 . 0 ( ) 15 . 0 20 . 0 (
) 08 . 0 ( 15 . 0
) a a ( ) a a (
a a
p
21 12 22 11
21 22
1
= =
+ +
=
+ +
= , and
13
6
p 1 p
1 3
= =
13
1
13 . 0
01 . 0
) 14 . 0 08 . 0 ( ) 15 . 0 20 . 0 (
14 . 0 15 . 0
) a a ( ) a a (
a a
q
21 12 22 11
12 22
2
= =
+ +
=
+ +
= and
13
12
q 1 q
1 3
= =
The value of the game is = V
13
188
) 14 . 0 08 . 0 ( ) 15 . 0 20 . 0 (
14 . 0 08 . 0 15 . 0 20 . 0
=
+ +
Arithmetic method yields the following conclusions
The optimum strategy for company: ] 0 , 13 / 12 , 13 / 1 , 0 [ ] q , q , q , q [
4 3 2 1
=
The optimum strategy for union: ] 0 , 13 / 6 , 0 , 13 / 7 [ ] p , p , p , p [
4 3 2 1
=
The value of the game is 13 / 188 . Rs V = .
EXERCISE
There are two competing departmental stores R and C in a city. Both stores have equal
reputation and the total number of customers is equally divided between the two. Both the stores
plan to run annual discount sales in the last week of December. For this, they want to attract
more number of customers by using advertisement through newspaper, radio and television. By
seeing the market trend, the store R constructed the following payoff matrix where the numbers
in the matrix indicate a gain or a loss of customers.
Store C
Store
R
40 50 -70
10 25 -10
100 30 60
(a) Check whether game is strictly determinable? If so find value of game.
41
(b) Use Dominance property to reduce the given payoff matrix to 2x2.
(c) Hence or otherwise, find optimal strategies for stores R and C and value of game.
6.7 GRAPHICAL METHOD FOR 2xn GAMES
Consider the Payoff matrix of following 2 x,n game in Table 6.31
TABLE 6.30 Payoff matrix of 2xn Game
B
1
B
2
B
3
. .B
n
A
1
A
2
(
n 2 23 22 21
n 1 13 12 11
a ... a a a
a ... a a a
Let the mixed strategy for player A be given by
(
=
p 1 p
A A
S
2 1
A
, 0 p
Now for each of the pure strategies available to B, expected payoff for player A would be as
follows.
B's pure move A's expected payoff function E(p)
B
I
) p 1 ( a p a ) p ( E
21 11 1
+ =
B
2
) p 1 ( a p a ) p ( E
22 12 2
+ =
B
3
) p 1 ( a p a ) p ( E
23 13 3
+ =
B
n
) p 1 ( a p a ) p ( E
n 2 n 1 n
+ =
The player B would like to choose that pure move B
j
against S
A
for which E
j
(p) is a minimum
for j= 1, 2 ... n. Let us denote this minimum expected payoff for A by.
V =Min (E
j
(P)) j = 1, 2 ... n.
The objective of player A is to select p in such a way that V is as large as possible. This may be
done by plotting the straight lines.
( )
j 2 j 2 j 1 j 2 j 1 j
a p a a ) p 1 ( a p a ) p ( E + = + = j=1, 2, ....n.
as linear functions of p
The highest point on the lower boundary of these lines will give maximum value among the
minimum expected payoffs on the lower boundary (lower envelope) and the optimum value of
probability p.
42
Now the two strategies of player B corresponding to those lines which pass through the
maxmin point can be determined. It helps in reducing the size of the game to (2 x 2).
Similarly, we can treat m x 2 games in the same way and get minmax point which will be the
lowest point on the upper boundary (upper envelope).
EXAMPLE 6.17
Solve the following 2 x 3 game graphically.
(
2 5 8
11 3 1
SOLUTION
Since the problem does not possess any saddle point let the player A play the mixed strategy
| | p 1 p P = , 0 p
Against player B.
