Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium Member Libraries http://www.archive.org/details/informationaggreOOange 4 n\b) Massachusetts Institute of Technology Department of Economics Working Paper Series Information Aggregation and Equilibrium Multiplicity Morris-Shin Meets Grossman-Stiglitz George-Marios Angeletos Ivan Werning Working Paper 04-1 February 2004 Room E52-251 50 Memorial Drive Cambridge, MA 02142 This paper can be downloaded without charge from the Social Science Research Network Paper Collection at http://ssrn.com/abstract=51 8282 mm? OF TECHNOLOGY APR ] 4 2004 LIBRARIES Information Aggregation and Equilibrium Multiplicity: Morris-Shin Meets Grossman-Stiglitz* George-Marios Angeletos MIT and NBER Ivan Werning MIT, NBER and UTDT angelet@mit edu iwerning@mit edu . . This draft: February 2004 Abstract This paper argues that adding endogenous information aggregation to situations where coordination is important - such as riots, self-ful filing currency crises, bank runs, debt crises or Qiancial crashes - yields novel insights into multiplicity and characterization of equilibria. Morris and Shin (1998) have highlighted the importance They also show that, with exogenous of the information structure for this question. information, multiplicity collapses enough idiosyncratic when individuals observe fundamentals with small noise. In the spirit of nize public information Grossman and by allowing individuals Stiglitz (1976), we endoge- to observe Qiancial prices or other noisy indicators of aggregate activity. In equilibrium these indicators imperfectly aggregate disperse private information without ever inducing common knowledge. Importantly, their informativeness increases with the precision of private information. that multiplicity Interestingly, ensured when may survive and characterize the conditions under which We it show obtains. endogenous information typically reverses the limit result: multiplicity is individuals observe fundamentals with small enough idiosyncratic noise. JEL Codes: D8, E5, F3, Gl. Keywords: Multiple equilibria, coordination, self-ful filing expectations, speculative at- tacks, currency crises, bank runs, financial crashes, rational-expectations, global games. Introduction 1 It's a love-hate relationship, economists are at once fascinated and uncomfortable with mul- tiple equilibria. *We thank ballero. On helpful the one hand, a variety of comments and discussion from phenomena seem characterized by large Daron Acemoglu, Francisco Buera and Ricardo Ca- and abrupt changes outcomes not obviously triggered by commensurate changes in fun- in Commentators often attribute these changes damentals. sentiments' or 'animal spirits'. Models with multiple Qiancial crashes, riots and political regime changes. is equilibria may formally capture these Prominent examples include self-fulQling bank runs, currency attacks, debt ideas. arises to arbitrary changes in 'market 1 In this class of models, multiplicity due to a coordination problem: attacking a 'regime' - beneCbial On if and only if crises, for instance, a currency peg - enough agents are expected to attack. the other hand, models with multiple equilibria are sometimes viewed as incomplete theories that should ultimately be extended in and Shin (1998) 2 have contributed Recently, Morris information structure away from survives when analysis is some dimension common to resolve the indeterminacy. to this perspective by enriching the knowledge. They show that a unique equilibrium individuals observe fundamentals with small enough idiosyncratic noise. Their particularly attractive because it can be viewed as a small perturbation around the original common-knowledge model. More generally, Morris and Shin introduce a useful framework tion heterogeneity affects the determinacy for studying and characterization of equilibria. how The informa- dispersion of valuable information about uncertain fundamentals plays a critical role for their unique- ness result. An earlier literature dealt with the transmission and aggregation of disperse In particular, Green (1973) and information in rational-expectations equilibria. (1981) highlight that prices ing that in may be some cases they can be Grossman excellent aggregators of dispersed information fully revealing, yielding common knowledge by show- of economic fundamentals. Morris-Shin abstracts from the role of Qiancial prices and other indicators as endogenous aggregators of information. Taking the information structure as exogenous step and helps isolate the critical role played by disperse beliefs. However, if is a useful Qst Qiancial prices convey information about the