Backward Induction Reasoning in Games with Incomplete Information

Size: px
Start display at page:

Download "Backward Induction Reasoning in Games with Incomplete Information"

Transcription

1 Backward Inducton Reasonng n Games wth Incomplete Informaton Antono Penta y Unversty of Wsconsn - Madson, Dept. of Economcs Ths verson: September 26, 2011 Abstract Backward Inducton s one of the central notons n game theory. A number of soluton concepts, such as subgame perfect and sequental equlbrum, are thought of as beng based on backward nducton reasonng. Yet, t s not clear what ths means precsely, partcularly n stuatons wth ncomplete nformaton, where the game cannot even be solved backwards. Ths paper ntroduces a soluton concept for games wth ncomplete nformaton, backward extensve form ratonalzablty ( BR for short), and proves several propertes that show how, n a precse sense, BR characterzes the mplcatons of backward nducton reasonng n games wth ncomplete nformaton. These results reconcle n a untary framework several deas tradtonally (though only nformally) assocated to the logc of backward nducton, such as the dea of devatons as mstakes, tremble-based equlbrum concepts, the belef-persstence hypothess, the noton of subgame consstency and the possblty of solvng the game backwards. JEL Codes: C72; C73; D82. 1 Introducton Backward nducton s one of the basc notons of game theory. Although backward nducton s only de ned for games wth perfect and complete nformaton, the logc of backward nducton has a much broader scope n game theory. So, for nstance, subgame perfect equlbrum s certanly vewed as the natural extenson of ths logc to games wth Some of the materal contaned n ths manuscrpt prevously crculated as part of my job market paper Robust Dynamc Mechansm Desgn. I m ndebeted to my advsor, George Malath. I also thank partcpants of several semnars and conferences. y emal: apenta@ssc.wsc.edu. 1

2 mperfect nformaton. But there s a sense n whch also soluton concepts for ncomplete nformaton games, such as sequental or tremblng-hand perfect equlbrum, are often thought of as havng a backward nducton avor. Yet, t s not clear what backward nducton means n games wth ncomplete nformaton. More broadly: What do we mean by Backward Inducton Reasonng? Despte ts central poston n game theory, there s no comprehensve, formal answer to ths queston. Fllng ths gap s the am of ths paper. In pursut of an answer, a good startng pont s to nspect the concepts that are normally assocated to the dea of backward nducton reasonng. Subgame Perfect Equlbrum (SPE) s certanly one of these. SPE s probably the sngle most common soluton concept for dynamc games wth complete nformaton. An n uental argument n support of SPE s provded by Harsany and Selten s (1988) noton of subgame consstency: It s natural to requre that a soluton functon for extensve games s subgame consstent n the sense that the behavor prescrbed on a subgame s nothng else than the soluton to the subgame (bd., p.90) Subgame consstency warrants SPE the recursve structure of backward nducton,.e. the possblty of determnng the soluton concept s predctons for a subgame by lookng at the subgame n solaton. Hence the possblty (n games wth nte horzon) to solve for the subgame perfect equlbra startng from the termnal nodes and proceedng backwards. Ths s extremely convenent, and certantly one of the man reasons for the promnence of SPE n appled work. SPE assumes that, even after observng a devaton, agents mantan ther belefs n the equlbrum contnuaton strateges. Ths assumpton, whch can be referred to as the belef persstence hypothess, often consttuted the man target for the numerous crtques to SPE and ts mplct assumptons on counterfactuals. 1 Several soluton concepts extend the deas of SPE to games wth ncomplete nformaton, and many of these nvolve trembles (e.g., tremblng-hand perfect equlbrum (Selten, 1975), sequental equlbrum (Kreps and Wlson, 1982), etc.). In these soluton concepts, trembles are a shortcut to formalze another dea typcally assocated to the logc of backward nducton: that o -equlbrum moves are mstakes, unntended devatons. 2 Beng the ncomplete nformaton counterparts of SPE, t s commonly accepted that these soluton concepts share a backward nducton avor. But, f ths s true at an ntu- 1 See e.g. Stalnaker (1996, 1998). 2 The vew of devatons as mstakes contrasts wth the logc of forward nducton, whch requres nstead that unexpected moves be ratonalzed (f possble) as purposeful devatons. 2

3 tve level, ts precse meanng s not clear, as no formal de nton of backward nducton s avalable for games wth ncomplete nformaton. One mportant d erence between SPE and ts ncomplete nformaton counterparts s that the latters lack the recursve structure of SPE. Under ncomplete nformaton, an equlbrum requres a spec caton of agents belefs about the opponents types at each nformaton set. But such belefs are endogenous, equlbrum objects, and must be jontly determned wth the equlbrum strateges. Wth ncomplete nformaton, contnuaton games cannot be consdered n solaton, and the equlbrum analyss requres the soluton of xed pont problems, often d cult to compute. Thus, on the one hand, the tremble-based soluton concepts supposedly embody the logc of backward nducton; but, on the other hand, they lack the recursve structure of SPE, whch seems almost a de nng feature of backward nducton reasonng. Ths paper puts forward a soluton concept for belef-free dynamc games called Backwards Extensve Form Ratonalzablty (BR for short). BR conssts of an terated deleton procedure for games n extensve form that at each round elmnates strateges that are not sequental best responses to conjectures that, at each pont n the game, must be concentrated on opponents contnuaton strateges that are consstent wth the prevous rounds of deleton. Through the followng results, t s further argued that BR characterzes the behavoral mplcatons of Backward Inducton Reasonng n games wth ncomplete nformaton: Result 1: BR can be computed by a convenent backwards procedure that combnes the logc of (normal form) ratonalzablty and backward nducton. The backwards procedure conssts of the terated applcaton of (normal form) ratonalzablty to the contnuaton games from each nformaton set consdered n solaton, startng from the end of the game and then proceedng backwards. Besdes smplfyng the computaton of the set of BR strateges, ths result mples that BR sats es a property analogous to subgame consstency, whch we may call contnuatongame consstency: the predctons of BR for each contnuaton game are nothng but the BR strateges of the contnuaton game. I ntroduce next an equlbrum concept for dynamc Bayesan games, nterm perfect equlbrum (IPE). Bayesan games are obtaned appendng a model of agents belefs,.e. a type space, to the belef-free game. IPE s the weakest equlbrum noton consstent wth sequental ratonalty and Bayesan updatng, and concdes wth SPE n complete nformaton games. I show that: Result 2: IPE s consstent wth a tremble-based re nement of Bayesan equlbrum, n whch trembles may be correlated wth anythng. 3

4 Result 3: The set of BR strateges n the belef-free game concdes wth the set of strateges played as part of some IPE for some type space. Result 3 says that BR characterzes the robust predctons of IPE, that s the IPE predctons that do not depend on assumptons on players exogenous belefs. Furthermore, whle for a gven type space the computaton of the IPE strateges remans a xed pont problem, Results 1 and 3 together mply that a property analogous to subgame consstency holds for the set IPE strateges: the robust predctons are contnuaton-game consstent. At a practcal level, these results show that rather than computng the set of IPE by solvng a large (possbly n nte, n fact) number of xed pont problems, the set of all IPE strateges can be computed by means of a tractable backwards procedure. These results therefore can be partcularly useful n applcatons. 3 Fnally, two epstemc characterzatons of BR are provded: Result 4: BR s characterzed by Ratonalty and the followng assumptonson agents belefs: Common Certanty of Future Ratonalty. That s, players share common certanty of ratonalty at the begnnng of the game. If an unexpected move s observed, players are wllng to accept the dea that the devaton was a mstake and mantan common certanty of ratonalty n the contnuaton game; or Common Certanty of Full Ratonalty and Belef Persstence. That s, at the begnnng of the game, players share common certanty of ratonalty and they never change ther belefs about the opponents contnuaton strateges. Full Ratonalty s a stronger noton than ratonalty, n that t refers to statements about agents ratonalty condtonal on counterfactual hypothess. (To accommodate such counterfactual propostons, t wll be necessary to nnovate on the exstng lterature by ntroducng rcher epstemc models.) Overall, the results above (formally) reconcle all the features that are (nformally) assocated to backward nducton reasonng: the recursve structure, the noton of contnuatongame consstency, the belef persstence hypothess, the dea of trembles and of devatons as unntended mstakes. There s thus a precse sense n whch IPE s the ncomplete nformaton counterpart of SPE embodyng the backwards nducton logc and nothng more. 3 See Penta (2009) for an applcaton to problems of robust dynamc mechansm desgn. 4

5 I thus argue that the epstemc assumptons of Common Certanty of Future Ratonalty (or, alternatvely, Common Certanty of Ratonalty and Belef Persstence) provde a comprehensve formal answer to the openng queston: What do we mean by Backward Inducton Reasonng? From an appled perspectve, ths paper provdes foundatons to a tractable backwards procedure that characterzes the set of equlbra of dynamc games wth ncomplete nformaton. The tractablty of the algorythm may prove useful n overcomng the d cultes typcally faced n appled and emprcal works. 2 Belef-Free Dynamc Games The analyss that follows concerns multstage games wth observable actons. 4 de ned by an extensve form and agents preferences and nformaton. Extensve Form. These are The game has L stages, ndexed by l = 1; 2; :::; L. Let h 0 denote the empty hstory, and for every player 2 N = f1; :::; ng, let A denote the ( nte) set of actons avalable to player throughout the game. At stage l = 1, agents 2 N smultaneously choose actons a 1 from the nte sets A (h 0 ) A (for each 2 N). The chosen acton pro le s publcly observed, hence the set H 1 = 2N A (h 0 ) denotes the set of hstores of length one. For every h 2 H 1, let A (h) A denote the set of actons avalable to player at hstory h. 5 For every l = 2; :::; L, the set H l of publc hstores of length l s de ned recursvely as follows: for any l and h l 1 2 H l 1, let A h l 1 A denote the set of player s actons avalable at hstory h l 1, and let A h l 1 = 2N A h l 1 and A h l 1 = j2nnfg A j h l 1. 6 H l = n h l 1 ; a l 2 H l 1 A l : a l 2N 2 A h l 1 o for every 2 N : The set of publc hstores s de ned as H = S L 1 t=0 Hl (where H 0 = fh 0 g), whle the set of termnal hstores s Z = H L. Wthout loss of generalty, sets A (h) are assumed non-empty for each h 2 H: player s nactve at h f ja (h) j = 1; he s actve otherwse. Ths setup allows ntely repeated games as a specal case, or games wth perfect nformaton f H s such that only one player s actve at each h. If H = fh 0 g, the game s statc. It wll be convenent to ntroduce the precedence relaton on H: h l h l+k f and only f there exsts h l+k = h l ; a k. k=1;:::;k a k k=1;:::;k such that 4 See Fudenberg and Trole, 3.2 and 8.2. At the expense of heaver notaton, the analyss can be easly adapted to all nte dynamc games wth perfect recall. 5 The set A (h) may vary wth hstory h, hence the game need not be a repeated game. 6 Smlar notaton wll be adopted for other product sets. 5

