game theory - london school of economics and political science · 2011-12-12 · game theory and...
TRANSCRIPT
Game Theoryand Electronic Communications Markets
Bernhard von Stengel
London School of Economics
1
Non-cooperative game theory
• Players motivated by individual incentives
• Interactions resulting in payoffs
Explains:
• Selfish but collectively damaging behaviour
• How to think strategically
• More than one possible equilibrium (stable outcome)
• Rules of the game matter
• Selfish behaviour in networks
2
Explain selfish behaviorPrice-setting game
high
high low
price
price
price price
low
1
2
1
3
2
0
1
3
02
3
Explain selfish behaviorPrice-setting game
high
high low
price
price
price price
low
1
2
1
3
2
0
1
3
02
3
Explain selfish behaviorPrice-setting game
high
high low
price
price
price price
low
1
2
1
3
2
0
1
3
02
3
Explain selfish behaviorFishing game
over−
fishing
sustainable
overfishing
fishing
fishing
sustainable
1
2
1
3
2
0
1
3
02
price competition positive for consumers, but same game betweenfishing nations detrimental for all!
3
Thinking strategically: Epson vs. HP
stay out enter (laser printer market)
lower prices
fight
HP
Epson
Epson
14
−2 −1
3
2
5
0keep prices
back out
4
Thinking strategically: Epson vs. HP
stay out enter (laser printer market)
lower prices
fight
HP
Epson
Epson
14
−2 −1
3
2
5
0keep prices
back out
4
Thinking strategically: Epson vs. HP
stay out enter (laser printer market)
HP
Epson
3
2
5
0keep prices
4
Thinking strategically: Epson vs. HP
stay out enter (laser printer market)
lower prices
fight
HP
Epson
Epson
14
−2 −1
3
2
5
0keep prices
back out
4
Thinking strategically: Epson vs. HP
lower prices
fight
Epson
14
−2 −1
back out
stay out enter (laser printer market)
HP
Epson
3
2
5
0keep prices
4
Thinking strategically: Epson vs. HP
lower prices
fight
Epson
14
−2 −1
back out
stay out enter (laser printer market)
HP
Epson
3
2
5
0keep prices
4
Thinking strategically: Epson vs. HP
lower prices
fight
Epson
14
−2 −1
back out
stay out enter (laser printer market)
HP
Epson
3
2
5
0keep prices
4
Thinking strategically: Epson vs. HP
lower prices
fight
Epson
14
−2 −1
back out
stay out enter (laser printer market)
HP
Epson
3
2
5
0keep prices
4
More than one possible equilibrium
Equilibrium = one strategy for each player, which is optimalwhen the other players’ strategies stay fixed.
A game may have more than one equilibrium:
For example, in the Epson vs. HP game:
Equilibrium 1: (stay out , lower prices)
Equilibrium 2: (enter / fight , keep prices)
not an equilibrium: (enter / fight , lower prices)
5
More than one possible equilibrium
Equilibrium = one strategy for each player, which is optimalwhen the other players’ strategies stay fixed.
A game may have more than one equilibrium:
For example, in the Epson vs. HP game:
Equilibrium 1: (stay out , lower prices)
Equilibrium 2: (enter / fight , keep prices)
not an equilibrium: (enter / fight , lower prices)
5
More than one possible equilibrium
Equilibrium = one strategy for each player, which is optimalwhen the other players’ strategies stay fixed.
A game may have more than one equilibrium:
For example, in the Epson vs. HP game:
Equilibrium 1: (stay out , lower prices)
Equilibrium 2: (enter / fight , keep prices)
not an equilibrium: (enter / fight , lower prices)
5
The bandwidth choice game
Which equilibrium?
Evolved from starting distribution:
bandwidth
bandwidthlow
high
high low
width
band−
width
band−
1
2
1
1
5
0
1
1
5 0
6
The bandwidth choice game
Which equilibrium?
Evolved from starting distribution:
bandwidth
bandwidthlow
high
high low
width
band−
width
band−
1
2
1
1
5
0
1
1
5 0
6
The bandwidth choice gameWhich equilibrium?
