A resurrection of the Condorcet Jury Theorem

Size: px
Start display at page:

Download "A resurrection of the Condorcet Jury Theorem"

Transcription

1 Theoretcal Economcs 4 (2009), / A resurrecton of the Condorcet Jury Theorem YUKIO KORIYAMA Département d Économe, École Polytechnque, Palaseau BALÁZS SZENTES Department of Economcs, Unversty College London Ths paper analyzes the optmal sze of a delberatng commttee where () there s no conflct of nterest among ndvduals and () nformaton acquston s costly. The commttee members smultaneously decde whether to acqure nformaton, and then make the ex-post effcent decson. The optmal commttee sze, k, s shown to be bounded. The man result of ths paper s that any arbtrarly large commttee aggregates the decentralzed nformaton more effcently than the commttee of sze k 2. Ths result mples that overszed commttees generate only small neffcences. KEYWORDS. Votng, nformaton aggregaton, costly nformaton. JEL CLASSIFICATION. D72, D INTRODUCTION The classcal Condorcet Jury Theorem (CJT) states that large commttees can aggregate decentralzed nformaton more effcently than small ones. Its orgn can be traced to the dawn of the French Revoluton, when Mare-Jean-Antone-Ncolas de Cartat, marqus de Condorcet (1785) nvestgated the decson-makng processes n socetes. 1 Recent lterature on commttee desgn has ponted out that f nformaton acquston s costly, the CJT fals to hold. The reasonng s that f the sze of a commttee s large, a commttee member realzes that the probablty that she can nfluence the fnal decson s small compared to the cost of nformaton acquston. As a result, she mght prefer to avod ths cost and free-rde on the nformaton of others. Therefore, large commttees mght generate lower socal welfare than small ones. These results suggest that n the presence of costly nformaton acquston, optmally choosng the sze of a commttee s both an mportant and delcate ssue. In ths paper, we characterze the Yuko Koryama: yuko.koryama@polytechnque.edu Balázs Szentes: b.szentes@ucl.ac.uk The authors are grateful to semnar partcpants at the Unversty of Chcago, Berkeley, Unversty of Rochester, École Polytechnque, Arzona State Unversty, Maastrcht Unversty, Unversty of Hawa, and Unversty College London for helpful comments. Fnancal support from the NSF s gratefully acknowledged. 1 Summares of the hstory of the CJT can be found n, for example, Grofman and Owen (1986), Mller (1986), and Gerlng et al. (2005). Copyrght c 2009 Yuko Koryama and Balázs Szentes. Lcensed under the Creatve Commons Attrbuton- NonCommercal Lcense 3.0. Avalable at

2 228 Koryama and Szentes Theoretcal Economcs 4 (2009) welfare loss assocated wth overszed commttees and show that ths loss s surprsngly small n certan envronments. Therefore, as long as the commttee sze s large enough, the careful desgn of a commttee mght not be as mportant as t was orgnally thought to be. In fact, f ether the nformaton structure s ambguous, or the commttee has to make decsons n varous nformatonal envronments, t mght be optmal to desgn the commttee to be as large as possble. Commttee desgn receves consderable attenton from economsts because, n many stuatons, groups rather than ndvduals make decsons. Informaton about the desrablty of the possble decsons s often decentralzed: ndvdual group members must separately acqure costly nformaton about the alternatves. A classcal example s a jury tral where a jury has to decde whether a defendant s gulty or nnocent. Each juror ndvdually obtans some nformaton about the defendant, at some cost of effort (payng attenton to the tral, nvestgatng evdence, etc.). Lkewse, when a frm s facng the decson whether to mplement a project, each member of the executve commttee can collect nformaton about the proftablty of the project (by spendng tme and exertng effort). Yet another example s the decson of an academc department to hre a new member. Each member of the recrutng commttee must revew the applcatons ndvdually before makng a collectve decson. What these examples have n common s the fact that nformaton acquston s costly and often unobservable. The exact setup analyzed n ths paper s descrbed as follows. A group of ndvduals has to make a bnary decson. There s no conflct of nterest among the group members, but they have mperfect nformaton about whch decson s best. Frst, k ndvduals are asked to serve on a commttee. Then the commttee members smultaneously decde whether to nvest n an nformatve sgnal. Fnally, the commttee makes the optmal decson gven the acqured nformaton. We do not explctly model how the commttee members communcate and aggregate nformaton. Instead, we smply assume that they end up makng the ex post effcent decson. 2 The only strategc choce an ndvdual must make n our model s whether to acqure a sgnal upon beng selected to serve on the commttee. The central queston of our paper s the followng: how does the commttee sze, k, affect socal welfare? Frst, for each k, we fully characterze the set of equlbra (ncludng asymmetrc and mxed-strategy equlbra). We show that there exsts k P ( ) such that whenever k k P, there s a unque equlbrum n whch each commttee member nvests n nformaton wth probablty one. Furthermore, the socal welfare generated by these equlbra s an ncreasng functon of k. If k > k P, then there are multple equlbra, many of whch nvolve randomzaton by the members. We show also that the socal welfare generated by the worst equlbrum n the game, where the commttee sze s k, s decreasng n k f k > k P. The optmal commttee sze k s defned as the smallest commttee sze such that there s an equlbrum n the commttee wth k members that maxmzes socal welfare among all equlbra n any commttee. We 2 Snce there s no conflct of nterest among the ndvduals, t s easy to desgn a mechansm that s ncentve compatble and effcently aggregates the sgnals. Alternatvely, one can assume that the collected nformaton s hard.

3 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 229 prove that the optmal commttee sze s ether k P or k P + 1. Ths mples that the CJT fals to hold: large commttees can generate smaller socal welfare than small commttees. We show, nevertheless, that f the sze of the commttee s larger than k, even the worst equlbrum generates hgher socal welfare than the unque equlbrum n the commttee of sze k 2. That s, the welfare loss due to an overszed commttee s qute small. For our results to hold, we need an assumpton on the dstrbuton of the sgnals. Ths assumpton, stated formally n the next secton, requres the margnal beneft from a sgnal to decrease fast. Lterature Revew Although the Condorcet Jury Theorem provdes mportant support for the bass of democratc decson makng, many of the premses of the theorem have been crtczed. Perhaps most mportantly, Condorcet assumes sncere votng. That s, each ndvdual votes as f she were the only voter n the socety. Ths means that an ndvdual votes for the alternatve that s best, condtonal on her sgnal. Austen-Smth and Banks (1996) show that n general, sncere votng s not consstent wth equlbrum behavor. Ths s because a ratonal ndvdual votes condtonal not only on her sgnal, but also on her beng pvotal. Feddersen and Pesendorfer (1998) show that as the jury sze ncreases, the probablty of convctng an nnocent can actually ncrease under unanmty rule. A varety of papers show, however, that even f the voters are strategc, n certan envronments the outcome of votng converges to the effcent outcome as the number of voters goes to nfnty. Feddersen and Pesendorfer (1997) nvestgate a model n whch preferences are heterogeneous and each voter has a prvate sgnal concernng whch alternatve s best. They construct an equlbrum for each populaton sze, such that the equlbrum outcome converges to the full nformaton outcome as the number of voters goes to nfnty. The full nformaton outcome s defned as the result of a votng game where all nformaton s publc. Myerson (1998) shows that asymptotc effcency can be acheved even f there s populaton uncertanty; that s, f a voter does not know how many other voters there are. In contrast, the Condorcet Jury Theorem mght fal to hold f the nformaton acquston s costly. Mukhopadhaya (2003) consders a model smlar to ours where voters have dentcal preferences, but nformaton acquston s costly. He shows by numercal examples that mxed-strategy equlbra n large commttees may generate lower expected welfare than pure-strategy equlbra n small commttees. 3 Martnell (2006) also ntroduces a cost of nformaton acquston nto a model where sgnals are bnary. He allows the precson of the sgnals to depend contnuously on the amount of nvestment and proves that f the cost and the margnal cost of the precson are zero at the zero level of precson, then the decson s asymptotcally effcent. More precsely, f the sze of the commttee converges to nfnty, then there s 3 The results are qute dfferent f the votng, rather than the nformaton acquston, s costly: see e.g. Börgers (2004).

4 230 Koryama and Szentes Theoretcal Economcs 4 (2009) a sequence of symmetrc equlbra n whch each member nvests only a lttle, and the probablty of a correct decson converges to one. 4 We thnk that Martnell (2006) contrbutes substantally towards the understandng of the effcency propertes of group decson makng when there s no fxed cost assocated wth nformaton acquston. However, we beleve that the exstence of a fxed cost s an essental feature of many envronments. Indeed, one has to pay the prce of a newspaper, even f t wll be thrown away later. The management of a company has to pay for a consultant, even f the consultant s work wll be completely gnored. Smlarly, a juror has to st through a tral, even f he decdes not to pay any attenton. What these examples have n common s that a postve cost must be nvested n the sgnals even f the precson s zero. These examples also suggest that exertng more effort mght lead to more accurate nformaton, though ths s not modeled n our paper. Hence, we thnk that the result n our paper s an mportant complement to Martnell (2006). If nformaton acquston has fxed as well as varable costs, then an asymptotc effcency result, lke the one n Martnell (2006), does not hold. However, we conjecture that our result s stll vald n some form. That s, the effcency loss due to large commttees s small. 5 Numerous papers analyze the optmal decson rules n the presence of costly nformaton. Persco (2004) dscusses the relatonshp between the optmal decson rules and the accuracy of the sgnals. He shows that a votng rule that requres a strong consensus n order to upset the status quo s optmal only f the sgnals are suffcently accurate. The ntuton for the extreme case, where the decson rule s unanmty rule, s the followng. Under unanmty rule, the probablty of beng pvotal s small. However, ths probablty ncreases as the sgnals become more accurate. Therefore, n order to provde a voter wth an ncentve to nvest n nformaton, the sgnals must be suffcently accurate. L (2001), Gerard and Yarv (2008), and Gershkov and Szentes (forthcomng) show that the optmal votng mechansm sometmes nvolves ex post neffcent decsons. That s, the optmal mechansm mght specfy neffcent decsons for certan sgnal profles. We beleve that n many stuatons such a commtment devce s not avalable, whch s why we smply restrct attenton to ex post effcent decson rules. We beleve that ths s the approprate assumpton n the context of a delberatng commttee n whch there s no conflct of nterest among ndvduals. Secton 2 descrbes the model. The man theorems are stated and proved n Secton 3. Secton 4 concludes. Some of the proofs are relegated to the Appendx. 2. THE MODEL There s a populaton consstng of N (> 1) ndvduals. The state of the world, ω, can take one of two values, 1 and 1, wth Pr[ω = 1] = π (0, 1). The socety must make a 4 In hs accompanyng paper, Martnell (2007) analyzes a model n whch nformaton has a fxed cost, voters are heterogeneous n ther costs, and abstenton s not allowed. He shows that f, on the one hand, the support of the cost dstrbuton s not bounded away from zero, asymptotc effcency can be acheved. If, on the other hand, the cost s bounded away from zero, and the number of voters s large, nobody acqures nformaton n any equlbrum. 5 The dffculty of solvng these models s that t s partcularly hard to characterze the set of all equlbra.