6 Preferences and Informaton. To model stuatons wth ncomplete nformaton, players preferences over the termnal nodes are parametrzed on a fundamental space of uncertanty = 0 1 ::: n. 7 Elements of are referred to as payo states, and payo functons are denoted by u : Z! R, for each 2 N. For each = 1; :::; n, s the set of player s payo types. 0 s referred to as the set of states of nature. For each, = j2nnfg j, so that = 0. (To avod unnecessary techncaltes, the set s assumed nte throughout.) When the true state s ( 0 ; 1 ; :::; n ), player s payo type s, prvately observed at the begnnng of the game. Hence, payo types represent agents nformaton about the payo o state: f s payo type s ^, knows that the true state belongs to the set 0 n^. The set 0 represents the resdual uncertanty that s left after poolng everybody s nformaton. The tuple 0 ; ( ; u ) 2N thus represents agents nformaton about payo s. It s assumed common knowledge and referred to as preference-nformaton structure (PIstructure). Specal cases of nterest are: complete nformaton ( k s a sngleton for all k = 0; 1; :::; n); prvate values (f, for all 2 N, u s constant n ( 0 ; )); no nformaton (f u s are constant on 0 ); dstrbuted knowledge (f u s are constant n 0 ) Belef-Free Games A belef-free dynamc game s thus de ned by a tuple = N; H; Z; 0 ; ( ; u ) 2N : Notce that ths s not a Bayesan game, as does not nclude a model of agents nteractve belefs over. Bayesan games wll be ntroduced n Secton 4.1. Strategc Forms. Pure strateges n the belef-free game are functons s : H! A such that for each h 2 H, s (h) 2 A (h). The set of player s strateges s denoted by S, and as usual we de ne the sets S = 2N S and S = j2nnfg S j. To dstngush them from those that wll be ntroduced for Bayesan games, elements of S are referred to as nterm (pure) strateges. Any strategy pro le s 2 S nduces a termnal hstory z () 2 Z. Hence, we can de ne strategc-form payo functons U : S! R as U (s; ) = u (z (s) ; ) for each s and. 7 General nformaton parttons on could be consdered, at the expense of heaver notaton. Restrctng attenton to product structures entals no essental loss of generalty. 8 Product structures wth dstrbuted knowledge are common n the lterature on robust mechansm desgn. (see, e.g., Bergemann and Morrs 2005, 2009) 6

7 It s useful to ntroduce notaton for the nterm mxed strateges (elements of (S )) and the nterm behavor strateges (elements of h2h (A )), whch wll be used for the analyss of Bayesan games (Secton 4.1). To save on notaton, we wll take advantage of Kuhn s (1953) equvalence theorem and use the same symbol, 2 P, to denote both knds of strateges (ts nterpretaton wll be clear from the context). Notaton [s ] refers to the probablty that mxed strategy attaches to strategy s ; whle (a jh) refers to the probablty of acton a at prvate hstory h accordng to the behavor strategy : For each publc hstory h and player, let S (h) denote the set of player s strateges that are consstent wth hstory h beng observed. It s also convenent to de ne strateges and payo functons for the contnuatons games: For each publc hstory h 2 H, let S h denote the set of strateges n the contnuaton game startng from h, and for each s 2 S, let s jh 2 S h denote the contnuaton of s from hstory h. The notaton z (sjh; ) refers to the termnal hstory nduced by strategy pro le s from the publc hstory h, when the realzed state s. Strategc-form payo functons can be de ned for contnuatons from a gven publc hstory: for each h 2 H and each (s; ) 2 S, let U (s; ; h) = u (z (sjh) ; ). (For the ntal hstory h 0, U (s; ) wll be wrtten nstead of U (s; ; h 0 ).) 3 Backwards Extensve Form Ratonalzablty Backwards Extensve Form Ratonalzablty (BR) s a soluton concept for belef-free games n extensve form, concept: n game not be consstent wth each other. Endogenous Belefs: Conjectures.. Smlar to ratonalzablty, BR s a non-equlbrum soluton agents form conjectures about everyone s behavor, whch may or may At every hstory, players hold conjectures about the state of nature, the opponents payo types and (everybody s) behavor. These are represented by condtonal probablty systems (CPS),.e. arrays of condtonal belefs, one for each hstory. These belefs d er from those that wll be ntroduced n Secton 4.1 n that they concern and depend on endogenous varables such as the opponents behavor. Hence, these are endegenous belefs. To avod confuson, we thus refer to ths knd of belefs as conjectures, retanng the term belefs for those ntroduced n Secton 4.1. For each hstory h 2 H, de ne the event [h] 0 S as: [h] = 0 S (h) : (Notce that, by de nton, [h] [h 0 ] whenever h follows h 0.) De nton 1 A conjecture for agent s a condtonal probablty system (CPS hereafter), that s a collecton = ( (h)) h2h of condtonal dstrbutons (h) 2 ( 0 S) 7

8 that satsfy the followng condtons: C.1 For all h 2 H, supp ( (h)) [h] ; C.2 For every measurable A [h] [h 0 ], (h) [A] (h 0 ) [h] = (h 0 ) [A]. The set of CPS over 0 S s denoted by H ( 0 S). Condton C.1 states that agents are always certan of what they know; condton C.2 states that agents conjectures are consstent wth Bayesan updatng whenever possble. Notce that n ths spec caton agents entertan conjectures about the payo state, the opponents and ther own strateges. The latter pont s not entrely standard: n smlar non-equlbrum soluton concepts for games n extensve form t s common practce to model conjectures on the opponents behavor only (see, e.g. Battgall and Snscalch, 2007, or Penta, 2010). We wll dscuss ths pont n some detal n Secton 4. Sequental Ratonalty. Strategy s s sequentally ratonal for type wth respect to conjectures f, at each hstory h 2 H, t prescrbes optmal behavor n the contnuaton game wth respect to conjectures (h). Formally: for any ^ 2, gven a CPS 2 H ( 0 S) and a hstory h, ^ s expected payo from s at h, gven, s de ned as: U s ; ; h; ^ = P marg 0 S (h) [ 0 ; ; s ] 0 ; ;s U s ; s ; 0 ; ^ (1) ; ; h De nton 2 Strategy s s sequentally ratonal for payo -type wth respect to 2 H ( 0 S), wrtten s 2 r ( ; ), f and only f for each h 2 H and each s 0 2 S the followng nequalty s sats ed: U s ; ; h; U s 0 ; ; h; : (2) If s 2 r ( ; ), we say that conjectures justfy strategy s for type. 3.1 Backwards Ratonalzablty n the Extensve Form We ntroduce next the soluton concept that, t wll be argued, characterzes backward nducton reasonng n ncomplete nformaton games: Backwards Extensve Form Ratonalzablty (BR). 8

9 De nton 3 For each 2 N let BR 0 = S. Recursvely, for k = 1; 2; :::, let BR k 1 = j2nnfg BR k 1 j, and for each 2, let, H ( 0 S) s.t. >< BR k ( ) = ^s 2 BR k 1 ( ) : (1) ^s 2 r ( ; ) (2) supp ( (h 0 )) 0 BR k 1 f^s g (3) for each h 2 H : ( ; s) 2 supp marg S (h) mples: (3.1) s jh = ^s jh, and >: (3.2) 9s 0 2 BR k 1 ( ) : s 0 jh = s jh >; BR k = ( ; s ) 2 S : s 2 BR k ( ), BR k = k 2N BR, and nally BR := T BR k. BR conssts of an terated deleton procedure. At each round, strategy ^s survves for type f t s just ed by conjectures that satsfy two condtons: condton (2) states that at the begnnng of the game, the agent must be certan of hs own strategy ^s and have conjectures concentrated on pars of the opponents type and strateges consstent wth the prevous rounds of deleton; condton (3) restrcts the agent s conjectures at unexpected hstores: condton (3.1) states that agent s always certan of hs own contnuaton strategy; condton (3.2) requres conjectures to be concentrated on opponents contnuaton strateges that are consstent wth the prevous rounds of deleton. Notce however that agents conjectures about 0 at unexpected hstores are unrestrcted. Thus, condton (3) embeds two conceptually dstnct knds of assumptons: the rst concernng agents conjectures about 0 ; the second concernng ther conjectures about the contnuaton behavor. For ease of reference, they are summarzed as follows: Unrestrcted-Inference Assumpton (UIA): At unexpected hstores, agents conjectures about 0 are unrestrcted. In partcular, agents are free to nfer anythng about the opponents prvate nformaton from the publc hstory. For example, condtonal conjectures may be such that marg (h) s concentrated on opponents types for whom some of the prevous moves n h would be rratonal, or mstakes. Nonetheless, condton (3.2) mples that t s beleved that such types wll behave ratonally n the future. From an epstemc vewpont, t can be shown that BR can be nterpreted as common certanty of future ratonalty at every hstory (for the formal statement, see Secton 6.3) Common Certanty n Future Ratonalty (CCFR): at every hstory (expected or not), agents share common certanty n future ratonalty. 9 >= ; k0

10 Fgure 1: Example 1 CCFR can be nterpreted as a condton of belef persstence on the contnuaton strateges. (A formal connecton between BR and the belef persstence hypothess s provded n Secton 6.4). Remark 1 Snce the game s nte, the terated deleton procedure stops after ntely many rounds: there exsts K < 1 such that BR K = BR K+1. It s also trval to show that BR sats es the followng xed pont characterzaton: Lemma 1 Strategy ^s 2 BR ( I ) f and only f 9 2 H ( 0 I S) s.t. (1) ^s 2 r ( ; I ); (2) supp( (h 0 )) 0 BR f^s g and (3) for each h 2 H: s 2supp(marg S (h)) mples: (3.1) s jh = ^s jh, and (3.2) 9 ; s 0 2 BR : s 0 jh = s jh. Example 1 Consder the game n gure 1, and let 1 = f10; 10g, whle 0 and 2 are sngletons (hence, player 1 knows the true state, whle player 2 has no nformaton). Let s apply BR: at the rst round, strateges nvolvng L 3 and R 3 are deleted for all types of player 1, whle strateges nvolvng l (resp. r) are deleted for type 1 = 10 (resp. 1 = 10). So, for nstance, after the rst round strategy rl 1 R 2 survves for type 1 = 10. Now, suppose that 2 s ntal conjectures are concentrated on payo state-strategy par ( 1 ; s 1 ) = ( 10; rl 1 R 2 ): then, l s unexpected, so after observng t we are free to specfy 2 s belefs, who could for example assgn probablty one on par ( 1 ; s 1 ) = (10; ll 1 R 2 ), hence play a 2, or on par ( 1 ; s 1 ) = ( 10; ll 1 R 2 ), hence play a 1. Smlarly, both b 1 and b 2 n the rght-most contnuaton game can be just ed f 2 s ntal conjectures assgn probablty one to 1 = 10. Hence, strateges that survve BR are fa 1 b 2 ; a 1 b 3 ; a 2 b 2 ; a 2 b 3 g for player 2, fll 1 R 1 ; ll 1 R 2 ; ll 2 R 1 ; ll 2 R 2 g for type 1 = 10 and frl 1 R 1 ; rl 1 R 2 ; rl 2 R 1 ; rl 2 R 2 g for type 1 =