underlying fundamentals, then the dispersion of beliefs is de- termined endogenously in equilibrium. Information aggregation can thus play an important role in determining multiple equilibria whether multiple equilibria may arise. Indeed, Atkeson (2000) notes that survive in the extreme case that Qiancial prices are fully revealing and restore common knowledge. Moreover, for most of the applications of interest, it seems unnatural to rule out endogenous information aggregators, such as Qiancial prices or other indicators of economic activity. In this paper, we make the Qst attempt, to the best of our knowledge, to incorporate ^or example, Diamond and Dybvig (1983), Obstfeld (1986, 1996), Velasco (1996), Calvo (1988), Cooper and John (1988), Cole and Kehoe (1996). 2 Morris and Shin (1998) build on Carlson and van Damme (1993) who developed the Global Games approach for the two-player two-action games. See also Morris and Shin (1998, 2000, 2001, 2003, 2004). endogenous information aggregation in coordination economies with heterogenous informa- Our model takes tion. as given the initially dispersed private information that is crucial in the Morris-Shin framework. nomic indicator restoring build on this by allowing individuals to observe an eco- that, in equilibrium, aggregates disperse information. Importantly, common knowledge by Grossman and mon We we avoid allowing enough 'noise' in the aggregation process, as in Stiglitz (1976, 1980). Thus, none of our results are driven by restoring com- knowledge, the main theme in Atkeson's comments. We consider a variety of endogenous information structures, including indicators other than Qiancial prices. Indeed, to condition their behavior we begin our on a noisy begin with this model for two reasons. exploration with a model that allows agents signal of the aggregate actions of other agents. First, this situation applications. For instance, during riots or bank runs, We seems directly relevant in many an important part of the story may be people are actively watching what others are doing. Ignoring this aspect is that missing an important piece of the puzzle. Second, this model parsimoniously highlights aspects of the problem that recur when Qiancial prices are the instruments We study two versions of the observable actions case. assumed to observe a noisy signal of for aggregating information. In the Qst version, agents are An contemporaneous aggregate actions. requires that agents choose optimally given the observed signal and, at the signal is be generated by the aggregation of individual choices. equilibrium same time, this Thus, our equilibrium concept novel and unavoidably at the crossroads of rational-expectations and game theory. The second version avoids the simultaneity in the signal and actions by dividing the population into two groups, 'early' and 'late' agents. Early agents to attack solely on their private information. move Qst and base their decisions Late agents move second and can observe a noisy signal about the aggregate activity of early agents. This non-simultaneous version only requires standard game-theoretic equilibrium concepts. second version converge to that of the Drst as the We next We show that the size of the early equilibria of this movers vanishes. study environments where individuals cannot observe others actions directly but instead can trade in a Qiancial asset market. their private information This market opens after they have received but prior to choosing whether to attack the regime. The rational- expectations equilibrium in the asset market generates imperfect public information that agents use in addition to their private information when they decide whether to attack. This framework opens up new modeling choices regarding the specittation of the and the preferences over its risky return. Indeed, we asset's payoff consider four different specittations that can be solved in closed-form. A common insight emerges from all the cases we study: information is the precision of endogenous public increasing in the exogenous precision of private information. We show that this implies that introducing endogenous sources of information the likelihood of uniqueness vs. we the six specifications result: multiplicity important for understanding Interestingly, for all multiplicity of equilibria. but one of study, endogenous information reverses the Morris-Shin limiting ensured is is when individuals observe fundamentals with small enough idiosyncratic noise. Conversely, uniqueness is ensured idiosyncratic noise we view our paper Despite the difference in the limiting result theme emphasized by Morris-Shin, that if is large enough. as underscoring the general multiplicity or uniqueness may depend on details of the information structure and that these are worth exploring. This paper has explored the importance of endogenous information aggregation. 