Decision Theory Game Theory
Decision Theory &
Game Theory
Game Theory
Elements of a game
Players: intelligent opponents (competitors
or enemies)
Strategies: choices of what to do to defeat
the opponents
Outcomes = Payoffs: functions of the
different strategies for each player
Payoff Matrix: Tableau (of the gain of the
Row Player)
Rule: how to allocate the payoffs to the
players
The Game: Example
Two Players: Player A (Row) and Player B
(Column)
Tossing a balanced coin
Possible outcomes: Head (H) and Tail (T)
Rule:
If
the outcome matches (player A selects H and the
outcome is also H or player A select T and the
outcomes is also T), Player A gets $1 from Player B;
Otherwise Player A loses $1 to Player B
The Game: Payoff Matrix
Player B
Player A
(Row Player)
H
T
H
1
–1
T
–1
1
Strategies of each player : H or T
An Optimum Solution is said to be reached if
neither player finds it beneficial to alter his
strategy
Optimum Solution
Optimum Solution can be achieved if
players choose to apply:
Pure strategy (e.g. select H or T)
A Mixture of pure strategies = Mixed Strategy
Two-Person Zero-Sum Game
A game with two players
A gain of one player equals a loss to the other
The focused player = the row player (Player A)
Player A maximize his minimum gain (why?)
Player B minimize her maximum loss (why?)
Optimum Solution is gained by Minimax-Maximin
criterion
Optimum Solution reflects that the game is
stable or in the a state of equilibrium
Two-Person Zero-Sum Game with
Saddle Point
1
Player B
2
3
4
Row
Min
1
8
2
9
5
2
Player A 2
6
5
7
18
5
3
7
3
–4
10
–4
Column Max 8
5
9
18
Minimax
Value
Maximin
Value
Two-Person Zero-Sum Game with
Saddle Point
Player A (Row): Maximin Value = Lower
Value of the game
Player B (Column): Minimax Value =
Upper Value of the game
Maximin value = Minimax value Saddle
point = Value of the game
Two-Person Zero-Sum Game with
Saddle Point
Saddle point leads to Optimum Solution
Saddle point indicates a stable game
Players apply Pure Strategy
In general
To maintain the “Optimality” of a game:
maximin value value of the game minimax value
OR
lower value value of the game upper value
Mixed Strategies
Used for solving a game that does not
have a saddle point
Used for solving a game that does not
have a saddle point
Optimum Solution is gained using:
Graphical Solution for (2 N) and (M 2)
Simplex for (M N) game
Unstable Game without Saddle Point
Player B
Player A
Column
Max
2
3
–10 9
Row
1
1
5
4
0
Min
–10
2
6
7
8
2
2
3
8
5
4
15
4 Maximin
4
7
4
–1
3
–1
8
7
9
15
Value
Minimax
Value
Minimax value = 7 > Maximin value = 4 sub-optimal
2 N game
2 N game:
Player A has 2 strategies
Player B has N ( 2) strategies
B
A
y1
y2
… yn
a11
a12
… a1n
x2 = 1 – x1 a21
a22
… a2n
x1
2 N game
B’s Pure Strategy A’s Expected Payoff
1
(a11 – a21)x1 + a21
2
(a12 – a22)x1 + a22
