Game Theory Games of strategy l Sequential games

Game Theory Games of strategy l Sequential games l Simultaneous decisions l Dominated strategies l Nash equilibrium l Prisoners’ dilemma l SEEM 3530 1

Sequential decisions l l l l Previously …. Sequential decisions with uncertainty Decision trees … with “chance” nodes but … “God does not play dice” – Albert Einstein “Subtle is the Lord, but malicious He is not. ” What about your competitors? SEEM 3530 2

A sequential “game” l l l Decisions made in sequence. Your decision depends on decision made previously by others, and others’ decisions follow and depend on yours, etc. Outcome/payoff depends on all decisions made by all. SEEM 3530 3

Lucy Van Pelt vs. Charlie Brown l Lucy Van Pelt holds a football on the ground and invites Charlie Brown to run up and kick it. At the last moment, Lucy pulls the ball away. Charles Brown, kicking air, lands on his back, and this gives Lucy great perverse pleasure. SEEM 3530 4

This time … SEEM 3530 5

Representing Decisions in a Game Tree Pull Ball Away Accept Lucy Charlie Let Charlie kick Reject SEEM 3530 , , , 6

And then … SEEM 3530 7

Games of Strategy Vijay Krishna (Harvard Business School): Any situation where the choices of two or more rational decision makers together leads to gains and losses for them is called a game. A game may simultaneously involve elements of both conflict and co-operation among the decision makers. SEEM 3530 8

Market Competition - HDTV Vizio considers entering a market now monopolised by Samsung can decide to respond by being accommodating or aggressively fight a price war. Profit outcomes for both firms depends on the strategies of both firms. As Vizio, you can analyse this problem using Decision Analysis by estimating probabilities of Samsung’s responses. SEEM 3530 9

Market Entry – Decision Tree for Vizio Accommodate $100, 000 to Vizio p Samsung Enter Vizio Keep out 1 -p -$200, 000 to Vizio Fight price war $0 to Vizio How to estimate the probabilities? What does p depend on? If no information, p=0. 5? SEEM 3530 Then Vizio will not enter market. 10

s nter 0, 10 not e -7, 2 re g g A Do SEEM 3530 5, 8 mo dat com Ac sive Samsung et Ma rk ter En Vizio Probability of Samsung’s response will depend on Samsung’s payoff in the different scenarios e Game Tree Representation 11

Market Entry – Game Tree Model Accommodate $100, 000 to Samsung -$200, 000 to Vizio Enter Vizio Keep out $100, 000 to Vizio Fight price war -$100, 000 to $0 to Vizio Samsung $300, 000 to Samsung SEEM 3530 12

Analysing Game Trees Rule 1: Look Ahead and Reason Back! l For this market alone, Vizio should choose enter because Samsung (rationally) will accommodate. l If Samsung worries that Vizio may enter other markets in the region after this, Samsung may take a tough stand. Vizio should not enter. The “payoff” should include all “benefits”. l SEEM 3530 13

Look Ahead & Reason Back 1. Formulate the game tree of the situation. l 2. Evaluate payoffs at the “leaves” of the tree. l 3. Identify your own and opponent’s strategy at each stage. This assumes: l Your opponent’s strategy can be observable. l Strategy must be irreversible. Think about what will happen at the end. Reason backward through the tree. l Identify the best strategy for each player at each stage, starting at the end. Note the essence of a game of strategy is interdependence. Your decision affects your opponent’s decision and your opponent’s affects yours. SEEMdecision 3530 14

More complex games Black-1 P-Q 4 Black-1 White-1 P-QB 4 White-2 P-K 4 Black-1 Theoretically, can map out all possible chess moves and then select the best sequence of moves SEEM 3530 to win the game! White-2 15

Chess - Human vs. Computers l l l Good chess players can “see” 14 moves ahead! (1968) David Levy: “No computer can beat him in 10 years” Deep Blue l l Chess playing machine built by IBM in the 1990’s 2 to 2. 5 million moves per second. (1996) Deep Blue 1 lost to world chess champion Gary Kasparov. (1997) Deep Blue 2 defeated Kasparov. SEEM 3530 16

Deep Blue vs. Kasparov 1996, game 1. The final position. 8 7 6 a b c d e f g h 8 7 6 5 4 3 2 1 a b c d e SEEM 3530 f g h 5 4 3 2 1 17

Homework? l l Draw the game tree for TIC-TAC-TOE. Sure-win strategies? SEEM 3530 18

