12. A Characterisation of 2 × 2 Symmetric Games A symmetric 2 × 2 matrix game can be described in the following form A B A (a, a) (c, b) B (b, c) (d, d) Since it suffices to give the payoff of the first player, this can be simplified to A B A a c B b d It will be assumed that a, b, c, d are all distinct values. 1 / 34 One-Shot, Repeated Games and Communication In this chapter, I will consider (without any great detail), how the possibility of repeated interactions and communication may affect behaviour. In a one-shot game, players assume that they will not interact again. In a repeated game, players assume that they interact many times and can adapt their behaviour to the previous behaviour of the other player. When players can communicate, they can co-ordinate their actions, i.e. they are not forced to take their actions independently of each other. Actions can be co-ordinated e.g. based on the result of a coin toss. 2 / 34 Pure Nash Equilibria in 2 × 2 Symmetric Games Result 1: All 2 × 2 symmetric games have at least one pure Nash equilibrium. This can be proved by leading to a contradiction. Assume there are no pure Nash equilibria. Since (A, A) and (B, B) are not Nash equilibria, we have c > a and b > d. It follows from this that both (B, A) and (A, B) are Nash equilibria. 3 / 34 Pure Nash Equilibria in 2 × 2 Symmetric Games Result 2: There can be at most two pure equilibria, which appear on a diagonal of the matrix describing the game. Suppose (A, A) is a Nash equilibrium. Hence, c < a. It follows that neither (A, B) nor (B, A) can be Nash equilibria. Now suppose (B, A) is a Nash equilibrium. Hence, c > a and b > d. It follows that neither (A, A) nor (B, B) can be Nash equilibria (the rest follows from an analogous argument). 4 / 34 a) A Single Pareto-Dominant Symmetric Pure Equilibrium Assume that i) c > a, d > b, also ii) d > a, d > c. The first pair of inequalities state that action B dominates action A. It follows that the strategy A can be eliminated and hence (B, B) is the only Nash equilibrium. From the final pair of inequalities, (B, B) Pareto dominates all of the other payoff vectors, i.e. both players have a greater payoff using this pair of strategies, than another of pair of strategies. 5 / 34 a) A Single Pareto-Dominant Symmetric Pure Equilibrium Game 1 in the set of experimental games was such a game A B A (2, 2) (4, 6) B (6, 4) (8, 8) 6 / 34 a) A Single Pareto-Dominant Symmetric Pure Equilibrium Switching the first three inequalities around, i.e. c < a, d < b, d < a, and assuming a > b, we obtain a game in which the (A, A) is the unique, Pareto dominant Nash equilibrium. In such a case, the only rational action to take is the one corresponding to the Nash equilibrium. 7 / 34 a) A Single Pareto-Dominant Symmetric Pure Equilibrium A variant of such games can be obtained by assuming b < d < c, such that b+c 2 > d. For example, the game A B A (2, 2) (6, 14) B (14, 6) (8, 8) In this case, when no communication (co-ordination) is possible, since B dominates A and (B, B) Pareto dominates (A, A), it seems natural that in a one-shot game (B, B) will be played. 8 / 34 a) A Single Pareto-Dominant Symmetric Pure Equilibrium However, when the game is repeated, by co-ordinating their actions, the players can obtain a higher mean reward by switching between (B, A) and (A, B). In this case, the players obtain an average reward of 10 per play ( 14+6 2 ), rather than 8. In such cases, players use a so called ”tit-for-tat” (I scratch your back, you scratch mine) strategy. Players alternate between (A, B) and (B, A) until at least one of them breaks such an agreement, from which point onwards the equilibrium of the ”one-shot” game (B, B) is played. 9 / 34 a) A Single Pareto-Dominant Symmetric Pure Equilibrium It should be noted that in the one-shot game with communication, the players can both obtain an expected reward of 10 by agreeing to toss a coin and play (A, B) if heads appear, otherwise play (B, A). However, there is a serious problem with regards to such an agreement. If the toss suggests that (A, B) is played, then Player 1 has an incentive to unilaterally defect from such an agreement (by playing B he will obtain 8 rather than 6). 10 / 34 a) A Single Pareto-Dominant Symmetric Pure Equilibrium Similarly, if the toss suggests that (B, A) is played, then Player 2 has an incentive to unilaterally defect from such an agreement (by playing B he will obtain 8 rather than 6). Hence, reciprocation rather than communication is key to obtaining the highest possible average payoff in such games. 