Game Theory
Game Theory
Introduction
Game Theory ▪ In previous topics, we have encountered many situations in which a
single decision maker chooses an optimal decision without reference
Industrial Engineering Department to the effect that the decision has on other decision makers (and
Institut Teknologi Sepuluh Nopember without reference to the effect that the decisions of others have on
Surabaya him or her).
▪ In many business situations, however, two or more decision makers
simultaneously choose an action, and the action chosen by each
player affects the rewards earned by the other players. For
example, each fast-food company must determine an advertising
and pricing policy for its product, and each company’s decision will
affect the revenues and profits of other fast-food companies.
Game theory is useful for making decisions in cases where two or
more decision makers have conflicting interests. Most of our study
of game theory deals with situations where there are only two decision
makers (or players), but we briefly study n-person (where n2) game
theory also.
2
Quantitative Modeling & Industrial Policy Analysis (Q-Lab)
1 2
Outline Definition
➢ Game Theory vs Decision Tree • Part of a rational behavior study of interdependency between
players.
➢ Pay-off Tree
• Mathematical approach to formulate situation of competition/conflict
➢ Zero-sum Game and cooperation between various interests.
➢ Decision making using Minimax and Maximin criteria. • Optimization method to optimize the problems of at least 2
➢ Dominant vs dominated Strategy individuals or group that have an equal interest in maximizing their
benefits as much as possible
➢ Prisoner’s Dilemma
• Optimal solution of the game → maximum benefit for players who
➢ Stable and unstable solution in a game win the game and minimum losses for players who lose the game
➢ Graphical solution in game theory – Zero sum game is said as a fair game, and vice versa
3 4
3 4
1
4/22/2020
5 6
5 6
7 8
7 8
2
4/22/2020
• In 1950, John Nash showed that finite games always have an equilibrium ❑ Each player plays rationally, assuming they have the same
point where all players choose the best action if they want the opponent’s intelligences and objectives, which is to maximize payoff
choice
with maximin and minimax criteria.
• In 1950s and 1960s, game theory is theoretically expanded and applied for
wars and politics. ❑ Each player chooses the strategy only for their own sake
• Since 1970s, game theory has triggered a revolution in economic theory ❑ Each player knows each other’s strategy
and applied for sociology and psychology case.
❑ Each player knows payoff value of each decision made
• Game theory received special attention in 1994 with the awarding of
Nobel Prizes for Nash, John Harsanyi, and Reinhard Selten in the field of
Economics.
9 10
9 10
11 12
11 12
3
4/22/2020
• a23 : The amount of loss of player 2 if player 1 takes strategy 2 and player Player B
Player A
2 takes strategy 3 Strategy B1 Strategy B2
Strategy A1 2 4
Strategy A2 -1 5
13 14
13 14
Player B Player B
Player A Player A
Strategy B1 Strategy B2 Strategy B1 Strategy B2
Strategy A1 2 4 Strategy A1 4, 3 3, 2
Strategy A2 -1 5 Strategy A2 2, -2 1, 1
• Player A (Row player) is positioned as the winner • If one player wins the game, then another player does not
• Player B (Column player) is positioned as the loser necessarily lose
• Value within matrix shows the quantity of win/loss • Consists of two numbers:
– Positive value → the quantity of wins for player A and loss for – Front number → belongs to Player A
player B as well – Rear number → belongs to Player B
• Number of wins for one player = number of loss for another player • Positive number indicates the win, yet negative shows the loss for
both Player A and Player B.
15 16
15 16
4
4/22/2020
17 18
17 18
19 20
19 20
5
4/22/2020
21 22
21 22
• Frequently, there are strategies which are always better than • Concept of dominated strategy is useful for:
other strategies regardless of the strategies taken by the
– Reducing the size of payoff matrix
opponent
– Identifying the optimal solution of game
• Strategies which are always better are maintained.
Otherwise, the strategies which are always worse are eliminated.
– For Player 1,
Strategy p is dominated by Strategy i → if apj ≤ aij for each j
– For Player 2,
Strategy q is dominated by Strategy j → if aiq ≥ aij for each i
• Dominated strategy is eliminated from further evaluation.