The A's expected payoff against B's pure move is given by
B's pure move A's expected payoff
B
I
8 p 7 ) p 1 ( 8 p ) p ( E
1
+ = + =
B
2
5 p 2 ) p 1 ( 5 p 3 ) p ( E
2
+ = + =
B
3
2 p 9 ) p 1 ( 2 p 11 ) p ( E
3
+ = + =
These expected payoff equations are then plotted as functions of p as shown in the Fig. 6.1,
which shows the payoffs of each column represented as points on two vertical axes 1 and 2 of
unit distance apart.
43
Fig 6.1 Player As Expected Payoff
Now since the player A wishes to maximize his minimum expected payoff, we consider the
highest point of intersection H on the lower envelope of A's expected payoff equation. The lines
B
3
and B
2
passing through H define the relevant moves B
2
and B
3
that alone need to play. The
solution to the origin 2 x 3 game reduces to
B
2
B
3
A
1
A
2
(
2 5
11 3
The optimum strategy for A and B is given by
11
3
11
3
) 5 11 ( 2 3
5 2
p =
=
+ +
=
11
9
11
9
) 5 11 ( 2 3
11 2
q
1
=
=
+ +
= and 1 q q
2 1
= + ,
11
2
q
2
=
(
=
11
8
11
3
P
*
and
(
(
(
(
(
(
=
11
2
11
9
0
Q
*
The value of the game is
11
49
11
55 6
V =
= .
44
EXAMPLE 6.18 Solve graphically
(
(
(
(
(
(
(
(
0 5
2 2
1 4
6 1
5 3
3 1
SOLUTION The given problem does not possess any saddle point. Therefore, let the player B
plays the mixed strategy
(
=
q 1
q
Q
against player A.
The expected payoff equations are plotted in the following Fig. 6.2 with two axes I and II
vertically at unit distance apart.
Fig. 6.2 Player Bs Expected Payoff
Since the player B wishes to minimise his maximum expected payoff. We consider the lowest
point of the upper boundary of B's expected payoff equation. The point H (intersection of lines
A
2
and A
4
) represents the minimax expected value of the game for player B. Hence, the solution
45
to the original 6 x 2 game therefore reduces to the 2 x 2 payoff matrix.
The optimal mixed strategy for A and B is given by
2. The companies A and B are competing for the same product. Their different strategies are
given in the following payoff matrix. Company B
Determine the best strategies for the two companies.
Formulating games with mixed strategies
Consider the following example.
Example Two players A and B play a game in which they both simultaneously choose an
integer between one and three. Suppose A chooses x and B chooses y. If x = y, then A wins $x
from B. If y x , then B wins $x from A.
Each player has three possible strategies. For A, let A
i
denote the strategy of choosing the
integer i, i = 1, 2,3; and for B, let Bj denote the strategy of choosing the integer j, j = 1, 2, 3.
Then the payoff matrix is given by the following table.
B
1
B
2
B
3
A
1
A
2
A
3
3 3 3
2 2 2
1 1 1
Definition The minmax criterion. In a two-person zero-sum game, each player will act to
minimize his maximum expected loss.
Using the minmax criterion, A will choose his mixed strategy ) p , p , p (
3 2 1
such that his
maximum expected loss is a minimum or, equivalently, such that his minimum expected gain is
as large as possible, where the minimum is taken over all possible strategies B
1
, B
2
, B
3
for B.
Similarly, B will wish to choose his mixed strategy ) q , q , q (
3 2 1
such that his maximum
expected loss is as small as possible, where the maximum is taken over all possible strategies A
1
,
A
2
, A
3
for A.
Formulating the linear programming model
We shall show how the problem of finding the optimal mixed strategy for each player can be
modelled as a pair of dual linear programming problems. The formulation of this model requires
46
that the guaranteed minimum gain to player A is positive. This will be the case when the matrix
D contains a positive row. When D does not have a positive row, we may add a positive real
number k to every entry of D to construct a new payoff matrix D* which does have a positive
row. Thus D* corresponds to playing a new game in which A always wins k more than in the
original game. Clearly the optimal mixed strategies for both players will be the same in both
games.