Evolved from starting distribution:
bandwidth
bandwidthlow
high
high low
width
band−
width
band−
1
2
1
1
5
0
1
1
5 0
6
The bandwidth choice gameWhich equilibrium? Evolved from starting distribution:
1
70 %30 %
1.5bandwidth
bandwidthlow
high
high low
width
band−
width
band−
1
2
1
1
5
0
1
1
5 0
6
The bandwidth choice gameWhich equilibrium? Evolved from starting distribution:
50 % 50 %
2.5
1
bandwidth
bandwidthlow
high
high low
width
band−
width
band−
1
2
1
1
5
0
1
1
5 0
6
The bandwidth choice gameWhich equilibrium? Evolved from starting distribution:
4
20 %80 %
1
bandwidth
bandwidthlow
high
high low
width
band−
width
band−
1
2
1
1
5
0
1
1
5 0
6
The bandwidth choice gameWhich equilibrium? Evolved from starting distribution:
100 % 0 %
5
1
bandwidth
bandwidthlow
high
high low
width
band−
width
band−
1
2
1
1
5
0
1
1
5 0
6
Rules of the game matter
The Quality Game
changed to a game with commitment
high
buy don’t
low
quality
quality
buy1
2
1
3
2
0
1
1
02
⇒ Commitment power can help both players!
7
Rules of the game matter
The Quality Game
changed to a game with commitment
high
buy don’t
low
quality
quality
buy1
2
1
3
2
0
1
1
02
⇒ Commitment power can help both players!
7
Rules of the game matter
The Quality Game
changed to a game with commitment
high
buy don’t
low
quality
quality
buy1
2
1
3
2
0
1
1
02
⇒ Commitment power can help both players!
7
Rules of the game matter
The Quality Game changed to a game with commitment
12 0
0 132
1
buybuy
2
1
buy buy
2
don’t don’t
low qualityhigh quality
high
buy don’t
low
quality
quality
buy1
2
1
3
2
0
1
1
02
⇒ Commitment power can help both players!
7
Rules of the game matter
The Quality Game changed to a game with commitment
12 0
0 132
1
buybuy
2
1
buy buy
2
don’t don’t
low qualityhigh quality
high
buy don’t
low
quality
quality
buy1
2
1
3
2
0
1
1
02
⇒ Commitment power can help both players!
7
Rules of the game matter
The Quality Game changed to a game with commitment
12 0
0 132
1
buybuy
2
1
buy buy
2
don’t don’t
low qualityhigh quality
high
buy don’t
low
quality
quality
buy1
2
1
3
2
0
1
1
02
⇒ Commitment power can help both players!
7
Rules of the game matter
The Quality Game changed to a game with commitment
12 0
0 132
1
buybuy
2
1
buy buy
2
don’t don’t
low qualityhigh quality
high
buy don’t
low
quality
quality
buy1
2
1
3
2
0
1
1
02
⇒ Commitment power can help both players!
7
Selfish routing in networks
A B
x delayed by x
1−x delayed by 1
flow 1
Selfish routing 33% worse than optimal, centrally planned routing
(33% longer delay is worst possible if delay functions are linear,e.g. flow x delayed by x)
8
Selfish routing in networks
optimal: 1−x = 0.5
optimal: x = 0.5
average delay: 0.25 + 0.5 = 0.75
A B
x delayed by x
1−x delayed by 1
flow 1
Selfish routing 33% worse than optimal, centrally planned routing
(33% longer delay is worst possible if delay functions are linear,e.g. flow x delayed by x)
8
Selfish routing in networks
selfish: x = 1.0
average delay: 1.0 + 0 = 1.0
selfish: 1−x = 0
A B
x delayed by x
1−x delayed by 1
flow 1
Selfish routing 33% worse than optimal, centrally planned routing
(33% longer delay is worst possible if delay functions are linear,e.g. flow x delayed by x)
8
Selfish routing in networks
selfish: x = 1.0
average delay: 1.0 + 0 = 1.0
selfish: 1−x = 0
A B
x delayed by x
1−x delayed by 1
flow 1
Selfish routing 33% worse than optimal, centrally planned routing
(33% longer delay is worst possible if delay functions are linear,e.g. flow x delayed by x)
8
Selfish routing can be optimal
D
C
BA
x delayed by x
1−x delayed by 1
flow 1
1−x delayed by 1−x
x delayed by 1
Increasing network capacity can worsen equilibrium congestion!
9
Selfish routing can be optimal
optimal: x = 0.5
optimal: 1−x = 0.5
optimal average delay: 0.75 + 0.75 = 1.5
D
C
BA
x delayed by x
1−x delayed by 1
flow 1
1−x delayed by 1−x
x delayed by 1
Increasing network capacity can worsen equilibrium congestion!