3 The rest of the paper is organized as follows. Section 2 introduces the basic model and reviews the Morris-Shin benchmark with exogenous public information. Section 3 introduces endogenous public information with a signal on aggregate actions. Section 4 studies the of Dnancial prices as role endogenous aggregators of information. Section 5 concludes. The Basic Model 2 We present an abstract general formulation of the basic model and then brieDy- discuss the various interpretations available in the literature. Actions, Outcomes and an alternative. There is Payoffs. There are two possible regimes, the status quo and a measure-one continuum of agents, indexed by can choose between an action that favorable to the status quo. All agents We move e {0, 1}, The if and where a; 1 = G {0, 1}, We represents collapse. where is is abandoned and R= represents survival of the similarly denote the action of an agent with represents "not attack" and a; payoff from not attacking is R = 1 represents "attack". normalized to zero. The payoff from attacking —c < Endogenizing the information structure is otherwise. Hence, the utility of agent = also the They examine how the information conveyed by how the Each agent these actions, respectively, "attack" and "not attack". call U{a i} R) 3 [0, 1]. simultaneously. R= the status quo € favorable to the alternative regime and an action denote the regime outcome with status quo at We is i is b > i is <n(bR-c). theme in Angeletos, Hellwig and Pavan (2003, 2004). active policy interventions may result to multiplicity, and evolution of information over time affects the dynamics of regime change. Dasgupta (2002), on the other hand, considers how social learning affects coordination. quo Finally, the status abandoned (R is = 1) if and only if A>9, where A = f atdi G denotes the mass of agents attacking and [0, 1] exogenous strength of the status quo £ = = and (or the quality of the vidual to attack if and only if and only is the heart of the model. parameterizes the economic fundamentals). Let if complements, since strategic it pays for an indi- the status quo collapses and, in turn, the status quo collapses a sufficiently large fraction of the agents attacks. This coordination problem Interpretations. This simple model can capture the role of coordination and mul- tiplicity of equilibria in a variety of interesting applications. self-fulQling currency crises (Obstfeld, 1986, 1996; Morris bank 1R 1. Note that the actions of the agents are if G For instance, in models of and Shin, 1998), there and a large number interested in maintaining a currency peg is a central of speculators, with Unite wealth, deciding whether to attack the currency or not. In this context, a "regime change" when occurs bank to a sufficiently large mass of speculators attacks the currency, forcing the central abandon the peg. In models of self-fulling bank runs, a "regime change" occurs once a sufficiently large number of depositors decide to to the system, forcing the bank withdraw to their deposits, relative to liquid resources available suspend its payments. Similarly, in models of self-fulQling debt crises (Calvo, 1988; Cole and Kehoe, 1996; Morris and Shin; 2003), a "regime change" occurs when Finally, may or a lender fails may riots. The potential rioters not overwhelm the police force in charge of containing social unrest depending rioters Information. Suppose 0, sufficiently large fraction of its creditors. Atkeson (2000) interprets the model as describing on the number of the < by a to obtain reQiancing and the strength for a moment of the police force. that were commonly known by the fundamentals are so weak that the regime unique equilibrium is > every agent attacking. For 9, is doomed with all agents. For certainty and the the fundamentals are so strong that the regime can survive an attack of any size and the unique equilibrium is every agent not attacking. For intermediate values, large attack G (0,0], and there are multiple the regime is sound but vulnerable to a equilibria sustained by self-fulQling expectations. In one equilibrium, individuals expect everyone else to attack, they then Old to attack, the status quo is sufficiently it individually optimal abandoned and expectations are vindicated. In another, uals expect no one else to attack, they then Dad it individ- individually optimal not to attack, the status quo The spared and expectations are again fulUled. is interval (0, 