ECC2610 Game Theory and Strategic Thinking Lecture 2: Simultaneous-Move Games Jackie So Monash University 1 August 2023 Jackie So ECC2610 Games with Simultaneous Moves A game is said to have simultaneous moves if each player must move without knowing what other players have chosen to do. This is true if players choose their actions at exactly the same time. But it is also true in a situation in which players make choices at different times but do not have any information about others’ moves. It is quite common in the business world where firms constantly choose what price to charge, which technology to adopt, what marketing strategies to implement, without knowing what each other are doing. Simultaneous-move games are also called static game (e.g. Gibbons, 1992) or strategic game (e.g. Osborne, 2004). Jackie So ECC2610 Games with Simultaneous Moves Robinson Crusoe and Friday are considering whether to work together to build a dam on their island. If both parties contribute their efforts, a good dam can be built. If a party contribute his effort while the other shirks, a dam will still be constructed, but it will be of low quality. If both shirk, then no dam will be constructed. A good dam is worth $7 to each party, while a low quality dam is worth $2. The cost of effort is $6 for each party. Our job is to study whether Robinson and Friday will contribute their efforts to build a dam. Jackie So ECC2610 Games with Simultaneous Moves The three elements of the game are: A set of players or agents (i.e. Robinson and Friday) A set of strategies (i.e. Contribute and Shirk) A set of payoff functions (i.e. Value of Dam minus Effort) The key is that the payoff of a player is not only a function of the player’s action but also a function of the actions of other players. Here is the payoffs of our example in table form Robinson Contribute Shirk Jackie So Friday Contribute Shirk 1, 1 -4, 2 2, -4 0, 0 ECC2610 Dominant Strategy Robinson Contribute Shirk Friday Contribute Shirk 1, 1 -4, 2 2, -4 0, 0 A strategy is said to be a dominant strategy for a player if it earns a higher payoff than any other strategy, no matter how that player’s opponents may play. Shirk is a dominant strategy for both Robinson and Friday. For example, regardless what Friday plays (Contribute or Shirk), it is better for Robinson to play Shirk. It is because 2 > 1 when Friday plays Contribute and 0 > −4 when Friday plays Shirk. The same happens to Friday. Thus it is reasonable to expect that both Robinson and Friday will shirk. Jackie So ECC2610 Successive Elimination of Dominated Strategies There is strong reason for players to play the dominant strategies. However, not all games have dominant strategies. For example, if the value of good quality dam is increased to $10, then we have Robinson Contribute Shirk Friday Contribute Shirk 4, 4 -4, 2 2, -4 0, 0 Shirking is no longer beneficial when the other player contributes. Robinson should choose to contribute if Friday contributes and to choose shirk if Friday shirks. Similarly, Friday should choose to contribute if Robinson contributes and to choose shirk if Robinson shirks. Jackie So ECC2610 Successive Elimination of Dominated Strategies A strategy is a dominated strategy for a player if it earns a lower payoff than some other strategies, no matter how that player’s opponents may play. Consider the following game without a dominant strategy: McDonald High Medium Low High 60, 60 70, 36 35, 36 KFC Medium 36, 70 50, 50 35, 30 Low 36, 35 30, 35 25, 25 Low price is a strictly dominated strategy for McDonald as well as for KFC because it is dominated by both high and medium price. For example, if KFC plays High, the payoff from Low (i.e. 35) is less than the payoffs from High (i.e. 60) and from Medium (i.e. 70). The same happens for Mcdonald. Jackie So ECC2610 Successive Elimination of Dominated Strategies If we eliminate Low from the game, McDonald High Medium High 60, 60 70, 36 KFC Medium 36, 70 50, 50 Then High will be dominated by Medium for both players. After eliminating High from the game, the only possible outcome left will be for both players choosing a medium price – (Medium, Medium). If