9
Selfish routing can be optimal
selfish =
selfish = optimal: x = 0.5
optimal: 1−x = 0.5
optimal average delay: 0.75 + 0.75 = 1.5
D
C
BA
x delayed by x
1−x delayed by 1
flow 1
1−x delayed by 1−x
x delayed by 1
Increasing network capacity can worsen equilibrium congestion!
9
Add a shortcut
1−x+z delayed by 1−x+z
x−z delayed by 1
z delayed by 0
D
C
BA
x delayed by x
1−x delayed by 1
flow 1
Increasing network capacity can worsen equilibrium congestion!
9
Add a shortcut
1−x+z delayed by 1−x+z
x−z delayed by 1
z delayed by 0
optimal average delay: 0.75 + 0.75 = 1.5
optimal: x = 0.5, z=0
optimal: 1−x = 0.5
D
C
BA
x delayed by x
1−x delayed by 1
flow 1
Increasing network capacity can worsen equilibrium congestion!
9
Add a shortcut
selfish: x = z = 1.0
average delay: 1.0 + 0 + 1.0 = 2.01−x+z delayed by 1−x+z
x−z delayed by 1
z delayed by 0
D
C
BA
x delayed by x
1−x delayed by 1
flow 1
Increasing network capacity can worsen equilibrium congestion!
9
Add a shortcut
selfish: x = z = 1.0
average delay: 1.0 + 0 + 1.0 = 2.01−x+z delayed by 1−x+z
x−z delayed by 1
z delayed by 0
optimal average delay: 0.75 + 0.75 = 1.5
D
C
BA
x delayed by x
1−x delayed by 1
flow 1
Increasing network capacity can worsen equilibrium congestion!
9
Braess’s paradox
selfish: x = z = 1.0
average delay: 1.0 + 0 + 1.0 = 2.01−x+z delayed by 1−x+z
x−z delayed by 1
z delayed by 0
optimal average delay: 0.75 + 0.75 = 1.5
D
C
BA
x delayed by x
1−x delayed by 1
flow 1
Increasing network capacity can worsen equilibrium congestion!
9
Summary: Game theory
• provides tools for analysing interactive decisions
• very suitable for: few players, or large number of similarplayers; competitive dynamics of markets; “thinking ahead”
• important are: incentives; rules of the game;information of the players
• central concept: equilibrium (not always unique)
• explain: selfish behaviour in competition, networks androuting
• many more applications: design of auctions, optimal bidding,. . .
10
Summary: Game theory
• provides tools for analysing interactive decisions
• very suitable for: few players, or large number of similarplayers; competitive dynamics of markets; “thinking ahead”
• important are: incentives; rules of the game;information of the players
• central concept: equilibrium (not always unique)
• explain: selfish behaviour in competition, networks androuting
• many more applications: design of auctions, optimal bidding,. . .
10
Summary: Game theory
• provides tools for analysing interactive decisions
• very suitable for: few players, or large number of similarplayers; competitive dynamics of markets; “thinking ahead”
• important are: incentives; rules of the game;information of the players
• central concept: equilibrium (not always unique)
• explain: selfish behaviour in competition, networks androuting
• many more applications: design of auctions, optimal bidding,. . .
10
Summary: Game theory
• provides tools for analysing interactive decisions
• very suitable for: few players, or large number of similarplayers; competitive dynamics of markets; “thinking ahead”
• important are: incentives; rules of the game;information of the players
• central concept: equilibrium (not always unique)
• explain: selfish behaviour in competition, networks androuting
• many more applications: design of auctions, optimal bidding,. . .
10
Summary: Game theory
• provides tools for analysing interactive decisions
• very suitable for: few players, or large number of similarplayers; competitive dynamics of markets; “thinking ahead”
• important are: incentives; rules of the game;information of the players
• central concept: equilibrium (not always unique)
• explain: selfish behaviour in competition, networks androuting
• many more applications: design of auctions, optimal bidding,. . .
10
Summary: Game theory
• provides tools for analysing interactive decisions
• very suitable for: few players, or large number of similarplayers; competitive dynamics of markets; “thinking ahead”
• important are: incentives; rules of the game;information of the players
• central concept: equilibrium (not always unique)
• explain: selfish behaviour in competition, networks androuting
• many more applications: design of auctions, optimal bidding,. . .
10