5 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 231 decson, d, whch s ether 1 or 1. There s no conflct of nterest among ndvduals. Each ndvdual has a beneft of u (d,ω) f decson d s made when the state of the world s ω. In partcular, 0 f d = ω u (d,ω) = q f d = 1 and ω = 1 (1 q) f d = 1 and ω = 1, where q ( (0, 1)), ndcates the severty of a type-i error. 6 Each ndvdual can purchase a sgnal at a cost c (> 0) at most once. Sgnals are d across ndvduals, condtonal on the realzaton of the state of the world. The ex post payoff of an ndvdual who nvests n nformaton s u c. Each ndvdual maxmzes her expected payoff. There are two stages of the decson-makng process. At Stage 1, k ( N ) members of the socety are desgnated at random to serve on a commttee. At Stage 2, the commttee members decde smultaneously and ndependently whether or not to nvest n nformaton. Then, the effcent decson s made gven the sgnals collected by the members. We do not model explctly how commttee members delberate at Stage 2. Snce there s no conflct of nterest among the members, t s easy to desgn a communcaton protocol that effcently aggregates nformaton. Alternatvely, one can assume that the acqured nformaton s hard. Hence, no communcaton s necessary for makng the ex post effcent decson. We focus merely on the commttee members ncentves to acqure nformaton. Next, we turn to the defnton of socal welfare. Frst, let µ denote the ex post effcent decson rule. That s, µ s a mappng from sets of sgnals nto possble decsons. If the sgnal profle s (s 1,..., s n ), where n s the number of sgnals acqured, then µ(s 1,..., s n ) = 1 ω [u (1,ω) s 1,..., s n ] ω [u ( 1,ω) s 1,..., s n ]. Socal welfare s measured by the expected sum of the payoffs of the ndvduals, s1,...,s n,ω[n u (µ(s 1,..., s n ),ω) c n], where the expectaton also takes nto account possble randomzaton by the ndvduals. That s, n can be a random varable. If the commttee s large, then a member mght prefer to save the cost of nformaton acquston and choose to rely on the opnons of others. By contrast, f k s too small, there s too lttle nformaton to aggregate, and thus the fnal decson s lkely to be neffcent. The questons are: What value of k maxmzes ex-ante socal welfare? How bg s the welfare loss n sub-optmal commttees? Next, we defne the optmal commttee sze formally. 6 In the jury context, where ω = 1 corresponds to the nnocence of the suspect, q ndcates the severty of the error of convctng an nnocent.

6 232 Koryama and Szentes Theoretcal Economcs 4 (2009) DEFINITION. The optmal sze of the commttee s the smallest commttee sze k ( ) such that there s an equlbrum n the commttee wth k members that maxmzes socal welfare among all equlbra n any commttee. In our model, the optmal commttee sze s k f () there s an equlbrum n the commttee wth k members that maxmzes socal welfare among all equlbra n any commttee, and () each member acqures nformaton wth postve probablty n ths equlbrum. Ths s because a member who does not nvest n nformaton can be elmnated from the commttee wthout changng the ncentves of the other members. We compute socal welfare n all equlbra n suboptmal commttees and compare the welfare loss n these commttees. Snce the sgnals are d condtonal on the state of the world, the expected beneft of an ndvdual from the ex post effcent decson s a functon of the number of sgnals acqured. We defne ths functon as follows: η(n) = s1,...,s n,ω[u (µ(s 1,..., s n ),ω)]. We assume that the sgnals are nformatve about the state of the world, but only mperfectly so. That s, as the number of sgnals goes to nfnty, the probablty of makng the correct decson s strctly ncreasng and converges to one. Formally, the functon η s strctly ncreasng and lm n η(n) = 0. An ndvdual s margnal beneft from collectng an addtonal sgnal, when n sgnals are already obtaned, s g (n) = η(n + 1) η(n). Note that lm n g (n) = 0. For our man theorem to hold, we need the followng assumpton. ASSUMPTION 1. The functon g s log-convex. 7 That s, g (n + 1)/g (n) s ncreasng n n ( ). Whether or not ths assumpton s satsfed depends only on the prmtves of the model that s, on the dstrbuton of the sgnals and on the parameters q and π. An mmedate consequence of the assumpton s the followng. REMARK 1. The functon g s decreasng. PROOF. Suppose to the contrary that there exsts an nteger n 0 such that g (n 0 +1) > g (n 0 ). Snce g (n + 1)/g (n) s ncreasng n n, t follows that g (n + 1) > g (n) whenever n n 0. Hence g (n) > g (n 0 ) > 0 whenever n > n 0. Ths mples that lm n g (n) 0, whch s a contradcton. Next, we explan that Assumpton 1 essentally means that the margnal value of a sgnal decreases rapdly. Notce that the functon g beng decreasng means that the 7 The standard defnton of convex functons requres the functons to have convex domans. Ths defnton, however, can be naturally extended to functons wth non-convex domans by requrng the convexty nequalty to hold only on the domans (see Peters and Wakker 1987).

7 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 233 margnal socal value of an addtonal sgnal s decreasng. We thnk that ths assumpton s satsfed n most economc and poltcal applcatons. How much more does Assumpton 1 requre? Snce g s decreasng and lm n g (n) = 0, there always exsts an ncreasng sequence {n m } m=1 such that g (n m ) g (n m + 1) s decreasng n m. Hence, t s stll natural to restrct attenton to nformaton structures where the second dfference n the socal value of a sgnal, g (n) g (n + 1), s decreasng. Recall that Assumpton 1 s equvalent to (g (n) g (n +1))/g (n) beng decreasng. That s, Assumpton 1 requres that the second dfference n the value of a sgnal not only decreases, but does so at an ncreasng rate. In general, t s hard to check whether ths assumpton holds because t s often dffcult (or mpossble) to express g (n) analytcally. The next secton provdes examples where the assumpton s satsfed. 2.1 Examples for the log-convexty assumpton Frst, suppose that the sgnals are normally dstrbuted around the true state of the world. The log-convexty assumpton s satsfed for the model where π + q = 1. That s, the socety would be ndfferent between the two possble decsons f nformaton acquston were mpossble. The assumpton s also satsfed even f π + q 1, f the sgnals are suffcently precse. Formally, we have the followng result, whch, lke the other results n ths secton, s proved n the Appendx. PROPOSITION 1. Suppose that s N (ω,σ). () If q + π = 1 then Assumpton 1 s satsfed. () For all q, π, there exsts ɛ q,π > 0 such that Assumpton 1 s satsfed f ɛ q,π > σ. In our next example the sgnal s ternary that s, ts possble values are { 1, 0, 1}. In addton, Pr(s = ω ω) = p r, Pr(s = 0 ω) = 1 r, and Pr(s = ω ω) = (1 p)r. Notce that r ( (0, 1)) s the probablty that the realzaton of the sgnal s nformatve, and p s the precson of the sgnal condtonal on beng nformatve. PROPOSITION 2. Suppose that the sgnal s ternary. Then for any r ( (0, 1)) there exsts a threshold p(r ) (0, 1) such that f p > p(r ), Assumpton 1 s satsfed. Next, we provde an example where the log-convexty assumpton s not satsfed. Suppose that the sgnal s bnary that s, s { 1, 1} and Pr(s = ω ω) = p, Pr(s = ω ω) = 1 p.

8 234 Koryama and Szentes Theoretcal Economcs 4 (2009) PROPOSITION 3. If the sgnal s bnary then Assumpton 1 s not satsfed. Snce bnary sgnals are commonly assumed n the lterature on commttee desgn, we further explan the negatve result n Proposton 3. For smplcty, assume that the pror s symmetrc. In ths case, the ex post optmal decson after recevng a set of sgnals s 1 f and only f there are (weakly) more sgnals 1 n the set than sgnals 1. The margnal socal value of a sgnal can be postve only f the ex post optmal decson s dfferent from the one wthout ths sgnal wth postve probablty. Ths can happen only f there are as many sgnals 1 as sgnals 1 pror to recevng the addtonal sgnal, and n partcular, f the number of sgnals s even. Ths mples that the margnal socal values of the second, fourth, sxth, etc. sgnals are all zero. Hence, the functon g s not even decreasng. Notce that n ths example, even f the cost of nformaton s zero there always exsts an equlbrum n whch only one ndvdual obtans a sgnal. Indeed, f an ndvdual knows that there s exactly one sgnal collected by the others, he has no ncentve to nvest n nformaton because he cannot have any effect on the optmalty of the fnal decson. Ths argument s due to the fact that each sgnal realzaton has the same strength. Were an ndvdual to know that hs sgnal would be potentally very nformatve, he would nvest f the cost s low enough, because, at least wth small probablty, hs nformaton nduces a more nformatve decson. Most of the lterature assumes that commttee members aggregate nformaton by a bnary votng procedure where members vote ether yes or no (or abstan). Assumng bnary sgnals n ths case s perhaps less problematc because even f a member has a strong sgnal he cannot communcate t va hs vote. We beleve, however, that allowng varaton n the strength of the sgnals s an mportant and realstc feature of the world. Allowng, for example, that a juror leans slghtly towards a gulty verdct and at the same tme another juror s sure that the defendant s nnocent mght have mportant consequences. We beleve also that even f there s bnary votng at the end of the delberaton, commttee members are able to communcate ther nformaton to the others, at least partally. Focusng on the delberaton mght be more mportant than analyzng the votng game. 3. RESULTS Ths secton s devoted to our man theorems. We frst characterze the set of equlbra for all k ( ). The next subsecton shows that f k s small, the equlbrum s unque, and each member ncurs the cost of nformaton (Proposton 4). Secton 3.2 descrbes the set of mxed-strategy equlbra for suffcently large k (Proposton 5). Fnally, Secton 3.3 states and proves the man results (Theorems 1 and 2). 3.1 Pure-strategy equlbrum Suppose that the sze of the commttee s k. If the frst k 1 members acqure nformaton, the expected gan from collectng nformaton for the k th member s g (k 1). She