11 4 Backwards Ratonalzablty and Equlbra Ths secton explores the connecton between BR and equlbrum predctons. Pursung an equlbrum approach n games wth ncompelte nformaton requres a spec caton of agents herarches of belefs. We wll follow the tradtonal approach (Harsany, ) of modelng such herarches of belefs mplctly, by means of type spaces. Appendng a type space to a belef-free game delvers a Bayesan game. 4.1 Bayesan Games De nton 4 A (-based) type space s a tuple T = (T ; ; ) 2N such that for each 2 N, T s a nte set of types, : T! s an onto functon assgnng a payo -type to each type, and : T! ( T ) assgns to each type a belef about the payo state and the opponents types. 9 The Bayesan game obtaned appendng type space T = (T ; ; ) 2N to the belef-free game s de ned as the tuple T = N; ; H; Z; (T ; ; ; ^u ) 2N where ^u : Z 0 T! R s such that that for each (z; 0 ; t) 2 Z 0 T, ^u (z; 0 ; t) = u (z; 0 ; (t)). To avod unnecessary notaton, n the followng we wll use u to denote both payo functons. Strateges n a Bayesan game are functons b : T! assgnng an nterm (mxed or behavor) strategy to each type n the type space. The notaton b a l ; t ; h l 1 refers to the probablty that behavor strategy b (t ) 2 assgns to acton a l at h l Interm Perfect Equlbrum De ne the set of nformaton sets of player n the Bayesan game as H = T H. 10 system of belefs conssts of collectons (p (h )) h 2H for each agent, such that p (h ) 2 ( 0 T ) for each h 2 H : a belef system represents agents condtonal belefs about the state of nature and the opponents types at each nformaton set. A strategy pro le and a belef system (b; p) form an assessment. For each agent, a strategy pro le b 9 The restrcton to nte type spaces s only made for smplcty of exposton. 10 Behavor strateges n the Bayesan game could be de ned as functons : H! A s.t. (t ; h) 2 (A (h)) for each h 2 H. Clearly, ths s equvalent to the de nton of strateges as functons b : T!. A 11

12 and condtonal belefs p nduce, at each prvate hstory h l 1 = t ; h l 1, a probablty dstrbuton P ;p a l jh l 1 over acton pro les at stage l: X Y P b;p ^a l ; h l 1 = b ^a l ; h l 1 p 0 ; t jh l 1 b j ^a l j; t j ; h l 1 (3) ( 0 ;t )2 0 T : De nton 5 Assessment (b; p) s weakly preconsstent f for each 2 N : 8t 2 T ; p t ; h 0 = (t ) (4) j6= p 8h l = h l 1 ; a l 2 H 0 ; t jh l = p 0 ; t jh l 1 P b;p ^a l ; h l 1 : (5) Condton (4) requres each agent s belefs condtonal on observng type t to agree wth that type s (exogenous) belefs as spec ed n the type space T ; condton (5) requres that the belef system p s consstent wth Bayesan updatng whenever possble. From the pont of vew of each, for each h = (t ; h) 2 H and strategy pro le b, the nduced termnal hstory s a random varable that depends on the opponents type pro le. Ths s denoted by z (bjh ; t ). As done for belef-free games (Secton 2), we can de ne strategc-form payo functons for the contnuaton games: U (b; 0 ; t ; t ; h ) = u (z (bjh ; t ) ; (t)). De nton 6 Fx a belef system p. Strategy pro le s sequentally ratonal wth respect to p f for every 2 N and every h t 2 H n fg, the followng nequalty s sats ed for every b 0 : T! : X p 0 ; t jh l U b 0 ; b ; 0 ; t ; t ; h l 0 T X p U b; 0 ; t ; t ; h l 0 T 0 ; t jh l. De nton 7 An assessment (; p) s an Interm Perfect Equlbrum (IPE) f t s weakly preconsstent and sequentally ratonal. If nequalty (6) s only mposed at prvate hstores of length zero, the soluton concept concdes wth nterm equlbrum (Bergemann and Morrs, 2005). IPE re nes nterm equlbrum mposng two natural condtons: rst, sequental ratonalty; second, preconsstency of the belef system. (6) 12

13 Notce that weak preconsstency mposes no restrctons on the belefs held at hstores that receve zero probablty at the precedng node. 11 Hence, even f agents ntal belefs admt a common pror, IPE s weaker than Fudenberg and Trole s (1991) perfect Bayesan equlbrum. Also, notce that any player s devaton s a zero probablty event, and treated the same way. In partcular, f hstory h l s precluded by b h l 1 alone, h l =2suppP ;p h l 1, and agent s belefs at h l are unrestrcted the same way they would be after an unexpected move of the opponents. Ths feature of IPE s not entrely standard, but t s key to the result that the set of IPE strateges (takng the unon over all type spaces) can be computed by means of a convenent backwards procedure : Treatng own devatons the same as the opponents s key to the possblty of consderng contnuaton games n solaton, necessary for the result. 12 Tremble-Based Formulaton. It can be shown that IPE s consstent wth a tremblnghand vew of unexpected moves, n whch no restrctons on the possble correlatons between trembles and other elements of uncertanty are mposed. 4.3 Characterzaton of the set of IPE As emphaszed above, n BR agents hold conjectures about both the opponents and ther own strateges. Frst, notce that condtons (2) and (3.2) n the de nton of BR mantan that agents are always certan of ther own strategy; furthermore, the de nton of sequental best response (def. 2) depends only on the margnals of the condtonal conjectures over 0 S. Hence, ths partcular feature of BR does not a ect the standard noton of ratonalty. The fact that conjectures are elements of H ( 0 S) rather than H ( 0 S ) corresponds to the assumpton that IPE treats all devatons the same; ts mplcaton s that both hstores arsng from unexpected moves of the opponents and from one s own devatons represent zero-probablty events, allowng the same set of condtonal belefs about 0 S, wth essentally the same freedom that IPE allows after anyone s devaton. Ths s the man nsght behnd the followng result (the proof s n Appendx A.1). 11 Unlke other notons of weak perfect Bayesan equlbrum, n IPE agents belefs are consstent wth Bayesan updatng also o -the-equlbrum path. In partcular, n complete nformaton games, IPE concdes wth subgame-perfect equlbrum. 12 In Penta (2009b) I consder a mnmal strengthenng of IPE, n whch agents belefs are not upset by unlateral own devatons, and I show how the analyss that follows adapts to that case: The backwards procedure to compute the set of equlbra across models of belefs must be mod ed, so to keep track of the restrctons the extensve form mposes on the agents belefs at unexpected nodes. The possblty of envsonng contnuaton games n solaton s thus lost. 13

14 Proposton 1 Fx a belef-free game. For each, s 2 BR ^ f and only f 9T, ^t 2 T and ^b; ^p such that: () ^b; ^p s an IPE of T ; () ^t = ^ and () ^s 2supp ^b ^t : An analogous result can be obtaned for the more standard re nement of IPE, n whch unlateral own devatons leave an agents belefs unchanged, applyng to a mod ed verson of BR: such mod caton entals assumng that agents only form conjectures about 0 S (that s, 2 H ( 0 S )) and by consequently adaptng condtons (2) and (3) n the de nton of BR. (See Penta, 2009b.) Hence, the assumpton that IPE treats anyone s devaton the same (and, correspondngly, that n BR agents hold conjectures about ther own strategy as well) s not crucal to characterze the set of equlbrum strateges across models of belefs. It s crucal nstead for the next result, whch shows that such set can be computed applyng a procedure that extends the logc of backward nducton to envronments wth ncomplete nformaton (proposton 2 below). 5 Contnuaton-game Consstency Ths Secton provdes a characterzaton of BR n terms of a recursve procedure, that solves the game backwards. Together wth proposton 1, the result n ths secton mples that the robust predctons of IPE satsfy a property analogous to Harsany and Selten s (1988) subgame consstency: for each h, the set of IP E contnuaton strateges from h concdes wth the set of IP E strateges of the contnuaton game consdered n solaton. The Backwards Procedure. The backwards procedure s descrbed as follows: Fx a publc hstory h L 1 of length L 1. For each payo -type 2 of each agent, the contnuaton game s a statc game, to whch we can apply belef-free ratonalzablty (e.g., Bergemann and Morrs, 2009). For each h L 1, let R hl 1 denote the set of pars ; s jh L 1 such that contnuaton strategy s jh L 1 s ratonalzable n the contnuaton game from h L 1 for type. We now proceed backwards: for each publc hstory h L 2 of length L 2, we apply agan ratonalzablty to the contnuaton game from h L 2 (n normal form), restrctng contnuaton strateges s jh L 2 2 S hl 2 to be ratonalzable n the contnuaton games from hstores of length h L 1. R hl 2 denotes the set of pars ; s jh L 2 such that contnuaton strategy s jh L 2 s ratonalzable n the contnuaton game from h L 2 for type. Inductvely, ths s done for each h l 1, l = L; :::; 1, untl the ntal node s reached. Before ntroducng the procedure formally, consder the followng example: Example 2 Consder the game n gure 1 agan. If we apply belef-free ratonalzablty to the contnuaton game followng r, R 3 s deleted at the rst round for both types of 14

15 player 1, and b 1 at the second round for player 2. In ths contnuaton game, the procedure selects contnuatons fb 2 ; b 3 g for player 2 and contnuatons fr 1 ; R 2 g for both types of player 1. Smlarly, after l, belef-free ratonalzablty selects fa 1 ; a 2 g for player 2, and fl 1 ; L 2 g for both types of player 1. So, now we apply belef-free ratonalzablty to the normal form n whch t s mantaned that contnuaton strateges are ratonalzable n the correspondng contnuatons,.e. the relevant strategy sets now are fa 1 b 2 ; a 1 b 3 ; a 2 b 2 ; a 2 b 3 g for player 2, and fll 1 R 1 ; ll 1 R 2 ; ll 2 R 1 ; ll 2 R 2 g [ frl 1 R 1 ; rl 1 R 2 ; rl 2 R 1 ; rl 2 R 2 g for (both types of) player 1: at ths stage, type 1 = 10 deletes all strateges nvolvng r at the rst round, and so does type 1 = 10 wth those nvolvng l, but player 2 doesn t delete anythng: f he expects 1 = 10 (hence 1 to play l), then both a 2 b 2 and a 2 b 3 are best responses n the normal form; smlarly, f he expects 1 = 10 (.e. 1 to play r), then both a 2 b 3 and a 1 b 3 are optmal. Notce that the resultng strateges are precsely those selected by BR n example 1. As proposton 2 shows, ths nsght has general valdty.} The backwards procedure s de ned recursvely, startng from the last stage of the game and proceedng backwards: [l = L 1] For each h L 1 2 H L 1, and for each 2, let R 0 ; h L 1 = S hl 1 For each k = 1; 2; :::, let h L 1 n = ; s hl 1 8 R k 1 : s hl 1 2 R k 1 ; h L 1o, s hl 1 2 R k 1 ; h L 1 : (R.1): R k 1 h L 1 >< R k ; h L 1 (R.2): for all s 0 2 S hl 1 = P 0 ; ;s hl 1 hu s hl 1 ; s hl 1 ; ; ; h L 1, >: U s 0 ; s hl 1 ; ; ; h T 1 R h L 1 1\ = R k h L 1 k=1 0 ; ; s hl 1 9 >= 0 >; 15