11 / 34 b) The Prisoner’s Dilemma Assume that i) c > a, d > b, also ii) d < a. The first pair of inequalities state that action B dominates action A. It follows that the strategy A can be eliminated and hence (B, B) is the only Nash equilibrium. However, (A, A) Pareto dominates (B, B). It follows that players can arrive at socially non-optimal solutions by acting in their own interests. The Prisoner’s Dilemma is used to model situations in which there are costs and benefits to cooperating. Each individual would prefer the benefits of cooperation, without incurring the costs of cooperation themselves. 12 / 34 b) The Prisoner’s Dilemma For example, consider the game (Game 3 on the experimental games sheet) A B A (5, 5) (0, 8) B (8, 0) (2, 2) In this case, the unique Nash equilibrium is (B, B), which gives both players a payoff of 2. However, if both players chose A, then they would both get a payoff of 5. 13 / 34 b) The Prisoner’s Dilemma In such a Prisoner’s dilemma, the action A is labelled ”cooperation” and the action B is labelled ”defection” (reversing the inequalities above, the roles of the actions are swapped). It is furthermore assumed that 2a > b + c, i.e. the players obtain more by always cooperating than by alternating between (A, B) and (B, A). In such a case, ”tit-for-tat” strategies can lead to conditional cooperation when interactions (games) are repeated. As before, communication is not useful in a one-shot game to achieve cooperation, since each player has an incentive to defect. 14 / 34 c) Anti-Coordination Games In this case, c > a and b > d. The Hawk-Dove game is an anti-coordination game in which it is additionally assumed that two ”doves” (who play B) obtain a greater payoff than two ”hawks” (who play A), i.e. d > a. There are two pure Nash equilibria (A, B) and (B, A). Hence, the name anti-coordination, since both players would prefer to choose the action that the other player does not choose. There is also a mixed Nash equilibria where both players use the mixed strategy qA + (1 − q)B, where q= b−d . (c − a) + (b − d) 15 / 34 c) Anti-Coordination Games The Hawk-Dove game is used to model aggression. Doves share resources. Each player would prefer the other to be non-aggressive and then by being aggressive, an aggressive player can gain resources. However, the worst situation occurs when both players are aggressive. 16 / 34 c) Anti-Coordination Games For example, the game (Game 4 on the experimental games sheet) A B A (1, 1) (10, 4) B (4, 10) (7, 7) There are three Nash equilibria in this game (A, B), (B, A) and a mixed equilibrium where A and B are chosen with equal probabilities. 17 / 34 c) Anti-Coordination Games In one-shot games without communication, it is unclear how the players should play unless some asymmetry is introduced. In the natural world, such an asymmetry can based on size/hierarchy or on the possession of a territory (owners are aggressive, intruders are not). When players are averse to risk, then it seems reasonable to play B (dove) which ensures a payoff of at least 4, rather than A (hawk) which can lead to a ”fight” (payoff of 1). 18 / 34 c) Anti-Coordination Games If communication is possible, the players could agree to play (A, B) if a coin toss results in heads, otherwise play (B, A). In this case, such an agreement is stable, since either player would lose by unilaterally changing their action (both of these action pairs are Nash equilibria). It is best to avoid randomized strategies, due to the possible costs of two-sided aggression. 19 / 34 c) Anti-Coordination Games In a repeated version of such a game, the players could alternate between (A, B) and (B, A). In addition, both players choosing B (dove) is stable, as long as there is the threat of switching to A (hawk) in retaliation to the other player choosing A. Such strategies are called ”tit-for-tat” strategies, i.e. one player continues to cooperate, as long as the other player cooperates. 20 / 34 d) Coordination Games In this case, a > c, d > b. There are two Nash equilibria (A, A) and (B, B). Hence, the name of such games, since both players prefer to choose the same action as the other. There is also a mixed equilibrium where both players use the mixed strategy qA + (1 − q)B. where q= d −b . (a − c) + (d − b) 21 / 34 d) Coordination Games In such one-shot games, the behaviour that we expect from rational players depends on the payoffs obtained at the two equilibria and the risk associated with both actions. (A, A) payoff dominates (B, B) if and only if a > d. It risk dominates (B, B) if and only if a − c > d − b. (B, B) payoff dominates (A, A) if and only if a < d. It risk dominates (A, A) if and only if a − c < d − b. The mixed equilibrium is always payoff dominated by both pure equilibria. 