GAME THEORY

The term game refers to a situation of conflict and competition in which two or more competitors (or players) are involved in decision-making in anticipation of certain outcomes over a period of time. Examples:
Pricing of products, where a firms ultimate sales are determined not only by the price levels it selects but also by the prices its competitors set. Management-labor negotiations War games Election canvassing

 Game theory is a theory of conflict. It is a mathematical theory which deals with competitive situations. Game theory is a form of decision theory where two or more rational opponents are involved under conditions of competition and conflicting interests. Theory of games became popular when Von Neumann along with Oscar Morgenstern published the book, Theory of Games and Economic Behavior in 1944.

 A game is defined as an activity between two or more players according to a set of rules, at the end of which each player receives some benefit or some loss. In a game there are two or more opposing parties with conflicting interests. They know their objectives and the rules of the game. An experienced player can usually predict with accuracy how his opponent will react if a particular strategy is adopted. When one player wins, the others lose.

 Features of a Game
1. There are finite number of competitors or players. 2. Each player has a finite number of possible courses of action. 3. A play is said to be played when each of the players chooses a single course of action from the list of actions available to him. 4. Each play is associated with an outcome known as payoff. Payoff determines a set of payments. 5. The possible gain or loss of each player depends not only on the choice made by him but also the choice made by his opponents.

 Assumptions of a Game
1. The players act rationally and intelligently. 2. Each player has a finite set of possible courses of action. 3. The players attempt to maximize gains or minimize losses. 4. All relevant information are available to each player. 5. The payers make individual decisions. 6. The players simultaneously select their courses of action. 7. The payoff is fixed and determined in advance.

 Strategy
The strategy of a player is the predetermined rule by which a player decides his courses of action during the game.
There are two types of strategies:
Pure Strategy A pure strategy is a decision, known in advance to a play, to always choose a particular course of action in a given situation. It is a predetermined course of action. The player knows it advance. Mixed Strategy A player is said to adopt mixed strategy when he does not adopt a single strategy for all the time, but would play different strategies in each situation. A mixed strategy is a decision to choose a course of action for each play in accordance with some probability distribution. In a mixed strategy we cannot say for certain which course of action the player would choose.

 Payoffs
The outcome of a game in the form of gains or losses to the players for choosing different strategies is known as the payoffs.

Payoff Matrix
In a game, the gains and losses, resulting from the different moves and counter-moves, when represented in form of a matrix, is known as the payoff matrix. Each element of the payoff matrix is the gain of the maximizing player when a particular course of action is chosen by him as against the course of action chosen buy the opponent.

B
A1

B1 5
1 8

B2 3
-2 -1

B3 2
0 1

A A2
A3

Here, A is the maximizing player. B is the minimizing player. Each element is the matrix is the gain for A when he chooses a course of action against which B chooses another course of action. For example, when A chooses A2 and B chooses B1, the gain for A is shown in the second row of first column.

Value of the Game:

The value of the game is the maximum guaranteed gain to the maximizing player (A) if both the players use their optimal strategies. It is the expected payoff of a game when all the players of the game follow their optimal strategies.
Minimizing and Maximizing Players

If there are two players A and B, generally the payoff given in a payoff matrix indicate gains to A for each of the possible outcomes of the game. That is, each of the outcome of a game results in a gain for A. All such gains are shown in the payoff matrix. Usually each row of the payoff matrix indicates gains for A for his particular strategy. A is called the maximizing player and B is called the minimizing player. The payoff values given in each column of payoff matrix indicated the losses for B for his particular course of action.

Minimizing and Maximizing Players

Therefore, if the element in the position (A1, B3) is a, then As gain is a; and Bs loss is a, when A chooses the strategy A1 and B chooses the strategy B3. Maximin and Minimax Rows in a payoff matrix represents payoffs in respect of every strategy of the maximizing player A. Similarly columns represent payoffs in respect of the minimizing player B. Maximin: Maximum of minimum payoffs in each row. Minimax: Minimum of maximum payoffs in each column. Maximin strategy for maximizing player Minimax strategy for minimizing player

B
A1 A A2 A3

B1 5 1 8

B2 3 -2 -1

B3 2 0 1

Minimum in rows: A1 = 2; A2 = -2; A3 = -1 Maximum of these minima = Maximin = 2 Maximum for columns: B1 = 8; B2 = 3; B3 = 2 Minimum of these maxima = Minimax = 2

Maximin Principle Here, the maximizing player (A) lists worst possible payoffs of all his potential (possible) strategies and then chooses that strategy which corresponds to the best. Minimax Principle Here, the minimizing player (B) lists his maximum losses from all his potential (possible) strategies and selects that strategy which corresponds to the least.