The Election of the Chief Executive for Hong Kong l l l The next Chief Executive of Hong Kong SAR Government will be “elected”. Mr. B is Beijing’s favourite candidate. Ms. C (the potential challenger) considers entering the race. Mr. B must determine whether to launch a preemptive advertising campaign against Ms. C (expensive) or not (cost-saving). Ms. C must determine whether to enter the race. SEEM 3530 19

Election Game Tree In B’s, C’s payoff 1, 1 C Advertise Out B In No Ad SEEM 3530 The larger the better 3, 2 2, 4 C Out 20 4, 2

Game Tree In B’s, C’s payoff 1, 1 C Ads Out B In No Ads SEEM 3530 3, 2 2, 4 C Out 21 4, 2

Game Tree In B’s, C’s payoff 1, 1 C Ads Out B In No Ads SEEM 3530 3, 2 2, 4 C Out 22 4, 2

Advantage due to Order of Decisions? l l l First-mover advantage? Mr. B (first) sets the stage for Ms. C (second). Mr. B can look ahead to Ms. C’s optimal response and make the move to his advantage. Can Ms. C improve her situation by acting first? SEEM 3530 23

Game Tree Adv C’s, B’s payoff 1, 1 B In No Adv C Adv Out SEEM 3530 4, 2 2, 3 B No Adv 24 2, 4

Better off being first? l l Is there a first-mover advantage? What about adoption of new technology? l l SEEM 3530 Better off as a technology leader? Better off as a technology follower? 25

Simultaneous Decisions In the chess example, the sequence of decisions alternate between the players. In other situations, the decision may not be sequential but simultaneous. Tic-tac-toe (sequential) Stone-paper-scissors (simultaneous) In simultaneous games, the payoffs to the players are still interdependent on chosen strategies of ALL players. SEEM 3530 26

Time vs. Newsweek l l l Each week, these magazines decide on what story to put on the cover. They do not know the other’s decision until publication. Suppose there are two “hot” stories: l l (A): Anna Chapman, the Russian Spy, (B): British Petroleum Oil Spill damage Newsstand buyers only purchase if story is on cover. 70% interested in (A) and 30% in (B). l SEEM 3530 Purchases evenly split the if both magazines have the same story. 27

Matrix Representation of Game Payoff for Time Newsweek A B A 35 = 70/2 70 B 30 15 = 30/2 Time What should Time do? SEEM 3530 28

Matrix Representation of Game for Newsweek Payoff to Newsweek A B A 35 30 B 70 15 Time SEEM 3530 No matter what Time does, Newsweek is better off putting (A) as cover story. 29

Dominant Strategies Payoffs Newsweek Time, Newsweek A B A 35, 35 70, 30 B 30, 70 15, 15 Time Choosing (A) is a dominant strategy for both Time and Newsweek! SEEM 3530 30

Dominant Strategy l l l A dominant strategy is one that makes a player better off than he would be if he used any other strategy, no matter what strategy his opponent uses. A strategy is dominated if there is another strategy that under no circumstances leads to a lower payoff, and sometimes yields a better payoff. Note: For some games, there may be no dominant strategy for some players. SEEM 3530 31

Properties of a dominant strategy 1: A dominant strategy dominates your other strategies, NOT your opponent! Even with your dominant strategy, your payoff could be smaller than your opponents. 2: A dominant strategy does not requires that the worst possible outcome of the dominant strategy is better than the best outcome of an alternative strategy. SEEM 3530 32

Pricing example Suppose there are just two possible pricing choices: $3 (a profit margin of $2 per copy) and $2 ($1 per copy). Customers will always buy the lower-priced magazine. Profits are split equally between the two. The total readership is 5 million if the price is $3, and rises to 8 million if the price is only $2. Time’s Sales Time’s Price SEEM 3530 Newsweek’s Price $2 $3 $2 4 million 8 million $3 0 million 5 million 33

Analysing Games Rule 2: If you have a dominant strategy, use it ! Rule 3: Eliminate any dominated strategies from consideration, and do so successively! SEEM 3530 34

Eliminating Dominated Strategies - Example l l American ship at A, Iraqi ship at I. Iraqi plans to fire a missile at American ship; American ship plans to fire a defense missile to neutralize the attack (simultaneously). Missiles programmed to A (possibly) turn every 20 seconds. If missile not neutralised in 60 seconds, American ship sinks! SEEM 3530 B C F E D G I H 35

Possible strategies (paths) l For American, l l l Similarly for Iraqi. l SEEM 3530 A 2, A 3 dominated by A 4, A 6, A 7 dominated by A 8, A 1 is dominated by A 8, A 5 is dominated by A 4, Only A 4 and A 8 not dominated. Only I 1 and I 5 are not dominated. 36