9 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 235 s wllng to nvest f ths gan exceeds the cost of the sgnal; that s, f c < g (k 1). (1) Ths nequalty s the ncentve compatblty constrant that guarantees that a commttee member s wllng to nvest n nformaton f the sze of the commttee s k. 8 PROPOSITION 4. Let k denote the sze of the commttee. There exsts k P such that there exsts a unque equlbrum n whch each member nvests n a sgnal wth probablty one f and only f k mn{k P, N }. Furthermore, the socal welfare generated by these equlbra s monotoncally ncreasng n k ( mn{k P, N }). PROOF. Recall from Remark 1 that g s decreasng and lm k g (k ) = 0. Therefore, for any postve cost 9 c < g (0), there exsts a unque k P such that g (k P ) < c < g (k P 1). (2) Frst, we show that f k k P then there s a unque equlbrum n whch each commttee member nvests n nformaton. Suppose that n an equlbrum, the frst k 1 members randomze accordng to the profle (r 1,..., r k 1 ), where r [0, 1] denotes the probablty that the th member nvests. Let I denote the number of sgnals collected by the frst k 1 members. Snce the members randomze, I s a random varable. Notce that I k 1 and r1,...,r k 1 [g (I )] g (k 1) because g s decreasng. Notce also that from k k P and (2), t follows that g (k 1) > c. Combnng the prevous two nequaltes, we get r1,...,r k 1 [g (I )] > c. Ths nequalty mples that no matter what the strateges of the frst k 1 members are, the k th member strctly prefers to nvest n nformaton. From ths observaton, the exstence and unqueness of the pure-strategy equlbrum follow. It remans to show that f k > k P, such a pure-strategy equlbrum does not exst. If k > k P, then g (k 1) < c. Therefore, the ncentve compatblty constrant, (1), s volated, and there s no equlbrum where each member ncurs the cost of the sgnal. Fnally, we must show that the socal welfare generated by these pure-strategy equlbra s ncreasng n k ( mn{k P, N }). Notce that snce N > 1, c < g (k 1) = η(k ) η(k 1) < N (η(k ) η(k 1)). 8 In what follows, we gnore the case where there exsts k such that c = g (k ). Ths equalty does not hold genercally, and would have no effect on our results. 9 If c > η(1) η(0), then nobody has an ncentve to collect nformaton, hence k P = 0.

10 236 Koryama and Szentes Theoretcal Economcs 4 (2009) g (k ) k c FIGURE 1. Expected gan g (k ) and the cost c. After addng N η(k 1) c k, we get N η(k 1) c(k 1) < N η(k ) c k. The left-hand sde s the socal welfare generated by the equlbrum n a commttee of sze k 1, whle the rght-hand sde s the socal welfare nduced by a commttee of sze k. Fgure 1 s the graph of g (k ) and c when s N (ω, 1), π =.3, q =.7, and c = The expected gan s decreasng and log-convex. In ths example, k P = 11. The amount of nformaton purchased n any equlbrum s neffcently small. Ths s because when a commttee member decdes whether or not to nvest, she consders her prvate beneft rather than the socety s beneft. Snce nformaton s a publc good, ts socal beneft s bgger than ts ndvdual beneft. Hence, the total number of sgnals acqured n an equlbrum s smaller than the socally optmal number. Ths s why the socal welfare s monotoncally ncreasng n the commttee sze k as long as k k P. Mukhopadhaya (2003) has proved a result that corresponds to the statement of Proposton 4 for the case where the sgnals are bnary. He has also shown by numercal examples that mxed-strategy equlbra can yeld lower expected welfare n large commttees than n small commttees. Our analyss goes further by analytcally comparng the expected welfare of all mxed-strategy equlbra.

11 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem Mxed-strategy equlbrum Suppose now that the sze of the commttee s larger than k P. We consder strategy profles n whch the commttee members can randomze when makng a decson about ncurrng the cost of nformaton acquston. The followng proposton characterzes the set of mxed-strategy equlbra (ncludng asymmetrc ones). We show that each equlbrum s characterzed by a par of ntegers (a,b). In the commttee, a members nvest n a sgnal wth probablty one and b members acqure nformaton wth a postve probablty that s less than one. The rest of the members, k (a + b) n number, do not ncur the cost. We call such an equlbrum a type-(a,b) equlbrum. PROPOSITION 5. Let the commttee sze be k (> k P ). There exsts an equlbrum where a members nvest for sure, b members nvest wth probablty r (0, 1), and k (a + b) members do not nvest, f and only f where the frst two nequaltes are strct whenever b > 0. a k P a + b k, (3) PROOF. Frst, we explan that f, n an equlbrum n whch one member nvests wth probablty r 1 (0, 1) and another nvests wth probablty r 2 (0, 1), then r 1 = r 2. Snce the margnal beneft from an addtonal sgnal s decreasng, our games exhbt strategc substtuton. That s, the more nformaton the others acqure, the less ncentve a member has to nvest. Hence, f r 1 < r 2, then the ndvdual who nvests wth probablty r 1 faces more nformaton n expectaton and has less ncentve to nvest than the ndvdual who nvests wth probablty r 2. On the other hand, snce r 1, r 2 (0, 1), both ndvduals must be exactly ndfferent between nvestng and not nvestng, a contradcton. Now, we formalze ths argument. Let r [0, 1] ( = 1,..., k ) be the probablty that the th member collects nformaton n an equlbrum. Suppose that r 1, r 2 (0, 1) and r 1 > r 2. Let I 1 and I 2 denote the number of sgnals collected by members 2, 3,..., k and by members 1, 3,..., k, respectvely. Notce that snce r 1 > r 2 and g s decreasng, r2,r 3,...,r k [g (I 1 )] > r1,r 3,...,r k [g (I 2 )]. (4) On the other hand, a member who strctly randomzes must be ndfferent between nvestng and not nvestng. Hence, for j = 1, 2, rj,r 3,...,r k [g (I j )] = c. Ths equalty mples that (4) should hold wth equalty, whch s a contradcton. Therefore, each equlbrum can be characterzed by a par (a,b) where a members collect nformaton for sure and b members randomze but collect nformaton wth the same probablty. It remans to show that there exsts a type-(a,b) equlbrum f and only f (a,b) satsfes (3). Frst, notce that whenever k > k P, n all pure-strategy equlbra k P members

12 238 Koryama and Szentes Theoretcal Economcs 4 (2009) nvest wth probablty one and the remanng members never nvest. In addton, the par (k P, 0) satsfes (3). Therefore, we have to show only that there exsts an equlbrum of type-(a,b) where b > 0 f and only f (a,b) satsfes a < k P < a + b k. (5) Suppose that a members nvest n nformaton for sure and b 1 nvest wth probablty r. Let G (r ; a,b) denote the expected gan from acqurng nformaton for the (a + b)th member. That s, b 1 b 1 G (r ; a,b) = r (1 r ) b 1 g (a + ). =0 We clam that there exsts a type-(a,b) equlbrum f and only f there exsts r (0, 1) such that G (r ; a,b) = c. Suppose frst that such an r exsts. We frst argue that there exsts a type-(a,b) equlbrum n whch b members nvest wth probablty r. Ths means that the b members, who are randomzng, are ndfferent between nvestng and not nvestng. The a members who nvest for sure strctly prefer to nvest because the margnal gan from an addtonal sgnal exceeds G (r ; a,b). Smlarly, those members who do not nvest, who number k (a +b), are strctly better off not nvestng because ther margnal gans are strctly smaller than G (r ; a,b). Next, we argue that f G (r ; a,b) = c does not have a soluton n (0, 1), then there exsts no type-(a,b) equlbrum. Ths mmedately follows from the observaton that f b members are strctly randomzng, they must be ndfferent between nvestng and not nvestng, and hence G (r ; a,b) = c. Therefore, t s suffcent to show that G (r ; a,b) = c has a soluton n (0, 1) f and only f (5) holds. Notce that G (r ; a,b) s strctly decreasng n r because g s strctly decreasng. Also observe that G (0; a,b) = g (a ) and G (1; a,b) = g (a + b 1). By the Intermedate Value Theorem, G (r ; a,b) = c has a soluton n (0, 1) f and only f G (1; a,b) < c < G (0; a,b), whch s equvalent to g (a + b 1) < c < g (a ). (6) Recall that k P satsfes g (k P ) < c < g (k P 1). Snce g s decreasng, (6) holds f and only f a < k P and a + b > k P. That s, the two strct nequaltes n (5) are satsfed. The last nequalty n (5) must hold because a + b cannot exceed the sze of the commttee, k. Fgure 2 graphcally represents the set of pars (a,b) that satsfy (3). Accordng to the prevous proposton, there are several equlbra n whch more than k P members acqure nformaton wth postve probablty. A natural queston to ask s: can these mxed-strategy equlbra be compared from the pont of vew of socal welfare? The next proposton partally answers ths queston. We show that f one fxes the number of members who acqure nformaton for sure, then as the number of randomzng members ncreases, the socal welfare generated by the equlbrum decreases. Ths proposton plays an mportant role n determnng the optmal sze of the commttee.

