1 / 53

Dynamic Games

Dynamic Games. Overview. In this unit we study: Combinations of sequential and simultaneous games Solutions to these types of games Repeated games How to use dynamics to build self-sustaining agreements. Sequential and Simultaneous Games.

reeves
Download Presentation

Dynamic Games

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Dynamic Games

  2. Overview • In this unit we study: • Combinations of sequential and simultaneous games • Solutions to these types of games • Repeated games • How to use dynamics to build self-sustaining agreements

  3. Sequential and Simultaneous Games • There are many situations where the strategic situation has both simultaneous and sequential elements • Examples: • Decision by a firm to enter a market followed by competition in pricing and advertising • Decisions by candidates to run for office followed by voting • Attempts at legal settlements followed by trial in the event no settlement is reached

  4. How to Analyze these Games? • In sequential games, we saw that it paid to look forward and reason back • Find the best decision a player can make on reaching a point in the game • “Prune” the game tree to eliminate worse (dominated) decisions • In simultaneous games, we looked for a best response to a best response (Nash equilibrium). • Project the strategy of a rival • Choose a best response to that strategy • Check if the rival would want to change his “projected” strategy in view of your move.

  5. New Elements to Dynamic Games • History matters • Strategy is now based not only on projections of the future and the present but also on the past. • The process by which you arrived at a point in the game might matter • Alternative futures matter • Histories of the game that no one contemplated as arising can play a key role in influencing outcomes

  6. Entry • Consider the following situation: • An incumbent firm is presently operating in some profitable market • A potential rival firm is considering entering this market • Upon entry, the rival and the incumbent simultaneously decide whether to fight or not fight • If they both fight, then the profitability of the market is such that the rival would be better off staying out • If they do not fight, the rival would be better off entering

  7. Game Tree in rival out Incumbent is row player. Payoffs are (incumbent, rival) 2a, e

  8. Description • One interpretation of this diagram is the following: • There are three levels of profitability in the market: • High (when no one fights) • Medium (when 1 firm fights) • Low (when both fight) • The product of fighting is to capture market share from the rival • If both fight or both don’t fight, market share is 50-50 • If one fights and the other doesn’t the fighter gains market share at the expense of the rival

  9. Goal • Your goal is to provide an analysis of under what conditions to enter this market. • What are the key things to think about in making this decision?

  10. Example 1: Large Market Share Capture • Suppose that the profitability of the market is: • Big (No one fights)= 32 • Medium (One firm fights)= 25 • Small (Both fight) = 16 • The outside option of a rival who does not enter is 11 • When 1 firm fights and the other does not, the fighter obtains 80% market share

  11. Game Tree: Example 1 in rival out 32, 11

  12. Analysis • If the rival enters: • Incumbent’s best response to don’t fight is to fight • Incumbent’s best response to fight is to fight • The situation is symmetric for the rival • Therefore: • If enter, then (fight, fight)

  13. Game Tree: Example 1 in rival out Since the rival anticipates a fight on entering, it is better not to enter 32, 11

  14. Game Tree: Example 1- Generalized in rival out • a > e > b (It only pays to enter absent a fight) • c > a • b > d 2a, e

  15. Game Tree: Example 1- Generalized • a > e > b (It only pays to enter absent a fight) • c > a • b > d in rival out Notice that the game after entry is a Prisoner’s dilemma In this case the incumbent uses the b, b outcome to successfully deter entry 2a, e

  16. Comments • Notice that what didn’t happen –entry – had a profound effect on what did • The rival could count on the fact that entry combined with the temptation to grab market share would lead to a fight • Therefore, it paid to stay out of this market.

  17. Example 2: Smaller Market Share Grab • Suppose that the market sizes are again • Big = 32 • Medium = 25 • Small = 16 • The outside option of a rival who does not enter is still 11 • When 1 firm fights and the other does not, the fighter obtains 60% market share • Now what happens?

