Set 2: Two-person, Zero-sum game with equilibrium points Erwin Widodo Introductory problems • Game 1 Your opponent You I II III A 5 -2 1 B 6 4 2 C 0 7 -1 Introductory problems • Game 2 Your opponent You I II III A -2 1 1 B -3 0 2 C -4 -6 4 Introductory problems • Game 3 Your opponent I II III IV A -3 17 -5 21 B 7 9 5 7 C 3 -7 1 13 D 1 -19 3 11 You Introductory problems • Game 4 Your opponent You I II III A 2 -5 -2 B 3 -1 -1 C -3 4 -4 Your opponent Rules You I II III A 5 -2 1 B 6 4 2 C 0 7 -1 • Pick a row (A, B, or C) and your opponent picks a column (I, II, or III) at the same time neither knows when choosing what the other has picked • The number where the column and row intersect is the amount your opponent pays you. • In Ex 1: If you pick A and your opponent picks III, you will get 1 • Assume that your opponent knows the rules and as intelligent as you • You must consider what your opponent is thinking Questions • • • • What would you do? Why? What should the outcome of this game be? If your choice depends on your opponent’s choice, how do you play when you don’t know what he/she will do? What about this payoff? • Game 5: missing payoff Your opponent You I II III A ? ? 3 B ? ? 4 C 7 6 5 Problem in Feb 1943 • Gen. George Churchill Kenney, Air Force Commander in Southwest Pacific had a problem. • Japanese were about to reinforce their army in New Guinea and had two alternative routes: – Sail north of New Britain with rainy weather – Sail south of New Britain with fair weather • In any case the journey takes 3 days Problem in Feb 1943 • Gen Kenney had do decide where to concentrate his spy aircraft. • The Japanese wanted their ships to have the least exposure to enemy. • Of course Gen. Kenney wanted the reverse. • The following matrix represents the expected number of days of bombing exposure Its payoff matrix Japanese choice Allies choice North South North 2 days 2 days South 1 day 3 days The difference • A more difficult game than previous one. • The critical different is here the players lack information. • Both players must decide simultaneously, so neither knows the other strategy when choosing his own. • However, the analysis is simple… The analysis • The Allies thought it would be best for them to take the same route as the Japanese. • But when they made decision, they did not know what the route would be. • Nonetheless the problem would be solved when they took the Japanese standpoint. • For the Japanese, the northern route minimized their exposure whatever the Allies did. • So after working this out, its is clear, the Allies decision was: Go North! GT analysis • The last example is an two-person, zero-sum game with equilibrium points • The term zero-sum (equivalently, constant sum) means the players have diametrically opposed (=sangat bertentangan) interests. • The term comes from parlor games like poker where there is fixed money around the table. • If you want to win some money others have to lose an equivalent amount. • Please contrast with a trading between two nations (both may simultaneously gain) • An equilibrium point is a stable outcome of a game associated with a pair of strategy. • It is considered stable because a player unilaterally (affecting only 1 side) picking a new strategy is hurt by the change. A political example • This year is an election year and 2 major political parties are busy in writing their platforms (=janji2 politik) • There is a dispute in district X and Y concerning certain water rights. • Each party must decide whether it will favor X or favor Y or evade the issue. • The parties will announce their decisions simultaneously. A political example • Citizens outside the two states are indifferent to the issue. • In X and Y, the voting behavior of the electorate (=pemilih) can be predicted from the past experience. • The regulars will support their party in any case. • Others will vote for the party supporting their state, or, if both parties take the same position on the issue, will simply abstain. A political example • The leaders of both parties calculate what will happen in each circumstance and come up with the following payoff matrix. • The entries are percentage of votes party A will get if each party follows the indicated strategy. • Ex: if A favors X and B dodges (=mengelak) the issue, A will get 40% of the vote. Its payoff matrix B’s platform A’s platform Favor X Favor Y Dodge issue Favor X 45% 50% 40% Favor Y 60% 55% 50% Dodge issue 45% 55% 40% Its analysis • This is the simplest example of this type of game. • Though both parties “have a hand” in determining how the electorate will vote, there is no point in one party trying to anticipate what the other will do. • Whatever A does, B does the best to dodge the issue; Whatever B does, A does the best to support Y. Its analysis • The predictable outcome is an even split. • If, for some reason, one of the parties deviated from the indicated strategy, this should have no effect on the other party’s actions. • A slightly more complicated situation arises if the percentages are changed a little as follows. Its modified matrix B’s platform A’s platform Favor X Favor Y Dodge issue Favor X 45% 10% 40% Favor Y 60% 55% 50% Dodge issue 45% 10% 40% Its following analysis • B’s decision is now a bit harder. • If B thinks A will favor Y, B should dodge the issue; otherwise, B should favor Y. • But the answer to the problem is in fact not far off. • A’s decision is clear-cut and easy for B to read: favor Y. • Unless A is foolish, B should realize that the chance of getting 90% of the vote is very slim indeed, and that it would do best: to dodge the issue! Its re-modified matrix B’s platform A’s platform Favor X Favor Y Dodge issue Favor X 35% 10% 60% Favor Y 45% 55% 50% Dodge issue 40% 10% 65% Its re-modified analysis • Neither player has an obviously superior strategy both players must think a little. • Each player’s decision hangs on what he expects the other will do. – If B dodges the issue, A should too. – If not, A should favor Y. – On the other hand, if A favor Y, B should favor X. – Otherwise, B should favor Y. Its GT analysis • A is favoring Y and B is favoring X are important enough to be given a name: Equilibrium Strategies. • The outcome resulting from the use of these strategies – the 45% vote for A – is called an Equilibrium point. • Two strategies are said to be equilibrium (they come in