Saddle Point
A Saddle point of a payoff matrix is that position in the payoff matrix where the maximin coincides with the minimax. Payoff at the saddle point is the value of the game. In a game having a saddle point, optimum strategy of maximizing player is always to choose the row containing the saddle point and for the minimizing player to choose the column containing the saddle point. If there are more than one saddle point, there will be more than one solution. A game for which maximin for A = minimax for B is called a game with a saddle point The element at the saddle point position is the value of the game denoted by v

B
A1

B1 3
-2 0
3

B2 2
1 -2
2

B3 4
-3 3
4

Row min

2 -3 -2

Maximin

A A2
A3
Column max

Minimax Saddle point is (A1, B2) Value of the game = v = 2

Types of Games
Games are categorized on the basis of : 1. Number of players
1. Two-person game 2. N-person game

2. Number of moves 3. Nature of payoff 4. Nature of rules

Zero-sum Game In a game if the algebraic sum of the outcomes (or gains) of all the players together is zero, the game is called a zero-sum game, otherwise it is called a non-zero-sum game. In a zero-sum game, the amounts won by the winner is equal to the amount lost by the loser.

Two-person Zero-sum Game Two-person Zero-sum game is the simplest of game models. Here, there will be two persons in the conflict and the sum of the payoffs of both together is zero. That is, the gain of one is at the expense of the other. The two-person zero-sum game may be:
Pure strategy Mixed strategy game

Two-person Zero-sum Game Basic Assumptions

There are two players The players have opposing interests The number of strategies available to each player is finite For each specific strategy, selected by one player, there results a payoff. The amount won by one player is exactly equal to the amount lost by the other.

Limitations of Game Theory

In reality a player have infinite strategies, but here we assume that there are only finite number of strategies for each player. It is assumed that each player has the knowledge of the opponents strategies. But in reality this may not be the case. The assumption that the gain of one player is the loss of the opposing player need not be true in all situations. Game theory usually ignores the presence of risk and uncertainty. It is assumed that the payoff is always known in advance. But in reality payoff of strategies may not be known in advance. In game theory it is assumed that the two players have equal intelligence. This need not be the case in reality.

Fair Game
A game is said to be fair if the value of the game is zero (v = 0).

Solution of Pure Strategy Games

The maximizing player arrives at his optimal strategy on the basis of maximin criterion, while the minimizing player arrives at his optimal strategy on the basis of minimax criterion. The game is solved by equating maximin value with the minimax value. In this type of problems saddle point exists.

1. A companys management and labor union are negotiating a settlement. Each has the strategies (1) Aggressive approach (2) Logical approach (2) Legalistic approach (4) Conciliatory approach. The cost to the company for every pair of strategic choices are given below:
Management Ag Log Leg 40 30 24 50 28 16 80 4 20 -10 8 22

Ag Union Log Leg C

C 70 20 10 0

Management MAg UAg MLog MLeg MC

Row min 24

Union ULog
ULeg UC
Column min

40 50 80 -10
80

30 28 4 8
30

24 16 20 22
24

70 20 10 0
70

16
4 -10

Maximin = Minimax = 24 at (UAg, MLeg) Saddle point = (UAg, MLeg) Unions strategy : Aggressive Managements strategy : Legalistic

2. For what value of , the game with the following matrix is determinable?
B A1 A2 A3 B1 -1 -2 B2 6 4 B3 2 -7

A1 A2 A3

B1 -1 -2
-1

B B2 6 4
6

B3 2 -7
2

Row min

-7
-2

Column max

The maximin value is 2 and the minimax value is -1. So the value of the game lies between -1 and 2. i.e. -1 <= <= 2 For a strictly determinable game , maximin = minimax So, for the game to be determinable, -1 <= <= 2

3. In a certain game, player A has three possible courses of action L, M and N, while B has two possible choices P and Q.
Payments to be made according to the choice made.
Choices L,P Payments A pays B Rs.3

L,Q
M,P M,Q N,P N,Q

B pays A Rs. 3
A pays B Rs.2 B pays A Rs.4 B pays A Rs.2 B pays A Rs.3

What are the best strategies for players A and B in this game? What is the value of the game for A and B?