  18. Game Tree: Example 2 in rival out 32, 11

  19. Analysis • If rival enters: • Incumbent’s best response to don’t fight is don’t fight • Incumbent’s best response to fight is don’t fight • Same for rival • So neither fight if rival enters • If rival does not enter • Incumbent is free to threaten to do whatever it likes • In particular, it can threaten to fight • In which case it pays for the rival to stay out

  20. Equilibria • Rival enters, neither firm fights • Rival doesn’t enter, incumbent threatens to fight if it did enter • Notice that now entry deterrence depends crucially on the rival’s beliefs about the incumbent’s response • If the rival is convinced that the incumbent will be aggressive, it should not enter • Since the rival chooses not to enter, choosing to actually be aggressive is a best response by incumbent

  21. Game Tree: Example 2 Generalized a > e > b in rival out 1. a > c 2. d > b 2a, e

  22. Game Tree: Example 2 Generalized a > e > b 1. a > c 2. d > b in rival out Even though it is a dominant strategy for incumbent to not fight It can deter entry by threatening. Since in the even of successful deterrence, the threat is no tested this is still a best response for the incumbent 2a, e

  23. Comments • So it would seem that if the incumbent can affect the rival’s beliefs, it is possible to deter entry even in this framework.

  24. Choosing Equilibria • The prospect that a threat which is costly to carry out might succeed in a situation like this posed a problem for game theory • Is there some rational means to choose between the equilibria?

  25. Subgame Perfect Equilibria • We’ll generalize the idea of look forward, reason back the following way: • Rationality Axiom 2: When presented with any history of the game (even an unexpected one), players should choose best responses to future beliefs • Formally, we require that players choose optimizing strategies everywhere in the game

  26. Example 2: Refined • Recall that not fighting was a dominant strategy for each of the players if the rival enters • Therefore, despite incumbent’s threats to the contrary • Rival should anticipate that its entry will not lead to fighting • Therefore, it pays to enter. • Entry deterrence is not credible in this case.

  27. Example 3: Shrinking Markets • Now consider a variation of example 2. • Suppose that when either firm fights, it still gains 60% market share but the profitability of the market shrinks to a greater extent than before. • Does this change rival’s view of the incumbent’s threat to fight?

  28. Example 3: Specifics • Suppose that the market sizes are now • Big = 32 • Medium = 18 • Small = 16 • The outside option of a rival who does not enter is 11 • When 1 firm fights and the other does not, the fighter still obtains 60% market share

  29. Game Tree: Example 3 in rival out 32, 11

  30. Analysis • The best response to not fighting is not to fight • The best response to fighting is to fight • Therefore, there are 2 equilibria following entry • If rival anticipates a fight, it should not enter • If it anticipates no fighting, it should • Hence there are 2 equilibria of the dynamic game: • Out -> Fight, Fight • In -> Don’t fight, Don’t fight

  31. Comments • Notice that making competition more disruptive on entry actually improves the credibility of the threat by the incumbent • In many situations it is possible to control how destructive competition will be in markets • The subtlety here is that it’s in the incumbent’s interest to make the destructiveness of competition more rather than less

  32. Game Tree: Example 3 Generalized a > e > b in rival out 1. a > c 2. b > d 2a, e

  33. Game Tree: Example 2 Generalized a > e > b 1. a > c 2. b > d in rival out Now this is a coordination game if the rival enters. Either both firms can coordinate on not fighting, or they can coordinate on fighting Either is self-sustaining and the fighting outcome deters entry. 2a, e

  34. Key Conclusions • The idea of subgame perfection is to assess the credibility of threats. • We’ll return to this issue in the next class • Threats which are not self-sustaining if carried out, should correctly be viewed with skepticism • To make a threat credible, it can sometimes serve the interest of the incumbent to destroy profitability of the market in the event of entry.

  35. Repeated Games • We turn now to repeated games • These are games where players are involved in the same (or similar) strategic situation for many periods in a row. • The key insight here will be that we can use the future to affect the outcome in the present.