8] thus represents the set of "critical fundamentals" for which multiple equilibria are possible under common knowledge. by Implicitly, each equilibrium is sustained With common knowledge, other agents do. different self-MQling expectations about what can perfectly forecast in equilibrium individuals each other actions and coordinate on multiple courses of action. Following Morris and Shin we assume (1998), that 6 private noisy information about may actions that 8. have a common uniform over the entire real where the idiosyncratic noise £ { Xi is that individuals instead have Private information serves as an anchor for individual's avoid the indeterminacy of expectations about others actions. Initially agents erate) common knowledge and never is about prior line. Agent is A/"(0, o~l) 8; for simplicity, i we be (degen- let this prior then observes a private signal with a x > and is independent of The 8. signal thus a sufficient statistic for the private information of an agent. Note that because there economy, (^i) ie ji r , is is a continuum of agents the information contained by the entire enough to dispersed throughout the population, which Finally, agents may fundamental infer the is However, this information 9. the key feature of the Morris-Shin framework. have access to some public information. also is the Morris-Shin benchmark, where the public signal is exogenous. We start by reviewing We then consider the endogenous public information generated by a noisy indicator of aggregate activity or prices 2.1 The Morris-Shin Benchmark: Exogenous Information In this subsection, we assume an exogenous addition to the private signals Xi = 8 + public signal z £ i; where ^ ~ = 8 + v, A/"(0, <t£). where v The ~ _A/"(0, o^) private noise £ and the public noise v are distributed independently of each other and independently of model then reduces to that of Morris in 8. Our and Shin (2000, 2001), with exogenous private and public information (see also Hellwig, 2002). In a monotone equilibrium, an agent attacks decreasing in if and only we if and only so that there 8, 8 if < 8*(z). A if for x is < any realization of x*(z). By z, there . In step 3, quo also a threshold 9*(z) such that the status monotone equilibrium we a threshold x*(z) such that implication, the aggregate size of the attack is identiCed by x* and 8* characterize the equilibrium 8* for given x*. In step 2, for given 8* is we . is is abandoned In step 1, below, characterize the equilibrium x* characterize both conditions and examine equilibrium existence and uniqueness. 6 Step 1. For given realizations of 8 and mass of agents who receive signals x < = a~ 2 x*(z). the aggregate size of the attack That so that regime change occurs if given by the is, the precision of private information. Note that A(8, z) is is = $(^(x*(z)-d)), A(d,z) where a x z, and only if 8 < where 8*(z) 8*(z), is decreasing in is 8, the unique solution to A(8*(z),z)=8*(z). Rearranging we obtain: x*{z)=d*(z) + -^=$- 1 {F{z)). (1) Jot* Step 2. Given that regime change occurs if and only < 8 if 8*(z), the payoff of an agent is = a(bPi[8 < E[U(a,R{8,e))\x,z] Let a x = a~ 2 and a z = a~ 2 mation. The posterior of the agent = a x /(a x + a z ) is x,z]-c). \ denote, respectively, the precision of private and public infor- 8\x,z where 5 8*(z) is ~M(6x + (l—5)z a' 1 , ) , the relative precision of private information and a = ax + az the overall precision of information. Hence, the posterior probability of regime change Pi[9<6*(z)\x,z] = which is monotonic in l-$ (^{5x + x. It follows that (1 the agent attacks if 8)z - 8*(z))) and only if is is , x < x*(z), where x* (z) solves the indifference condition bPv[8<8*{z) | x*(z),z] =c. Substituting the expression for the posterior and the deQiition of 5 and a, + ^-z-8*( z ))) ^(v^T^ (^r-x*(z) a +a \a + a z \ Step if 3. and only Combining if it x (1) and (2), z x we conclude z JJ we b = —^. obtain: (2) b that 8*{z) can be sustained in equilibrium solves G(8*(z),z)=g, (3) = where g + a z /a x § y/l l (1 - and c/b) ^ = -^(z~8) + G(8,z) 1 (8) 'a. With 6*{z) given by (3), x*(z) then given by is We (1). now are in a position to establish existence and determinacy of the equilibrium by considering the properties of the function G. that, for every Note zGE, G(6, z) is which implies that there necessarily This establishes existence; continuous in exists we now turn dG(8, max^R <j){w) = 1/\/2tt then to uniqueness. z) a z /^/a^ < which implies a unique solution to 8, and there in 8 £ if and only We (8, 8). y^ (8)) (1) G we have that \/27r a z /^fcTx > is strictly increasing in V2tt, then G is non-monotonic admits multiple solutions 9*(z) whenever We