Simplified Game American C B Iraqi A I 1 IFCB I 5 IHGD A 4 ABED O H A 8 ADEB H O vs. Iraqi American SEEM 3530 F E D G I H No dominant strategy for either player! 37

Nash Equilibrium A set of strategies constitute a Nash Equilibrium if: no player can benefit by changing her strategy while the other players keep their strategies unchanged. l Each player’s strategy is the “best-response” to the other players’ set of strategies. SEEM 3530 38

Dominant Strategy Equilibrium l l l Higher viewership TVB payoff means more advertising revenues for both TV stations. ATV payoff Each TV station has a dominant strategy. In this case, the equilibrium for this game is obvious. SEEM 3530 TVB ATV Soap. Opera News & Analysis Soap. Opera 55%, 45% 52%, 48% News & Analysis 50%, 50% 45%, 55% 39

Dominant Strategy Equilibrium l If, in a game, each player has a dominant strategy, and each player plays the dominant strategy, then that combination of (dominant) strategies and the corresponding payoffs are said to constitute the dominant strategy equilibrium for that game. SEEM 3530 40

Nash Equilibrium Payoffs Newsweek Time, Newsweek A B A 42, 28 70, 30 B 30, 70 18, 12 Time No dominant strategy for Newsweek. Unique Nash equilibrium. SEEM 3530 41

Example - “Chicken” James Dean Swerve Don’t swerve Mad Swerve C, C C, H Max Don’t swerve H, C D, D H>C>D l. No dominant strategy l. Two Nash equilibria SEEM 3530 42

Choosing among Multiple Equilibria l l Some games have multiple equilibria. “Rule of the road” l l l Brit Drive on left right Hong Kong, Britain, Australia, Japan (left) China, Europe, Mexico, USA (right) The social convention of the locale determines which equilibrium to choose. Drive on left , D, D Drive on right D, D , Yank Sweden switch from left to right in 1967. SEEM 3530 43

In-class exercise (Texas A&M) Each of you owns a production plant and can choose to produce 1 or 2 units of a product. l More total production will lower price and hence profit. What would you do? l SEEM 3530 # of “ 1” Payoff to Payoffs to “ 1” firms “ 2” firms 0 $0. 50 1 $0. 04 $0. 54 2 $0. 08 $0. 58 : : : 29 $1. 16 $1. 66 30 $1. 20 $1. 70 : : : 59 $2. 36 $2. 86 60 $2. 40 $2. 90 : : : 44

Is a Nash equilibrium “good” for the players? Just because a game has an equilibrium does not mean that those strategies are “best” for the players. l Prisoners’ dilemma: l l l Two burglars, Bob and Al, are captured at the scene of a burglary and interrogated separately by the police. Each has to choose whether or not to confess. Outcomes: l l l SEEM 3530 If neither man confesses, then both will serve only one year. If both confesses, both will go to prison for 10 years. However, if one burglar confesses and implicates the other, and the other burglar does not confess, the one who has collaborated with the police will go free, while the other burglar will go to prison for 20 years. 45

1950 – Dresher & Flood (Rand) A. W. Tucker Prisoners’ dilemma Punishment Bob Al confess deny confess 10, 10 0, 20 deny 20, 0 1, 1 l l • SEEM 3530 For each player, the dominant strategy is to confess! Unique Nash equilibrium! Both play the dominant strategy but create mutually disastrous outcome! Both would be better off by denying! 46

Cartels l l l Companies or countries form an alliance to jointly make price and production decisions. World Trade Organisation (WTO) / General Agreement on Tariffs and Trade (GATT) The Organization of Petroleum Exporting Countries (OPEC) is a cartel. l SEEM 3530 the mission of OPEC is to coordinate and unify the policies of its Member Countries and ensure the stabilization of oil markets in order to secure a regular supply of petroleum to consumers, a steady income to producers … 47

OPEC – Maintaining a Cartel l Total output: 4 mb 6 mb 8 mb Price per barrel: $25 $10 Extraction costs: Iran: $2/barrel; Iraq: $4/barrel Dominant strategy: produce at higher level ! SEEM 3530 48

Ensuring Co-operation l l The dominant strategy equilibrium results in each producing 4 million barrels and achieving 56 million in total joint profit. Suppose OPEC countries have agreed to maintain production at 2 mb per day. If members produces 2 million barrels each (as agreed), they will make 88 million in total joint profit. Is it possible to achieve cooperation, when the dominant strategy is to cheat? SEEM 3530 49