…
…
n
(a1n – a2n)x1 + a2n
2 N game: Example
B
A
y1
y2
y3
y4
x1
2
2
3
–1
x2
4
3
2
6
2 N game
B’s Pure Strategy
A’s Expected Payoff
1
– 2x1 + 4
2
– x1 + 3
3
x1 + 2
4
–7 x1 + 6
Optimum Solution: Graphical Solution
Graphical Solution x1 = 0 and x1 = 1 = x2
6
5
Maximin
1
4
3
3
2
2
1
4
x1 = 1
x1 = 0
-1
x*1 =1/2
Optimum Solution for Player A
Intercept between lines (2), (3) and (4)
(x1* = ½, x2*= ½)
(2) – x1 + 3 = – ½ + 3 = 5/2
v*
(3) x1 + 2 = ½ + 2 = 5/2
(4) –7 x1 + 6 = – 7/2 + 6 = 5/2
Player B can mix all the 3 strategies
Player A’s gains = 5/2
Optimum Solution for Player B
Combination (2), (3) and (4):
(2,3) y1 and y4 = 0, y3 = y2 –1 (y2* = y3*)
(2,4) y1 and y3 = 0, y4 = y2 –1 (y2* = y4*)
(3,4) y1 and y2 = 0, y4 = y3 –1 (y3* = y4*)
Optimum Solution for Player B
(2,3) y1 and y4 = 0, y3 = y2 –1 (y2* = y3*)
A’s Pure
Strategy
2
B’s Expected Payoff
3
y2 + 2
– y2 + 3 = y2 + 2
– 2 y2 = – 1
y2* = 1/2 dan y3* = 1/2
– y2 + 3
B’s loss = 5/2
Optimum Solution for Player B
(2,4) y1 and y3 = 0, y4 = y2 –1 (y2* = y4*)
A’s Pure
Strategy
2
B’s Expected Payoff
4
– 7y2 + 6
– y2 + 3 = –7y2 + 6
6 y2 = 3
y2* = 1/2 dan y4* = 1/2
– y2 + 3
B’s loss = 5/2
Optimum Solution for Player B
(3,4) y1 and y2 = 0, y4 = y3 –1 (y3* = y4*)
A’s Pure
Strategy
3
B’s Expected Payoff
4
– 7y3 + 6
y3 + 2 = –7y3 + 6
8 y3 = 4
y3* = 1/2 dan y4* = 1/2
y3 + 2
B’s loss = 5/2
M 2 game
M 2 game:
Player A has M ( 2) strategies
Player B has 2 strategies
B
x1
A
x2
…
y1
a11
a21
…
y2= 1 – y1
a12
a22
…
xm
am1
am2
M 2 game
A’s Pure Strategy B’s Expected Payoff
1
(a11 – a12)y1 + a12
2
(a21 – a22)y1 + a22
…
…
m
(am1 – am2)y1 + am2
M 2 game: Example
B
A
y1
y2
x1
2
4
x2
3
2
x3
–2
6
M 2 game
A’s Pure Strategy
B’s Expected Payoff
1
– 2 y1 + 4
2
y1 + 2
3
– 8 y1 + 6
Optimum Solution: Graphical Solution
Graphical Solution y1 = 0 and y1 = 1 = y2
6
Minimax
5
3
4
2
3
1
2
y1 = 1
1
-1
-2
y1* = y3* = 1/3
Optimum Solution for Player B
Intercept between lines (1) and (3)
(y1* = 1/3, y3*= 1/3)
(1) – 2y1 + 4 = – 2/3 + 4 = 10/3
v*
(3) – 8y1 + 6 = – 8/3 + 6 = 10/3
Player B can mix all the 2 strategies
Player B’s loss = 10/3
Optimum Solution for Player A
Combination (1) and (3):
(1,3) x2 and x4 = 0, x3 = x1 –1 (x1* = x3*)
Optimum Solution for Player A
(1,3) x2 and x4 = 0, x3 = x1 –1 (x1* = x3*)
B’s Pure Strategy
A’s Expected Payoff
1
– 2x1 + 4
3
– 8x1 +6
–2x1 + 4 = – 8x1 +6
6 x1 = 2
x1* = 1/3 dan x3* = 1/3
A’s gain = 10/3
M N Games: Simplex
Focus on Row (Player A)
Duality Problem
Objective Function: maximize
w = Y1 + Y2 + . . . Yn
M N Games: Simplex
Subject to (Constraints):
a11 Y1 + a12 Y2 + . . . + a1nYn 1
a21 Y1 + a22 Y2 + . . . + a2nYn 1
…
…
…
am1 Y1 + am2 Y2 + . . . + amnYn 1
Y1, Y2, . . . , Yn 0
w = 1/v
v* = 1/w
Yj = Yi /v,
j = 1,2,. . . , n
M N Games: Simplex
Ensure the tableau does not contain any
zero and negative value