In the game described in Example , the guaranteed minimum gain to A is $(-1), and hence
we add 2 to each element of D, to form the matrix D*, in which the first row is positive. (We
could have added any constant k > 1 to create a positive row. We choose the smallest integer
value of k in order to make D* as nice as possible.) This gives the new payoff matrix
|
|
|
.
|
\
|
=
5 1 1
0 4 0
1 1 3
* D
Let v
2
denote A's minimum expected gain when he plays the mixed strategy ) p , p , p (
3 2 1
in the
revised game with payoff matrix D*. Then, we can calculate the expected payoff to A when B
chooses his strategy Bj, for j = 1, 2, 3. Since each of these expected payoffs must be at least A's
minimum expected gain v
2
, we obtain the following inequalities.
2 3 1
2 3 2 1
2 3 1
v p 5 p
v p p 4 p
v p p 3
+
(1)
Since ) p , p , p (
3 2 1
are probabilities, the numbers p
i
, i = 1, 2, 3, satisfy the following additional
constraints.
1 p p p
3 2 1
= + + (2)
0 p , p , p
3 2 1
(3)
Now A's problem is to find R p , p , p
3 2 1
to maximize v
2
, subject to the constraints (1), (2)
and (3). In order to express this problem as an LPP in standard form, we have to make a change
of variable. Since in the revised game, A has a strategy A
1
that guarantees a gain of at least $1,
we can add the constraint V
2
> 1 without changing the system. Thus we may assume that V
2
is
47
positive and so we can divide through each of the constraints (1), (2) and (3) by v
2
while
preserving the direction of the inequalities. We then put
i
2
i
y
v
p
= , i = 1, 2, 3, and
2
2
z
v
1
= .
Then constraint (2) becomes
2 2 3 2 1
z v / 1 y y y = = + +
.
Now maximizing v
2
is equivalent to minimizing z
2
. Hence we can formulate A's problem as the
following linear programming problem.
Find R y , y , y
3 2 1
to minimize
3 2 1
y y y z + + =
subject to
2 3 1
2 3 2 1
2 3 1
v y 5 y
v y y 4 y
v y y 3
+
We can formulate the problem of finding an optimal mixed strategy ) q , q , q (
3 2 1
for B in a
similar way. Let v
1
denote B's maximum expected loss when he plays the mixed strategy
) q , q , q (
3 2 1
in the revised game with payoff matrix D*. Then the expected loss to B when A
chooses his strategy Ai, for i = 1, 2, 3, must be at most B's maximum expected loss of v
1
. Thus
we obtain the following inequalities.
1 3 2 1
1 2
1 3 2 1
v q 5 q q
v q 4
v q q q 3
+
+ +
(5)
where
1 q q q
3 2 1
= + + (6)
0 q , q , q
3 2 1
(7)
Now B's problem is to find R q , q , q
3 2 1
to minimize v
1
, subject to the constraints (5), (6)
and (7). By construction, a row of D* is positive (in this case, row 1). From the corresponding
48
constraint in (5), v
l
is also positive. Thus, we can divide through each of the constraints (5), (6)
and (.7) by v
1
without reversing the direction of the inequalities, and then put
i 1 i
x v / q , i = 1, 2, 3, and 1/v
i
= z
1
. Then constraint (6) becomes
1 1 3 2 1
z v / 1 x x x = = + +
Now minimizing v
l
is equivalent to maximizing z
1
. Hence we can formulate B's problem as a
linear programming problem as follows.
Find R x , x , x
3 2 1
to maximize z
1
= x
1
+ x
2
+ x
3
,
subject to
1 x 5 x x
1 x 4
1 x x x 3
3 2 1
2
3 2 1
+
+ +
0 x , x , x
3 2 1
We complete the solution of the problem described in Example .
