Pareto Optimality

Game Theory Course: Jackson, Leyton-Brown & Shoham

Game Theory Course: Jackson, Leyton-Brown & Shoham

Pareto Optimality

Analyzing Games
Weve defined some canonical games, and thought about how

to play them. Now lets examine the games from the outside From the point of view of an outside observer, can some outcomes of a game be said to be better than others?

Game Theory Course: Jackson, Leyton-Brown & Shoham

Pareto Optimality

Analyzing Games
Weve defined some canonical games, and thought about how

to play them. Now lets examine the games from the outside From the point of view of an outside observer, can some outcomes of a game be said to be better than others?
cant say one agents interests are more important than anothers imagine trying to find the revenue-maximizing outcome when you

dont know what currency is used to express each agents payoff Are there ways to still prefer one outcome to another?

Game Theory Course: Jackson, Leyton-Brown & Shoham

Pareto Optimality

Pareto Optimality
Idea: sometimes, one outcome o is at least as good for every

agent as another outcome o , and there is some agent who strictly prefers o to o

in this case, it seems reasonable to say that o is better than o we say that o Pareto-dominates o .

Game Theory Course: Jackson, Leyton-Brown & Shoham

Pareto Optimality

Pareto Optimality
Idea: sometimes, one outcome o is at least as good for every

agent as another outcome o , and there is some agent who strictly prefers o to o

in this case, it seems reasonable to say that o is better than o we say that o Pareto-dominates o .

. An outcome o is Pareto-optimal if there is no other outcome that Pareto-dominates it. .

Definition (Pareto Optimality)

Game Theory Course: Jackson, Leyton-Brown & Shoham

Pareto Optimality

Pareto Optimality
Idea: sometimes, one outcome o is at least as good for every

agent as another outcome o , and there is some agent who strictly prefers o to o

in this case, it seems reasonable to say that o is better than o we say that o Pareto-dominates o .

. An outcome o is Pareto-optimal if there is no other outcome that Pareto-dominates it. .

can a game have more than one Pareto-optimal outcome?

Definition (Pareto Optimality)

Game Theory Course: Jackson, Leyton-Brown & Shoham

Pareto Optimality

Pareto Optimality
Idea: sometimes, one outcome o is at least as good for every

agent as another outcome o , and there is some agent who strictly prefers o to o

in this case, it seems reasonable to say that o is better than o we say that o Pareto-dominates o .

. An outcome o is Pareto-optimal if there is no other outcome that Pareto-dominates it. .

can a game have more than one Pareto-optimal outcome? does every game have at least one Pareto-optimal outcome?
Game Theory Course: Jackson, Leyton-Brown & Shoham Pareto Optimality

Definition (Pareto Optimality)

3.5.

Pareto Optimal Outcomes in Example Games

Left Left Right Right

1, 1 0, 0

0, 0 1, 1

Figure 3.5: Coordination game.

m games

her end of the spectrum from pure coordination games lie zero-sum games, earing in mind the comment we made earlier about positive afne transns) are more properly called constant-sum games. Unlike common-payoff onstant-sum games are meaningful primarily in the context of two-player ot necessarily two-strategy) games.
. Game Theory Course: Jackson, Leyton-Brown & Shoham Pareto Optimality

3.5.

Pareto Optimal Outcomes in Example Games

Left Left Right Right
B B F 2, 1 0, 0 F 0, 0 1, 2

Figure 3.6

Rock, Paper, Scissors game.

1, 1 0, 0

0, 0 1, 1

Figure 3.5: Coordination game.Figure 3.7 Battle of the Sexes game.