13 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 239 k b k P + 1 k P 1 k P k P + 1 k a FIGURE 2. The set of mxed-strategy equlbra. PROPOSITION 6. Suppose that k s such that there are both type-(a,b) and type- (a, b + 1) equlbra. Then the type-(a, b) equlbrum generates strctly hgher socal welfare than the type-(a, b + 1) equlbrum. In order to prove ths proposton we need the followng result, whch s proved n the Appendx. LEMMA 1. () G (r ; a,b) > G (r ; a,b + 1) for all r (0, 1]. () r a,b > r a,b+1, where r a,b and r a,b+1 are the solutons for r of G (r ; a,b) = c and G (r ; a,b + 1) = c respectvely. PROOF OF PROPOSITION 6. Suppose that a members collect nformaton wth probablty one and b members do so wth probablty r. Let f (r ; a,b) denote the beneft of an ndvdual; that s, b b f (r ; a,b) = r (1 r ) b η(a + ). Clearly =0 f (r ; a,b) b b 1 b b = r 1 (1 r ) b η(a + ) r (b )(1 r ) b 1 η(a + ). r Notce that =1 b b 1 = b 1 and =0 b b 1 (b ) = b.

14 240 Koryama and Szentes Theoretcal Economcs 4 (2009) Therefore the rght-hand sde of the prevous equalty can be rewrtten as b b 1 b 1 b 1 b r 1 (1 r ) b η(a + ) b r (1 r ) b 1 η(a + ). 1 =1 After changng the notaton n the frst summaton, ths can be further rewrtten as b 1 b 1 b 1 b 1 b r (1 r ) b 1 η(a + + 1) b r (1 r ) b 1 η(a + ) =0 Ths last expresson s just bg (r ; a,b), and hence we have Next, we show that Snce f (0; a,b) = f (0; a,b + 1) = η(a ), f (r a,b ; a,b) f (r a,b+1 ;a,b + 1) =0 =0 b 1 b 1 = b r (1 r ) b 1 η(a + + 1) η(a + ). =0 f (r ; a,b) = bg (r ; a,b). r f (r a,b ; a,b) f (r a,b+1 ; a,b + 1) > b(r a,b r a,b+1 )c. (7) = f (r a,b ; a,b) f (0; a,b) f (r a,b+1 ; a,b + 1) f (0; a,b + 1) = b ra,b 0 G (r ; a,b)d r b ra,b+1 By part () of Lemma 1, ths last dfference s larger than b ra,b 0 G (r ; a,b)d r b ra,b G (r ; a,b)d r = b G (r ; a,b + 1) d r. ra,b r a,b+1 G (r ; a,b)d r. By part () of the lemma, we know that r a,b+1 < r a,b. In addton, snce G s decreasng n r, ths last expresson s larger than b(r a,b r a,b+1 )G (r a,b ; a,b). Recall that r a,b s defned such that G (r a,b ; a,b) = c and hence we can conclude (7). Let S(a, b) denote the socal welfare n the type-(a, b) equlbrum; that s, Then S(a,b) = N f (r a,b ; a,b) c(a + b r a,b ). S(a,b) S(a,b + 1) = N f (r a,b ; a,b) c(a + b r a,b ) [N f (r a,b+1 ; a,b + 1) c(a + b r a,b+1 )] > N b(r a,b r a,b+1 )c cb(r a,b r a,b+1 ) = (N 1)cb(r a,b r a,b+1 ) > 0, where the frst nequalty follows from (7), and the last one follows from part () of Lemma 1.

15 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem The man theorems Frst, we show that the optmal commttee sze s ether k P or k P + 1. Second, we prove that f k > k, then even the worst possble equlbrum yelds hgher socal welfare than the unque equlbrum n the commttee of sze k 2. THEOREM 1. The optmal commttee sze, k, s ether k P or k P + 1. We emphasze that for a certan set of parameter values, the optmal sze s k = k P, and for another set, k = k P + 1. PROOF. Suppose that k s the optmal sze of the commttee and the equlbrum that maxmzes socal welfare s of type-(a,b). By the defnton of optmal sze, a + b = k. If b = 0, then all of the commttee members nvest n nformaton n ths equlbrum. From Proposton 4, k k P follows. In addton, Proposton 4 states that the socal welfare s ncreasng n k as long as k k P. Therefore, k = k P follows. Suppose now that b > 0. If there exsts an equlbrum of type-(a,b 1), then, by Proposton 6, k s not the optmal commttee sze. Hence f the sze of the commttee s k, there does not exst an equlbrum of type-(a,b 1). By Proposton 5, ths mples that the par (a,b 1) volates the nequalty chan (3) wth k = k. Snce the frst and last nequaltes n (3) hold because there s a type-(a,b) equlbrum, t must be the case that the second nequalty s volated. That s, k P a + b 1 = k 1. Ths mples that k k P + 1. Agan, from Proposton 4, t follows that k = k P or k P + 1. Next, we turn our attenton to the potental welfare loss due to overszed commttees. THEOREM 2. In any commttee of sze k (> k ), all equlbra nduce hgher socal welfare than the unque equlbrum n the commttee of sze k 2. The followng lemma plays an mportant role n the proof. We pont out that the proof of ths lemma s the only place where we use the log-convexty assumpton (Assumpton 1). For our prevous results, we need the functon g only to be decreasng, whch s a consequence of Assumpton 1. LEMMA 2. For all k 1 and, g (k 1){g ( ) g (k )} {g (k ) g (k 1)}{η( ) η(k )}, (8) wth equalty f and only f = k or k 1. PROOF OF THEOREM 2. Recall that S(a, b) denotes the expected socal welfare generated by an equlbrum of type-(a,b). Usng ths notaton, we have to prove that S(k 2, 0) < S(a,b). From Theorem 1, we know that k = k P or k P +1. By Proposton 4, S(k P 2, 0) < S(k P 1, 0). Therefore, n order to establsh S(k 2, 0) < S(a,b), t s enough to show that for all pars of (a,b) that satsfy (3). S(k P 1, 0) < S(a,b) (9)

16 242 Koryama and Szentes Theoretcal Economcs 4 (2009) Notce that f a + members nvest n nformaton, whch happens wth probablty b r a,b (1 r a,b ) b n a type-(a,b) equlbrum, the socal welfare s N η(a + ) c(a + ). Therefore, b b S(a,b) = r a,b (1 r a,b ) b N η(a + ) c(a + ) =0 b b = r a,b (1 r a,b ) b N η(a + ) c c a =0 b b = N r a,b (1 r a,b ) b η(a + ) c(a + b r a,b ). =0 In the last equaton, we use the dentty b b =0 ra,b (1 r a,b ) b = b r a,b. Therefore, (9) can be rewrtten as b b N η(k P 1) c(k P 1) < N r a,b (1 r a,b ) b η(a + ) c(a + b r a,b ). =0 Snce a k P 1 by (3) and b N, the rght-hand sde of ths nequalty s larger than b b N r a,b (1 r a,b ) b η(a + ) c(k P 1 + N r a,b ). =0 Hence t suffces to show that b b N η(k P 1) c(k P 1) < N r a,b (1 r a,b ) b η(a + ) c(k P 1 + N r a,b ). =0 After addng c(k P 1) to both sdes and dvdng through by N, we have b b η(k P 1) < r a,b (1 r a,b ) b η(a + ) c r a,b. (10) =0 The left-hand sde s the payoff of an ndvdual f k P 1 sgnals are acqured by others, whle the rght-hand sde s the payoff of an ndvdual who s randomzng n a type-(a,b) equlbrum wth probablty r a,b. Snce ths ndvdual s ndfferent between randomzng and not collectng nformaton, the rght-hand sde of (10) can be rewrtten as b 1 b 1 r a,b (1 r a,b ) b 1 η(a + ). =0 Hence (10) s equvalent to b 1 b 1 η(k P 1) < r a,b (1 r a,b ) b 1 η(a + ). (11) =0

17 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 243 By Lemma 2, b 1 b 1 g (k P 1) r a,b (1 r a,b ) b 1 g (a + ) g (k P ) =0 g (k P ) g (k P 1) b 1 b 1 r a,b (1 r a,b ) b 1 η(a + ) η(k P ). =0 (12) Notce that b 1 b 1 r a,b (1 r a,b ) b 1 g (a + ) = c < g (k P 1), (13) =0 where the equalty guarantees that a member who s randomzng s ndfferent between nvestng and not nvestng, and the nequalty holds by (2). Hence from (12) and (13), g (k P 1){g (k P 1) g (k P )} > g (k P ) g (k P 1) b 1 b 1 r a,b (1 r a,b ) b 1 η(a + ) η(k P ). Snce g (k P 1) g (k P ) > 0, ths nequalty s equvalent to =0 b 1 b 1 g (k P 1) > η(k P ) r a,b (1 r a,b ) b 1 η(a + ). =0 Fnally, snce η(k P ) g (k P 1) = η(k P 1), ths nequalty s just (11). The two graphs of Fgure 3 show the socal welfare n the worst equlbrum as a functon of the commttee sze for an example. In the example, the pror s symmetrc, and the parameters are N = 100, s N (ω, 1), π =.3, p =.7, and c = In ths case we have k P = 11, and k = 12. The two graphs are the same except that the scalngs of the vertcal axes dffer. One can see that the welfare loss due to overszed commttees s qute small. 4. CONCLUSION In ths paper, we dscuss the optmal commttee sze and the potental welfare losses assocated wth overszed commttees. We focus on envronments n whch there s no conflct of nterest among ndvduals, but nformaton acquston s costly. Frst, we confrm that the optmal commttee sze s bounded. In other words, the Condorcet Jury Theorem fals to hold: larger commttees mght nduce smaller socal welfare. However, we show also that the welfare loss due to overszed commttees s surprsngly small. In an arbtrarly large commttee, even the worst equlbrum generates welfare hgher than does an equlbrum n a commttee wth two less members than the optmal commttee. Our results suggest that carefully desgnng commttees mght be not as mportant as has been thought.