We add slack variables S
1
, S
2
, S
3
to the first, second and third constraints respectively. Then,
after several iterations of the simplex algorithm, we obtain the following tableau.
3 2 1 3
2 2
3 2 1 1
3 2 1
S
16
3
S
32
1
S
16
1
32
9
x
S
4
1
4
1
x
S
16
1
S
32
3
S
16
5
32
3
x
S
8
1
S
16
3
S
8
3
16
11
z
=
=
+ + =
=
Thus, the optimal solution is xl = 5/32, x2 = 1/4, x3 = 9/32, and the maximum value of z1 is
11/16. Using the change of variables z1 = 1/vl and xi = qi/vi, i = 1, 2, 3, we deduce that the
optimal mixed strategy for B is (ql, q2, q3) = (5/22, 4/11, 9/22), and the minimum value of v1 is
16/11.
Similarly, we may deduce from this tableau that the optimal solution to the dual problem to
is
yi
= 3/8,
y2
= 3/16, y3 = 1/8, and the minimum value of z2 is 11/16. Using the change of
variables z2 = 1/v2 and
i i i
v / p v =
,
i = 1, 2, 3, we deduce that the optimal mixed strategy for A
is (pl, p2, p3) = (6/11,3/11,2/11) and the maximum value of v2 is 16/11.
Hence, the value of the game with pay-off matrix D* is 16/11. Since D* was obtained from D by
49
adding 2 to each entry, it follows that the value of the original game is given by
v = 16/11 - 2 = -6/11. 0
The negative value for v obtained for this game shows that it is not a fair game. We say that it is
biased in favour of B. The games described in Example 5.1 and Example 5.3 were both biased in
favour of A.
The optimal mixed strategy for A and B is given by
2. The companies A and B are competing for the same product. Their different strategies are
given in the following payoff matrix. Company B
Determine the best strategies for the two companies.
Since no row (or column) dominates another row (or column). The 3 x 2 game could now be
solved by graphical method. Since the player B wishes to minimise his maximum loss, we find
the lowest point of the upper boundary. The expected payoff equations are then plotted as show
in Fig. 17.3.
Figure 17.3
The lowest point in the upper boundary is given by the intersection of lines A, and A,. The
solutionTn the original game is reduced to 2 x 2 matrix
B, B
2
A,
C
3 -2 A
3
2 2
The optimum strategy for A and B is given by
Example 17.11 Solve the following game
Player B
1 7 2
Player A 6 2 7
5 1 6
Solution Since all the elements in the third row are less than or equal to the corresponding
elements of second row. Therefore, row III is dominated by row II. Delete this dominated row.
The reduced payoff matrix is given by
The optimum strategy for A and B is given by
Value of 3x2+(-2)x(2) _ 2
game y = 3+2-(-2+2) 5
50
Example 17.11 Solve the following game
Player B
1 7 2
Player A 6 2 7
5 1 6
Solution Since all the elements in the third row are less than or equal to the corresponding
elements of second row. Therefore, row III is dominated by row II. Delete this dominated row.
The reduced payoff matrix is given by
Solution Since the game has no saddle point it is not a stable one. All the elements of the first
row and a second row are < to the corresponding elements of third row. Hence, these two rows
are dominated rows. Deleting these two rows from the payoff matrix, The reduced payoff matrix
is given by
Player B
Player A 13 4
1
1 4]
In this modified payoff matrix, we observe that all the elements of the second column are r to
the corresponding elements of the fourth column. Hence, this dominated column (2nd column) is
deleted from the payoff matrix. The reduced payoff matrix is given by
Player B
8 15 1
Player A
3 -1 4
Now we observe that no row or column dominates another row or column. However, we note
that a convex combination of 2"" and 3"' column is given by
15 x 2+ 1 x 2= 8<, 8 -lxl+4x1
=3
<3
2 2 2 2
and hence the elements of the first column is greater than or equal to the corresponding elements
of this combination. Deleting this dominated column, the reduced payoff matrix is given by
51
Player B
Player _A 15 1 -1 4