3.2.2

Strategies in normal-form games

m games

We have so far dened the actions available to each player in a game, but not yet his her end of the spectrum set of pure coordinationavailable choices. Certainly one kind of strategy is to select from strategies, or his games lie zero-sum games, pure mind a single action and play it; we call such a strategy earing in strategythe comment we made earlier about positive afne trans-a pure strategy, and we will use the notation we have already developed for actions to ns) are more properly called constant-sum games. Unlike common-payoff represent it. There is, however, onstant-sum games are another, less obvious type thestrategy; a player can choose to randomize over the set of meaningful primarily in of context of two-player available actions according to some probability distribution; such a strategy is called ot necessarily two-strategy) games. mixed strategy a mixed strategy. Game Theory Course: Jackson, Leyton-Brown & ShohamAlthough it may not be immediatelyOptimality Pareto obvious why a player should
.

of pure competition; one players gain must come at the expense of the 3.5. Figure 3.6 Rock, Paper, Scissors yer. This property requires that there be exactly two agents. Indeed, if game. w more agents, any game can be turned into a zero-sum game by adding Right B player whose actions do notLeft impact the payoffs to the other agents, F and wn payoffs are chosen to make the payoffs in each outcome sum to zero. ical example of a zero-sum game is the0, 0 of Matching Pennies. In this game B 2, 1 0, 0 Left 1, 1 ch of the two players has a penny and independently chooses to display ads or tails. The two players then compare their pennies. If they are the Right 0, 0 1, 1 n player 1 pockets both, and otherwise player 2 pocketsFthem.0, 0 payoff The 1, 2 shown in Figure 3.6. Figure 3.5: Coordination game.Figure 3.7 Battle of the Sexes game. Heads Tails

Pareto Optimal Outcomes in Example Games

m games

We have so far dened the actions available to each player in a game, but not yet his her end of the spectrum set of pure coordinationavailable choices. Certainly one kind of strategy is to select from strategies, or his games lie zero-sum games, 1, 1 pure mind Tails a single action and play it; we call such a strategy earing in strategythe comment we made 1, 1 about positive afne trans-a pure strategy, and we will use earlier the notation we have already developed for actions to ns) are more properly called constant-sum games. Unlike common-payoff represent it. There is, however, onstant-sum games are another, less obvious type thestrategy; a player can choose to randomize over the set of meaningful Pennies game. Figure 3.6: Matching primarily in of context of two-player available actions according to some probability distribution; such a strategy is called ot necessarily two-strategy) games. mixed strategy a mixed strategy. Game Theory Course: Jackson, Leyton-Brown & ShohamAlthough it may not be immediatelyOptimality Pareto obvious why a player should
.

3.2.2 Strategies in normal-form games Heads 1, 1 1, 1

of pure competition; one players gain must come at the expense of the when congestion occurs. You 3.6 Rock, Indeed, strategies: 3.5. Figure have two Paper, Scissors yer. This property requires that there be exactly two agents. possible if game. C (for using a Correct implementation) and Da (for using a Defective one). If both you and your colleague w more agents, any game can be turned into zero-sum game by adding then Right B player whose actions adopt Cimpact the average packet delay is 1ms (millisecond). If you both adopt D the do notLeft your payoffs to the other agents, F and delay is 3ms, because of additional overhead at the network router. Finally, if one of wn payoffs are chosen to make the payoffs in each outcome sum to zero. you adopts D and the other adopts C then the D adopter will experience no delay at all, ical example of a zero-sum game is the0, 0 of Matching Pennies. In this game 0, Left the1, 1 but C adopter will experience B delay2, 1 4ms. 0 a of ch of the two players has a penny and independently chooses to display These consequences are shown in Figure 3.1. Your options are the two rows, and ads or tails. The two players then compare their pennies. If they In each cell, the rst number represents are the your colleagues1, 1 options are the columns. Right 0, 0 n player 1 pockets both, andpayoff (or, minus your delay),them.0, 0 payoff otherwise player 2 pocketsF and the second2 The 1, number represents your colleagues your shown inusers 3.6. payoff.1 TCP Figure Figure 3.5: Coordination game.Figure 3.7 Battle of the Sexes game. game Heads Tails C D
.