successive elimination of dominated strategies does yield a unique equilibrium, the game is said to be dominance solvable. Jackie So ECC2610 Successive Elimination of Dominated Strategies Consider one more example: Row Row U D U M D L 4, 11 3, 4 3, 10 Column C R 3, 6 5, 12 2, 8 4, 6 4, 6 3, 8 Column L R 4, 11 5, 12 3, 10 3, 8 U U D Column L R 4, 11 5, 12 Jackie So ECC2610 L 4, 11 3, 10 Column C R 3, 6 5, 12 4, 6 3, 8 U Column R 5, 12 The Maximin Method Consider the following football game which is not dominance solvable Offence Run Short Pass Medium Pass Long Pass Run 2 6 6 10 Defence Pass 5 5.6 4.5 3 Blitz 13 10.5 1 -2 Offence wants the outcome to be a cell with as high a number (yard advanced) as possible, and Defence wants the outcome to be a cell with as low a number as possible. We call it a zero-sum game because one’s gain is equivalent to another’s loss (the payoffs of the two players in each cell add up to zero). Jackie So ECC2610 The Maximin Method The minimax method of solution relies on the view that players in such games are pessimistic about their chances of achieving good outcomes. You expect that your opponent is going to do the best she can. In a zero-sum game, this implies that she will effectively pick a strategy that makes you as bad off as she can make you. Similarly, your opponent figures that you will pick a strategy from among your choices that minimizes her payoff from whatever strategy she chooses. Maximin means maximizing the minimum gain. It is called minimax when minimizing the maximum loss. Jackie So ECC2610 The Maximin Method Offence Run Short Pass Medium Pass Long Pass Run 2 6 6 10 max = 10 Defence Pass 5 5.6 4.5 3 max = 5.6 Blitz 13 10.5 1 -2 max = 13 min min min min = = = = Offence figures that for each of her rows, Defence will choose the column with the lowest number in that row. Therefore, Offence should choose the row that gives her the highest among these lowest numbers, or the maximum among the minima – the maximin for short. Jackie So ECC2610 2 5.6 1 -2 The Maximin Method Offence Run Short Pass Medium Pass Long Pass Run 2 6 6 10 max = 10 Defence Pass 5 5.6 4.5 3 max = 5.6 Blitz 13 10.5 1 -2 max = 13 min min min min = = = = Defence reckons that for each of her columns, Row will choose the row with the largest numbers in that column. Defence should choose the column with the smallest number among these largest ones, or the minimum among the maxima – the minimax. Jackie So ECC2610 2 5.6 1 -2 The Maximin Method In the football game, the minimax method yields the outcome (Short Pass, Pass). It is worthwhile to remind you again that the minimax method is only applicable to zero-sum games. But not all zero-sum games can be solved by the maximin method, e.g. Paper-Scissors-Rock Row Player Paper Scissors Rock Paper 0, 0 1, -1 -1, 1 min = -1 Jackie So Column Player Scissors Rock -1, 1 1, -1 0, 0 -1, 1 1, -1 0, 0 min = -1 min = -1 ECC2610 min = -1 min = -1 min = -1 Nash Equilibrium Both the successive elimination of dominated strategies method and the maximin method are applicable to some games only. Thus we need a concept that is applicable to a very wide range of games – Nash equilibrium. Nash equilibrium is a more general concept because solutions derived from successive elimination of dominated strategies and maximin method are also Nash equilibria. We start with two widely studied non zero-sum games that are not dominance solvable – the battle of sexes and the chicken game. Jackie So ECC2610 The Battle of Sexes A husband and wife are supposed to choose between going to a boxing match and going to a ballet. The husband prefers the boxing match and the wife prefers ballet. So both players want to go out, but prefer different events. The husband might get a payoff of 2 and the wife a payoff of 1 from going to a boxing match, and the other way around from going to a ballet. Wife Ballet Boxing Jackie So Husband Ballet Boxing 2, 1 0, 0 0, 0 1, 2 ECC2610 The Chicken Game James and Dean take their cars to opposite ends of a street and start to drive toward each other. The one who swerves to prevent a collision is the ”chicken”, and the one who keeps going