B L M N P -3 -2 2 2 Q 3 4 3 4 -3 -2 2
Maximin

Minimax

Optimal strategies for A and B are:

AN BP Value of the game is v = +2

4) Find the ranges of values of p and q, which will render the entry (2,2) a saddle point for the game. B I II III

I 2 10 4

II 4 7 p

III 5 q 6

Let us ignore the values of p and q and find the row minimum and column maximum.

A
Column maximum

I II III
10

I 2 10 4 7

II 4 7 p 6

B III 5 q 6

Row minimum

2 7 4

Maximin value = 7 = Minimax value. This means that p 7 i.e. column maximum and q 7 i.e. row minimum. Hence the range of p and q will be p 7 and q 7.

Problem for Solving

Assume that two firms are competing for market share for a particular product. Each firm is considering what promotional strategy to employ for the coming period. Assume that the following payoff matrix describes the increase in market share of Firm A and the decrease in the market share for Firm B, Determine the optimal strategies for each firm.
Firm A No promotion Mod. Promotion Heavy Promotion No Promotion 5 10 20 Firm B Moderate Promotion 0 6 15 Heavy Promotion -10 2 10

1. Which firm would be the winner, in terms of market share?

2. Would the solution strategies necessarily maximize profits for either of the firms?
3. What might the two firms do to maximize their profits?

Problem for Solving

Two competitive manufacturers are producing a new toy under license from a patent holder. In order to meet the demand they have the option of running the plant for 1 shift, 2 shifts, or 3 shifts per day. As the length of production increases so does the cost. One of the manufacturers A has set up the matrix given below, in which he estimates the percentage of the market that he could capture and the different production schedule:
Firm B Firm A 1-Shift 2-Shifts 3-Shifts 1-Shift 60% 63% 83% 2-Shifts 56% 60% 72% 3-Shifts 34% 55% 60%

1. At which level should each produce? 2. What percentage of the market will B have?

Solution of Mixed Strategy Problems

When there is no saddle point for a game problem, the minimax and maximin principle cannot be applied to solve that problem. In such cases, the concept of chance move is introduced. Here, the choice among the number of strategies is not a decision of the player but it is decided by some chance (probability)mechanism. The strategies thus used are called mixed strategies.

Solutions to mixed strategy problem can be arrived at by any of the following methods: 1. Probability method (equal gain method) 2. Graphic method 3. Linear Programming method

Probability Method This method is applied when there is no saddle point, and the payoff matrix has two rows and two columns only. The player may adopt mixed strategies with certain probabilities. Hence the problem is to determine the probabilities of both players and the expected value of the game.

Consider the following payoff matrix: Player B B1 B2 a b c d

Player A A1 (1-p) A2
p

Let p be the probability for A using strategy A1 Let (1-p) be the probability for A using strategy A2 Therefore, the expected gain of A if B chooses B1 strategy = = ap + c(1-p)

Player A A1 (1-p) A2
p

Player B B1 B2 a b c d

The expected gain of A if B chooses B2 strategy = = bp + d(1-p) ap + c(1-p) = bp + d(1-p) Solving we get,

p =

(d-c)/((a+d)-(b+c))

Player B
q (1-q)

Player A A1 A2

B1 a c

B2 b d

Let q be the probability for B using strategy B1 Let (1-q) be the probability for B using strategy B2 Therefore, the expected gain of B if A chooses A1 strategy = = aq + b(1-q)

Player B
q (1-q)

Player A A1 A2

B1 a c

B2 b d

Let q be the probability for B using strategy B1 Let (1-q) be the probability for B using strategy B2 Therefore, the expected gain of B if A chooses A2 strategy = = cq + d(1-q)

Player B
q (1-q)

Player A A1 A2

B1 a c

B2 b d

aq + b(1-q) = cq + d(1-q) Solving we get,

q = (d-b)/((a+d)-(b+c))

q p (1-p)

(1-q)

a c

b d

Expected value of the game (v) = x. prob = a.pq + b.p(1-q) + c.(1-p)q + d. (1-p)(1-q) Substituting the values of p and q and simplifying:

v = (ad-bc)/((a+d)-(b+c))

Therefore the solution is:

p
q v

=
= =

(d-c)/((a+d)-(b+c))
(d-b)/((a+d)-(b+c)) (ad-bc)/((a+d)-(b+c))

Where, a, b, c, and d are the payoffs.

Principle of Dominance
The rules of dominance are used to reduce the size of the payoff matrix. These rules help in deleting certain rows and/or columns of the payoff matrix which are inferior (less attractive) to at least one of the remaining rows and/or columns (strategies) in terms of payoffs to both the players. Rows and/or columns once deleted will never be used for determining the optimum strategy for both the players.