23 24
23 24
6
4/22/2020
Player 1
Player 1
25 26
25 26
Payoff Matrix – Dominated Strategy (3) Payoff Matrix – Dominated Strategy (4)
Player 2 Player 2
Player 1 Player 1
27 28
27 28
7
4/22/2020
• Player 2 → Strategy 1 ▪ In the payoff table/matrix, the large positive value actually shows the
large loss for Player 2, so that Maximin becomes MINIMAX
• Payoff value of game = 1 → Saddle Point
• Choosing other strategies will not guarantee in increasing the
win of Player 1 and decreasing the loss of Player 2
29 30
29 30
maximum
Player 1 → Strategy 1
Player 2 → Strategy 1
31 Point
Payoff Value of Game) = 1 → Saddle 32
31 32
8
4/22/2020
Player B
Player A Minimum
Strategy B1 Strategy B2 Strategy B3
Strategy A1 -3 -2 6 -3
Maximum 5 0 6
Minimax value
Zero Sum Game → Fair
• Optimal Solution : Player A → Strategy A2
Player B → Strategy B2
33 34
33 34
35 36
35 36
9
4/22/2020
(B) Two-Person Constant Sum Game (2) (C) Two-Person Non-Constant-sum Game
• Two local TV stations are considering to determine which program will be
aired on prime time. • Most game-theory models of business situations are not
• Total audience is 100 million people → so, this is 2-person constant sum constant-sum games, because it is unusual for business
game with c = 100 (million) competitors to be in total conflict
• If a TV station shows comedy and another one shows soap opera, then there • Prisoner’s Dilemma → a well-known concept to discuss the
will be 14 million people watching comedy and the rest is 100-14 = 86 analysis of two-person non-constant-sum game in which
million people watching soap opera
cooperation between the players is not allowed.
37 38
37 38
❑ Two people are caught by police and charged with being ❑ Then, both of them will be interrogated, and before being
unlawful interrogated, they are informed that:
– If one confesses and another does not confess, then the one admitting
❑ Both of them are arrested in the separated rooms, so that will be free and another who does not admit will be sentenced for 20
they can not communicate with each other years.
– If both of them do not confess, then they will be sentenced for 1 year..
– If those Prisoners confess, then they will be sentenced for 5 years.
39 40
39 40
10
4/22/2020
5 years 5 years 0 year 20 years → Whatever strategies Prisoner 2 takes, the best
Remain option for Prisoner 1 is to confess!
Silent
41 42
41 42
43 44
43 44
11
4/22/2020
What will happen if they are allowed to communicate What will happen if they are allowed to communicate
with each other? with each other?
• They will cooperate and both will remain silent, so that each
of them will be sentenced only for 1 year.
Prisoner B
Confess
Prisoner A
Remain
Silent
45 46
47 48
12
4/22/2020
Characteristics:
• Methods used in mixed strategy are:
• Used when value of Maximin ≠ Minimax → no Saddle Point
a) Matrix Algebra
max (row minimum) = min (column maximum) b) Analytical Method
all rows all columns
c) Graphical Method
• Unstable game
d) Linear Programming
Note:
Value of game follows this equation:
49 50
49 50
Player 1
51 52
51 52
13
4/22/2020
Player 1’s Viewpoint (Maximin Problem) Player 2’s Viewpoint (Minimax Problem)
• Player 1 chooses Xi which maximizes minimum expected payoff • Player 2 chooses Yj which minimizes maximum expected payoff
53 54
53 54
Player B
Player A Minimum
Strategy B1 Strategy B2 Strategy B3
Strategy A1 2 5 7 2 Maximin
Strategy A2 -1 2 4 -1
Strategy A3 6 1 9 1
Maximum 6 5 9
Minimax
55 56
55 56
14
4/22/2020
• If Player B uses strategy B1, then prob. of Player A’s profit: = -4(0.625) + 6
57 58
