ECON 3210/4210 Decisions, Markets and Incentives Lecture notes 21.09.05 Nils-Henrik von der Fehr STATIC GAMES OF COMPLETE INFORMATION Introduction Multi-person decision problems Strategic interaction Examples oligopolistic interaction bargaining and negotiations principal-agent theory political economy In this part we consider games of the following form: first players simultaneously choose actions (simultaneous moves); then players receive payoffs that depend on the combination of actions chosen; each player’s payoff function, as well as the rules of the game, is common knowledge (complete information) Topics how to describe a game (representation) how to solve a game (equilibrium concepts) An example Consider two firms who each may choose to introduce a customer loyalty programme, such as a frequent flyer programme common in passenger air transport. The programme itself is costly, but if it succeeds in attracting and maintaining a larger market share the increase in sales revenues may more than outweigh the additional costs. Let payoffs to the two firms be given as follows, where the first number gives the payoff of the Firm 1 and the second number gives the payoff of Firm 2: Firm 2 No loyalty programme Loyalty programme No loyalty programme 0,0 -2,1 Loyalty programme 1,-2 -1,-1 Firm 1 The game is commonly referred to as a Prisoners’ Dilemma type game. The original story has two suspects on prison who may either confess or not. If neither confesses, they are convicted for minor offences only. If both confess, both will be sentenced to jail. If one confesses, but the other does not, the confessor is released while the other gets an additional sentence for obstructing justice. Note that only the outcome (-1,-1) is reasonable (i.e., in which both firms choose to introduce a loyalty programme or both prisoners confess). Normal-form representation The normal-form representation of a game specifies: 1) the players in the game; 2) the strategies available to each player; 3) the payoff received by each player for each combination of strategies that could be chosen by players. We make the following definitions: there are n players, with an arbitrary player called player i, i = 1,2,...,n; Si denotes the set of strategies available to player i (strategy space), while si denotes an arbitrary member of this set (so si ∈ Si ); ( s1,..., sn ) is a combination of strategies for all players (a strategy profile); 2 ui ( s1,..., sn ) denotes the payoff to player i for the strategy combination ( s1,..., sn ) . The normal-form representation of an n-player game specifies the players’ strategy spaces S1,..., Sn and their payoff functions u1,..., un . Such a game is denoted G = {S1,..., Sn ; u1,..., un } . An interpretation of the simultaneous-move structure is that each player chooses his or her action without knowledge of the others’ choices. Iterated elimination of dominated strategies Note that in the above example, introducing a consumer loyalty programme is preferable whatever the competitor chooses to do. In other words, the strategy ‘no loyalty programme’ is dominated by the strategy ‘loyalty programme’. More generally, we have the following definition: In the normal-form game G = {S1,..., Sn ; u1,..., un } , the strategy si′ ∈ Si is strictly dominated by the strategy si′′ ∈ Si if, for each ( s1,..., si −1, si +1,..., sn ) , sk ∈ Sk , ui ( s1,..., si −1, si′, si +1,..., sn ) < ui ( s1,..., si −1, si′′, si +1,..., sn ) . In other words, the strategy si′′ yields a higher payoff than si′ whatever the strategy choices of the opponents. Rational players do not play strictly dominated strategies! Note that this leads to a unique equilibrium in the Prisoners’ Dilemma game considered above. More generally, when, for all players, all strategies except one is dominated by one particular strategy we have what is called a dominant-strategy equilibrium. In the Prisoners’ Dilemma game, the strategy combination (loyalty programme, loyalty programme) is a dominant-strategy equilibrium. A weaker solution concept is based on iterated elimination of strictly dominated strategies. Consider the following example: Player 2 Left Middle Right Up 1,0 1,2 0,1 Down 0,3 0,1 2,0 Player 1 3 Note that, for Player 2, the strategy Right is strictly dominated by the strategy Middle. Consequently, Player 1, knowing the preferences of Player 2, should consider it impossible that Player 2 will play Right and hence may eliminate this strategy from the strategy space of Player 2. This leads to the following game: Player 