straight is the winner. If both maintains a straight course, there is a collision in which both cars are damaged and both players injured. James Swerve (Chicken) Straight (Tough) Dean Swerve (Chicken) Straight (Tough) 0, 0 -1, 1 1, -1 -2, -2 Jackie So ECC2610 John F. Nash Jr. Completed his PhD at the age of 22. His 28 pages thesis ”Non-Cooperative Games” introduces the notion of Nash equilibrium which vastly expanded the scope of game theory. Shared the 1994 Nobel Prize with John Harsanyi and Reinhard Seltan. His life became the basis for the film A Beautiful Mind, in which Nash was portrayed by Russell Crowe. Died in 2015 at the age of 86. Jackie So ECC2610 A quick review of sets and functions Definition A set is a collection of irreducible objects, which we call elements. E.g. {Up, Down} is the set of actions that a row player can choose. Let SR = {Up, Down} be a set as defined above. We say Up is in SR (or Up ∈ SR ) if and only if Up is an element of SR We say Left is not in SR or Left ∈ / SR if Left is not an element of SR . The order of elements in a set does not matter, thus SR = {Up, Down} = {Down, Up} Jackie So ECC2610 A quick review of sets and functions Definition The Cartesian product of two nonempty sets X and Y , denoted as X × Y , is the set of all ordered pairs (x, y ) where x comes from X and y comes from Y . An ordered pair is an ordered list (x, y ) consisting of two objects x and y . Ordered pairs are not sets but elements (i.e. (x, y ) 6= {x, y }). Since the order matters, (x, y ) 6= (y , x) unless x = y . Recall the set SR = {Up, Down}, let SC = {Left, Right}, SR × SC = {(Up, Left), (Up, Right), (Down, Left), (Down, Right)} Jackie So ECC2610 A quick review of sets and functions Definition A subset R of X × Y is called a (binary) relation from X to Y . If (x, y ) is in R, then we think of R as associating the object x with y . Definition A function f that maps X onto Y , denoted as f : X → Y , is a relation f ⊂ X × Y such that for every x ∈ X , there exists exactly one y ∈ Y such that (x, y ) ∈ f or xfy . The function UR : SR × SC → R maps the joint actions of the Row and Column players into a real number. For example, we write UR (Up, Left) = 1 or ((Up, Left), 1) ∈ UR . The function BR : SC → SR maps the actions of the Column player onto an action of the Row player. For example, we write BR (Left) = Up or (Left, Up) ∈ BR . Jackie So ECC2610 A quick review of sets and functions Definition A set Y is a subset of another set X if and only if for every element of Y is also an element of X . Functions can be considered as sets. Let BR = {(Left, Up), (Right, Up)} be the function BR : SC → SR , then we have BR ⊆ SC × SR . Note that (Left, Down) and (Right, Down) are not in BR since they are not related. Similarly, we can define BC = {(Up, Left), (Down, Left)} ⊆ SR × SC be the function BR : SR → SC . Rewrite BC to BC0 = {(Left, Up), (Left, Down)} ⊆ SC × SR The union of sets BR and BC0 , denoted by BR ∪ BC , is {(Left, Up), (Right, Up), (Left, Down)}. The intersect of BR and BC0 , denoted by BR ∩ BC , is {(Left, Up)}. Jackie So ECC2610 Nash Equilibrium and Best Response Functions Recall the game between Robinson and Friday Robinson Contribute Shirk Friday Contribute Shirk 1, 1 -4, 2 2, -4 0, 0 Let A = {Robinson, Friday } be the set of players. In this game, the set of actions available to Robinson is S1 = {Contribute, Shirk}. The set of actions available to Friday is S2 = {Contribute, Shirk}. s = (s1 , s2 ) is called a strategy profile. In this case, the set of all strategy profiles is the Cartesian product of S1 and S2 : S1 × S2 = {(Contribute, Contribute), (Contribute, Shirk), (Shirk, Contribute), (Shirk, Shirk)} Jackie So ECC2610 Nash Equilibrium and Best Response Functions The player’s payoff is determined by his action and the actions chosen by other players. That is, a player’s payoff is a function of the strategy profile chosen by the players. Let U1 and U2 be the payoff functions of Robinson and Friday, we have U1 : S1 × S2 → R and U2 : S1 × S2 → R Each function is mapping a strategy profile onto the real line. For example, U1 (Contribute, Shirk) = 2 − 6 = −4 and U2 (Contribute, Shirk) = 2 − 0 = 2. Jackie So ECC2610 Nash Equilibrium and Best Response Functions For any s2 , s1∗ is called a best response to s2 if and only if U1 (s1∗ , s2 ) ≥ U1 (s10 , s2 ) ∀s10 ∈ S1 Robinson Contribute Shirk Friday Contribute Shirk 1, 1 -4, 2 2, -4 0, 0 In the above example, Shirk is Robinson’s best response to Contribute since U1 (Shirk, Contribute) = 2 > U1 (Contribute, Contribute) = 1 Shirk is also Robinson’s best response to Shirk since U1 (Shirk, Shirk) = 0 > U1 (Contribute, Shirk) = −4 Jackie So ECC2610 Nash Equilibrium and Best Response Functions The best response function gives the best response for each possible strategy of other players. Let B1 and B2 be the best response functions for Robinson and Friday, then we have B1 : S2 → S1 and B2 : S1 → S2 . Writing B1 as a set, we have B1 = {(Contribute, Shirk), (Shirk, Shirk)} ⊂ S2 × S1 A dominant strategy happens when the best response function is a constant. Since B1 (Contribute) = B1 (Shirk) = Shirk Shirk is a dominant strategy for Robinson. A dominated strategy is a strategy which is not in the image of the best response function. That is a dominated strategy does not appear in the set of best responses. Jackie So ECC2610 Nash Equilibrium and Best Response Functions In a two player game, a strategy profile s ∗ = (s1∗ , s2∗ ) is a pure strategy Nash equilibrium if and only if U1 (s1∗ , s2∗ ) ≥ U1 (s10 , s2∗ ) ∀s10 ∈ S1 and U2 (s1∗ , s2∗ ) ≥ U2 (s1∗ , s20 ) ∀s20 ∈ S2 In general, in an n-player game, Nash equilibrium is a strategy profile such that each player’s strategy is a best response to the strategies of all other players. That is the intersection of all players’ best response functions. A Nash equilibrium is a strategy selection such that no player can gain by deviating, given the strategy of his opponents. When both Robinson and Friday are playing their dominant strategy (i.e. Shirk), the game is at its Nash equilbiurm. Jackie So ECC2610 The Chicken Game Revisited James Swerve (Chicken) Straight (Tough) Dean Swerve (Chicken) Straight (Tough) 0, 0 -1, 1 1, -1 -2, -2 Let A = {J, D} be the set of agents. Let S = {C , T } be the strategies of the two agents where C refers to Swerve and T refers to Straight. The strategy space is defined as S = S 2 which is the two-fold Cartesian product of S (i.e. S = {(C , C ), (C , T ), (T , C ), (T , T )} where the first entry of an element refers to the strategy plays by James and the second entry refers to the strategy plays by Dean). Jackie So ECC2610 The Chicken Game Revisited James Swerve (Chicken) Straight (Tough) Dean Swerve (Chicken) Straight (Tough) 0, 0 -1, 1 1, -1 -2, -2 The payoff for each agent is a function Ui : S → R where i ∈ A = {J, D}. Observe that UJ (C , T ) ≥ UJ (T , T ) and UJ (T , C ) ≥ UJ (C , C ), and UD (T , C ) ≥ UJ (T , T ) and UJ (C , T ) ≥ UJ (C , C ) A best response function Bi for each agent is then a subset of S (i.e. BJ = {(C , T ), (T , C )} and BD = {(C , T ), (T , C )}). The Nash equilibrium is the intersect of all agents’ best response functions (i.e. BJ ∩ BD = {(C , T ), (T , C )}). Jackie So ECC2610 Prisoner’s Dilemma Consider the following game Prisoner 1 Confess Deny Prisoner 2 Confess Deny -10, -10 -1, -25 -25, -1 -3, -3 The best response functions for them are both {(Confess, Confess), (Confess, Deny )}. The Nash equilibrium (which is also a dominant strategy for both players) involves both of the players to choose Confess and each of them will end up in prison for 10 years which yields a utility of -10. However, if both of the players choose to Deny, then both of them receive a higher utility (i.e. -3). (Deny , Deny ) cannot be sustainable because the prisoner can cheat and earn a higher playoff if the opponent plays Deny. Jackie So ECC2610 Prisoner’s Dilemma In public goods games, each N players can invest resources ci from their endowment ei in a public good that is shared by everyone P and has a total per-unit value of m. Player i earns ei − ci + m( k ck ). Assuming 1 > m > 1/N, the payoff-maximizing outcome is to contribute nothing (ci = 0). If everyone contributed, however, the players would collectively earn the most. Suppose there are 