conclude that monotone equilibrium is unique ~ results in the following proposition. (Morris- Shin) Suppose agents observe an exogenous public and private 1 Let a x and a z denote the standard deviations of the private and the public noise, respectively. Monotone equilibria exist Finally, consider the limits as ctz/^foLc for all z. — > and y/(a x + az ) — ax /a x and are unique — > > 1. for given az Condition , (3) if or and only az — » if a x /a z2 < v2tt- oo for given a x In either case, . then implies that 8*(z) — > 8 = 1 — c/b, This proves the following result, which we refer to as the Morris-Shin limit Proposition 2 (Morris-Shin limit) In oo for given a x 3 oo, a z /-sfa x < y2n. if Proposition only = z) az 1 instead an interval (z,z) such that summarize these signal. (3). If and a unique solution otherwise. z (z, z) is and G{8, Note that 1 0($" if = -co z) a solution and any solution satisCes 8*(z) e 88 Since with G(8, 8, , there is a unique ifd<6, where 8 = 1 - c/b e the limit as either ax — monotone equilibrium in which > for given a z , result: or az the regime changes if — > and (6, 6) Endogenous Information on Actions We now study the case where public information is endogenous. Agents no longer receive the public signal z as assumed in section 2.1. Instead, individuals are able to observe a public noisy signal of the aggregate actions of others. We study two versions of such a model. In the dst version, contemporaneous actions are observed with noise. Thus, our equilibrium concept crossroads of rational-expectations and The second game is novel and unavoidably at the theory. version, in Section 3.3, has non-simultaneous and lation into two groups, 'early' moves by dividing the popu- movers. Individuals in the early group 'late' make their decisions to attack or not based solely on their private information. Individuals in the late group move and are able to observe a noisy signal of the early group's aggregate action. This non-simultaneous version only requires standard game-theoretic equilibrium concepts. show that the equilibria of this second version converge to that of the Qst We as the size of the early movers vanishes. Equilibrium with Endogenous Information 3.1 We assume that agents can condition their behavior on a noisy indicator of the contempora- neous aggregate attack: y where e is random ously so that y is noise and s : [0, 1] x = s(A,e) R —* 1R. All agents are assumed o- move simultane- a signal about contemporaneous actions. For reasons of tractability we specify the signal function as s(A,e) noise s as J\f(0, to 2 e) with a e > 0. The common fundamentals 6 and the idiosyncratic noise £. noise e As we is = <£ -1 (A) +e and the distributed independently of the will see, the above speciCbation allows the equilibrium to preserves normality of the information structure, which in turn permits closed-form solutions. This convenient specittation was introduced by Dasgupta (2001) in a different setup. The information structure is parameterized by the pair of standard deviations (a x ,a £ ). In any symmetric equilibrium agents are distinguished solely by their information, summarized by their observation of the private signal action chosen by such an agent. %i and the public signal y. Let a(x{,y) denote the A symmetric rational-expectations equilibrium is deQied as follows. A symmetric equilibrium consists of an endogenous signal y = Y(9,e), an individual attack strategy a(x,y), and an aggregate attack A(9,y), that satisfy: = a(x, y) max E arg [ U{a, R(9, y)) x, j y (4) ] a£[0,l] A(9,y) y for all (9, e, x, y) Condition change occurs . Where and only if t y)d^^^y (5) = $-\A(9,y))+s = i?(0, y) means that a(x,y) (4) if R4 e = Ja(x is > A(9, y) 1 if A(0, y) (6) > 9 and E(0, y) = otherwise. the optimal strategy for the agent given that regime whereas condition 9, means that A(9,y) (5) is simply the aggregate across agents. Of course, the aggregate public signal y must be consistent with individual actions which gives condition (6). This is the rational-expectations feature in our equilibrium concept. we For tractability focus on symmetric equilibria where the information structure normally distributed and the strategy of the agents we is monotone shall see below, normality of the information structure simultaneous model of section 3.2 3.3. monotone as and the regime changes with the if and and only mappings triplet of construct the set of x*, 9* (4)-(6). In 0*(y) such that if 9 < A 9*(y). monotone equilibria, for an agent attacks if any realization and only monotone equilibrium is if as given optimal. In Step monotone and use condition 3, < x*(y) thus identiCed equilibria in four