2 Left Middle Up 1,0 1,2 Down 0,3 0,1 Player 1 Now, given that Player 1 can rule out that Player 2 will ever play Right, for him the strategy Down is strictly dominated by the strategy Up. Consequently, Player 2, knowing that Player 1 will reason that Player 2 will never play Right, should consider it impossible that Player 1 will play Down and hence may eliminate this strategy from the strategy space. This leads to: Player 2 Player 1 Up Left Middle 1,0 1,2 Now, for Player 2, the strategy Left is strictly dominated by the strategy Middle. We conclude that, in this case, iterated elimination of strictly dominated strategies leads to a unique solution of the game, namely the strategy combination (Up, Middle). Although the process of iterated elimination of dominated strategies is based on the appealing idea that rational players do not play strictly dominated strategies, it has two drawbacks: it requires ever further assumptions about what players know about each other’s rationality. If the process is to be applied for an arbitrary number of steps, we need to assume that it is common knowledge that players are rational; that is, all players are rational, all players know that all players are rational, all players know that all players know that all players are rational, and so on ; it often produces very imprecise prediction of play of the game; i.e., the process often stops before all but one strategy for each player are eliminated. 4 To illustrate the last point, consider the following example. A new type of consumer product is about to be introduced (relevant examples include music machines and game consoles). There are two competing technologies, controlled by different firms. Both firms would like there to be one standard, as this would increase total sales. However, each firm would like its own technology to become the standard, as this would mean higher sales for itself. The strategic choice involves whether to choose ones own technology or that of its competitor. Payoffs are as follows: Firm 2 Own technology Competitor’s technology Own technology 0,0 2,1 Competitor’s technology 1,2 0,0 Firm 1 In this game, no strategy is strictly dominated by another. Hence, the process of elimination of strictly dominated strategies has no bite and hence provides no solution to how the players will play this game. Nash Equilibrium Motivation: in order to form an equilibrium, a strategy combination should be such that each player consider his or her strategy a best response to what everybody else is expected to play (no regrets), or, play should be strategically stable or self-enforcing. Definition: in the n-player normal-form game G = {S1,..., Sn ; u1,..., un } the ( ) strategies s1* ,..., sn* constitute a Nash equilibrium if, for each player i, ( ) ( ) ui s1* ,..., si*−1, si* , si*+1,..., sn* ≥ ui s1* ,..., si*−1, si , si*+1,..., sn* , si ∈ Si In other words, the strategy si* is a best response to the strategies ( s ,..., s * 1 * i −1 ) , si*+1,..., sn* of the n-1 other players and solves ( ) max si ∈Si ui s1* ,..., si*−1, si , si*+1,..., sn* . 5 It follows that, by definition, for a strategy combination ( s1,..., sn ) that does not constitute a Nash equilibrium, at least one player will have an incentive to deviate to another strategy. Note also that players’ strategies in a Nash equilibrium always survive iterated elimination of strictly dominated strategies, but the converse is not true. Hence, Nash equilibrium is a stronger solution concept. In the Prisoner’s Dilemma game, the dominant-strategy equilibrium is also a Nash equilibrium. We can check this by first demonstrating that for all other strategy combinations at least one player would want to deviate. Second, at equilibrium, playing the equilibrium strategy is a best response. Alternatively, we may apply the result that if iterated elimination of strictly dominated strategies eliminate all but one strategy combination, this combination is the unique Nash equilibrium. Consider next the technology-choice example introduced immediately above. This is an example of a so-called coordination game and is in this particular form often referred to as a Battle of the Sexes type game. The original story has a man and a woman having to choose (without communicating!) whether to go to a boxing match or to the opera. They would both like to go together, but the man prefers the boxing match and the woman the opera. In this game there are two Nash equilibria: in the technology-choice interpretation these are (own technology, competitor’s