 The rules of dominance are especially used for the evaluation of two-person zero-sum games without a saddle (equilibrium). Dominance principles are: 1. For the minimizing player (B), if each element in any column Cr is greater than or equal to the corresponding elements in another column Cs in the payoff matrix, then the column Cr is said to be dominated by column Cs and therefore, column Cr can be deleted from the payoff matrix. In other words, the minimizing player (B) will never use the strategy corresponding to column Cr because he will lose more by choosing such strategy.

2. For the maximizing player (A), if each element in any Row Rr is less than or equal to the corresponding elements in another Row Rs in the payoff matrix, then the Row Rr is said to be dominated by Row Rs and therefore, Row Rr can be deleted from the payoff matrix. In other words, the maximizing player (A) will never use the strategy corresponding to column Rr because he will gain less by choosing such strategy.

1 -2

3 1

4 4

5 0

Here, every element in the first row is more than or equal to the corresponding elements of the second row. Therefore, the first row dominates and so the second row can be ignored.

2 -3 -1 4

2 1 1 5

Here, every element in the second column is greater than or equal to the corresponding elements of the first column. Therefore, the first column dominates and so the second column can be ignored.

Problem:
The following is the payoff matrix for players A and B:

Player A

I 2 3 4

I 2 5 6 4

Player B II III 4 3 6 3 7 9 2 8

IV 3 7 8 4

V 4 8 7 3

Using dominance property, obtain the optimum strategies for both the players and determine the value of the game

All the elements of Column-IV are greater than or equal to the corresponding elements of Column-I. So, we eliminate Column-IV. Player B I II III IV V I 2 4 3 3 4 2 5 6 3 7 8 Player A 3 6 7 9 8 7 4 4 2 8 4 3 All the elements of Row-4 are less than or equal to the corresponding elements of Row-3. So, we eliminate Row -4.

 So we have
I 2 5 6 Player B II III 4 3 6 3 7 9 V 4 8 7

Player A

I 2 3

In the above payoff matrix, all elements of column V were greater than the corresponding elements of column I. So, we deleted column V. In the reduced matrix, Row-1 and Row-2 are both less than the corresponding elements of Row-3 . So, we delete both Row-1 and Row-2.

The resulting matrix is:

I 6 Player B II 7 III 9

Player A

Column-I dominates columns II and III. So deleting them we have: Player B I Player A 3 6 Optimum strategies: Player A will choose strategy 3 and Player B will choose strategy I. Value of the game = 6

Problem

Players A and B play a game in which each has three coins, each with a value of Rs.5. Rs.10 and a Rs.20. Each player selects a coin from his set without the knowledge of the others choice. If the sum of the coins is an odd amount, then A wins Bs coin. But if the sum is even, then B wins As coin. Find the best strategy for each player and the value of the game.

Solution
The payoff matrix is given below:
5 -5 5 5
5

Player A

5 10 20

Player B 10 20 10 20 -10 -10 -20 -20

10 20

-5
-10 -20

It is clear that payoff matrix has no saddle point.

Therefore, we further reduce the size of the given payoff matrix.

Solution
The reduction of the payoff matrix is given below:
5 -5 5 5 Player B 10 10 -10 -20 20 20 -10 -20

Player A

5 10 20

The reduced payoff matrix is:

Player B 5 10 -5 10 5 -10
5 Minimax 10

5 Player A 10

-5
-10

Maximin

Solution
The reduced payoff matrix is:
Player B 5 10 -5 10 5 -10
5 10 Minimax

Player A

5 10

-5
-10

Maximini

In the above reduced payoff matrix, the maximin value is not equal to the minimax value. Hence there is no saddle point and one cannot find the point of equilibrium. Hence, for this type of a game situation, it is possible to obtain a solution by applying the mixed strategy.

Graphical Method
If the payoff matrix is of order 2 x n or m x 2, graphic method can be applied.

Solution for 2 x n Games

Here, the maximizing player (A) has only two strategies and the minimizing player (B) has many (n) strategies. Player B B1 B2 B3 B4 Bn A1 2 -4 6 -3 5 Player A A2 -3 4 -4 1 0 Let p and (1-p) be the probabilities with which player A uses his pure strategies. Player A will select the value of p that maximizes the minimum expected value of the game.