• If Player A uses strategy A1, then prob. of Player B’s loss: = -3(0.5) + 5 Thus,
= 3.5 Profit of Player A = Loss of Player B
2q + 5 (1-q) = -3q + 5 (1.3)
Player A → profit of 2 to 3.5
• If Player A uses strategy A3, then prob. of Player B’s loss : Player B → loss of 5 to 3.5
6q + 1 (1-q) = 5q + 1 (1.4) Saddle Point = 3.5
59 60
59 60
15
4/22/2020
Lower Envelope
1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 00.0
Probabiliti (Selection
Probability (Pemilihan Strategi 1)
of Strategy 1)
61 62
61 62
• Payoff Matrix 2 x n • Expected Payoff of Player 1 related to each pure strategy of j from Player 2:
Player 2
Player 1
𝑥𝑖 = 1 Player 2
Player 2
𝑖=1
𝑥2 = 1 − 𝑥1
Player 1
Player 1
63 64
63 64
16
4/22/2020
Company A
Company B
Solution: Minimum
Company A Maximin
Maximum
65 66
Minimax
65 66
Company A Player 1
Company B
Expected Payoff
Pure Strategy of Player 2
of Player 1
Company A
Player 2
Player 1
67 68
67 68
17
4/22/2020
maximin point), then there are more than 1 solution for Player 2:
– Combination 2 and 3
– Combination 3 and 4
Maximin Area – Combination 2 and 4 → is excluded because it gives non-optimal solution
(slope with the same sign)
𝒙𝟏
Strategy 1 Strategy 2 Strategy 3 Strategy 4
69 70
69 70
Player 2 Player 2
Player 1 71 Player 1 72
71 72
18
4/22/2020
Solution of Player 1’s Viewpoint using Player 2 Solution of Player 1’s Viewpoint (Payoff Matrix)
Payoff Matrix – Dominated Strategy Player 2
Player 1
Player 1
→ Strategy 1 of Player 2 is
dominated by Strategy 2, Player 2
then Strategy 1 is Player 2
eliminated
Player 1
Player 1
73 74
73 74
75 76
75 76
19
4/22/2020
• Each player has 2 strategies • To determine X1, X2, Y1, and Y2, the following formula is used:
Y1 Y2 = 1–Y1
• Player 1 tends to choose Strategy 2 that will give Maximin Value
Player II
• Player II tends to choose Strategy 2 that will give Minimax Value Player I
Strategy 1 Strategy 2 Maximin
Player II X1 Strategy 1 12 -8 -8
Player I
Strategy 1 Strategy 2 Maximin X2 = 1–X1 Strategy 2 -3 6 -3
Strategy 1 12 -8 -8 Minimax 12 6
Strategy 2 -3 6 -3
Player 1’s Viewpoint Player 2’s Viewpoint
Minimax 12 6
12X1 – 3(1–X1) = –8X1 + 6(1–X1) 12Y1 – 8(1–Y1) = – 3Y1 + 6(1–Y1)
12X1 – 3 + 3X1 = –8X1 + 6 – 6X1 12Y1 – 8 + 8Y1 = –3Y1 + 6 – 6Y1
Solution → using Mixed strategy (Maximin ≠ Minimax)
15X1 –3 = –14X1 + 6 20Y1 – 8 = –9Y1 + 6
So, it has to be found the probability of strategy chosen by both players
29X1 = 9 29Y1 = 14
➢ Probabilities of strategy chosen by Player 1 = X1 and X2
X1 = 9/29 Y1 = 14/29
➢ Probabilities of strategy chosen by Player 2 = Y1 and Y2
X2 = 20/29 Y2 = 15/29
82 83
82 83
V* = 15X1 – 3 V* = 20Y1 – 8
Solution → using Mixed strategy (Maximin ≠ Minimax)
V* =15(9/29) – 3 V* = 20(14/29) – 8 So, it has to be found the probability of strategy chosen by both players
V* = 48/29 V* = 48/29 ➢ Probabilities of strategy chosen by Player 1 = X1 and X2
➢ Probabilities of strategy chosen by Player 2 = Y1, Y2, Y3, and Y4
84 85
84 85
20
4/22/2020
86 87
86 87
88 89
88 89
21
4/22/2020
90 91
92 93
22
4/22/2020
94 95
94 95
• Linear programming can be used to find the value and optimal • Player 1’s Viewpoint (Maximin Problem)
strategies for any two-person zero-sum game.
• Example:
▪ Two players must simultaneously utter one of the three words
stone, paper, or scissors and show corresponding hand signs.
▪ If both players utter the same word, then the game is a draw.
▪ Otherwise one player wins $1 from the other player according to
the these rules
− Scissors defeats (cuts) paper
− Paper defeats (covers) stone
− Stone defeats (breaks) scissors
▪ Find the optimal strategies for this two-person zero-sum game.