16 [l = L 2; :::0] For each h l 2 H l, and for each 2, let R 0 ; h l n = s hl 2 S hl : 8a l+1 2 A l+1 h l, s hl j h l ; a l+1 2 R ; h l ; a l+1o : For each k = 1; 2; :::, let 8 s hl 2 R k 1 ; h l : (R.1): R k 1 h l >< R k ; h l (R.2): for all s 0 2 S hl, = P hu s hl ; s hl ; ; ; h l R h l = >: 1\ k=1 0 ; ;s hl U s 0 ; s hl ; ; ; h l R k h l Proposton 2 BR = R (h 0 ) for each. 0 ; ; s hl 9 >= 0 >; The propertes UIA and CCFR dscussed n Secton 3 provde the basc nsght behnd ths result. Frst, notce that under UIA, the set of belefs agents are allowed to entertan about the state of nature and the opponents payo -types s the same at every hstory ( 0 ). Hence, n ths respect, ther nformaton about the opponents types n the contnuaton game from hstory h s the same as f the game started from h. Also, CCFR mples that agents assumptons about everyone s behavor n the contnuaton s also the same at every hstory. Thus, UIA and CCFR combned mply that, from the pont of vew of BR, a contnuaton from hstory h s equvalent to a game wth the same space of uncertanty and strategy spaces equal to the contnuaton strateges, whch just es the possblty of analyzng contnuaton games n solaton. 6 Epstemc Characterzatons Ths Secton provdes two alternatve epstemc characterzatons of BR, one n terms of Common Certanty of Future Ratonalty, the other n terms of Common Certanty of Full Ratonalty and n Belef Persstence. The epstemc models adopted here have one mportant non-standard feature. exstng epstemc models (e.g., Battgall and Snscalch 2002, 2007), a state of the world s a tuple! = 0 ; ( ; s ; ) 2N,.e. a descrpton of a state of nature (0 ), and for each player, hs payo type, strategy s and epstemc type. Each epstemc type nduces a CPS over the states of world. state of the world s of the form! = In the epstemc models consdered here, a 0 ; ( ; x ; ) 2N, where x :! S denotes player s ex-ante strategy. Ex-ante strateges represent a full theory of s behavor: 16 In

17 n state! =! 0 ; (! ; x! ;! ) 2N 2, player s actual dsposton to act s gven by strategy x! (! ), but x! also represents player s dsposton to act under the hypothess (counterfactual at!) that hs payo type s d erent from!. Epstemc types stll nduce a CPS over the states of the world!, but ths means that now they express rcher nformaton. For nstance, consder a state at whch s belefs are concentrated on a par ^ ; ^x. Ths means that beleves that the opponents type s ^, and that hs actual strategy wll be ^x ^. But t also expresses s vew of how the opponents behavor would be under the hypothess that 0 6=. Ths enrchment of the state space allows to dsentangle epstemc counterfactuals concernng the opponents behavor from counterfactuals concernng the opponents prvate nformaton, necessary for the characterzaton of BR n terms of Common Certanty of Full Ratonalty and Belef Persstence. 13 The characterzaton n terms of Common Certanty of Future Ratonalty doesn t explot the nformaton contaned n such counterfactual statements. The epstemc characterzatons below wll be of the form BR = f( ; s ) : 9! 2 E s.t. (! ; x! (! )) = ( ; s )g, (7) for some measurable event E. Snce each X can be seen as a subset of S, equaton (7) s wrtten for short as BR = proj S E. 6.1 Condtonal Probablty Systems Let be a metrc space and A ts Borel sgma-algebra. Fx a non-empty collecton of subsets C An;, to be nterpreted as relevant hypothess. A condtonal probablty system (CPS hereafter) on (; A; C) s a mappng : A C! [0; 1] such that: Axom 1 For all B 2 C, (B) [B] = 1 Axom 2 For all B 2 C, (B) s a probablty measure on (; A). Axom 3 For all A 2 A, B; C 2 C, f A B C then (B) [A] (C) [B] = (C) [A]. The set of CPS on (; A; C), denoted by C (), can be seen as a subset of [ ()] C (.e. mappngs from C to probablty measures over (; A)). CPS s wll be wrtten as = ( (B)) B2C 2 C (). The subsets of n C are the condtonng events, each nducng belefs over ; () s endowed wth the topology of weak convergence of measures and 13 In envronments wth complete nformaton, Stalnaker (1996, 1998) dscusses the necessty of ncorporatng d erent knds of counterfactual propostons for the analyss of backward nducton reasonng. The ntroducton of ex-ante strateges n the de nton of the states of the world s one way of accomodatng the rcher set of counterfactuals requred by the presence of ncomplete nformaton. 17

18 [ ()] C s endowed wth the product topology. In the belef-free games of Secton 2.1, player s CPS were obtaned settng = 0 S, and the set of condtonng events was naturally provded by the set of hstores H: for each publc hstory h 2 H, the correspondng event [h] was de ned as [h] = 0 S (h). 6.2 Epstemc Models Ex ante strateges assgn an nterm strategy to every payo type. The set of ex ante strateges s denoted by X = S. For each x 2 X, let x jh denote the contnuaton strategy x jh :! S h s.t. for each 2, x jh ( ) = x ( ) jh. De nton 8 A (dynamc) type space on s a tuple D = N; H; ( ; X ; ; ; g ) 2N s.t. for every 2 N, s a Polsh space, and 1. X s a closed subset such that proj X = X 2. g :! H ( 0 ) s a contnuous mappng. For any 2 N, the elements of the set are referred to as Player s epstemc types. A type space s compact f all the sets ( ) 2I are compact. Let = 0 1 ::: n denote the set of states of the world, and let A be ts Borel sgma-algebra. Thus, at any possble world! 2 = 0 1 ::: n, we specfy a payo state (! 0 ;! 1 ; :::;! n) as well as each player s dspostons to act (hs strategy s! = x! (! )) and hs dspostons to beleve (hs CPS g (! ) = (g ;h (! )) h2h ). As dscussed above, x! also spec es s belefs about hs own strategy under the counterfactual (at!) hypothess that 0 6=!. Furthermore, notce that these dspostons also nclude what a player would do and thnk at hstores that are nconsstent wth! (.e. h =2 H (s! )). As standard n nteractve epstemology lterature, we assume that players also have belefs about ther own strategy, and complete a player s system of condtonal belefs by assumng that he s certan of hs true ex ante strategy, nformaton type and epstemc type. More spec cally, we assume that for every state of the world! and every hstory h, player 2 N would be certan of! and! gven h, and would also be certan of x! gven h, provded that x! (! ) 2 S (h). We also assume that f x! (! ) =2 S (h), player would stll be certan that hs contnuaton strategy agrees wth x! () jh for each. Formally, player s condtonal belefs on (; H) are gven by a contnuous mappng g = g;h : h2h! H () derved from g by the followng formula: for all ( ; x ; ) 2, for all h 2 H and E 2 A, g ;h ( ; x ; ) = g ;h ( ) (f( 0 ;! ) 2 0 : ( 0 ;! ; ( ; x 0 ; )) 2 E; for some x 0 2 X s.t. x 0 ( 0 ) jh = x ( 0 ) jh for all 0 2 g) : (8) 18

19 Type spaces encode a collecton of n nte herarches of CPSs for each epstemc type of each player. Battgall and Snscalch (1999) constructed the unversal type space for CPS,.e. a type space whch encodes all concevable herarchcal belefs. Consder the followng de nton: De nton 9 A belef-complete type space s a type space D such that, for every 2 N, g :! H ( 0 ) s onto. Battgall and Snscalch (1999) showed that a belef-complete type space may always be constructed (for all nte games, and also a large class of n nte games) by takng the sets of epstemc types to be the collecton of all possble herarches of condtonal probablty systems that satsfy certan ntutve coherency condtons. Also, every type space may be vewed as a belef-closed subspace of the space of n nte herarches of condtonal belefs. Fnally, snce we assume that the set of external states s nte and hence compact, the belef-complete type space thus constructed s also compact. Sequental h Ratonalty. Fx a type space D. For every player 2 N, let f = (f ;h ) h2h :! ( 0 X S ) H denote hs rst-order belef mappng, that s, for all! 2 and h 2 H, f ;h (! ) [ 0 ; ; x ; s ] = Z!2 s.t. (! 0 ;! ;x! ;x! (! ))=( 0; ;x ;s ) dg ;h (! ) : (9) It s easy to see that f s contnuous and that, for every! 2, f (! ) s ndeed a CPS over X S, where for each h 2 H, the correspondng condtonng event [h] ( 0 X S ) s de ned as [h] = f( 0 ; ; x ; s ) : (x ( ) ; s ) 2 S (h)g (10) For any CPS 2 H ( 0 X S ), de ne the map : H ( 0 X S )! H ( 0 S) so that, for each 2 H ( 0 X S ), () = ;h () h2h and for each h 2 H and [ 0 ; ; s ; s ] 2 0 S, ;h () [ 0 ; ; s ; s ] = x 0 :x0 ( X (h) 0 ; ; x 0 ; s )=s In words, () represents the payo relevant component of, obtaned by gnorng the nformaton on counterfactual belefs encoded by x! 6=!. It s easy to verfy that () thus de ned s a CPS whenever s. (11) 19

20 Fnally, we can de ne ratonalty: we say that player s ratonal at a state! f and only f (! ; x! (! )) 2 r ( (f (! ))). Then the event Rat = f! 2 : (! ; x! (! )) 2 r ( (f (! )))g (12) corresponds to the statement player s ratonal. De ne also the event everybody s ratonal : R = T 2N R : Belefs Operators. The next buldng block s the epstemc noton of (condtonal) probablty one belef (or certanty). Recall that an epstemc type encodes the belefs a player would hold, should any one of the possble non-termnal hstores occur. Ths allows us to formalze statements such as, Player would be certan that Player j s ratonal, were he to observe hstory h. Gven a type space D, for every 2 N, h 2 H and E 2 A, de ne the event B ;h (E) =! 2 : g;h (! ) [E] = 1 whch corresponds to the statement Player would be certan of E, were he to observe hstory h. Observe that ths de nton ncorporates the natural requrement that a player only be certan of events whch are consstent wth her own (contnuaton) strategy and epstemc type (recall how g was de ned, equaton 8). Recallng that h 0 s the empty hstory, B ;h 0 (E) s the event Player beleves E at the begnnng of the game. For each player 2 I and hstory h 2 H, the de nton dent es a set-to-set operator B ;h : A! A whch sats es the usual propertes of fals able belefs (see, for example, Chapter 3 of Fagn et al. (1995)); n partcular, t sats es: Conjuncton: For all events E; F 2 A, B ;h (E \ F ) = B ;h (E) \ B ;h (F ); Monotoncty: For all events E; F 2 A, E F mples B ;h (E) B ;h (F ). The mutual belef operator s de ned as B h (E) = T 2N B ;h (E). We wll be nterested n terated belefs operators. In general, x a self-map on A, : A! A. For any E 2 A, let 1 (E) = (E), and for k = 2; 3; :::, de ne k (E) = k 1 (E). The event common belef n E at h s thus de ned as T k1 Bk h (E). 6.3 Common Certanty of Future Ratonalty For every h, every and 2 H ( 0 S), let r ( jh) denote the set of pars payo type-strategy that are sequental best response to CPS from hstory h onwards. 20