steps. (5) and In Step 1, we start (6) to characterize (4) to we study the Cked compute the threshold point x* = x**. x** that Finally, in Step 4, with an the implied aggregate attack A, the resulting 9* and the possible public signals Y. In Step Y x and Y. by the agents and use conditions arbitrary x* used and equilibria. Equilibrium Analysis of y, there exist thresholds x*(y) 9* As an implication of the non- We refer to such equilibria simply We now study the equilibrium conditions We is in private information. is 2, we take is individually we consider the deter minacy of Y. Step 1. function x*. In a monotone equilibrium, a(x,y) The aggregate attack is = 1 if and only if x < x*(y), for some then A(9,y) = <S>(^-x (x*(y)-9)), 10 (7) . = where a x that A{9,y) a x 2 Note that A(9,y) > 9 if and only if 9 < decreasing in 9 so there exists a function 9*(y) such is . The 9*{y). threshold 9*(y) solves A(0*(y),y) = 6*(y), or equivalently **(y) = ^(y) + -4=*- 1 (^(y)). (8) \/CXt. Equilibrium condition that the signal signal must satisfy, (6) implies y = ^faT x [x*{y) -8]+e, or equivalently x*(y)-a x y For any (9,e) € z. R2 let z , = Z(6,e) = 9 = 9-ax e. — a x e and (9) note that a relation between y and (9) is DeQie the correspondence y(z) In Step Now we show 4, is non-empty and examine when take any function Y(z) that Y(z) £ y{z) see that y(z) = {yeR\x*(y)-ax y = z}. and for all z, let is 2. We now map 9* = the signal be Y(6,e) and Y | single- or multi-valued. Y(Z(9,e)) i.e., = Y{9 - a x e). As we to x**. Given that regime change occurs x,y] is if = a{6Pr [ < 9 That is, it is as if x*(y) - axy = <t~ 2 , ae = we conclude 9 - ax e = the agents observe a public signal z about that each agent also observes a private signal about ax = cr~ 2 , and a z = a x ae = (a x a £ )~ that the posterior of an agent 2 . 9, = ax + a z is if 9* (y) \ x, y } z 9, with noise A/"(0, cr^cr with idiosyncratic noise Combining the two sources 2 A/"(0, ). <j Recall x ). Let of information is 9\x,y ~ Af(5x + (l-S)Z(y) where a and only \x,y]-c}. 9*{y) [ = shall given by We thus need to consider the determination of the posterior probability Pr 9 < The observation of y = Y(9,e) = Y(z) is equivalent to the observation of Z(y) such that of the information structure. 9 <-9*(y), the expected payoff for the agent E[U(a,R(9,e)) it is a selection from this correspondence, any such selection preserves normality Step (10) , the total precision of information and S 11 a' 1 ) , = a x /{a x + a z ) is the precision of x relative to the total. It follows that Pr [6 <d*(y)\x,y] Note that the above is monotonic = l-$ in (yfa (Sx + - (1 Z(y) 5) x so the agent attacks if - 6*(y))) and only if x . < x**(y), where x**(y) solves bPv[9<6*(y) \x**(y),y] = c. Combining the above two conditions and substituting Z(y) x**(y) = x*(y) — y/.y/ax~, we did that must solve $ (y/Z Step (5x**(y) + (1-5) 3. In equilibrium, x** = x* and ~ - -j=y\ - P(y)X\ = (x*(y) (11) (11) reduces to 1-5 \\ b-c = ${^(x *(y)-6\y)-±-=y\\ Combining the above with we (8), using 5 = a x /(a x + a z and a ) = ax + az , and rearranging, obtain: 6 -(y) = • (-*-» + X^Z*-(^i)), \a +a b a +a x **(y) where a z and = a x aE . Hence, for 9*(y) all + -^t>- x z \ y. 1 (0*(y)), (13) a x and ae the equilibrium x* and 6* are , Finally, 9*(y) does not e*(y)-+$(y) asa £ Step is 4. We (12) JJ irrespectively of the selected equilibrium signal Y. Moreover, increasing in this = V z depend on a x , 6*(y) — > determined uniquely both 0*(y) and x*(y) are 1 — c/6 as cr e — > oo and ^0. Qially need to consider the equilibrium correspondence y(z). Recall that given by the set of solutions to x*(y) r=zy 12 = z. Using (12) and the above reduces to (13), \a x + a z where as y A= —> \Ja x j (a x + a z )Q~ +co, and F(y) an equilibrium always for all z,. which true is Differentiating F — +oo > exist. 1 (l as ' > > To examine if F we ask whether y(z) multiplicity, monotonic is = in y. +A --*=-(l--£=#Y-5-V y/a x \a x + a x + a \ a> z z follows that the determinacy of equilibrium hinges on the ratio Shin benchmark. therefore y(z) is max^gK <j){w) Since {z,~z). are, respectively, = single- valued for all z, V2tt, there are thresholds z values for z € 1/v2tt, if we have and only ~ F = F(y) and that the equilibrium is unique if = is like in = y, and a z / sfcTx > instead for z ~z the Morris- decreasing in \f2ir. If and z such that y(z) takes one value the lowest and highest solution with F'(y) , that ' These thresholds are given by z a z = a £ a x we conclude a z / y/oi^, a z j yfa x < if Unlike the Morris-Shin benchmark, however, the ratio The single-valued is we obtain fto) It a x + az V — c/b). Note that F(y) is continuous in y, and