96 97
96 97
23
4/22/2020
• Player 1’s Viewpoint (Maximin Problem) • Player 2’s Viewpoint (Minimax Problem)
Subject to:
unrestricted
98 99
98 99
• Player 2’s Viewpoint (Minimax Problem) Relationship of Maximin and Minimax Problem
• Both problems (Maximin & Minimax) optimize variable V,
which is the expected value of game
Subject to: • Problem of Player 2 (Minimax problem) is the DUAL Problem
of Player 1 (Maximin problem)
• Optimal solution for one problem yields optimal solution for
another.
unrestricted
100 101
100 101
24
4/22/2020
Player 2
Player 1
unrestricted
102 103
102 103
• The reward matrix is • The row player has chosen a mixed strategy (x1, x2, x3) then the optimal
strategy can be found by solving the LP
Column Player
Row Player Stone Paper Scissors Row
minimum
max z = v
Stone 0 -1 +1 -1
s.t. v ≤ x2 – x3 (Stone constraint)
Paper +1 0 -1 -1
v ≤ -x1 + x3 (Paper constraint)
Scissors -1 +1 0 -1
v ≤ x1 – x2 (Scissor constraint)
Column
x1 + x2 +x3 = 1
minimum +1 +1 +1 x1, x2, x3 ≥ 0; v urs
104 105
25
4/22/2020
• The column player has chosen a mixed strategy (y1, y2, y3) then the
optimal strategy can be found by solving the LP
min z = w
s.t. w ≥ –y2 – y3 (Stone constraint)
w ≥ y1 – y3 (Paper constraint)
w ≥ –y1 + y2 (Scissor constraint)
y1 + y2 +y3 = 1
y1, y2, y3 ≥ 0; w urs
106 107
106 107
108 109
108 109
26
4/22/2020
– The value of Stone, Paper, Scissors is v’ -1 = 0. • LINDO or LINGO can be used to solve for the value and the
– The optimal strategy for the row player is ( 1/3, 1/3, 1/3). optimal strategies in a two-person zero-sum game.
– The optimal strategy for the column player is ( 1/3, 1/3, 1/3). • Simply type in either the row or column player’s problem.
111 112
111 112
CASE 1
• One of types of Game Theory classification
• Table of Payoff Matrix of 2-person non-zero sum gam
• Shows the quantity of win for Player 1 is different with
quantity of loss for other Players Company A
Company B
Strategy B1 Strategy B2
Strategy A1 4%, 3% 3%, 2%
Strategy A2 2%, -2% 1%, 1%
• In the cell of Strategy A1 and B1, Player A wins of 4% and Player B also
wins of 3%
• In the cell of Strategy A2 and B1, Player A wins of 2% and Player B loses of
2%
113 114
113 114
27
4/22/2020
(D) Non-Zero Sum Game –Example (2) (D) Non-Zero Sum Game – Example (3)
CASE 2 CASE 3
• Table of Payoff Matrix of 2-person non-zero sum game in competing • Table of Payoff Matrix of 2-person non zero sum game in competing the
market share, in which A1 dominates A2, and B1 dominates B2 market share, in which there is domination, but not optimal (maximum).
Company B Company B
Company A Company A
Strategy B1 Strategy B2 Strategy B1 Strategy B2
Strategy A1 4%, 3% 3%, 2% Strategy A1 3%, 3% 1%, 4%
Strategy A2 2%, 2% 1%, 1% Strategy A2 4%, 1% 2%, 2%
• In competing market share: • Dominant strategy will take Company A and B to choose Strategy A2 and
B2 respectively, but the most optimal strategies for both are Strategy A1
Company A → Strategy A1 : proft of 4%
and B1
Company B → Strategy B1 : profit of 3%
• Thus, in this situation every company will accept that strategy.
115 116
115 116
Conclusion Reference
• Game Theory had been developed for decision making in conflict situation • Winston, Wayne L., (2003), Operations Research; Applications and
between 2 parties/interests, so that it will be obtained the optimal solution Algorithms, 4thedition, Duxbury Press
• Intended optimal solution is to choose the alternative solution that will give
maximum profits for the winner, but minimum loss for the loser
• In daily practices, game theory is not always able to be applied properly
(because the assumptions used to underlie this theory can not always be
fulfilled)
• Moreover, problem complexity, often gives its own difficulties for the
application of this theory.
117 119
117 119
28
4/22/2020
120
29