18 244 Koryama and Szentes Theoretcal Economcs 4 (2009) Socal welfare Commttee sze (k ) Commttee sze (k ) Socal welfare FIGURE 3. Socal welfare as a functon of the commttee sze k for N = 100, π = 0.3, q = 0.7, σ = 1, c = k P = 11, k = 12. APPENDIX LEMMA 3. Suppose that η ( C 1 ( + )) s absolutely contnuous and strctly ncreasng, and η (k + 1)/η (k ) s strctly ncreasng for k > ɛ, where ɛ 0. Let g (k ) = η(k + 1) η(k ) for all k 0. Then g (k + 1)/g (k ) < g (k + 2)/g (k + 1) for k ɛ. PROOF. Fx k ( ɛ). Notce that η (k + 2)/η (k + 1) < η (t + 2)/η (t + 1) s equvalent to η (k + 2)η (t + 1) < η (k + 1)η (t + 2). Therefore η (k + 2) k +1 k η (t + 1) d t < η (k + 1) k +1 k η (t + 2) d t η (k + 2)[η(k + 2) η(k + 1)] < η (k + 1)[η(k + 3) η(k + 2)].

19 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 245 It follows that η (k + 2) η (k + 1) η(k + 3) η(k + 2) g (k + 2) < = η(k + 2) η(k + 1) g (k + 1). (14) Smlarly, for all t (k, k + 1), η (k + 2)/η (k + 1) > η (t + 1)/η (t ) s equvalent to η (k + 2)η (t ) > η (k + 1)η (t + 1). Therefore η (k + 2) k +1 k It follows that η (t ) d t > η (k + 1) k +1 k η (t + 1) d t η (k + 2)[η(k + 1) η(k )] > η (k + 1)[η(k + 2) η(k + 1)]. η (k + 2) η (k + 1) From (14) and (15) t follows that > η(k + 2) η(k + 1) η(k + 1) η(k ) g (k + 1) g (k + 2) < g (k ) g (k + 1) g (k + 1) =. (15) g (k ) for all k ɛ. PROOF OF PROPOSITION 1. The sum of normally dstrbuted sgnals s also normal: k =1 s N (ωk,σ k ). The densty functon of k =1 s condtonal on ω s k 1 σ k φ =1 s ωk σ k where φ(x) = (2π) 1/2 exp( x 2 /2). The ex post effcent decson rule s gven by 1 f s s k θ µ(s 1,..., s k ) = 1 f s s k < θ, where θ = (σ 2 /2) log[(1 q)(1 π)/qπ] s the cut-off value. Hence, for k \ {0}, η(k ) = qπpr µ(s 1,..., s k ) = 1 ω = 1 (1 q)(1 π) Pr µ(s 1,..., s k ) = 1 ω = 1 θ k = qπφ σ (1 q)(1 π)φ θ + k k σ (16) k where Φ s the cdf of standard normal dstrbuton. If k = 0, η(0) = max{ qπ, (1 q)(1 π)}. (17) Notce that the rght-hand sde of (16) converges to that of (17) as k goes to zero. Part(). If q + π = 1, then qπ = (1 q)(1 π) and θ = 0. Hence k η(k ) = 2qπΦ σ and η (k ) = qπ σ 1 k φ for k > 0. k σ

20 246 Koryama and Szentes Theoretcal Economcs 4 (2009) Therefore η (k + 1) η = (k ) k k + 1 exp 1 2σ 2 s ncreasng n k (> 0). From Lemma 3, settng ɛ to be zero, t follows that g (k + 1)/g (k ) s ncreasng n k. Part(). By the defnton of θ, θ k qπφ σ = (1 q)(1 π)φ θ + k k σ. k Applyng ths to take the dervatve of (16), we get θ k θ k η (k ) = qπφ σ k k σ k θ k 1 = qπφ σ k σ k. + θ + k k σ k Next, we argue that for any ɛ (> 0), η (k + 1)/η (k ) s ncreasng for all k > ɛ f σ s suffcently small. For k > 0, η (k + 1) k φ θ (k +1) σ k +1 η = (k ) k + 1 φ k θ k = σ k + 1 exp 1 k k 1 θ 2 = k + 1 exp 2σ 2 k (k + 1) 1 k L 2 = k + 1 exp σ 2 1 exp 8k (k + 1) 2σ 2 exp 1 θ 2 2θ + k + 1 2σ 2 k +1 θ 2 2σ 2 k 2θ + k where L = log{(1 q)(1 π)/(qπ)}. Now suppose that k > ɛ. The last term n (18) has no nfluence on whether η (k + 1)/η (k ) s ncreasng. The second term converges to 1 as σ goes to 0. Obvously, the frst term s strctly ncreasng n k. Hence, η (k + 1)/η (k ) s ncreasng n k (> ɛ) f σ s suffcently small. By settng ɛ (0, 1) and usng Lemma 3, we have shown that g (k + 1)/g (k ) < g (k + 2)/g (k + 1) for all k 1. It remans to show that g (1)/g (0) < g (2)/g (1). From the argument n the proof of Lemma 3, t follows that η (2)/η (1) < η (t + 1)/η (t ) for all t (1, 2) mples η (2)/η (1) < g (2)/g (1). Hence t s enough to show that g (1)/g (0) < η (2)/η (1) for suffcently small σ. Snce lm σ 0 g (0) = η(0) > 0, t s enough to show that lm σ 0 [g (1)/{η (2)/η (1)}] = 0. In order to establsh ths equalty, t s obvously enough to show that (18) η(k ) lm σ 0 η (2)/η = 0 for k {1, 2}. (19) (1) Remember L = log{(1 q)(1 π)/(qπ)}. By (16), for k > 0, Lσ k η(k ) = qπ Φ 2 k + exp(l)φ Lσ k σ 2 k, σ

21 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 247 whch mples η(k ) O E (Φ( k /σ)) as σ Usng (18), η (2)/η (1) O E (exp( 1/2σ 2 )) = O E (φ(1/σ)) as σ 0. By l Hôptal s Rule, for k {1, 2}, Φ k /σ φ k /σ k /σ 2 lm = lm σ 0 φ(1/σ) σ 0 φ(1/σ)(1/σ 3 = 0, ) whch mples (19). PROOF OF PROPOSITION 2. Frst, we clam that the ex post effcent decson rule µ : { 1, 0, 1} k { 1, 1} s the cut-off rule 1 f k =1 µ(s 1,..., s k ) = s θ, 1 f k =1 s < θ (20), where θ = log[(1 q)(1 π)/qπ]/ log[p/(1 p)]. (s 1,..., s k ) s a permutaton of Suppose that the sgnal sequence Then µ(s 1,..., s k ) = 1 f 1,..., 1, 0,..., 0, 1,..., 1. (21) }{{}}{{}}{{} a k a b b ω [u (ω, 1) s 1,..., s k ] = (1 q)(1 π) Pr[s 1,..., s k ω = 1] Pr[s 1,..., s k ] In addton, Hence, µ(s 1,..., s k ) = 1 f or equvalently, > ω [u (ω, 1) s 1,..., s k ] = qπ Pr[s 1,..., s k ω = 1]. Pr[s 1,..., s k ] Pr[s 1,..., s k ω = 1] = (p r ) a (1 r ) k a b (r (1 p)) b Pr[s 1,..., s k ω = 1] = (p r ) b (1 r ) k a b (r (1 p)) a. (1 q)(1 π)p b (1 p) a > qπp a (1 p) b, log (1 q)(1 π) qπ a b > log p = θ. 1 p Snce k =1 s = a b, (20) follows. Now we consder the case where p converges to 1. Let ɛ denote 1 p and let Pr[a,b] denote the probablty of a sgnal sequence that s a permutaton of (21). Then Pr[a,b ω = 1] = C k (a,b)(1 ɛ) a ɛ b Pr[a,b ω = 1] = C k (a,b)(1 ɛ) b ɛ a, 10 O E s a verson of Landau s O, whch descrbes the exact order of the expresson. Formally, f (x) O E (g (x )) as x a f and only f there exsts M > 0 such that lm x a f (x)/g (x) = M.

22 248 Koryama and Szentes Theoretcal Economcs 4 (2009) where C k (a,b) = [k!/(a!b!(k a b)!)]r a +b (1 r ) k a b. 11 Notce that C k (a,b) s ndependent of ɛ and symmetrc wth respect to a and b. We have 12 Pr[a b 1 ω = 1] = C k (0, 1)ɛ +O(ɛ 2 ) Pr[a b 0 ω = 1] = C k (0, 0) + {C k (1, 0) + C k (1, 1)}ɛ +O(ɛ 2 ). Observe that θ < 1 f p s close enough to one. Wthout loss of generalty, assume that q + π 1. Then 1 < θ 0. Hence η(k ) = qπpr[a b 1 ω = 1] (1 q)(1 π) Pr[a b 0 ω = 1] = qπc k (0, 1)ɛ (1 q)(1 π)[c k (0, 0) + {C k (1, 0) + C k (1, 1)}ɛ] +O(ɛ 2 ). Then g (k ) = η(k + 1) η(k ) = A(k ) + B(k )ɛ +O(ɛ 2 ), where A(k ) = (1 q)(1 π)d k (0, 0) B(k ) = qπd k (0, 1) (1 q)(1 π)[d k (1, 0) + D k (1, 1)] and D k (a,b) = C k +1 (a,b) C k (a,b). Usng ths notaton, g (k + 1) = A(k + 1) + B(k + 1)ɛ +O(ɛ2 ) A(k + 1) B(k +1) 1 + A(k +1) ɛ +O(ɛ2 ) g (k ) A(k ) + B(k )ɛ +O(ɛ 2 = ) A(k ) 1 + B(k ) A(k ) ɛ +O(ɛ2 ) A(k + 1) B(k + 1) = 1 + A(k ) A(k + 1) B(k ) ɛ +O(ɛ 2 ). A(k ) We want to show that g (k + 1)/g (k ) s ncreasng n k f ɛ s suffcently small. Snce A(k + 1)/A(k ) = 1 r, t s suffcent to show that B(k )/A(k ) s convex n k. It s straghtforward to see that D k (0, 1) D k (0, 0) = D k (1, 0) D k (0, 0) = (k + 1)r (1 r )k k r (1 r ) k 1 (k + 1)r (1 r ) k r (1 r ) k +1 (1 r ) k = (1 r ) 2 (1 r ) s a polynomal of k wth degree 1, hence t has no nfluence on the convexty of B(k )/A(k ). On the other hand, D k (1, 1) D k (0, 0) = (k + 1)k r 2 (1 r ) k 1 k (k 1)r 2 (1 r ) k 2 (1 r ) k +1 (1 r ) k = (k + 1)k r 2 (1 r ) k (k 1)r 2 (1 r ) 3 (1 r ) 2 has a postve coeffcent of k 2. Hence we conclude that B(k )/A(k ) s convex n k. 11 Defne C k (a,b) = 0 f k < a + b. 12 f (x ) O(g (x )) as x 0 f and only f there exsts δ > 0, M > 0 such that x < δ mples f (x)/g (x) < M.