21 That s: for any 2 H ( S), ( ; s ) 2 r jh f and only f 8h 0 h, s 2 arg max s 0 2S U s 0 ; ; h 0 ; : 14 For every and h 2 H, the event s ratonal n the contnuaton from h s de ned as follows: Rat ;h f! 2 : (! ; x! (! )) 2 r ( (f (! )) jh)g ; (13) The event ratonalty n the contnuaton from h s de ned as Rat h = T Rat ;h. (Clearly, Rat = Rat h 0) The event common certanty of future ratonalty (CCFR) corresponds to the assumpton that, at each pont n the game, there s common certanty that everybody s ratonal n the contnuaton game. It s formally de ned as: CCF R = \! \ Bh k (Rat h ) : : h2h k1 The next proposton shows that BR corresponds to the epstemc assumptons of ratonalty and common certanty of future ratonalty. Proposton 3 BR = proj S Rat \ CCF R Proof. (See Appendx ) 6.4 Belef Persstence We ntroduce here an alternatve characterzaton of BR. De nton 10 For each 2 N let BP 0 2, let BP k 1 = j2nnfg BP k 1 8 >< BP k ( ) = >: ^s 2 BP k 1 : j, 2N = S. Recursvely, for k = 1; 2; :::, and 9 2 H ( 0 X S ) s.t. (A) ^s 2 r () ; (B) supp marg X S (h 0 ) BP k 1 f^s g (C) for each h 2 H: supp marg X h S h (h) = supp marg X h S h (h 0 ) BP k = ( ; s ) 2 S : s 2 BP k (t ), BP k = k 2N BP, and nally BP := T 14 For a de nton of U s ; ; h 0 ;, see equaton 1, p. 8. k0 9 >= >; BP k. 21

22 Condton (C) can be nterpreted as a belef persstence hypothess: at each hstory (whether t s reached wth postve probablty or not), the support of the conjectures about the contnuaton strateges (one s own and the opponents ) must not change. Condton (C) thus conforms to what n the phlosophy lterature s known as the conservatvty prncple. Such prncple states that When changng belefs n response to new evdence, you should contnue to beleve as many of the old belefs as possble (Harman, 1986, p. 46). 15 Here, agent s conjectures over X h represent hs theory of everyone s (future) behavor n the game, both n the states ( 0 ; ) n the support of and those that are not (smlarly, conjectures over S h regard s belefs about hs own contnuaton). De nton 11 requres that upon observng hstory h, agent revses hs belefs so to accomodate the new nformaton but wthout changng hs theory about evreyone s behavor n the contnuaton game. Notce though that no restrcton are mposed on the agent s belefs about the opponents prvate nformaton,.e. on marg (h). For example, t may concentrate belefs on a payo type pro le for whch the observed hstory s rratonal, but the restrcton on the belefs on the contnuaton strateges entals that t s beleved that type wll behave ratonally n the future. In ths sense, belef persstence here only refers to the endogenous varables (strateges), not the exogenous ones (payo -types). Proposton 4 For each, and k, BR k = BP k. Proof. (See Appendx) Common Certanty of full Ratonalty and Belef Persstence. The event player s ratonal, de ned n equaton 12, conssts of the set of states of the world! n whch s actual behavor x! (! ) s a (sequental) best response to s belefs at that state. No restrctons are mposed on the counterfactual behavor, x! ( 0 ) for 0 6=!. We ntroduce next a stronger noton of ratonalty, whch also restrcts s counterfactual behavor. Rat = f! 2 R : 8 0 2, ( 0 ; x! ( 0 )) 2 proj S Rat g (14) Rat can thus be nterpreted as the set of states of the world n whch not only s ratonal, but also hs dsposton to act n the (counterfactual) hypothess that he observed a d erent payo -type s consstent wth ratonalty. The event Rat thus descrbes states of the world n whch s ex-ante strategy s ratonal. Rat s referred to as the event s fully ratonal. As above, the event everyone s fully ratonal s de ned as Rat = T Rat : 15 The conservatvty prncple, or the prncple of belef persstence, s one of the gudng prncples adopted n the phlosophy lterature on theores of ratonal belef changes (see, e.g., Gardenfors, 1988). Battgall and Bonanno (1991) provde a semantc and synctatc characterzaton of the prncple. 2N 22

23 De nton 11 Say that -restrctons satsfy the belef persstence hypothess f: 8 2 N, 8 2 BP = 2 H ( 0 X S ) : 8h 2 H, o supp marg (h) X h = supp marg S h X h S h (h0 ) : For gven collecton = (( ) 2 ) 2N where H ( 0 X S ), de ne the event [ ] = f( 0 ; ; x ; ;! ) 2 : f ( ; x ; ) 2 g ; [ ] = \ \ [ ] and [] = [ ] 2 2N Proposton 5 For any belef-complete type space: 1. BR 1 = proj S Rat \ BP 2. for every k 1, BR k+1 = proj S Rat \ BP \ k T =1 B h 0 Rat \ BP ; 3. f the type space s also compact, then BR = proj S Rat\ BP \ T 7 Further Remarks on the Soluton Concepts. Backwards procedure, Subgame-Perfect Equlbrum and IPE. k=1 B k h 0 Rat \ BP. In games wth complete and perfect nformaton, the backwards procedure R (h 0 ) concdes wth the backward nducton soluton, hence wth subgame perfecton. 16 The next example (borrowed from Perea, 2009) shows that f the game has complete but mperfect nformaton, strateges played n Subgame-Perfect Equlbrum (SPE) may be a strct subset of R (h 0 ): Example 1 Consder the game n the followng gure: 16 For the specal case of games wth complete nformaton, Perea (2009) ndependently ntroduced a procedure that s equvalent to R, and showed that R concdes wth the backward nducton soluton f the game has perfect nformaton. 23

24 R 1 (h 0 ) = fbc; bd; acg and R 2 (h 0 ) = ff; gg. The game though has only one SPE, n whch player 1 chooses b: n the proper subgame, the only Nash equlbrum entals the mxed (contnuaton) strateges 1 2 c d and 3 4 f g, yeldng a contnuaton payo of 11 4 for player 1. Hence, player 1 chooses b at the rst node. In games wth complete nformaton, IPE concdes wth SPE, but R (h 0 ) n general s weaker than subgame perfecton. At rst glance, ths may appear n contradcton wth propostons 1 and 2, whch mply that R (h 0 ) characterzes the set of strateges played n IPE across models of belefs. The reason s that even f the envronment has no payo uncertanty ( s a sngleton), the complete nformaton model n whch B s a sngleton for every s not the only possble: models wth redundant types may exst, for whch IPE strateges d er from the SPE-strateges played n the complete nformaton model. The source of the dscrepancy s analogous to the one between Nash equlbrum and subjectve correlated equlbrum (Aumann, 1974). We llustrate the pont constructng a model of belefs and an IPE n whch strategy (ac) s played wth postve probablty by some type of player Let payo s be the same as n example 1, and consder the model T such that T 1 = n o t bc 1 ; t bd 1 ; t ac 1 and T 2 = t f 2; t g 2, wth the followng belefs: ( h 1 (t 1 ) t f 1 f t 1 = t bc 1 ; t ac 1 2 = 0 otherwse and 2 (t g 2) t ad 1 = 1, 2 t f 2 t bc 1 = 1 The equlbrum strategy pro le s such that 8; 8t, b (t s ) = s. The system of belefs agrees wth the model s belefs at the ntal hstory, hence the belefs of types t g 2 and t ac 1 17 It s easy to see that such d erence s not merely due to the possblty of zero-probablty types. Also the relaxaton of the common pror assumpton s not crucal. 24

25 are unquely determned by Bayesan updatng. For types t s 6= t g 2; t ac 1, t s su cent to set p (t s ; a ) = (t s ) (.e. mantan whatever the belefs at the begnnng of the game were) Then, t s easy to verfy that (; p) s an IPE. On the other hand, f jj = 1 and the game has perfect nformaton (no stage wth smultaneous moves), then R (h 0 ) concdes wth the set of SPE-strateges. Hence, n envronments wth no payo uncertanty and wth perfect nformaton, only SPE-strateges are played n IPE for any model of belefs. 8 Concludng Remarks. On the Soluton Concepts. Proposton 1 can be seen as the dynamc counterpart of Brandenburger and Dekel s (1987) characterzaton of correlated equlbrum. The weakness of IPE (relatve to other notons of perfect Bayesan equlbrum) s key to that result: the heart of proposton 2 s BR s property of contnuaton-game consstency, whch allows to analyze contnuaton games n solaton, n analogy wth the logc of backward nducton. The CCFR and UIA assumptons (p. 9) provde the epstemc underpnnngs of the argument. To understand ths pont, t s nstructve to compare BR wth Battgall and Sncalsch s (2007) weak and strong versons of extensve form ratonalzablty (EFR), whch correspond respectvely to the epstemc assumptons of (ntal) common certanty of ratonalty (CCR) and common strong belef n ratonalty (CSBR): BR s stronger than the rst, and weaker than the latter. The strong verson of EFR fals the property of subgame consstency because t s based on a forward nducton logc, whch nherently precludes the possblty of envsonng contnuatons n solaton : by takng nto account the possblty of counterfactual moves, agents may draw nferences from ther opponents past moves and re ne ther conjectures on the behavor n the contnuaton. The weak verson of EFR fals contnuaton-game consstency for opposte reasons: an agent can make weaker predctons on the opponents behavor n the contnuaton than he would make f he envsoned the contnuaton game n solaton, because no restrctons on the agents belefs about ther opponents ratonalty are mposed after an unexpected hstory. Thus, the form of backward nducton reasonng mplct n IPE (whch generalzes the dea of subgame perfecton) s based on stronger (respectvely, weaker) epstemc assumptons than CCR (respectvely, CSBR). Appendx 25

Subjective Uncertainty Over Behavior Strategies: A Correction

Subjective Uncertainty Over Behavior Strategies: A Correction Subjectve Uncertanty Over Behavor Strateges: A Correcton The Harvard communty has made ths artcle openly avalable. Please share how ths access benefts you. Your story matters. Ctaton Publshed Verson Accessed

More information

Transparent Restrictions on Beliefs and Forward-Induction Reasoning in Games with Asymmetric Information

Transparent Restrictions on Beliefs and Forward-Induction Reasoning in Games with Asymmetric Information Transparent Restrctons on Belefs and Forward-Inducton Reasonng n Games wth Asymmetrc Informaton Perpaolo Battgall y Andrea Prestpno z Fnal draft, 2013 Abstract We analyze forward-nducton reasonng n games

More information

Interim Correlated Rationalizability 1

Interim Correlated Rationalizability 1 Interm Correlated Ratonalzablty Edde Dekel Northwestern Unversty and Tel Avv Unversty Drew Fudenberg Harvard Unversty Stephen Morrs Prnceton Unversty Frst Draft: May 2003. Ths Draft: November 2006 Ths

More information

Higher Order Beliefs in Dynamic Envinronments.