F(y) — — oo y — — oo. Thus, the correspondence y(z) is non-empty and and only if JoTx J ' (z,z) but three ^ F(y), where y and y 0. a z / ^/a x ~ and only if is endogenous. a ey/a x < ' Using \f2ir. following proposition summarizes these results. A monotone Proposition 3 (Morris-Shin meet Grossman-Stiglitz) equilibrium is characterized by a triplet of mappings (Y,x*,6*) such that the endogenous signal satisDes y = Y(9,e), a agent attacks if and only if x < and regime change occurs x*(y), if and only tf0<9*(y). A monotone If a\a x < equilibrium exists for all (a x ,a 6 ) and l/\/27r, the equilibrium signal function is Y unique if and only if o\o~ x > 1/V27r. indeterminate, but the equilibrium is threshold functions x* and 9* remain unique and independent of the selected Y. We conclude that the equilibrium is unique only if there of information, the exogenous information of the agents aggregate activity. is enough noise in both sources and the endogenous signal about Multiple equilibria survive as long as either source of information is sufficiently precise. Interestingly, when multiplicity arises with respect to individual behavior. knowledge limit (a x = ae — 0), in it is with respect to aggregate outcomes but not To understand which case x 13 — 9 this result, and y = ^~ 1 consider the common- (A), so that the agent learns to attack if and only a(x,y) for the agent A values of A= and A by observing x and learns 9 perfectly A > if is by observing or equivalently 9, = The of indeterminacy remains. any given observation x and A = is uniquely determined for A and y for e. formation economy of Morris-Shin, it is economy is different from the exogenous '. However, note that, in equilibrium, the equivalent to the information generated by z Our endogenous-information economy is in- interesting that the determinacy of equilibrium in both cases hinges on exactly the same ratio a z /-s/ax is both (9,6], but there can be multiple equilibrium values of Finally, since our endogenous-information information generated by y E o^ and a e are non-zero, the same nature equilibrium behavior of the agent y, any given realization of 6 and When optimal However, the equilibrium, $(y). for every 9 and y are not uniquely determined. Instead, it Here, the equilibrium strategy $(y). uniquely determined with x*(y) 4 can be sustained in equilibrium. 1 < x The agent then Qids y. = Z(y) = 6 — a x e. thus related to an exogenous-information economy = ay = ae a x az with precision of public information given by . Indeed, substituting this expression into the criterion for multiplicity from proposition, ?? that o x ja\ the criterion for multiplicity in proposition 3, < that a^a x 1/\/2tt. As the > v27r, yields precision of private information becomes inLnite so does the precision of public information. Proposition 3 establishes that, for any given level of noise in the agents' private infor- mation, multiple equilibria exist if and only an increase sufficiently small. Intuitively, in the noise in the macroeconomic indicator if is a £ reduces the public information generated by the observation of y and thus reduces the ability of the market to coordinate on multiple courses of action. Indeed, the equilibrium conditions (12) and (13) imply that, for every y, we have 9*(y) — > 1 — c/b = 9 and x*(y) — > 6 + a x ^~ 1 (9) Proposition 4 (Limit a £ —> oo) Fix a x and if and only if 9 < 9, where 9 The Morris-Shin outcome becomes = 1 is arbitrarily large. This — c/b G let ay — » as at — > oo. oo. In the limit, the regime changes (9, 9). obtained as the noise in the observation of aggregate activity is no information intuitive, for in this case the observation of y and the endogeneity of public information is of is generated by no importance. Consider next the limit of the precision of agents' private information, for given level of noise in y. Proposition (3) establishes that, for given only if o~ x is sufficiently small. The a e multiple , equilibria exist interval (z,z) represents the region of multiplicity if and and a 4 1 If A — 0, then y = — oo and x*{y) = <5~ (— oo) = 8, in which case all agents attack whenever 6 < 6 and no agent attacks whenever 8 > 9_. If instead A = 1, then y = +oo and x*(y) = $ _1 (+oo) = 9, in which case all agents attack whenever 8 < 8 and no agent attacks whenever 8 > 8. In the former case, a regime change is triggered if and only if 8 < 9; in the latter, if and only 14 if 8 < 8. > . reduction in a x reduces z and increases z making larger the multiplicity region. Indeed, as a x —* we can show that any outcome — Proposition 5 (Limit a x > 0) Fix possible for is and o~ e ax let — G 9 all (9, 9). There exists an