technology) and (competitor’s technology, own technology) (in the traditional interpretation they are (boxing match, boxing match) and (opera, opera)). More generally, the Nash solution concept may produce one, many or no pure-strategy equilibria (if we allow for mixed-strategy equilibria, in which players randomise between pure strategies, there always exists an equilibrium in these kinds of games). In the case of many Nash equilibria, it may sometimes be possible to invoke additional criteria for selecting a particular equilibrium. Consider the following variant of the technology-choice example: Firm 2 Own technology Competitor’s technology Own technology 0,0 3,2 Competitor’s technology 1,2 0,0 Firm 1 The interpretation may be the following: Firm 1’s technology is superior to Firm 2’s and hence payoffs are higher when this technology becomes the 6 standard (in fact, here (own technology, competitor’s technology) Pareto dominates (competitor’s technology, own technology) from the point of view of the two firms). It may then seem reasonable to choose (own technology, competitor’s technology) as the solution to the game. In other cases, there is no natural ‘focal point’ and hence the Nash equilibrium does not provide a solution to the game. Application: Tragedy of the Commons There are n fishermen. Fisherman i, i = 1,2,...,n, must choose his fishing capacity (size and type of boat, crew etc.), which, for simplicity, we think of as a one-dimensional continuous variable si ∈ Si [0, ∞ ) . The unit cost of capacity is constant and equal to c. The value of fisherman i’s catch, when the total n capacity of the fleet is s = ∑ i =1 si , is v ( s ) si . It follows that the payoff to fisherman i, when the strategy combination is ( s1,..., sn ) , is ⎛ n ⎞ π i ( s1,..., si ) = v ⎜ ∑ si ⎟ si − csi . ⎝ i =1 ⎠ We assume that the function v – value of catch per capacity unit – is decreasing and convex; that is, v ′ < 0 and v ′′ < 0 (value of catch may be decreasing because a larger fishing fleet – and hence competition among fishermen – increases the effort – i.e. hours at sea – required to obtain a certain volume of catch; also, a higher catch may depress prices on fish). We further assume that V ( s ) = v ( s ) s – the value of total catch – is increasing everywhere; that is, V ′ ( s ) > 0 . The figure below provides an example: 7 ( ) At the Nash equilibrium s1* ,...sn* , Fisherman i has chosen capacity so as to maximise payoff, given the choices of all other fishermen. The first-order condition for the optimality of Fisherman i’s choice is ⎛ n ⎞ ⎛ n ⎞ v ⎜ ∑ s *j ⎟ + v ′ ⎜ ∑ s *j ⎟ si* = c . ⎝ j =1 ⎠ ⎝ j =1 ⎠ On the left-hand side is the gross gain from marginally increasing capacity. This consists of the extra catch provided by the marginal capacity unit ( v ) less the reduction in catch on all inframarginal units as a result of an overall increase in capacity ( v ′ ⋅ si ). This marginal gain should equal marginal cost (c). Summing over all i = 1,2,...,n, and dividing by n, gives (*) v (s *) + 1 v ′ (s * ) s* = c . n We may contrast the outcome of the game (the decentralised solution) with the total capacity that maximises the net value of total catch, i.e. v ( s ) s − cs . The first-order condition for this problem is (**) v (s ) + v ′ (s ) s = c . The solution is illustrated by the point s’ in the figure above. 8 Comparing (*) and (**), we find that s * > s ' , that is, the Nash equilibrium results in a total capacity that exceeds the capacity that maximises the value of total catch. The intuition is that each fisherman only considers the reduction in value on unit catch on his own capacity and not that of all other fishermen. In other words, increasing individual capacity involves a negative externality on all others. A formal proof of the above result may proceed as follows. Assume, for contradiction, that s * ≤ s ' . Then c = v (s ') + v ′ (s ') s ' ≤ v (s * ) + v ′ (s * ) s * < v (s * ) + 1 v ′ (s *) s * . n The equality follows from (**); the first inequality follows from the assumptions that s * ≤ s ' and v is decreasing and convex, which implies that V ′ is decreasing also; and the last inequality follows from the assumption that n > 1 and v ′ < 0 . This result, however, contradicts (*). It follows that the assumption s * ≤ s ' must be wrong; that is, s * > s ' . QED. The model illustrates the more general idea that, if individuals respond to private incentives only, public (i.e. freely accessible) resources will be overutilised (and public goods under-provided). 9