Higher Order Beliefs in Dynamic Envinronments. Hgher Order Belefs n Dynamc Envnronments. Antono Penta y Dept. of Economcs, Unversty of Pennsylvana May 2008 (prelmnary and ncomplete verson) Abstract Ths paper explores the role of hgher order belefs

More information

A note on the one-deviation property in extensive form games

A note on the one-deviation property in extensive form games Games and Economc Behavor 40 (2002) 322 338 www.academcpress.com Note A note on the one-devaton property n extensve form games Andrés Perea Departamento de Economía, Unversdad Carlos III de Madrd, Calle

More information

Hierarchies of Beliefs and the Belief-invariant. Bayesian Solution

Hierarchies of Beliefs and the Belief-invariant. Bayesian Solution Herarches of Belefs and the Belef-nvarant Bayesan Soluton Qanfeng Tang March 26, 2015 Abstract The belef-nvarant Bayesan soluton s a noton of correlated equlbrum n games wth ncomplete nformaton proposed

More information

Incomplete Information and Robustness in Strategic Environments

Incomplete Information and Robustness in Strategic Environments Unversty of Pennsylvana ScholarlyCommons Publcly Accessble Penn Dssertatons Sprng 5-17-2010 Incomplete Informaton and Robustness n Strategc Envronments Antono Penta Unversty of Pennsylvana, apenta@ssc.wsc.edu

More information

Beliefs, Plans, and Perceived Intentions in Dynamic Games.

Beliefs, Plans, and Perceived Intentions in Dynamic Games. Belefs, Plans, and Perceved Intentons n Dynamc Games. Perpaolo Battgall Department of Decson Scences and IGIER, Boccon Unversty perpaolo.battgall@unboccon.t Ncodemo De Vto Department of Decson Scences,

More information

HIERARCHIES OF BELIEF AND INTERIM RATIONALIZABILITY

HIERARCHIES OF BELIEF AND INTERIM RATIONALIZABILITY HIERARCHIES OF BELIEF AND INTERIM RATIONALIZABILITY JEFFREY C. ELY AND MARCIN PESKI Abstract. In games wth ncomplete nformaton, conventonal herarches of belef are ncomplete as descrptons of the players

More information

Genericity of Critical Types

Genericity of Critical Types Genercty of Crtcal Types Y-Chun Chen Alfredo D Tllo Eduardo Fangold Syang Xong September 2008 Abstract Ely and Pesk 2008 offers an nsghtful characterzaton of crtcal types: a type s crtcal f and only f

More information

Correlated Equilibrium in Games with Incomplete Information

Correlated Equilibrium in Games with Incomplete Information Correlated Equlbrum n Games wth Incomplete Informaton Drk Bergemann y Stephen Morrs z Frst Verson: October Current Verson: May 4, Abstract We de ne a noton of correlated equlbrum for games wth ncomplete

More information

Robust Implementation: The Role of Large Type Spaces

Robust Implementation: The Role of Large Type Spaces Robust Implementaton: The Role of Large Type Spaces Drk Bergemann y Stephen Morrs z Frst Verson: March 2003 Ths Verson: Aprl 2004 Abstract We analyze the problem of fully mplementng a socal choce functon

More information

Cowles Foundation for Research in Economics at Yale University

Cowles Foundation for Research in Economics at Yale University Cowles Foundaton for Research n Economcs at Yale Unversty Cowles Foundaton Dscusson Paper No. 666 ROBUST IMPLEMENTATION IN GENERAL MECHANISMS Drk Bergemann and Stephen Morrs Month 28 An author ndex to

More information

CS286r Assign One. Answer Key

CS286r Assign One. Answer Key CS286r Assgn One Answer Key 1 Game theory 1.1 1.1.1 Let off-equlbrum strateges also be that people contnue to play n Nash equlbrum. Devatng from any Nash equlbrum s a weakly domnated strategy. That s,

More information

Epistemic Game Theory: Beliefs and Types

Epistemic Game Theory: Beliefs and Types Epstemc Game Theory: Belefs and Types Marcano Snscalch March 28, 2007 1 Introducton John Harsany [19] ntroduced the formalsm of type spaces to provde a smple and parsmonous representaton of belef herarches.

More information

Robust Implementation: The Role of Large Type Spaces

Robust Implementation: The Role of Large Type Spaces Robust Implementaton: The Role of Large Type Spaces Drk Bergemann y Stephen Morrs z Frst Verson: March 23 Ths Verson: June 25 Abstract A socal choce functon s robustly mplemented f every equlbrum on every

More information

Topologies on Types: Connections

Topologies on Types: Connections Topologes on Types: Connectons Y-Chun Chen y Syang Xong z May 28, 2008 Abstract For d erent purposes, economsts may use d erent topologes on types. We characterze the relatonshp among these varous topologes.

More information

Difference Equations

Difference Equations Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1

More information

DIFFERENTIAL SCHEMES

DIFFERENTIAL SCHEMES DIFFERENTIAL SCHEMES RAYMOND T. HOOBLER Dedcated to the memory o Jerry Kovacc 1. schemes All rngs contan Q and are commutatve. We x a d erental rng A throughout ths secton. 1.1. The topologcal space. Let

More information

The Second Anti-Mathima on Game Theory

The Second Anti-Mathima on Game Theory The Second Ant-Mathma on Game Theory Ath. Kehagas December 1 2006 1 Introducton In ths note we wll examne the noton of game equlbrum for three types of games 1. 2-player 2-acton zero-sum games 2. 2-player

More information

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and Ths artcle appeared n a journal publshed by Elsever. The attached copy s furnshed to the author for nternal non-commercal research and educaton use, ncludng for nstructon at the authors nsttuton and sharng

More information

Hierarchies of belief and interim rationalizability

Hierarchies of belief and interim rationalizability Theoretcal Economcs 1 (2006), 19 65 1555-7561/20060019 Herarches of belef and nterm ratonalzablty JEFFREY C. ELY Department of Economcs, Northwestern Unversty MARCIN PESKI Department of Economcs, Unversty

More information

Edge Isoperimetric Inequalities

Edge Isoperimetric Inequalities November 7, 2005 Ross M. Rchardson Edge Isopermetrc Inequaltes 1 Four Questons Recall that n the last lecture we looked at the problem of sopermetrc nequaltes n the hypercube, Q n. Our noton of boundary

More information

Limited focus in dynamic games

Limited focus in dynamic games Internatonal Journal of Game Theory https://do.org/10.1007/s00182-018-0642-x ORIGINAL PAPER Lmted focus n dynamc games Andrés Perea 1 Elas Tsakas 2 Accepted: 29 September 2018 The Author(s) 2018 Abstract

More information

Games in Preference Form and Preference Rationalizability

Games in Preference Form and Preference Rationalizability Games n Preference Form and Preference Ratonalzablty Stephen Morrs Satoru Takahash September 8, 2012 Abstract We ntroduce a game n preference form, whch conssts of a game form and a preference structure,

More information

Algorithms for cautious reasoning in games

Algorithms for cautious reasoning in games Algorthms for cautous reasonng n games Ger B. Ashem a Andrés Perea b 12 October 2018 Abstract We provde comparable algorthms for the Dekel-Fudenberg procedure, terated admssblty, proper ratonalzablty and

More information

Perfect Competition and the Nash Bargaining Solution

Perfect Competition and the Nash Bargaining Solution Perfect Competton and the Nash Barganng Soluton Renhard John Department of Economcs Unversty of Bonn Adenauerallee 24-42 53113 Bonn, Germany emal: rohn@un-bonn.de May 2005 Abstract For a lnear exchange

More information

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4) I. Classcal Assumptons Econ7 Appled Econometrcs Topc 3: Classcal Model (Studenmund, Chapter 4) We have defned OLS and studed some algebrac propertes of OLS. In ths topc we wll study statstcal propertes

More information

Critical Types. September 24, Abstract

Critical Types. September 24, Abstract Crtcal Types Jeffrey C. Ely Marcn Pesk September 24, 2010 Abstract How can we know n advance whether smplfyng assumptons about belefs wll make a dfference n the conclusons of game-theoretc models? We defne

More information

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg prnceton unv. F 17 cos 521: Advanced Algorthm Desgn Lecture 7: LP Dualty Lecturer: Matt Wenberg Scrbe: LP Dualty s an extremely useful tool for analyzng structural propertes of lnear programs. Whle there

More information

Interim Rationalizability. Eddie Dekel, Drew Fudenberg and Stephen Morris. Working Paper No September, 2005

Interim Rationalizability. Eddie Dekel, Drew Fudenberg and Stephen Morris. Working Paper No September, 2005 Interm Ratonalzablty By Edde Dekel, Drew Fudenberg and Stephen Morrs Workng Paper No. 6-2005 September, 2005 The Foerder Insttute for Economc Research and The Sackler Insttute of Economc Studes Interm

More information

Payoff Information and. Self-Confirming Equilibrium 1

Payoff Information and. Self-Confirming Equilibrium 1 Payoff Informaton and Self-Confrmng Equlbrum 1 Frst verson: Aprl 25, 1995 Ths revson: July 12, 1999 Edde Deel Drew Fudenberg Davd K. Levne 2 1 We than Faru Gul, an assocate edtor, two referees, and semnar

More information

The Folk Theorem for Games with Private Almost-Perfect Monitoring

The Folk Theorem for Games with Private Almost-Perfect Monitoring The Folk Theorem for Games wth Prvate Almost-Perfect Montorng Johannes Hörner y Wojcech Olszewsk z October 2005 Abstract We prove the folk theorem for dscounted repeated games under prvate, almost-perfect

More information

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that Artcle forthcomng to ; manuscrpt no (Please, provde the manuscrpt number!) 1 Onlne Appendx Appendx E: Proofs Proof of Proposton 1 Frst we derve the equlbrum when the manufacturer does not vertcally ntegrate

More information

2.3 Nilpotent endomorphisms

2.3 Nilpotent endomorphisms s a block dagonal matrx, wth A Mat dm U (C) In fact, we can assume that B = B 1 B k, wth B an ordered bass of U, and that A = [f U ] B, where f U : U U s the restrcton of f to U 40 23 Nlpotent endomorphsms

More information

Structure and Drive Paul A. Jensen Copyright July 20, 2003

Structure and Drive Paul A. Jensen Copyright July 20, 2003 Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.