2 players with ei = 20 and m = 0.6, then we can easily see that the public goods game is just a prisoner’s dilemma: Player 1 c1 = 20 c2 = 0 Jackie So Player 2 c1 = 20 c2 = 0 24, 24 12, 32 32, 12 20, 20 ECC2610 Prisoner’s Dilemma Prisoner’s dilemma game is non-cooperative – players make their decisions and implement their choices individually. If the two players could discuss, choose, and play their strategies jointly – there would be no difficulty about their achieving the outcome that both would prefer. The essence of the question of whether, when, and how a prisoners’ dilemma can be resolved is the difficulty of achieving a cooperative outcome through non-cooperative actions. In experiments, players cooperate in prisoners’ dilemma games about half the time and and contribute about half their endowments in public good game (Behavioral Game Theory, Camerer (2003)). Jackie So ECC2610 A quick review of differentiation and optimization Rules of Differentiation If f (x) = c, then f 0 (x) = 0. If f (x) = cx, then f 0 (x) = c. If f (x) = x n , then f 0 (x) = nx n−1 . If f (x) = g (x) + h(x) then f 0 (x) = g 0 (x) + h0 (x). If f (x) = g (x) − h(x) then f 0 (x) = g 0 (x) − h0 (x). To solve the problem max f (x) x we use the first order condition f 0 (x) = 0 Jackie So ECC2610 Cournot Model Two firms produce the same good. Suppose both of them simultaneously choose the outputs to maximize their profits. Consider the following demand and cost functions: P = 40 − Q = 40 − (q1 + q2 ) c1 = 4q1 and c2 = 4q2 The strategy set for both players is R+ with payoff functions equal to their profit functions, π1 (q1 , q2 ) = (40 − q1 − q2 )q1 − 4q1 π2 (q2 , q1 ) = (40 − q1 − q2 )q2 − 4q2 Jackie So ECC2610 Cournot Model Both firms choose their outputs to maximize profit given another firm’s output, Firm 1’s problem: max(40q1 − q12 − q2 q1 − 4q1 ) q1 Firm 2’s problem max(40q2 − q22 − q1 q2 − 4q2 ) q2 Thus the best response functions are simply the first order conditions for the above two maximization problems: ∂π1 1 = 40 − 2q1 − q2 − 4 = 0 ⇒ q1∗ = B1 (q2 ) = 18 − q2 ∂q1 2 ∂π2 1 = 40 − q1 − 2q2 − 4 = 0 ⇒ q2∗ = B2 (q1 ) = 18 − q1 ∂q2 2 Jackie So ECC2610 Cournot Model Recall that Nash equilibrium is a strategy profile such that each player’s strategy is a best response to the strategies of another player. Let (q1∗ , q2∗ ) be such strategy profile, then it must satisfy both B1 (q2 ) and B2 (q1 ). That can be found by solving the intersection of the two best response functions, 1 1 q1 = 18 − q2 18 − q1 ⇒ q1∗ = 12 2 2 and q2∗ = 12. The profits for the two firms are 40(12) − 122 − (12)(12) − 4(12) = 144 Jackie So ECC2610 Cournot Model q2 36 A q = B (q ) 1 2 A 1 A A A A 18 H Nash Equilibrium HHA A H 12 H A H A HH HH q = B (q ) A 2 1 HH2 A A H q1 12 18 Jackie So 36 ECC2610 Cournot Model Joint profit maximization max Π = (40 − Q)Q − 4Q Q First-order condition: ∂π = 0 ⇒ 40 − 2Q = 4 ⇒ Q = 18 ∂Q Consider the collusive agreement q1 = q2 = 9 If q2 = 9, firm 2’s best response will be 1 q2 = 18 − (9) = 13.5 > 9 2 The collusive agreement is not self-enforcing. Both firms will have an incentive to deviate. Jackie So ECC2610 Cournot Model Social optimal occurs at the output level where P = MC 40 − Q = 4 ⇒ Q = 36, P = 4 To summarise Joint Profit Maximization Cournot Equilibrium Social Optimum Output (9,9) Price 22 Profits (162,162) Consumer Surplus 162 (12,12) 16 (144,144) 288 (18,18) 4 (0,0) 648 CS = (40 − P) ∗ Q/2 Jackie So ECC2610 Cournot Model q2 36 q1 = B1 (q2 ) A So A A A al im pt lO cia A A 18 H HHA HA H A H A HH q2 = B2 (q1 ) HH A HH A A H q1 y ol op on M 18 Jackie So 36 ECC2610 Cournot Model The Nash equilibrium of the Cournot game (sometimes referred to as the Cournot Nash equilibrium) is worse than the collusive outcome from the perspective of the firms. It resembles a prisoner’s dilemma from the players’ perspective if the game allows only for the collusive and the Cournot output levels. Firm 1 Q=9 Q = 12 Firm 2 Q=9 Q = 12 162, 162 135, 180 180, 135 144, 144 However, the equilibrium is preferred to the collusive outcome from the perspectives of the consumers. Jackie So ECC2610 References Dixit, Chapter 4 - 5 Jackie So ECC2610