  36. Example 1 Revisited • Suppose that entry has occurred and that the situation is as in example 1

  37. Example 1: Game Table • Suppose • a > b • Competition is destructive to profitability • c > a • Market share grabs are profitable • b > d • Fighting back is better than being a victim of a grab

  38. Analysis • We determined that fighting was the inevitable outcome • With consequent decrease in the profitability of the market • Suppose that the firms will compete for 2 periods instead of 1? • Firm 1 and 2 agree to the following: • Don’t fight in either period • If either of us fights in the first period, fight in the second • Will this work?

  39. Carrots and Sticks • Each firm is holding out a carrot—the promise of a in both periods • And a stick, the threat of b in the second period • To try to deter the temptation to grab market share (and get c) • This could work if 2a > c + b

  40. Look Forward Reason Back • But there’s a problem here: • In period 2, there’s no stick and no carrot • So each firm will be tempted to fight and succumb to that temptation • And this “reverberates” back to period 1 • Each firm knows that there is no “carrot” in the second period---only the “stick” • So there’s nothing to deter the temptation to fight in period 1 • Hence the firms will fight in both periods

  41. Generalizing • The principle applies for any set number of periods. • Since there’s nothing to promote good behavior at the end of the game, firms will fight then • And this reverberates backward throughout the game • The conclusion is a sad one: • If the game has any set ending time, the firms will fight in every period

  42. Infinite Repetition • Suppose there is no fixed endpoint to the game • Instead the firms expect the game to be infinitely repeated • This is an abstraction---think of it as a game being repeated for a really long time with no one knowing exactly when it will end • Now can the firms cooperate?

  43. Tit-for-Tat Strategies • Suppose the firms make the following deal: • We agree not to fight • If my rival fights, I’ll fight in the next period • Then the war is over and we’ll resume not fighting • Will this work? • Now in every period the firm must weigh the gains from cheating (i.e. the temptation) • c – a • Versus the cost in the future (the carrot and stick) • a – b • If the cost exceeds the temptation, each firm will refrain from fighting. • a – b > c – a

  44. Variations • Suppose the carrot and stick in the above agreement are not enough to overcome the temptation • i.e., a – b < c – a • Is all lost? • No. Because there is always a future, the size of the “stick” can be increased

  45. Tit-for-2 Tats • Suppose the firms make the following deal: • We agree not to fight • If my rival fights, I’ll fight for the next 2 periods • Then the war is over and we’ll resume not fighting • Will this work? • Now in every period the firm must weigh the gains from cheating (i.e. the temptation) • c – a • Versus the cost in the future (the carrot and stick) • 2 x (a – b) • If the cost exceeds the temptation, each firm will refrain from fighting. • 2 x (a – b) > c – a

  46. Key Point • The larger is the temptation or the weaker the punishment available in any one period • The longer the threat of “war” needs to be to deter cheating. • The size of the stick needs to be calibrated to the upside from cheating. • Obviously the promise of infinite war leads to the largest possible “stick”

  47. Discounting • All of this assumed that profits in the future were worth the same as those in the present • Of course, they’re not • Suppose that we discount profits in the future by the real interest rate r. • How does this change the analysis?

  48. Tit for n Tats • If we play tit-for-tat: • NPV of temptation = c – a • NPV of the threat = (a – b)/(1 + r) • Notice that as the interest rate increases, the punishment associated with the threat declines

  49. Perpetual Punishment • In a world with positive real interest rates, even the threat of perpetual punishment may not be enough to stave off fighting • Recall the perpetuity formula: • What is the NPV of an asset has a cash flow C starting 1 period in the future and lasting in perpetuity? • NPV = C/r

  50. Analysis • Suppose we threaten punishment forever after a defection. • NPV of Temptation = c – a • NPV of punishment = (a – b)/r • For higher real interest rates, even the most severe possible punishment loses efficacy • If this threat is insufficient to deter cheating, nothing will deter it.

More Related