More information

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number

More information

CMS-EMS Center for Mathematical Studies in Economics And Management Science. Discussion Paper #1586

CMS-EMS Center for Mathematical Studies in Economics And Management Science. Discussion Paper #1586 CMS-EMS Center for Mathematcal Studes n Economcs And Management Scence Dscusson Paper #1586 "When Do Types Induce the Same Belef Herarchy?" Andres Perea and Wllemen Kets Northwestern Unversty December

More information

Understanding Reasoning Using Utility Proportional Beliefs

Understanding Reasoning Using Utility Proportional Beliefs Understandng Reasonng Usng Utlty Proportonal Belefs Chrstan Nauerz EpCenter, Maastrcht Unversty c.nauerz@maastrchtunversty.nl Abstract. Tradtonally very lttle attenton has been pad to the reasonng process

More information

Constant Best-Response Functions: Interpreting Cournot

Constant Best-Response Functions: Interpreting Cournot Internatonal Journal of Busness and Economcs, 009, Vol. 8, No., -6 Constant Best-Response Functons: Interpretng Cournot Zvan Forshner Department of Economcs, Unversty of Hafa, Israel Oz Shy * Research

More information

20. Mon, Oct. 13 What we have done so far corresponds roughly to Chapters 2 & 3 of Lee. Now we turn to Chapter 4. The first idea is connectedness.

20. Mon, Oct. 13 What we have done so far corresponds roughly to Chapters 2 & 3 of Lee. Now we turn to Chapter 4. The first idea is connectedness. 20. Mon, Oct. 13 What we have done so far corresponds roughly to Chapters 2 & 3 of Lee. Now we turn to Chapter 4. The frst dea s connectedness. Essentally, we want to say that a space cannot be decomposed

More information

(1 ) (1 ) 0 (1 ) (1 ) 0

(1 ) (1 ) 0 (1 ) (1 ) 0 Appendx A Appendx A contans proofs for resubmsson "Contractng Informaton Securty n the Presence of Double oral Hazard" Proof of Lemma 1: Assume that, to the contrary, BS efforts are achevable under a blateral

More information

Implementation in Mixed Nash Equilibrium

Implementation in Mixed Nash Equilibrium Department of Economcs Workng Paper Seres Implementaton n Mxed Nash Equlbrum Claudo Mezzett & Ludovc Renou May 2012 Research Paper Number 1146 ISSN: 0819-2642 ISBN: 978 0 7340 4496 9 Department of Economcs

More information

Affine transformations and convexity

Affine transformations and convexity Affne transformatons and convexty The purpose of ths document s to prove some basc propertes of affne transformatons nvolvng convex sets. Here are a few onlne references for background nformaton: http://math.ucr.edu/

More information

Continuous Implementation

Continuous Implementation Contnuous Implementaton Maron Oury HEC Pars Olver Terceux Pars School of Economcs and CNRS Abstract It s well-known that mechansm desgn lterature makes many smplfyng nformatonal assumptons n partcular

More information

APPENDIX A Some Linear Algebra

APPENDIX A Some Linear Algebra APPENDIX A Some Lnear Algebra The collecton of m, n matrces A.1 Matrces a 1,1,..., a 1,n A = a m,1,..., a m,n wth real elements a,j s denoted by R m,n. If n = 1 then A s called a column vector. Smlarly,

More information

Bayesian predictive Configural Frequency Analysis

Bayesian predictive Configural Frequency Analysis Psychologcal Test and Assessment Modelng, Volume 54, 2012 (3), 285-292 Bayesan predctve Confgural Frequency Analyss Eduardo Gutérrez-Peña 1 Abstract Confgural Frequency Analyss s a method for cell-wse

More information

THE CHINESE REMAINDER THEOREM. We should thank the Chinese for their wonderful remainder theorem. Glenn Stevens

THE CHINESE REMAINDER THEOREM. We should thank the Chinese for their wonderful remainder theorem. Glenn Stevens THE CHINESE REMAINDER THEOREM KEITH CONRAD We should thank the Chnese for ther wonderful remander theorem. Glenn Stevens 1. Introducton The Chnese remander theorem says we can unquely solve any par of

More information

Endogenous timing in a mixed oligopoly consisting of a single public firm and foreign competitors. Abstract

Endogenous timing in a mixed oligopoly consisting of a single public firm and foreign competitors. Abstract Endogenous tmng n a mxed olgopoly consstng o a sngle publc rm and oregn compettors Yuanzhu Lu Chna Economcs and Management Academy, Central Unversty o Fnance and Economcs Abstract We nvestgate endogenous

More information

A Folk Theorem For Stochastic Games with Finite Horizon.

A Folk Theorem For Stochastic Games with Finite Horizon. A Folk Theorem For Stochastc Games wth Fnte Horzon. Chantal Marlats January 11, 2010 Abstract Ths paper provdes condtons for a lmt folk theorem to hold n stochastc games wth nte horzon. If asymptotc assumptons

More information

A SURVEY OF PROPERTIES OF FINITE HORIZON DIFFERENTIAL GAMES UNDER ISAACS CONDITION. Contents

A SURVEY OF PROPERTIES OF FINITE HORIZON DIFFERENTIAL GAMES UNDER ISAACS CONDITION. Contents A SURVEY OF PROPERTIES OF FINITE HORIZON DIFFERENTIAL GAMES UNDER ISAACS CONDITION BOTAO WU Abstract. In ths paper, we attempt to answer the followng questons about dfferental games: 1) when does a two-player,

More information

Games of Threats. Elon Kohlberg Abraham Neyman. Working Paper

Games of Threats. Elon Kohlberg Abraham Neyman. Working Paper Games of Threats Elon Kohlberg Abraham Neyman Workng Paper 18-023 Games of Threats Elon Kohlberg Harvard Busness School Abraham Neyman The Hebrew Unversty of Jerusalem Workng Paper 18-023 Copyrght 2017

More information

An Explicit Approach to Modeling Finite-Order Type Spaces and Applications

An Explicit Approach to Modeling Finite-Order Type Spaces and Applications An Explct Approach to Modelng Fnte-Order Type Spaces and Applcatons Cheng-Zhong Qn and Chun-Le Yang November 18, 2009 Abstract Every abstract type of a belef-closed type space corresponds to an nfnte belef

More information

Every planar graph is 4-colourable a proof without computer

Every planar graph is 4-colourable a proof without computer Peter Dörre Department of Informatcs and Natural Scences Fachhochschule Südwestfalen (Unversty of Appled Scences) Frauenstuhlweg 31, D-58644 Iserlohn, Germany Emal: doerre(at)fh-swf.de Mathematcs Subject

More information

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:

More information

RATIONALIZABLE IMPLEMENTATION. Dirk Bergemann and Stephen Morris. May 2009 COWLES FOUNDATION DISCUSSION PAPER NO. 1697

RATIONALIZABLE IMPLEMENTATION. Dirk Bergemann and Stephen Morris. May 2009 COWLES FOUNDATION DISCUSSION PAPER NO. 1697 RATIONALIZABLE IMPLEMENTATION By Drk Bergemann and Stephen Morrs May 2009 COWLES FOUNDATION DISCUSSION PAPER NO. 1697 COWLES FOUNDATION FOR RESEARCH IN ECONOMICS YALE UNIVERSITY Box 208281 New Haven, Connectcut

More information

42. Mon, Dec. 8 Last time, we were discussing CW complexes, and we considered two di erent CW structures on S n. We continue with more examples.

42. Mon, Dec. 8 Last time, we were discussing CW complexes, and we considered two di erent CW structures on S n. We continue with more examples. 42. Mon, Dec. 8 Last tme, we were dscussng CW complexes, and we consdered two d erent CW structures on S n. We contnue wth more examples. (2) RP n. Let s start wth RP 2. Recall that one model for ths space

More information

Comprehensive Rationalizability

Comprehensive Rationalizability Comprehensve Ratonalzablty Avad Hefetz Martn Meer y Burkhard C. Schpper z Prelmnary & Incomplete: August 12, 2010 Abstract We present a new soluton concept for strategc games called comprehensve ratonalzablty.

More information

The Order Relation and Trace Inequalities for. Hermitian Operators

The Order Relation and Trace Inequalities for. Hermitian Operators Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence

More information

The Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction

The Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction ECONOMICS 5* -- NOTE (Summary) ECON 5* -- NOTE The Multple Classcal Lnear Regresson Model (CLRM): Specfcaton and Assumptons. Introducton CLRM stands for the Classcal Lnear Regresson Model. The CLRM s also

More information

Online Appendix: Reciprocity with Many Goods

Online Appendix: Reciprocity with Many Goods T D T A : O A Kyle Bagwell Stanford Unversty and NBER Robert W. Stager Dartmouth College and NBER March 2016 Abstract Ths onlne Appendx extends to a many-good settng the man features of recprocty emphaszed

More information

Foundations of Arithmetic

Foundations of Arithmetic Foundatons of Arthmetc Notaton We shall denote the sum and product of numbers n the usual notaton as a 2 + a 2 + a 3 + + a = a, a 1 a 2 a 3 a = a The notaton a b means a dvdes b,.e. ac = b where c s an

More information

Bounded Reasoning and Higher-Order Uncertainty

Bounded Reasoning and Higher-Order Uncertainty Bounded Reasonng and Hgher-Order Uncertanty Wllemen Kets September 4, 2014 Frst verson: November 2011 Abstract Expermental evdence suggests that ndvduals cannot reason about ther opponents belefs up to

More information

COWLES FOUNDATION FOR RESEARCH IN ECONOMICS YALE UNIVERSITY

COWLES FOUNDATION FOR RESEARCH IN ECONOMICS YALE UNIVERSITY ROBUST MECHANISM DESIGN By Drk Bergemann and Stephen Morrs Aprl 2004 COWLES FOUNDATION DISCUSSION PAPER NO. 1421R COWLES FOUNDATION FOR RESEARCH IN ECONOMICS YALE UNIVERSITY Box 208281 New Haven, Connectcut

More information

k t+1 + c t A t k t, t=0

k t+1 + c t A t k t, t=0 Macro II (UC3M, MA/PhD Econ) Professor: Matthas Kredler Fnal Exam 6 May 208 You have 50 mnutes to complete the exam There are 80 ponts n total The exam has 4 pages If somethng n the queston s unclear,

More information

Volume 31, Issue 1. The Stackelberg equilibrium as a consistent conjectural equilibrium

Volume 31, Issue 1. The Stackelberg equilibrium as a consistent conjectural equilibrium Volume 3, Issue The Stackelberg equlbrum as a consstent conjectural equlbrum Ludovc A. Julen LEG, Unversté de Bourgogne Olver Musy EconomX, Unversté Pars Ouest-Nanterre La Défense Aurélen W. Sad Informaton

More information

Economics 101. Lecture 4 - Equilibrium and Efficiency

Economics 101. Lecture 4 - Equilibrium and Efficiency Economcs 0 Lecture 4 - Equlbrum and Effcency Intro As dscussed n the prevous lecture, we wll now move from an envronment where we looed at consumers mang decsons n solaton to analyzng economes full of

More information

Bounded Reasoning and Higher-Order Uncertainty

Bounded Reasoning and Higher-Order Uncertainty Bounded Reasonng and Hgher-Order Uncertanty Wllemen Kets February 1, 2012 Abstract Standard models of games wth ncomplete nformaton assume that players form belefs about ther opponents belefs about ther