Detection of Cheating l l Co-operation is difficult when the reward for cheating is high. How to tell if some member cheated and produced more? The price is US$25 per barrel only if members maintained low production. If price drops below $25, then someone has cheated! What if demand actually decreased? SEEM 3530 50

Identifying cheaters l In a two-player game, an honest party knows who cheated. l l l Still, the cheating party may deny they cheated. When there are many players, even when cheating has been detected, it may be difficult to identify who cheated ! If voluntary cooperation is not possible, how about making use of punishment? l SEEM 3530 In a one-period game, there is no solution to achieve reciprocal co-operation. 51

Punishment – Credible Threat? l l l If the game repeats, cooperation may be enforced. Suppose Iran begins to cheat and produces 4 million barrel per day secretly. l Iran’s profit goes up from 46 to 52 million per day. When Iraq finds out, Iraq also produces 4 million barrels. l Iran’s profit goes down to 46 to 32 million per day. Assume it takes a month for Iraq to know. l Iran’s total profit through cheating: 6 x 30= 180 million Iraq retaliates by increasing production. l Iran’s cheating gain will be wiped out in 13 days (i. e. , 180 million / 14 million) SEEM 3530 52

Competition or Collusion? l l l DVD player vendors: Fortress Broadway wholesale: $1500, retail: $3000 Broadway: lowest price guarantee: “refund double the price difference” l Should Fortress cut its price to $2750? l What will the consumer do? l How will Broadway respond? SEEM 3530 53

“Implicit” Cartel l l If Fortress tries to increase its market share by lowering its price to $2750. Customers will buy from Broadway and claim from a $500 rebate. The “selling price” for Broadway is effectively $2500; lower than Fortress’ price of $2750. In response, Broadway will not give away rebates but lower its price to $2750. Fortress becomes worse off … so why bother? Collusion is enforced by “announcing” the punishment! SEEM 3530 54

Sustaining Co-operation l l l Mechanism must Detect cheating and Deter cheating. Which Punishment? Simplicity and clarity l l Certainty l l Easy for potential cheaters to evaluate consequences. Players have confidence that defection will be punished and co-operation rewarded. Severity l not to “fit the crime” but for deterrence! Risk of mistakes? SEEM 3530 55

Tit-for-tat Exodus 21: 22 -25 l If men who are fighting hit a pregnant woman and she gives birth prematurely but there is no serious injury, the offender must be fined whatever the woman’s husband demands. l But if there is a serious injury, you are to take life for a life, eye for eye, tooth for tooth, hand for hand, burn for burn, wound for wound, bruise for bruise. SEEM 3530 56

Tit-for-tat Strategy Co-operates in the first period, thereafter mimics the rival’s action from previous rounds l l l SEEM 3530 Clarity (simple to implement) Niceness (does not initiates cheating) Provocability (it never lets cheating go unpunished) Forgiveness (does not hold a grudge, willing to restore cooperation) “Chain reaction” of mistakes? 57

Hatfields and Mc. Coys l l l This is one of best-documented stories on inter-family feud (1878 – 1891). Early settlers in the Tug Valley on the Kentucky and West Virginia border. Feud started over the disputed ownership of a pig! Kentucky SEEM 3530 West Virginia 58

Tit-for-tat with misperception Mis-perception leads to perpetual retaliation! Nuclear conflict? Cuban missile crisis (1962). SEEM 3530 59

Tit-for-tat Strategy l When misperceptions are possible, in the long run tit -for-tat will spend half the time cooperating and half of it defecting. l l When the probability of a misperception is small, it will take a lot longer for this phenomenon to occur. When the probability is 50%, whatever you do will not have any affect on your opponent. § l l Opponent will perceive aggression with 0. 5 probability. When the probability is 50%, there is no hope of achieving co-operation. One should always attack! Feud never ends … Should one be more forgiving? SEEM 3530 60

中庸之道 (The Moderate Chinese Way) l l Tit-for-tat is quick to punish opponent who has a long history of cooperation. Other responses: (Matthew 5: 38): “But I tell you, do not resist an evil person. If someone strikes you on the right cheek, turn to him the other also. ” A more forgiving tit-for-tat: l l l SEEM 3530 Begin cooperating Continue cooperating, but keep count of how many times the other side appears to be have defected while you have cooperated. If the this percentage becomes unacceptable, revert to titfor-tat. 61

Summary l l l Games of strategy Sequential games Simultaneous decisions Dominated strategies Nash equilibrium Prisoners’ dilemma SEEM 3530 62