 By dominance property, B5 dominates B1. B5 can be eliminated. So the reduced payoff matrix is: Player B B1 B2 B3 B4 Player A A1 2 -4 6 -3 A2 -3 4 -4 1 This is a 2 x n matrix, where n = 4 First draw the vertical lines A2 and A1. On both A2 and A1 mark points. Then draw lines representing B1, B2, B3 and B4. To draw the line B1, mark 2 on the A2 line and mark -3 on the A1 line. Similarly draw B2, B3 and B4.

Lower area bounded by the lines B1, B2, B3 and B4 is LMNOP. Of this, N is the required point as it is the maximum point. N is the intersection of B1 and B4. So ignoring B2 and B3. Thus the reduced matrix is obtained by avoiding B2 and B3. B1 B4 A1 2 -3 A2 -3 1 As optimal strategy for choosing A1 and A2 is given by: p = 1-(-3)/((1+2)-(-3-3)) = 4/9 1-p = 5/9 q = 1-(-3)/((1+2)-(-3-3)) = 4/9 1=q = 5/9 v = ad-bc /((a+d) (b+c)) = 2-9 /3-(-6) = -7/9 As strategy = (4/9, 5/9) Bs strategy = (4/9, 0, 0, 5/9)

Solution for m x 2 Games

Here, the minimizing player (B) has only two strategies and the maximizing player (A) has many (m) strategies. Player B B1 B2 A1 -6 7 A2 4 -5 Player A A3 1 -2 A4 2 5 A5 7 -6 We draw two vertical lines to represent B1 and B2. Draw the lines A1, A2, A3.Am by joining the respective points on B1 and B2.

 The upper area bounded by the lines A1, A2, A3, A4 and A5 gives the maximum of the loss and lowest point of this upper boundary is the minimax point. The lines passing through the minimax point represent the choices of A. Thus we get a 2 x 2 matrix. B1 -2 7 B4 5 -6

A4 A5

p = -6-(7)/((-2-6)-(5+7)) = -13/-20 = 13/20 (1-p) = 7/20 q = -6-(5)/((-2-6)-(5+7)) = -11/-20 = 11/20 (1-q) = 9/20 v = ad-bc /((a+d) (b+c)) = 12-35 /-8-12 = -23/-20 = 23/20 As strategy = (0, 0, 0, 13/20, 7/20) Bs strategy = (11/20, 9/20)

Reducing Game to a LP Problem

Every finite two persons zero-sum game can be expressed as a Linear Programming problem. For a m x n rectangular game where m or n is greater than 2, LP technique can be applied. Consider a game problem of m x n payoff matrix given by: a11 a12 a13a1n a21 a22 a23a2n am1 am2 am3amn

 Let the course of action for the maximizing player (A) be A1, A2..Am. Let the course of action for the minimizing player (B) be B1, B2..Bn. Let p1, p2, ..pm be the probabilities of A choosing the strategies A1, A2, Am. Let q1, q2, qn be the probabilities of B choosing the strategies B1, B2, Bn. Therefore, p1 + p2 + .+pm = 1 q1 + q2 + .+qn = 1

 Let v be the value of the game. The problem is to find the p1, p2, .pm that maximizes v or minimize (I/v) subject to: a11p1 + a21p2 + + am1pm >= v a12p1 + a22p2 + + am2pm >= v a1n p1+ a2np2 + .. + amnpm >= v p1, p2, ..pm >= 0 Dividing the inequalities by v, and substituting (p1/v)= x1 etc., the inequalities become,

a11x1 + a21x2 + + am1xm >= 1 a12x1 + a22x2 + + am2xm >= 1 a1n x1+ a2nx2 + .. + amnxm >= 1 x1 + x2 + + xm = (p1 +p2 + .+ pm )/v = 1/v Minimizing 1/v means minimizing x1 + x2 + + xm

Thus, we have the LP problem Minimize z = x1 + x2 + + xm Subject to a11x1 + a21x2 + + am1xm >= 1 a12x1 + a22x2 + + am2xm >= 1 a1n x1+ a2nx2 + .. + amnxm >= 1

Similarly if we consider player B, LPP is, Maximize Z = y1 + y2 + + yn Subject to a11y1 + a12y2 + + ainyn <= 1 a21y1 + a22y2 + + a2nyn <= 1 am1 y1+ am2y2 + .. + amnyn <= 1 y1, y2, yn >=0

 Value of the game =

= 1/(y1 + y2 +..+ yn) = 1/ (x1 + x2 +. + xm) or