Use K (a constant value) to make sure that
the tableau does not contain any zero and
negative value
K > negative of the maximin value
K > negatif of the most negative value
M N Games: Simplex
If K is used in the tableau,
v* = 1/w – K
z=w
X1* = X1/z, X2* = X2/z, . . . , Xm* = Xm/z
M N Games: Example
B
A
Column
Row
1
2
3
Min
1
3
–1
–3
–3
2
–3
3
–1
–3
3
–4
–3
3
–4
Max
3
3
3
K=5
B
A
Column
Row
1
2
3
Min
1
8
4
2
2
2
2
8
4
2
3
1
2
8
1
Max
8
8
8
Objective Function:
Maximize: w = Y1 + Y2 + Y3
B
A
Column
Row
1
2
3
Min
1
8
4
2
2
2
2
8
4
2
3
1
2
8
1
Max
8
8
8
Subject to :
8Y1 + 4Y2 + 2Y3 1
2Y1 + 8Y2 + 4Y3 1
1Y1 + 2Y2 + 8Y3 1
Y1, Y2,Y3 0
Subject to :
8Y1 + 4Y2 + 2Y3 1 8Y1 + 4Y2 + 2Y3 + S1 = 1
2Y1 + 8Y2 + 4Y3 1 2Y1 + 8Y2 + 4Y3 + S2 = 1
1Y1 + 2Y2 + 8Y3 1 1Y1 + 2Y2 + 8Y3 + S3 = 1
Y1, Y2,Y3 0
Objective Function:
Maximize: w = Y1 + Y2 + Y3 + S1+S2+S3
Basic Y1 Y2 Y3 S1 S2
S3 Solution
w
-1
-1
-1
0
0
0
0
S1
8
4
2
1
0
0
1
S2
S3
2
1
8
2
4
8
0
0
1
0
0
1
1
1
Final Optimal Tableau
Basic Y1 Y2 Y3
S1
S2
S3
w
0
0
0
5/49 11/196 1/14
Y1
1
0
0
1/7
Y2
Y3
0
0
1
0
0
1
-1/14
0
-3/98 31/196 -1/14
-1/98 -3/98
1/7
Solution
45/196
1/14
11/96
5/49
Solution for B
w = 45/196
v* = 1/w – K = 196/45 – 225/45 = –29/45
y1* = Y1/w = (1/14)/(45/196) = 14/45
y2* = Y2/w = (11/196)/(45/196) = 11/45
y3* = Y3/w = (5/49)/(45/196) = 20/45
Solution for A
z = w = 45/196
X1 = 5/49
X2 = 11/196
X3 = 1/14
x1* = X1/z = (5/49)/(45/196) = 20/45
x2* = X2/z = (11/196)/(45/196) = 11/45
x3* = X3/z = (1/14)/(45/196) = 14/45
The End
This is the end of Chapter 8B
Game Theory
Game Theory
Elements of a game
Players: intelligent opponents (competitors
or enemies)
Strategies: choices of what to do to defeat
the opponents
Outcomes = Payoffs: functions of the
different strategies for each player
Payoff Matrix: Tableau (of the gain of the
Row Player)
Rule: how to allocate the payoffs to the
players
The Game: Example
Two Players: Player A (Row) and Player B
(Column)
Tossing a balanced coin
Possible outcomes: Head (H) and Tail (T)
Rule:
If
the outcome matches (player A selects H and the
outcome is also H or player A select T and the
outcomes is also T), Player A gets $1 from Player B;
Otherwise Player A loses $1 to Player B
The Game: Payoff Matrix
Player B
Player A
(Row Player)
H
T
H
1
–1
T
–1
1
Strategies of each player : H or T
An Optimum Solution is said to be reached if
neither player finds it beneficial to alter his
strategy
Optimum Solution
Optimum Solution can be achieved if
players choose to apply:
Pure strategy (e.g. select H or T)
A Mixture of pure strategies = Mixed Strategy
Two-Person Zero-Sum Game
A game with two players
A gain of one player equals a loss to the other
The focused player = the row player (Player A)
Player A maximize his minimum gain (why?)
Player B minimize her maximum loss (why?)