pairs, one for each player) if neither player gains by changing strategy unilaterally. • The outcome (sometimes called payoff) corresponding to this pair of strategies is defined as equilibrium point. Its GT analysis • As the name suggests, equilibrium points are very stable (once a player settled, there is no reason to leave it) • If A knew in advance that B would favor X, A would still favor Y • Similarly, B would not change strategy if he knew A would favor Y • There may be more than 1 equilibrium point, but if there is, they will all have the same payoff. • In a two-person, non-zero-sum game, equilibrium points need not have the same payoff (check later in Prisoner’s dilemma) Its GT analysis • Assume that B knows A’s strategy in advance. • Since B would choose the minimum payoff of any row A choose, A should choose a strategy that yield the maximum of these minima this value is called the maximin. • It is the very least that A can be sure of getting. • If A plays “favor X”, “favor Y”, and “dodge issue”, these minima are 10, 45, and 10 respectively. • Thus, the maximin is 45. Its GT analysis • Now imagine the rules are changed so that A knows B’s strategy in advance. • A would be expected to choose the maximum of any column, so B should choose the column that minimizes these maxima this value is called as minimax. • This is the very optimistic that B can avoid. • If B plays “favor X”, “favor Y”, and “dodge issue”, these maxima are 45, 55, and 65 respectively. • Thus, the minimax is 45. Its GT graphical analysis • If the minimax equals the maximin, the payoff is an equilibrium point the corresponding strategies are an equilibrium strategy pair. Equilibrium point: the match of the smallest in row and theFavor X largest in column A’s platform B’s platform Favor X Favor Y Dodge issue 35% 10% 60% Favor Y 45% 55% 50% Dodge issue 40% 10% 65% 45 is the largest value in the column 45 is the smallest value in the row Solution • When an equilibrium point exists in a twoperson, zero-sum game, it is called a solution. • Rational players should adopt the equilibrium strategies and the outcome should be the payoff associated with the equilibrium point – the value of the game. • In previous game: – The equilibrium strategies were “favor Y” for A and “favor X” for B – The value of the game was 45 The reason why equilibrium points = solutions? • By playing his equilibrium strategy, a player will get at least the value of the game. In previous example, A gets at least 45 whatever B does, if A plays “Favor Y” • By playing his equilibrium strategy, an opponent can stop a player from getting any more than the value of the game. By playing “Favor X”, B can limit A’s payoff to 45 whatever A does • Since the game is zero-sum, a player’s opponent is motivated to minimize the player’s payoff. When A gets 45, B gets 55; if A gets any more, it must be because B obtained that much less Note on equilibrium point • In games with equilibrium points, payoffs that are not associated with either equilibrium strategy have no bearing on the outcome. • If the 2 payoff of 10, payoff of 60 and 65 were changed in any way, the player should pick the old same equilibrium strategies and outcome. B’s platform A’s platform Favor X Favor Y Dodge issue Favor X 35% 10% 60% Favor Y 45% 55% 50% Dodge issue 40% 10% 65% Domination • It is often possible to simplify a game by eliminating dominated strategies. • Strategy A dominates strategy B if a player’s payoff with strategy A is: – always at least as much as that of strategy B (whatever other players do) and, – at least some of the time actually better than strategy B Illustrating payoff matrix Your opponent You I II III A 7 9 8 B 9 10 12 C 8 8 8 Its analysis • For you, strategy B dominates both strategies A and C because it always yields a higher payoff. • Your opponent’s strategy I dominates strategies II and III (recall your opponent wants the payoff values small) • Although your opponent doesn’t always do better with I than with II and III, he always does at least and sometimes does better. Possible assumptions in zero-sum game • You will never pick a dominated strategy – why pick a dominated strategy when you can do at least as well using the strategy that dominates it? • Your opponent will never pick a dominated strategy – and this for the same reason that you won’t Consider the following matrix Your opponent You I II III A 19 0 1 B 11 9 3 C 23 7 -3 Its analysis • In this game none of your strategies are dominated initially. • However, since III dominates I for your opponent, you can eliminate I from your consideration. • With I eliminated, B dominates A and C. • With A and C eliminated, III dominates II • The only undominated strategies, B and III, make up an equilibrium strategy pair and the value of the game is 3. Let’s go back to Feb 1943 Japanese choice Allies choice North South North 2 days 2 days South 1 day 3 days • The northern route dominated the southern one for Japanese. • After eliminating the Japanese southern route, we eliminated the Allies southern route for the same reason. • The equilibrium strategies were North for both armies; the equilibrium point is 2 days Let’s go back to political example B’s platform A’s platform • • • • • Favor X Favor Y Dodge isssue Favor X 35% 10% 60% Favor Y 45% 55% 50% Dodge issue 40% 10% 65% “Favor X” dominated “Dodge issue” for B “Favor Y” dominated everything for A “Favor X” dominated “Favor Y” for B The equilibrium strategies were “Favor Y” for A and “Favor X” for B The equilibrium point was 45 No equilibrium point Opponent Heads Tails Heads -1 +1 Tails +1 -1 You • Since no strategy is dominated and there is no equilibrium • It is hard to see how you can play such a game rationally. • However, there is a theory that enables you to play such games intelligently. (next chapter) Solutions to our previous games Game 1: ES are B and III with EP is 2 Game 2: ES are A and I with EP is -2 Game 3: ES are B and III with EP is 5 Game 4: ES are B and III with EP is -1 In each of games 1 through 4, both players will not lose even though they announce their strategies in advance to the opponent • Game 5: whatever the values of the missing payoffs, you can be sure of getting 5 by playing C; your opponent can be sure of losing no more than 5 by playing III. Since neither one of you can enforce the payoff of 5, this is a plausible outcome. • • • • •