23 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 249 PROOF OF PROPOSITION 3. rule s defned by As n the proof of Proposton 2, the ex post effcent decson µ(s ) = 1 f s θ 1 otherwse, where θ = log[(1 q)(1 π)/qπ]/ log[p/(1 p)]. By symmetry, we can assume θ 0 wthout loss of generalty. Frst, suppose θ > 1. Then η(k ) = qπ for k < θ, and the margnal beneft from an addtonal sgnal s zero for k < θ 1. Therefore g (k +1)/g (k ) s not well-defned. Second, suppose 0 θ < 1. We consder two dfferent cases dependng on whether k s even or odd. Case 1: Suppose k = 2m, where m. Then the (2m + 1)-st sgnal makes a dfference f and only f m of the frst 2m sgnals are postve and m are negatve, and the (2m +1)-st sgnal s postve (denote ths stuaton as pv e ). In such a case, the socal decson changes from 1 to 1. Hence the gan s q f ω = 1 and the loss s (1 q) f ω = 1. Therefore, the expected margnal beneft s g (2m) = q Pr[ω = 1, pv e ] (1 q)pr[ω = 1, pv e ] 2m 2m = q π p m (1 p) m p (1 q) (1 π) p m (1 p) m (1 p) m m 2m = {pqπ (1 p)(1 q)(1 π)} p m (1 p) m. m Case 2: Suppose k = 2m + 1, where m. Then the (2m + 2)-nd sgnal makes a dfference f and only f the frst 2m + 1 sgnals contan m + 1 postve and m negatve sgnals and the (2m + 2)-nd sgnal s negatve (denote ths stuaton as pv o ). In such a case, the socal decson changes from 1 to 1. Hence the loss s q f ω = 1, and the gan s 1 q f ω = 1. Therefore the expected margnal beneft s g (2m + 1) = q Pr[ω = 1, pv o ] + (1 q)pr[ω = 1, pv o ] 2m + 1 = q π p m +1 (1 p) m +1 m 2m (1 q) (1 π) p m +1 (1 p) m +1 m 2m + 1 = { qπ + (1 q)(1 π)} p m +1 (1 p) m +1. m Recall that 0 θ < 1, whch s equvalent to pqπ (1 p)(1 q)(1 π) > 0 and (1 q)(1 π) qπ 0. If θ > 0, g (2m + 2) g (2m + 1) pqπ (1 p)(1 q)(1 π) =, (1 q)(1 π) qπ whch s a constant functon of m, so that Assumpton 1 does not hold. If θ = 0, then g (2m + 1) = 0 and g (2m + 2)/g (2m + 1) s not well-defned.

24 250 Koryama and Szentes Theoretcal Economcs 4 (2009) PROOF OF LEMMA 1. (). Notce that b 1 b 1 G (r ; a,b) = r (1 r ) b 1 g (a + ). =0 Snce r (1 r ) b 1 = r (1 r ) b + r +1 (1 r ) b 1, Snce g s decreasng, b 1 b 1 r G (r ; a,b) = (1 r ) b + r +1 (1 r ) b 1 g (a + ). =0 b 1 b 1 b 1 b 1 G (r ; a,b) > r (1 r ) b g (a + ) + r +1 (1 r ) b 1 g (a + + 1) =0 b 1 b 1 b b 1 = r (1 r ) b g (a + ) + r (1 r ) b g (a + ) 1 = =0 =0 =1 b b 1 b 1 + r (1 r ) b g (a + ), 1 =0 where the frst equalty holds because we have just redefned the notaton n the second summaton, and the second equalty holds because, by conventon, n n 1 = n+1 = 0 for all n. Fnally, usng b 1 b 1 b + 1 =, we have G (r ; a,b) > b b r (1 r ) b g (a + ) = G (r ; a,b + 1). =0 (). By the defntons of r a,b and r a,b+1, we have and by part () of ths lemma, c = G (r a,b ; a,b) = G (r a,b+1 ; a,b + 1), G (r a,b+1 ; a,b + 1) < G (r a,b+1 ; a,b). Therefore G (r a,b ; a,b) < G (r a,b+1 ; a,b). Snce G (r ; a,b) s strctly decreasng n r, r a,b > r a,b+1 follows. PROOF OF LEMMA 2. The statement of the lemma s obvous f {k 1, k }. It remans to show that (8) holds wth strct nequalty whenever / {k 1, k }. Frst, notce that for any postve sequence {a j } 0, f a j +1/a j < a j +2 /a j +1 for all j, then k a k j = +1 > a j a k 1 for all k > 1 and for all {0,..., k 2}. k 1 j = a j

25 Theoretcal Economcs 4 (2009) A resurrecton of the Condorcet Jury Theorem 251 Assumpton 1 allows us to apply ths result to the sequence a j = g (j ), and hence, for all k 1, k g (k ) g (k 1) > j = +1 g (j ) η(k + 1) η( + 1) k 1 = for all {0,..., k 2}. j = g (j ) η(k ) η( ) Snce η(a ) > η(b) f a > b, ths mples that for all {0,..., k 2}, g (k )[η( ) η(k )] < g (k 1)[η( + 1) η(k + 1)]. (22) Smlarly, for a postve sequence {a j } 0, f a j +1/a j < a j +2 /a j +1 for all j, then a k a k 1 < j =k +1 a j 1 j =k a j for all k 1 and for all k + 1. Agan, by Assumpton 1, we can apply ths result to the sequence a j = g (j ) and get g (k ) g (k 1) < j =k +1 g (j ) η( + 1) η(k + 1) 1 j =k g (j ) = for all > k. η( ) η(k ) Multplyng through by g (k 1)(η( ) η(k )), we get (22). That s, (22) holds whenever / {k 1, k }. After subtractng g (k 1)(η( ) η(k )) from both sdes of (22), we get (8). REFERENCES Austen-Smth, Davd and Jeffrey S. Banks (1996), Informaton aggregaton, ratonalty, and the Condorcet jury theorem. Amercan Poltcal Scence Revew, 90, [229] Börgers, Tlman (2004), Costly votng. Amercan Economc Revew, 94, [229] Condorcet, marqus de (Mare-Jean-Antone-Ncolas de Cartat) (1785), Essa sur l applcaton de l analyse à la probablté des décsons rendues à la pluralté des vox. Imprmere Royale, Pars. [227] Feddersen, Tmothy and Wolfgang Pesendorfer (1997), Votng behavor and nformaton aggregaton n electons wth prvate nformaton. Econometrca, 65, [229] Feddersen, Tmothy and Wolfgang Pesendorfer (1998), Convctng the nnocent: The nferorty of unanmous jury verdcts under strategc votng. Amercan Poltcal Scence Revew, 92, [229] Gerard, Dno and Leeat Yarv (2008), Informaton acquston n commttees. Games and Economc Behavor, 62, [230] Gerlng, Kerstn, Hans Peter Grüner, Alexandra Kel, and Elsabeth Schulte (2005), Informaton acquston and decson makng n commttees: A survey. European Journal of Poltcal Economy, 21, [227]

A Resurrection of the Condorcet Jury Theorem

A Resurrection of the Condorcet Jury Theorem A Resurrecton of the Condorcet Jury Theorem Yuo Koryama Balázs Szentes Department of Economcs, Unversty of Chcago June 13, 2007 Abstract Ths paper analyzes the optmal sze of a delberatng commttee where,

More information

A Resurrection of the Condorcet Jury Theorem

A Resurrection of the Condorcet Jury Theorem A Resurrecton of the Condorcet Jury Theorem Yuo Koryama and Balázs Szentes yz Septemer 19, 2008 Astract hal-00391197, verson 1-3 Jun 2009 Ths paper analyzes the optmal sze of a deleratng commttee where,

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

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

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

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

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

Welfare Properties of General Equilibrium. What can be said about optimality properties of resource allocation implied by general equilibrium?

Welfare Properties of General Equilibrium. What can be said about optimality properties of resource allocation implied by general equilibrium? APPLIED WELFARE ECONOMICS AND POLICY ANALYSIS Welfare Propertes of General Equlbrum What can be sad about optmalty propertes of resource allocaton mpled by general equlbrum? Any crteron used to compare

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

Assortment Optimization under MNL

Assortment Optimization under MNL Assortment Optmzaton under MNL Haotan Song Aprl 30, 2017 1 Introducton The assortment optmzaton problem ams to fnd the revenue-maxmzng assortment of products to offer when the prces of products are fxed.

More information

Market structure and Innovation

Market structure and Innovation Market structure and Innovaton Ths presentaton s based on the paper Market structure and Innovaton authored by Glenn C. Loury, publshed n The Quarterly Journal of Economcs, Vol. 93, No.3 ( Aug 1979) I.