Optimum Solution is gained by Minimax-Maximin
criterion
Optimum Solution reflects that the game is
stable or in the a state of equilibrium
Two-Person Zero-Sum Game with
Saddle Point
1
Player B
2
3
4
Row
Min
1
8
2
9
5
2
Player A 2
6
5
7
18
5
3
7
3
–4
10
–4
Column Max 8
5
9
18
Minimax
Value
Maximin
Value
Two-Person Zero-Sum Game with
Saddle Point
Player A (Row): Maximin Value = Lower
Value of the game
Player B (Column): Minimax Value =
Upper Value of the game
Maximin value = Minimax value Saddle
point = Value of the game
Two-Person Zero-Sum Game with
Saddle Point
Saddle point leads to Optimum Solution
Saddle point indicates a stable game
Players apply Pure Strategy
In general
To maintain the “Optimality” of a game:
maximin value value of the game minimax value
OR
lower value value of the game upper value
Mixed Strategies
Used for solving a game that does not
have a saddle point
Used for solving a game that does not
have a saddle point
Optimum Solution is gained using:
Graphical Solution for (2 N) and (M 2)
Simplex for (M N) game
Unstable Game without Saddle Point
Player B
Player A
Column
Max
2
3
–10 9
Row
1
1
5
4
0
Min
–10
2
6
7
8
2
2
3
8
5
4
15
4 Maximin
4
7
4
–1
3
–1
8
7
9
15
Value
Minimax
Value
Minimax value = 7 > Maximin value = 4 sub-optimal
2 N game
2 N game:
Player A has 2 strategies
Player B has N ( 2) strategies
B
A
y1
y2
… yn
a11
a12
… a1n
x2 = 1 – x1 a21
a22
… a2n
x1
2 N game
B’s Pure Strategy A’s Expected Payoff
1
(a11 – a21)x1 + a21
2
(a12 – a22)x1 + a22
…
…
n
(a1n – a2n)x1 + a2n
2 N game: Example
B
A
y1
y2
y3
y4
x1
2
2
3
–1
x2
4
3
2
6
2 N game
B’s Pure Strategy
A’s Expected Payoff
1
– 2x1 + 4
2
– x1 + 3
3
x1 + 2
4
–7 x1 + 6
Optimum Solution: Graphical Solution
Graphical Solution x1 = 0 and x1 = 1 = x2
6
5
Maximin
1
4
3
3
2
2
1
4
x1 = 1
x1 = 0
-1
x*1 =1/2
Optimum Solution for Player A
Intercept between lines (2), (3) and (4)
(x1* = ½, x2*= ½)
(2) – x1 + 3 = – ½ + 3 = 5/2
v*
(3) x1 + 2 = ½ + 2 = 5/2
(4) –7 x1 + 6 = – 7/2 + 6 = 5/2
Player B can mix all the 3 strategies
Player A’s gains = 5/2
Optimum Solution for Player B
Combination (2), (3) and (4):
(2,3) y1 and y4 = 0, y3 = y2 –1 (y2* = y3*)
(2,4) y1 and y3 = 0, y4 = y2 –1 (y2* = y4*)
(3,4) y1 and y2 = 0, y4 = y3 –1 (y3* = y4*)
Optimum Solution for Player B
(2,3) y1 and y4 = 0, y3 = y2 –1 (y2* = y3*)
A’s Pure
Strategy
2
B’s Expected Payoff
3
y2 + 2
– y2 + 3 = y2 + 2
– 2 y2 = – 1
y2* = 1/2 dan y3* = 1/2
– y2 + 3
B’s loss = 5/2
Optimum Solution for Player B
(2,4) y1 and y3 = 0, y4 = y2 –1 (y2* = y4*)
A’s Pure
Strategy
2
B’s Expected Payoff
4
– 7y2 + 6
– y2 + 3 = –7y2 + 6
6 y2 = 3
y2* = 1/2 dan y4* = 1/2
– y2 + 3
B’s loss = 5/2
Optimum Solution for Player B
(3,4) y1 and y2 = 0, y4 = y3 –1 (y3* = y4*)
A’s Pure
Strategy
3
B’s Expected Payoff
4
– 7y3 + 6
y3 + 2 = –7y3 + 6
8 y3 = 4
y3* = 1/2 dan y4* = 1/2
y3 + 2
B’s loss = 5/2
M 2 game
M 2 game:
Player A has M ( 2) strategies
Player B has 2 strategies