Pareto Optimal Outcomes in Example Games

Prisoners dilemma game 3.2.2

m games

4, 0 We have so far dened the actions available to each player in a game, but not yet his her end of the spectrum set of pure coordinationavailable choices. Certainly one kind of strategy is to select from strategies, or his games lie zero-sum games, 1, 1 D pure mind Tails a single action and play it; we call 0, 4 a strategy a such earing in strategythe comment we made 1, 1 about positive afne 3, 3 pure strategy, and we will use earlier transthe notation we have already developed for actions to ns) are more properly called constant-sum games. Unlike common-payoff represent it. There is, however, onstant-sum games are another, less obvious type thestrategy; a player canPrisoners) Dilemma. over the set of meaningful Pennies game. context of two-player Figure 3.6: Matching primarily 3.1 ofThe TCP users (aka the choose to randomize Figure in available actions according to some probability distribution; such a strategy is called ot necessarily two-strategy) games. mixed strategy a mixed strategy. Game Theory Course: Jackson, Leyton-Brown & ShohamAlthough it may not be immediatelyOptimality Pareto obvious why a player should
.

Heads

Strategies in normal-form games 1, 1 1, 1 C 1, 1

of pure competition; one players gain must come at the expense of the when congestion occurs. You 3.6 Rock, Indeed, strategies: 3.5. Figure have two Paper, Scissors yer. This property requires that there be exactly two agents. possible if game. C (for using a Correct implementation) and Da (for using a Defective one). If both you and your colleague w more agents, any game can be turned into zero-sum game by adding then Right B player whose actions adopt Cimpact the average packet delay is 1ms (millisecond). If you both adopt D the do notLeft your payoffs to the other agents, F and delay is 3ms, because of additional overhead at the network router. Finally, if one of wn payoffs are chosen to make the payoffs in each outcome sum to zero. you adopts D and the other adopts C then the D adopter will experience no delay at all, ical example of a zero-sum game is the0, 0 of Matching Pennies. In this game 0, Left the1, 1 but C adopter will experience B delay2, 1 4ms. 0 a of ch of the two players has a penny and independently chooses to display These consequences are shown in Figure 3.1. Your options are the two rows, and ads or tails. The two players then compare their pennies. If they In each cell, the rst number represents are the your colleagues1, 1 options are the columns. Right 0, 0 n player 1 pockets both, andpayoff (or, minus your delay),them.0, 0 payoff otherwise player 2 pocketsF and the second2 The 1, number represents your colleagues your shown inusers 3.6. payoff.1 TCP Figure Figure 3.5: Coordination game.Figure 3.7 Battle of the Sexes game. game Heads Tails C D
.

Pareto Optimal Outcomes in Example Games

Prisoners dilemma game 3.2.2

m games

4, 0 We have so far dened the actions available to each player in a game, but not yet his her end of the spectrum set of pure coordinationavailable choices. Certainly one kind of strategy is to select from strategies, or his games lie zero-sum games, 1, 1 D pure mind Tails a single action and play it; we call 0, 4 a strategy a such earing in strategythe comment we made 1, 1 about positive afne 3, 3 pure strategy, and we will use earlier transthe notation we have already developed for actions to ns) are more properly called constant-sum games. Unlike common-payoff represent it. There is, however, onstant-sum games are another, less obvious type thestrategy; a player canPrisoners) Dilemma. over the set of meaningful Pennies game.The dilemma: The paradox of 3.1 of context of two-player Figure 3.6: Matching primarilyPrisonersTCP users (aka the choose to randomize Figure in available actionsis the only to some probability distribution; such a strategy is called ot necessarily (DS) Nash equilibrium according non-Pareto-optimal outcome! the two-strategy) games. mixed strategy a mixed strategy. Game Theory Course: Jackson, Leyton-Brown & ShohamAlthough it may not be immediatelyOptimality Pareto obvious why a player should
.

Heads

Strategies in normal-form games 1, 1 1, 1 C 1, 1