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

ON THE EQUIVALENCE OF ORDINAL BAYESIAN INCENTIVE COMPATIBILITY AND DOMINANT STRATEGY INCENTIVE COMPATIBILITY FOR RANDOM RULES

ON THE EQUIVALENCE OF ORDINAL BAYESIAN INCENTIVE COMPATIBILITY AND DOMINANT STRATEGY INCENTIVE COMPATIBILITY FOR RANDOM RULES ON THE EQUIVALENCE OF ORDINAL BAYESIAN INCENTIVE COMPATIBILITY AND DOMINANT STRATEGY INCENTIVE COMPATIBILITY FOR RANDOM RULES Madhuparna Karmokar 1 and Souvk Roy 1 1 Economc Research Unt, Indan Statstcal

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

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

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

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

(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

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

Hila Etzion. Min-Seok Pang

Hila Etzion. Min-Seok Pang RESERCH RTICLE COPLEENTRY ONLINE SERVICES IN COPETITIVE RKETS: INTINING PROFITILITY IN THE PRESENCE OF NETWORK EFFECTS Hla Etzon Department of Technology and Operatons, Stephen. Ross School of usness,

More information

Implementation and Detection

Implementation and Detection 1 December 18 2014 Implementaton and Detecton Htosh Matsushma Department of Economcs Unversty of Tokyo 2 Ths paper consders mplementaton of scf: Mechansm Desgn wth Unqueness CP attempts to mplement scf

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

MMA and GCMMA two methods for nonlinear optimization

MMA and GCMMA two methods for nonlinear optimization MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons

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

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

Pricing and Resource Allocation Game Theoretic Models

Pricing and Resource Allocation Game Theoretic Models Prcng and Resource Allocaton Game Theoretc Models Zhy Huang Changbn Lu Q Zhang Computer and Informaton Scence December 8, 2009 Z. Huang, C. Lu, and Q. Zhang (CIS) Game Theoretc Models December 8, 2009

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

Marginal Effects in Probit Models: Interpretation and Testing. 1. Interpreting Probit Coefficients

Marginal Effects in Probit Models: Interpretation and Testing. 1. Interpreting Probit Coefficients ECON 5 -- NOE 15 Margnal Effects n Probt Models: Interpretaton and estng hs note ntroduces you to the two types of margnal effects n probt models: margnal ndex effects, and margnal probablty effects. It

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

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

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 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

/ n ) are compared. The logic is: if the two

/ n ) are compared. The logic is: if the two STAT C141, Sprng 2005 Lecture 13 Two sample tests One sample tests: examples of goodness of ft tests, where we are testng whether our data supports predctons. Two sample tests: called as tests of ndependence

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

Introduction. 1. The Model

Introduction. 1. The Model H23, Q5 Introducton In the feld of polluton regulaton the problems stemmng from the asymmetry of nformaton between the regulator and the pollutng frms have been thoroughly studed. The semnal works by Wetzman

More information

Lecture 10 Support Vector Machines II

Lecture 10 Support Vector Machines II Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed

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

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

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

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011 Stanford Unversty CS359G: Graph Parttonng and Expanders Handout 4 Luca Trevsan January 3, 0 Lecture 4 In whch we prove the dffcult drecton of Cheeger s nequalty. As n the past lectures, consder an undrected

More information

ECE559VV Project Report

ECE559VV Project Report ECE559VV Project Report (Supplementary Notes Loc Xuan Bu I. MAX SUM-RATE SCHEDULING: THE UPLINK CASE We have seen (n the presentaton that, for downlnk (broadcast channels, the strategy maxmzng the sum-rate

More information

Lecture 12: Discrete Laplacian

Lecture 12: Discrete Laplacian Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly

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

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

Maximizing the number of nonnegative subsets

Maximizing the number of nonnegative subsets Maxmzng the number of nonnegatve subsets Noga Alon Hao Huang December 1, 213 Abstract Gven a set of n real numbers, f the sum of elements of every subset of sze larger than k s negatve, what s the maxmum

More information

Abstract Single Crossing and the Value Dimension

Abstract Single Crossing and the Value Dimension Abstract Sngle Crossng and the Value Dmenson Davd Rahman September 24, 2007 Abstract When auctonng an ndvsble good wthout consumpton externaltes, abstract sngle crossng s necessary and suffcent to mplement

More information

NUMERICAL DIFFERENTIATION

NUMERICAL DIFFERENTIATION NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the

More information

Infinitely Split Nash Equilibrium Problems in Repeated Games

Infinitely Split Nash Equilibrium Problems in Repeated Games Infntely Splt ash Equlbrum Problems n Repeated Games Jnlu L Department of Mathematcs Shawnee State Unversty Portsmouth, Oho 4566 USA Abstract In ths paper, we ntroduce the concept of nfntely splt ash equlbrum

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

Lecture 4. Instructor: Haipeng Luo

Lecture 4. Instructor: Haipeng Luo Lecture 4 Instructor: Hapeng Luo In the followng lectures, we focus on the expert problem and study more adaptve algorthms. Although Hedge s proven to be worst-case optmal, one may wonder how well t would

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

Conjectures in Cournot Duopoly under Cost Uncertainty

Conjectures in Cournot Duopoly under Cost Uncertainty Conjectures n Cournot Duopoly under Cost Uncertanty Suyeol Ryu and Iltae Km * Ths paper presents a Cournot duopoly model based on a condton when frms are facng cost uncertanty under rsk neutralty and rsk

More information

Kernel Methods and SVMs Extension

Kernel Methods and SVMs Extension Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general

More information

ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EQUATION

ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EQUATION Advanced Mathematcal Models & Applcatons Vol.3, No.3, 2018, pp.215-222 ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EUATION

More information

Copyright (C) 2008 David K. Levine This document is an open textbook; you can redistribute it and/or modify it under the terms of the Creative

Copyright (C) 2008 David K. Levine This document is an open textbook; you can redistribute it and/or modify it under the terms of the Creative Copyrght (C) 008 Davd K. Levne Ths document s an open textbook; you can redstrbute t and/or modfy t under the terms of the Creatve Commons Attrbuton Lcense. Compettve Equlbrum wth Pure Exchange n traders

More information

Credit Card Pricing and Impact of Adverse Selection

Credit Card Pricing and Impact of Adverse Selection Credt Card Prcng and Impact of Adverse Selecton Bo Huang and Lyn C. Thomas Unversty of Southampton Contents Background Aucton model of credt card solctaton - Errors n probablty of beng Good - Errors n

More information

Limited Dependent Variables

Limited Dependent Variables Lmted Dependent Varables. What f the left-hand sde varable s not a contnuous thng spread from mnus nfnty to plus nfnty? That s, gven a model = f (, β, ε, where a. s bounded below at zero, such as wages

More information

A Note on Preference Uncertainty and Communication in Committees 1

A Note on Preference Uncertainty and Communication in Committees 1 A Note on Preference Uncertanty and Communcaton n Commttees 1 Davd Austen-Smth MEDS, Kellogg Graduate School of Management Northwestern Unversty Evanston, IL 60208 Tm Feddersen MEDS, Kellogg Graduate School

More information

Complete subgraphs in multipartite graphs

Complete subgraphs in multipartite graphs Complete subgraphs n multpartte graphs FLORIAN PFENDER Unverstät Rostock, Insttut für Mathematk D-18057 Rostock, Germany Floran.Pfender@un-rostock.de Abstract Turán s Theorem states that every graph G

More information

10-801: Advanced Optimization and Randomized Methods Lecture 2: Convex functions (Jan 15, 2014)

10-801: Advanced Optimization and Randomized Methods Lecture 2: Convex functions (Jan 15, 2014) 0-80: Advanced Optmzaton and Randomzed Methods Lecture : Convex functons (Jan 5, 04) Lecturer: Suvrt Sra Addr: Carnege Mellon Unversty, Sprng 04 Scrbes: Avnava Dubey, Ahmed Hefny Dsclamer: These notes

More information

Department of Statistics University of Toronto STA305H1S / 1004 HS Design and Analysis of Experiments Term Test - Winter Solution

Department of Statistics University of Toronto STA305H1S / 1004 HS Design and Analysis of Experiments Term Test - Winter Solution Department of Statstcs Unversty of Toronto STA35HS / HS Desgn and Analyss of Experments Term Test - Wnter - Soluton February, Last Name: Frst Name: Student Number: Instructons: Tme: hours. Ads: a non-programmable

More information

Polynomials. 1 More properties of polynomials

Polynomials. 1 More properties of polynomials Polynomals 1 More propertes of polynomals Recall that, for R a commutatve rng wth unty (as wth all rngs n ths course unless otherwse noted), we defne R[x] to be the set of expressons n =0 a x, where a

More information

FREQUENCY DISTRIBUTIONS Page 1 of The idea of a frequency distribution for sets of observations will be introduced,

FREQUENCY DISTRIBUTIONS Page 1 of The idea of a frequency distribution for sets of observations will be introduced, FREQUENCY DISTRIBUTIONS Page 1 of 6 I. Introducton 1. The dea of a frequency dstrbuton for sets of observatons wll be ntroduced, together wth some of the mechancs for constructng dstrbutons of data. Then

More information

Lecture 3: Probability Distributions

Lecture 3: Probability Distributions Lecture 3: Probablty Dstrbutons Random Varables Let us begn by defnng a sample space as a set of outcomes from an experment. We denote ths by S. A random varable s a functon whch maps outcomes nto the

More information

A Note on Preference Uncertainty and Communication in Committees 1

A Note on Preference Uncertainty and Communication in Committees 1 A Note on Preference Uncertanty and Communcaton n Commttees 1 Davd Austen-Smth MEDS, Kellogg Graduate School of Management Northwestern Unversty Evanston, IL 60208 Tm Feddersen MEDS, Kellogg Graduate School

More information

Stat260: Bayesian Modeling and Inference Lecture Date: February 22, Reference Priors

Stat260: Bayesian Modeling and Inference Lecture Date: February 22, Reference Priors Stat60: Bayesan Modelng and Inference Lecture Date: February, 00 Reference Prors Lecturer: Mchael I. Jordan Scrbe: Steven Troxler and Wayne Lee In ths lecture, we assume that θ R; n hgher-dmensons, reference

More information

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty Addtonal Codes usng Fnte Dfference Method Benamn Moll 1 HJB Equaton for Consumpton-Savng Problem Wthout Uncertanty Before consderng the case wth stochastc ncome n http://www.prnceton.edu/~moll/ HACTproect/HACT_Numercal_Appendx.pdf,

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

Basically, if you have a dummy dependent variable you will be estimating a probability.

