Strategy and Game Theory • Game theory is decision making in a strategic setting. • An agent’s optimal choice may depend on the choices or actions of others. • For example, suppose a company is deciding to choose a price for its product. It might take its rival’s pricing decision into account. Or a country is deciding to go to war, it may take the adversary’s decision into account (will they retaliate? Surrender?) • Many economic situations can be analysed using game theory. It is regarded as a more realistic and nuanced mode of analysis than mere static optimisation which depends on a set of rigid assumptions. • However, game theory is a vast and evolving topic which can get very complicated. • In this chapter we will only be scratching the surface of game theory. We may devote three or even four weeks to it. • William Spaniel’s youtube channel on game theory (Game Theory 101) is very informative. I highly recommend it. The Basics • A game is a model of a strategic situation. A game has three essential elements. (1) Players (2) Strategies and (3) Payoffs. • Some add a fourth element, information, but it can be subsumed under (2). • Let us look at each of the three elements separately. • Players: • Each decision maker in a game is a player. They could be individuals, firms or even entire nations. • A player has to choose from among a set of possible actions. • The number of players is usually fixed throughout the game. • We can have two, three up to n-player games. • Our primary focus would be on two-player games. • Strategies: • Each course of action is a strategy. Strategies could be complex or simple. • Let S1 denote the set of strategies open to player 1, S2 to player 2 and Si to player i. Thus, s1 ϵ S1 is a particular strategy chosen by player 1 s2 ϵ S2 is a particular strategy chosen by player 2 si ϵ Si is a particular strategy chosen by player i • A strategy profile is a listing of particular strategies chosen by each of a group of players. • Payoffs: • Payoffs are the final returns to the players at the conclusion of a game. • Payoffs can be measured in many ways – levels of utility, money, risk, prestige, emotion and so on. • Players are assumed to prefer higher payoffs to lower payoffs. • In a two player game, u1(s1,s2) represents player 1’s payoffs given he chooses s1 while the other player chooses s2. Similarly, u2(s2,s1) represents player 2’s payoffs. • In an n-player game, the payoff of generic player i is u i(si,s-i), where si is player i‘s own strategy and s-i = (s1,…,si-1,si+1,…,sn) of all players other than i. The Prisoner’s Dilemma • The first game we will encounter is a well known game in game theory called the prisoner’s dilemma (henceforth, PD). • P1 and P2 are the two players (prisoners). • Betray and Silence are the actions available to them. • The payoff structure of the game is as follows: • If P1 betrays and P2 is silent, P1 gets 1 year in prison while P2 gets 4 years in prison. • If P2 betrays and P1 is silent, P2 gets 1 year in prison while P1 gets 4 years in prison. • If P2 betrays and P1 also betrays, both get 3 years in prison. • If both remain silent, they both get 2 years in prison (due to lack of evidence). • Let loss of freedom (i.e. years in prison) be represented by negative numbers, then we can construct the following Payoff Matrix for the two prisoners. P2
represents the payoff to P1. • The exact same information can be represented in an equivalent table. • If we let four years in prison be equivalent to zero years of freedom, then three years in prison is equivalent to 1 year of freedom and so on. We get the following table: P2 Betray Silent P1 Betray 1,1 3, 0 Silent 0, 3 2, 2
• Confirm that it is equivalent to the previous table.
• When information is represented in tables it is called a “Normal Form” representation. • There is another way of representing the same information: (1 , 1) Betray This way of representing a 2 game is called an Betray (3 , 0) Silent “Extensive Form” 1 (0 , 3) representation. Betray The dashed oval Silent 2 indicates that P2 Silent (2 , 2) is not aware of P1’s actions.
Again, the first entry in brackets
always represents the payoff to P1. • Let us deal with positive payoffs. Now, think about what the prisoners might do. Neither is aware of the others actions. • If both are silent, the payoff is highest for both: (2,2). But there is no guarantee that the other prisoner will remain silent if one of them is silent. • If P1 betrays and P2 is silent, P1 is better-off: (3,0). On the other hand, if P2 were to betray, P1 is still better-off betraying: (1,1), because if he had remained silent the payoff would have been (0,3). • The same logic applies to P2. Betrayal always guarantees an extra year of freedom. Betrayal is a Best Response strategy. • When we have a payoff matrix, we can underline best response strategies. • Start with P1. If P1 betrays, what is P2’s best response? The best response would be the one which yields the highest payoff against P1’s action. Underline it. P2 When P1 plays Betray, P2 should play Betray and get a payoff of 1. If Betray Silent P2 had played Silent instead, he P1 Betray 1,1 3, 0 would have received a payoff of 0. Silent 0, 3 2, 2 Therefore, Silent does not represent a best response if P1 plays Betray. • If P1 is silent, what is P2’s best response? Again, the best response would be the one which yields the highest payoff against P1’s action. Underline it. P2 When P1 plays Silent, P2 should play Betray Silent Betray and get a payoff of 3. If P2 had played Silent instead, he would P1 Betray 1,1 3, 0 have received a payoff of 2. Silent 0, 3 2, 2 Therefore, Silent does not represent a best response if P1 plays Silent. • Now repeat the process with P2. If P2 betrays, what is P1’s best response? The best response would be the one which yields the highest payoff against P2’s action. Underline it. P2 When P2 plays Betray, P1 should Betray Silent play Betray and get a payoff of 1. If P1 had played Silent instead, he P1 Betray 1,1 3, 0 would have received a payoff of 0. Silent 0, 3 2, 2 Therefore, Silent does not represent a best response if P2 plays Betray. • If P2 is silent, what is P1’s best response? Again, the best response would be the one which yields the highest payoff against P2’s action. Underline it. P2 When P2 plays Silent, P1 should play Betray Silent Betray and get a payoff of 3. If P1 had played Silent instead, he would P1 Betray 1,1 3, 0 have received a payoff of 2. Silent 0, 3 2, 2 Therefore, Silent does not represent a best response if P2 plays Silent. • Now that we have underlined the best responses of both players against the other player’s actions we end up with a payoff matrix as follows. P2 Betray Silent P1 Betray 1,1 3, 0 Silent 0, 3 2, 2
• Notice that both the payoffs in the cell
corresponding to (Betray, Betray) are underlined. The strategy (Betray, Betray) is a best response strategy (also making it a dominant strategy). • Best Response: si is a best response for player i to rival’s strategies s-i , denoted s-i ϵ BRi (s-i) if ui(si , s-i) ≥ ui(s’i , s-i) for all s’i ϵ Si .
• Dominant Strategy: is a strategy si* for player i
that is a best response to all strategy profiles of other players. That is si* ϵ BRi (s-i) for all s-i . • We now introduce a very important equilibrium concept in economics. • Nash Equilibrium (NE): A NE is a strategy profile (s1*, s2*, … , sn*) such that, for each player i = 1, 2, … , n; si* is a best response to other player’s equilibrium strategies s-i* i.e. si*ϵ Bri(s-i*). • In a two-player game, (s1*,s2*) is a NE if s1* and s2* are mutual best responses against each other. u1(s1*, s2*) ≥ u1(s1 , s2*) for all s1ϵ S1 u2(s2*, s1*) ≥ u2(s2 , s1*) for all s2 ϵ S2 • There is a slight difference between a NE strategy and a dominant strategy. A dominant strategy is a best response to all other strategies, including NE strategies. • If all players in a game have a dominant strategy, we say that the game has a dominant strategy equilibrium. • In the PD game we just saw, (Betray, Betray) was a dominant strategy as well as a NE strategy.