More information

Appendix for Causal Interaction in Factorial Experiments: Application to Conjoint Analysis

Appendix for Causal Interaction in Factorial Experiments: Application to Conjoint Analysis A Appendx for Causal Interacton n Factoral Experments: Applcaton to Conjont Analyss Mathematcal Appendx: Proofs of Theorems A. Lemmas Below, we descrbe all the lemmas, whch are used to prove the man theorems

More information

Physics 5153 Classical Mechanics. D Alembert s Principle and The Lagrangian-1

Physics 5153 Classical Mechanics. D Alembert s Principle and The Lagrangian-1 P. Guterrez Physcs 5153 Classcal Mechancs D Alembert s Prncple and The Lagrangan 1 Introducton The prncple of vrtual work provdes a method of solvng problems of statc equlbrum wthout havng to consder the

More information

Tit-For-Tat Equilibria in Discounted Repeated Games with. Private Monitoring

Tit-For-Tat Equilibria in Discounted Repeated Games with. Private Monitoring 1 Tt-For-Tat Equlbra n Dscounted Repeated Games wth Prvate Montorng Htosh Matsushma 1 Department of Economcs, Unversty of Tokyo 2 Aprl 24, 2007 Abstract We nvestgate nfntely repeated games wth mperfect

More information

Higher-Order Expectations *

Higher-Order Expectations * Hgher-Order Expectatons * Benjamn Golub Stephen Morrs August 31, 2017 Abstract We study hgher-order expectatons parallelng the Harsany (1968) approach to hgher-order belefs takng a basc set of random varables

More information

UNIVERSITY OF NOTTINGHAM. Extensive Games of Imperfect Recall and Mind Perfection

UNIVERSITY OF NOTTINGHAM. Extensive Games of Imperfect Recall and Mind Perfection UNIVERSITY OF NOTTINGHAM DISUSSION PAPERS IN EONOMIS No. 98/ Extensve Games of Imperfect Recall and Mnd Perfecton Matt Ayres Abstract In ths paper we examne how the addton of mperfect recall as a perturbaton

More information

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009 College of Computer & Informaton Scence Fall 2009 Northeastern Unversty 20 October 2009 CS7880: Algorthmc Power Tools Scrbe: Jan Wen and Laura Poplawsk Lecture Outlne: Prmal-dual schema Network Desgn:

More information

Anti-van der Waerden numbers of 3-term arithmetic progressions.

Anti-van der Waerden numbers of 3-term arithmetic progressions. Ant-van der Waerden numbers of 3-term arthmetc progressons. Zhanar Berkkyzy, Alex Schulte, and Mchael Young Aprl 24, 2016 Abstract The ant-van der Waerden number, denoted by aw([n], k), s the smallest

More information

12 MATH 101A: ALGEBRA I, PART C: MULTILINEAR ALGEBRA. 4. Tensor product

12 MATH 101A: ALGEBRA I, PART C: MULTILINEAR ALGEBRA. 4. Tensor product 12 MATH 101A: ALGEBRA I, PART C: MULTILINEAR ALGEBRA Here s an outlne of what I dd: (1) categorcal defnton (2) constructon (3) lst of basc propertes (4) dstrbutve property (5) rght exactness (6) localzaton

More information

The RepeatedPrisoners DilemmawithPrivate Monitoring: a N-player case

The RepeatedPrisoners DilemmawithPrivate Monitoring: a N-player case The RepeatedPrsoners DlemmawthPrvate Montorng: a N-player case Ichro Obara Department of Economcs Unversty of Pennsylvana obara@ssc.upenn.edu Frst verson: December, 1998 Ths verson: Aprl, 2000 Abstract

More information

Analysis of Information Feedback and Selfcon rming Equilibrium

Analysis of Information Feedback and Selfcon rming Equilibrium Analyss of Informaton Feedback and Selfcon rmng Equlbrum P. Battgall, S. Cerrea-Voglo, F. Maccheron, M. Marnacc Unverstà Boccon Draft of May 2013 Abstract Recent work of us (Battgall, Cerrea-Voglo, Maccheron

More information

Game Theory. Lecture Notes By Y. Narahari. Department of Computer Science and Automation Indian Institute of Science Bangalore, India February 2008

Game Theory. Lecture Notes By Y. Narahari. Department of Computer Science and Automation Indian Institute of Science Bangalore, India February 2008 Game Theory Lecture Notes By Y. Narahar Department of Computer Scence and Automaton Indan Insttute of Scence Bangalore, Inda February 2008 Chapter 10: Two Person Zero Sum Games Note: Ths s a only a draft

More information

Folk Theorem in Stotchastic Games with Private State and Private Monitoring Preliminary: Please do not circulate without permission

Folk Theorem in Stotchastic Games with Private State and Private Monitoring Preliminary: Please do not circulate without permission Folk Theorem n Stotchastc Games wth Prvate State and Prvate Montorng Prelmnary: Please do not crculate wthout permsson Takuo Sugaya Stanford Graduate School of Busness December 9, 202 Abstract We show

More information

Numerical Heat and Mass Transfer

Numerical Heat and Mass Transfer Master degree n Mechancal Engneerng Numercal Heat and Mass Transfer 06-Fnte-Dfference Method (One-dmensonal, steady state heat conducton) Fausto Arpno f.arpno@uncas.t Introducton Why we use models and

More information

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix Lectures - Week 4 Matrx norms, Condtonng, Vector Spaces, Lnear Independence, Spannng sets and Bass, Null space and Range of a Matrx Matrx Norms Now we turn to assocatng a number to each matrx. We could

More information

Problem Set 9 Solutions

Problem Set 9 Solutions Desgn and Analyss of Algorthms May 4, 2015 Massachusetts Insttute of Technology 6.046J/18.410J Profs. Erk Demane, Srn Devadas, and Nancy Lynch Problem Set 9 Solutons Problem Set 9 Solutons Ths problem

More information

COS 521: Advanced Algorithms Game Theory and Linear Programming

COS 521: Advanced Algorithms Game Theory and Linear Programming COS 521: Advanced Algorthms Game Theory and Lnear Programmng Moses Charkar February 27, 2013 In these notes, we ntroduce some basc concepts n game theory and lnear programmng (LP). We show a connecton

More information

Common Belief Foundations of Global Games

Common Belief Foundations of Global Games Common Belef Foundatons of Global Games Stephen Morrs Prnceton Unversty smorrs@prnceton.edu Hyun Song Shn Prnceton Unversty hsshn@prnceton.edu October 2007 Abstract We provde a characterzaton of when an

More information

j) = 1 (note sigma notation) ii. Continuous random variable (e.g. Normal distribution) 1. density function: f ( x) 0 and f ( x) dx = 1

j) = 1 (note sigma notation) ii. Continuous random variable (e.g. Normal distribution) 1. density function: f ( x) 0 and f ( x) dx = 1 Random varables Measure of central tendences and varablty (means and varances) Jont densty functons and ndependence Measures of assocaton (covarance and correlaton) Interestng result Condtonal dstrbutons

More information

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton

More information

Module 9. Lecture 6. Duality in Assignment Problems

Module 9. Lecture 6. Duality in Assignment Problems Module 9 1 Lecture 6 Dualty n Assgnment Problems In ths lecture we attempt to answer few other mportant questons posed n earler lecture for (AP) and see how some of them can be explaned through the concept

More information

A combinatorial problem associated with nonograms

A combinatorial problem associated with nonograms A combnatoral problem assocated wth nonograms Jessca Benton Ron Snow Nolan Wallach March 21, 2005 1 Introducton. Ths work was motvated by a queston posed by the second named author to the frst named author

More information

NP-Completeness : Proofs

NP-Completeness : Proofs NP-Completeness : Proofs Proof Methods A method to show a decson problem Π NP-complete s as follows. (1) Show Π NP. (2) Choose an NP-complete problem Π. (3) Show Π Π. A method to show an optmzaton problem

More information

Power law and dimension of the maximum value for belief distribution with the max Deng entropy

Power law and dimension of the maximum value for belief distribution with the max Deng entropy Power law and dmenson of the maxmum value for belef dstrbuton wth the max Deng entropy Bngy Kang a, a College of Informaton Engneerng, Northwest A&F Unversty, Yanglng, Shaanx, 712100, Chna. Abstract Deng

More information

4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA

4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA 4 Analyss of Varance (ANOVA) 5 ANOVA 51 Introducton ANOVA ANOVA s a way to estmate and test the means of multple populatons We wll start wth one-way ANOVA If the populatons ncluded n the study are selected

More information

Inner Product. Euclidean Space. Orthonormal Basis. Orthogonal

Inner Product. Euclidean Space. Orthonormal Basis. Orthogonal Inner Product Defnton 1 () A Eucldean space s a fnte-dmensonal vector space over the reals R, wth an nner product,. Defnton 2 (Inner Product) An nner product, on a real vector space X s a symmetrc, blnear,

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 12 10/21/2013. Martingale Concentration Inequalities and Applications

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 12 10/21/2013. Martingale Concentration Inequalities and Applications MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.65/15.070J Fall 013 Lecture 1 10/1/013 Martngale Concentraton Inequaltes and Applcatons Content. 1. Exponental concentraton for martngales wth bounded ncrements.

More information

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals Smultaneous Optmzaton of Berth Allocaton, Quay Crane Assgnment and Quay Crane Schedulng Problems n Contaner Termnals Necat Aras, Yavuz Türkoğulları, Z. Caner Taşkın, Kuban Altınel Abstract In ths work,

More information

More metrics on cartesian products

More metrics on cartesian products More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of

More information

Vapnik-Chervonenkis theory

Vapnik-Chervonenkis theory Vapnk-Chervonenks theory Rs Kondor June 13, 2008 For the purposes of ths lecture, we restrct ourselves to the bnary supervsed batch learnng settng. We assume that we have an nput space X, and an unknown

More information

On C 0 multi-contractions having a regular dilation

On C 0 multi-contractions having a regular dilation SUDIA MAHEMAICA 170 (3) (2005) On C 0 mult-contractons havng a regular dlaton by Dan Popovc (mşoara) Abstract. Commutng mult-contractons of class C 0 and havng a regular sometrc dlaton are studed. We prove

More information

Ryan (2009)- regulating a concentrated industry (cement) Firms play Cournot in the stage. Make lumpy investment decisions

Ryan (2009)- regulating a concentrated industry (cement) Firms play Cournot in the stage. Make lumpy investment decisions 1 Motvaton Next we consder dynamc games where the choce varables are contnuous and/or dscrete. Example 1: Ryan (2009)- regulatng a concentrated ndustry (cement) Frms play Cournot n the stage Make lumpy

More information

Common Belief Foundations of Global Games

Common Belief Foundations of Global Games Common Belef Foundatons of Global Games Stephen Morrs Prnceton Unversty Hyun Song Shn Bank for Internatonal Settlements November 2015 Muhamet Yldz M.I.T. Abstract We study coordnaton games under general

More information