Basically, if you have a dummy dependent variable you will be estimating a probability. ECON 497: Lecture Notes 13 Page 1 of 1 Metropoltan State Unversty ECON 497: Research and Forecastng Lecture Notes 13 Dummy Dependent Varable Technques Studenmund Chapter 13 Bascally, f you have a dummy

More information

Feature Selection: Part 1

Feature Selection: Part 1 CSE 546: Machne Learnng Lecture 5 Feature Selecton: Part 1 Instructor: Sham Kakade 1 Regresson n the hgh dmensonal settng How do we learn when the number of features d s greater than the sample sze n?

More information

Lecture 14: Bandits with Budget Constraints

Lecture 14: Bandits with Budget Constraints IEOR 8100-001: Learnng and Optmzaton for Sequental Decson Makng 03/07/16 Lecture 14: andts wth udget Constrants Instructor: Shpra Agrawal Scrbed by: Zhpeng Lu 1 Problem defnton In the regular Mult-armed

More information

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

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017 U.C. Berkeley CS94: Beyond Worst-Case Analyss Handout 4s Luca Trevsan September 5, 07 Summary of Lecture 4 In whch we ntroduce semdefnte programmng and apply t to Max Cut. Semdefnte Programmng Recall that

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

Econ Statistical Properties of the OLS estimator. Sanjaya DeSilva

Econ Statistical Properties of the OLS estimator. Sanjaya DeSilva Econ 39 - Statstcal Propertes of the OLS estmator Sanjaya DeSlva September, 008 1 Overvew Recall that the true regresson model s Y = β 0 + β 1 X + u (1) Applyng the OLS method to a sample of data, we estmate

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

Week3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity

Week3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity Week3, Chapter 4 Moton n Two Dmensons Lecture Quz A partcle confned to moton along the x axs moves wth constant acceleraton from x =.0 m to x = 8.0 m durng a 1-s tme nterval. The velocty of the partcle

More information

Dirichlet s Theorem In Arithmetic Progressions

Dirichlet s Theorem In Arithmetic Progressions Drchlet s Theorem In Arthmetc Progressons Parsa Kavkan Hang Wang The Unversty of Adelade February 26, 205 Abstract The am of ths paper s to ntroduce and prove Drchlet s theorem n arthmetc progressons,

More information

Primer on High-Order Moment Estimators

Primer on High-Order Moment Estimators Prmer on Hgh-Order Moment Estmators Ton M. Whted July 2007 The Errors-n-Varables Model We wll start wth the classcal EIV for one msmeasured regressor. The general case s n Erckson and Whted Econometrc

More information

Investment Secrecy and Competitive R&D

Investment Secrecy and Competitive R&D BE J. Econ. nal. Polcy 2016; aop Letter dt Sengupta* Investment Secrecy and Compettve R&D DOI 10.1515/beeap-2016-0047 bstract: Secrecy about nvestment n research and development (R&D) can promote greater

More information

Competitive Experimentation and Private Information

Competitive Experimentation and Private Information Compettve Expermentaton an Prvate Informaton Guseppe Moscarn an Francesco Squntan Omtte Analyss not Submtte for Publcaton Dervatons for te Gamma-Exponental Moel Dervaton of expecte azar rates. By Bayes

More information

Deterministic versus Stochastic Mechanisms in Principal Agent Models

Deterministic versus Stochastic Mechanisms in Principal Agent Models Dscusson Paper No. 26 Determnstc versus Stochastc Mechansms n Prncpal Agent Models Roland Strausz* September 2004 *Roland Strausz, Free Unversty Berln, Department of Economcs, Boltzmannstr. 20, D-14195

More information

a b a In case b 0, a being divisible by b is the same as to say that

a b a In case b 0, a being divisible by b is the same as to say that Secton 6.2 Dvsblty among the ntegers An nteger a ε s dvsble by b ε f there s an nteger c ε such that a = bc. Note that s dvsble by any nteger b, snce = b. On the other hand, a s dvsble by only f a = :

More information

Graph Reconstruction by Permutations

Graph Reconstruction by Permutations Graph Reconstructon by Permutatons Perre Ille and Wllam Kocay* Insttut de Mathémathques de Lumny CNRS UMR 6206 163 avenue de Lumny, Case 907 13288 Marselle Cedex 9, France e-mal: lle@ml.unv-mrs.fr Computer

More information

Open Systems: Chemical Potential and Partial Molar Quantities Chemical Potential

Open Systems: Chemical Potential and Partial Molar Quantities Chemical Potential Open Systems: Chemcal Potental and Partal Molar Quanttes Chemcal Potental For closed systems, we have derved the followng relatonshps: du = TdS pdv dh = TdS + Vdp da = SdT pdv dg = VdP SdT For open systems,

More information

P exp(tx) = 1 + t 2k M 2k. k N

P exp(tx) = 1 + t 2k M 2k. k N 1. Subgaussan tals Defnton. Say that a random varable X has a subgaussan dstrbuton wth scale factor σ< f P exp(tx) exp(σ 2 t 2 /2) for all real t. For example, f X s dstrbuted N(,σ 2 ) then t s subgaussan.

More information

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 30 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 2 Remedes for multcollnearty Varous technques have

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

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U) Econ 413 Exam 13 H ANSWERS Settet er nndelt 9 deloppgaver, A,B,C, som alle anbefales å telle lkt for å gøre det ltt lettere å stå. Svar er gtt . Unfortunately, there s a prntng error n the hnt of

More information

CHAPTER 17 Amortized Analysis

CHAPTER 17 Amortized Analysis CHAPTER 7 Amortzed Analyss In an amortzed analyss, the tme requred to perform a sequence of data structure operatons s averaged over all the operatons performed. It can be used to show that the average

More information

Module 17: Mechanism Design & Optimal Auctions

Module 17: Mechanism Design & Optimal Auctions Module 7: Mechansm Desgn & Optmal Auctons Informaton Economcs (Ec 55) George Georgads Examples: Auctons Blateral trade Producton and dstrbuton n socety General Setup N agents Each agent has prvate nformaton

More information

Information Acquisition in Global Games of Regime Change

Information Acquisition in Global Games of Regime Change Informaton Acquston n Global Games of Regme Change Mchal Szkup and Isabel Trevno y Abstract We study costly nformaton acquston n global games of regme change (that s, coordnaton games where payo s are

More information

A note on almost sure behavior of randomly weighted sums of φ-mixing random variables with φ-mixing weights

A note on almost sure behavior of randomly weighted sums of φ-mixing random variables with φ-mixing weights ACTA ET COMMENTATIONES UNIVERSITATIS TARTUENSIS DE MATHEMATICA Volume 7, Number 2, December 203 Avalable onlne at http://acutm.math.ut.ee A note on almost sure behavor of randomly weghted sums of φ-mxng

More information

Yong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 )

Yong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 ) Kangweon-Kyungk Math. Jour. 4 1996), No. 1, pp. 7 16 AN ITERATIVE ROW-ACTION METHOD FOR MULTICOMMODITY TRANSPORTATION PROBLEMS Yong Joon Ryang Abstract. The optmzaton problems wth quadratc constrants often

More information

Chapter 13: Multiple Regression

Chapter 13: Multiple Regression Chapter 13: Multple Regresson 13.1 Developng the multple-regresson Model The general model can be descrbed as: It smplfes for two ndependent varables: The sample ft parameter b 0, b 1, and b are used to

More information

Volume 29, Issue 4. Incomplete third-degree price discrimination, and market partition problem. Yann Braouezec ESILV

Volume 29, Issue 4. Incomplete third-degree price discrimination, and market partition problem. Yann Braouezec ESILV Volume 29, Issue 4 Incomplete thrd-degree prce dscrmnaton, and market partton problem Yann Braouezec ESILV Abstract We ntroduce n ths paper the "ncomplete" thrd-degree prce dscrmnaton, whch s the stuaton

More information

Remarks on the Properties of a Quasi-Fibonacci-like Polynomial Sequence

Remarks on the Properties of a Quasi-Fibonacci-like Polynomial Sequence Remarks on the Propertes of a Quas-Fbonacc-lke Polynomal Sequence Brce Merwne LIU Brooklyn Ilan Wenschelbaum Wesleyan Unversty Abstract Consder the Quas-Fbonacc-lke Polynomal Sequence gven by F 0 = 1,

More information

Axiomatizations of Pareto Equilibria in Multicriteria Games

Axiomatizations of Pareto Equilibria in Multicriteria Games ames and Economc Behavor 28, 146154 1999. Artcle ID game.1998.0680, avalable onlne at http:www.dealbrary.com on Axomatzatons of Pareto Equlbra n Multcrtera ames Mark Voorneveld,* Dres Vermeulen, and Peter

More information

A Comparative Study for Estimation Parameters in Panel Data Model

A Comparative Study for Estimation Parameters in Panel Data Model A Comparatve Study for Estmaton Parameters n Panel Data Model Ahmed H. Youssef and Mohamed R. Abonazel hs paper examnes the panel data models when the regresson coeffcents are fxed random and mxed and

More information

Optimal Allocation with Costly Verification 1

Optimal Allocation with Costly Verification 1 Optmal Allocaton wth Costly Verfcaton 1 Elchanan Ben-Porath 2 Edde Dekel 3 Barton L. Lpman 4 Frst Draft August 2012 1 We thank Rcky Vohra and numerous semnar audences for helpful comments. We also thank

More information