B
x1
A
x2
…
y1
a11
a21
…
y2= 1 – y1
a12
a22
…
xm
am1
am2
M 2 game
A’s Pure Strategy B’s Expected Payoff
1
(a11 – a12)y1 + a12
2
(a21 – a22)y1 + a22
…
…
m
(am1 – am2)y1 + am2
M 2 game: Example
B
A
y1
y2
x1
2
4
x2
3
2
x3
–2
6
M 2 game
A’s Pure Strategy
B’s Expected Payoff
1
– 2 y1 + 4
2
y1 + 2
3
– 8 y1 + 6
Optimum Solution: Graphical Solution
Graphical Solution y1 = 0 and y1 = 1 = y2
6
Minimax
5
3
4
2
3
1
2
y1 = 1
1
-1
-2
y1* = y3* = 1/3
Optimum Solution for Player B
Intercept between lines (1) and (3)
(y1* = 1/3, y3*= 1/3)
(1) – 2y1 + 4 = – 2/3 + 4 = 10/3
v*
(3) – 8y1 + 6 = – 8/3 + 6 = 10/3
Player B can mix all the 2 strategies
Player B’s loss = 10/3
Optimum Solution for Player A
Combination (1) and (3):
(1,3) x2 and x4 = 0, x3 = x1 –1 (x1* = x3*)
Optimum Solution for Player A
(1,3) x2 and x4 = 0, x3 = x1 –1 (x1* = x3*)
B’s Pure Strategy
A’s Expected Payoff
1
– 2x1 + 4
3
– 8x1 +6
–2x1 + 4 = – 8x1 +6
6 x1 = 2
x1* = 1/3 dan x3* = 1/3
A’s gain = 10/3
M N Games: Simplex
Focus on Row (Player A)
Duality Problem
Objective Function: maximize
w = Y1 + Y2 + . . . Yn
M N Games: Simplex
Subject to (Constraints):
a11 Y1 + a12 Y2 + . . . + a1nYn 1
a21 Y1 + a22 Y2 + . . . + a2nYn 1
…
…
…
am1 Y1 + am2 Y2 + . . . + amnYn 1
Y1, Y2, . . . , Yn 0
w = 1/v
v* = 1/w
Yj = Yi /v,
j = 1,2,. . . , n
M N Games: Simplex
Ensure the tableau does not contain any
zero and negative value
Use K (a constant value) to make sure that
the tableau does not contain any zero and
negative value
K > negative of the maximin value
K > negatif of the most negative value
M N Games: Simplex
If K is used in the tableau,
v* = 1/w – K
z=w
X1* = X1/z, X2* = X2/z, . . . , Xm* = Xm/z
M N Games: Example
B
A
Column
Row
1
2
3
Min
1
3
–1
–3
–3
2
–3
3
–1
–3
3
–4
–3
3
–4
Max
3
3
3
K=5
B
A
Column
Row
1
2
3
Min
1
8
4
2
2
2
2
8
4
2
3
1
2
8
1
Max
8
8
8
Objective Function:
Maximize: w = Y1 + Y2 + Y3
B
A
Column
Row
1
2
3
Min
1
8
4
2
2
2
2
8
4
2
3
1
2
8
1
Max
8
8
8
Subject to :
8Y1 + 4Y2 + 2Y3 1
2Y1 + 8Y2 + 4Y3 1
1Y1 + 2Y2 + 8Y3 1
Y1, Y2,Y3 0
Subject to :
8Y1 + 4Y2 + 2Y3 1 8Y1 + 4Y2 + 2Y3 + S1 = 1
2Y1 + 8Y2 + 4Y3 1 2Y1 + 8Y2 + 4Y3 + S2 = 1
1Y1 + 2Y2 + 8Y3 1 1Y1 + 2Y2 + 8Y3 + S3 = 1
Y1, Y2,Y3 0
Objective Function:
Maximize: w = Y1 + Y2 + Y3 + S1+S2+S3
Basic Y1 Y2 Y3 S1 S2
S3 Solution
w
-1
-1
-1
0
0
0
0
S1
8
4
2
1
0
0
1
S2
S3
2
1
8
2
4
8
0
0
1
0
0
1
1
1
Final Optimal Tableau
Basic Y1 Y2 Y3
S1
S2
S3
w
0
0
0
5/49 11/196 1/14
Y1
1
0
0
1/7
Y2
Y3
0
0
1
0
0
1
-1/14
0
-3/98 31/196 -1/14
-1/98 -3/98
1/7
Solution
45/196
1/14
11/96
5/49
Solution for B
w = 45/196
v* = 1/w – K = 196/45 – 225/45 = –29/45
y1* = Y1/w = (1/14)/(45/196) = 14/45
y2* = Y2/w = (11/196)/(45/196) = 11/45
y3* = Y3/w = (5/49)/(45/196) = 20/45
Solution for A
z = w = 45/196
X1 = 5/49
X2 = 11/196
X3 = 1/14
x1* = X1/z = (5/49)/(45/196) = 20/45
x2* = X2/z = (11/196)/(45/196) = 11/45
x3* = X3/z = (1/14)/(45/196) = 14/45
The End
This is the end of Chapter 8B