22 / 34 d) Coordination Games The action A of a player minimax dominates an action B of the same player, if the minimum payoff obtainable from taking action A is greater than the minimum payoff obtainable from playing B. If one pure equilibrium (A1 , A2 ) both risk dominates and payoff dominates the other, where the action A1 minimax dominates the other actions of Player 1 and the action A2 minimax dominates the other actions of Player 2, then we expect players to choose the dominant equilibrium. 23 / 34 d) Coordination Games For example, the game (Game 2 on the experimental games sheet) A B A (10, 10) (4, 6) B (6, 4) (7, 7) There are three Nash equilibria in this game (A, A), (B, B) and a mixed equilibrium where A and B are chosen with probabilities 3/7 and 4/7, respectively. 24 / 34 d) Coordination Games The Nash equilibrium (A, A) payoff dominates the equilibrium (B, B) (a = 10 > d = 7). In addition, the risk associated with A (3/7) is lower than the risk associated with B (4/7) [a − c = 10 − 6 > d − b = 7 − 4]. However, for both players, action B (minimum possible payoff 6) minimax dominates A (minimum possible payoff 4). Hence, it is unclear what action rational players will take. 25 / 34 d) Coordination Games Another example is given by Game 5 on the experimental games sheet) A B A (5, 5) (−5, 4) B (4, −5) (2, 2) There are three Nash equilibria in this game (A, A), (B, B) and a mixed equilibrium where A and B are chosen with probabilities 7/8 and 1/8, respectively. 26 / 34 d) Coordination Games The Nash equilibrium (A, A) payoff dominates the equilibrium (B, B) (a = 5 > d = 2). However, the risk associated with A (7/8) is much higher than the risk associated with B (1/8) [a − c = 5 − 4 < d − b = 2 − (−5)]. Hence, it is unclear what strategy will be played by rational players. The very large risk and the possible ”losses” associated with A (minimum payoff -5) would suggest that many players would choose B. 27 / 34 d) Coordination Games Intuitively, coordination games are easy to solve using communication. The players should simply agree to play the payoff maximising equilibrium (neither has any incentive to change their action). Similarly, in repeated games, it seems natural that the payoff maximising equilibrium should be played, since by eventually coordinating on this equilibrium will give a bettter sum of payoffs than playing the other equilibrium (at least in the long run). 28 / 34 e) A Two Player Game with no Pure Equilibria As shown at the beginning of this chapter, every symmetric 2 × 2 game has at least one pure Nash equilibrium. However, if a game is asymmetric, i.e. the players have specific/different roles, then it is possible that no pure equilibrium exists. For example, consider the matching pennies game in which both players choose action A or B. Player 1 wins if he chooses the same action as Player 2, otherwise Player 2 wins. 29 / 34 e) A Two Player Game with no Pure Equilibria In this case, the payoff matrix can be given by A B A (1, 0) (0, 1) B (0, 1) (1, 0) 30 / 34 e) A Two Player Game with no Pure Equilibria In this case, it is intuitively clear that neither of the players wishes the other to know what action will be taken, hence no pure equilibrium exists. Note that if (A, A) or (B, B) is played, then Player 2 wishes to switch. Otherwise, if (A, B) or (B, A) is played then Player 1 wishes to switch. 31 / 34 e) A Two Player Game with no Pure Equilibria Intuitively, at equilibrium, both players should choose each action with probability 0.5. This can be shown as follows: Assume that Player 1 uses pA + (1 − p)B at equilibrium. In this case, Player 2 is indifferent between actions A and B. Playing A, Player 2 has an expected reward of p × 0 + (1 − p) × 1 = 1 − p = v2 . Playing B, Player 2 has an expected reward of p × 1 + (1 − p) × 0 = p = v2 . 32 / 34 e) A Two Player Game with no Pure Equilibria At equilibrium v2 = p = 1 − p ⇒ p = 0.5, v1 = 0.5. Analogously, assume that Player 2 uses qA + (1 − q)B at equilibrium. In this case, Player 1 is indifferent between actions A and B. Playing A, Player 1 has an expected reward of q × 1 + (1 − q) × 0 = q = v1 . Playing B, Player 1 has an expected reward of q × 0 + (1 − q) × 1 = 1 − q = v1 . Thus, at equilibrium v1 = q = 1 − q ⇒ q = 0.5, v2 = 0.5. 33 / 34 e) A Two Player Game with no Pure Equilibria Hence, at equilibrium both players choose an action at random. The unique value of the game is (0.5, 0.5). It should be noted that a symmetric 3 × 3 game may have no pure Nash equilibria. For example, the Rock-Scissors-Paper game has no pure equilibrium. Choosing an action at random is the only Nash equilibrium in this game. 34 / 34
© Copyright 2026 Paperzz