equilibrium with 0. the probability of regime change converging to zero for all 9 G as well as (9, 9), with the probability of regime change converging to one for every 9 G an equilibrium (#, 9) — +oo as a x — 0. Next, note that both \o e u x — (j)(y)\ and \o\o x — <f)(y)\ vanish. Since lim^-oo <f>{y)y = lim^ +00 (j)(y)y = 0, the latter implies and a x y — 0. Hence, z —> $(— oo) = 9 and z — $(+oo) = 9 as x — 0. Moreover, xy — for every 9 and e, 6 — a x e — as x — 0. It follows that Proof. o~ First, note that y > — — oo > > let Y(9,e) — ax e G = cr > o" > Pr that 9 > > 1 Next, J2 and y [ 9 - ax e £ (z,z). > (z,z) | — ^e) and min3^(^ 9 e {9,6) Y(9,e) Note that y(^,e) < y ] -> = max3 <y < (12), 0*(y) is $(+co) = 9 as y Pr[ 9 < — > independent of a x +oo. It ; ] -> and Pr [ < (6 l cj x -» 0. — a x e) and consider (^,e) such Y(9,e) and therefore <&(— oo) follows that, as long as 9 9* (Y(9,e)) which establishes the —* 9*{y) , as 1 F(0,e)—>— oo and y(#,£)—>+oo From » G ox as = ^ — > 0. — — oo, as y > and 6*{y) (9, 9), 5* (F(d,e)) ] - 1 as a x -> 0, result. This result stands in stark contrast to the Morris-Shin limit result in Proposition With exogenous reason is 2. public information a unique equilibrium survives as the noise in private information vanishes. present with — With endogenous common knowledge public information as modeled here the multiplicity obtains as the noise in private information vanishes. As the once again the endogeneity of public information: The precision of private information increases, the precision of public information also increases, and indeed at the same 3.3 The rate, so that common knowledge is recovered in the limit. Non-Simultaneous Signal analysis so far has assumed that agents can condition noisy indicator of contemporaneous aggregate behavior. 15 their decision to attack We now on a consider an alternative . that introducing to break the simultaneity of signals some simple dynamics a result, the equilibrium concept in this model is There are two types of agents, "early" and and As actions. entirely game-theoretic. "late" . Early agents move [1st, on the basis Late agents move second, on the basis of their private of their private information alone. information as well as a noisy public signal about the aggregate activity of early agents. Neither group can observe contemporaneous activity, but late agents can condition their behavior on the activity of early agents. Let and agents, ^ ii 4 1 is + e fj, (1 A2 A denote the fraction of early agents, (0, 1) The regime changes the aggregate activity of late agents. - fi)A 2 > The 6. signal generated by the aggregate activity of early Y early agents if and observed only by and only if late agents given by = $-\A + e, y1 where e In a x\. ~ A/"(0, a^) is independent of 6 and monotone equilibrium, an The aggregate £. early agent attacks attack of early agents A 1 (15) 1) is if and only if x < x\ for , some threshold thus = $( /^[9-x* ]). (9) y (16) 1 Hence, in equilibrium m = &The observation of y is 1 (A, (9)) +e= ^T x [x\ - 9} + e. which is contingent on his private signal x and the public signal y. thus equivalent to the observation of a public signal z, dedied by -y = - a x e. y/ot x The strategy of a late agent Since y and z has the is same informational of a late agent as a function of and only if x attack of late agents is attacks if < x and x\(z), for z. content, we can equivalently express the strategy Hence, in a monotone equilibrium, a late agent some threshold function x\. It follows that the A2 (6,z) = $(^x-[6-xl(z)}). Combining (16) and A (17), (6, z) = we obtain the ijl§ overall size of attack: {^Tx [9-xl]) + (l- /i)$ (Vc£ [9 - x*2 {z]\) 16 aggregate (17) . It follows that the regime changes and only if < 8 if 8*(z), where 9*(z) solves A (8*(z), z) = 9*(z), or equivalently //$ [6*{z) {yfe - x*]) + (1 - //)$ (Vo~ [8*(z) - 6Pr [8 to the late agents but < 8* {z)\x* {z) z] 2 , = where 5 &Pr [9 < cx. x /{ol x 9*(y)\x{] = c, $ (yfc(5x'2 (z) + az — c, unknown ) 8*(z). (18) The to the early agents. is threshold x 2 (z) thus solves or equivalently + {l-6)z- = ax + a and a z The . 8*(z))) = ~, (19) threshold x\, on the other hand, solves or equivalently f$(^[d*(z) where we used the = Note that the realization of z Next, consider the optimal behavior of the agents. known x*2 (z)}) x\]) fact that, conditional ^^(^[[z - x])dz = on x, z is ^, (20) distributed normal with precision ot\ = a x a£ /{\ + a £ ). 5 Solving (18) for x*2 {z) gives = 9*{z) - y^®- U(z) + -^— 1 x*2 {z) Substituting the above into (19) and using 5 = a x /(a x + a z T(8*(z),z,x*1 where g = y/l + a z /a x § r(e,z, Xl ,n) x (1 — For any x\ G E T(8,z,xl,fi) = —00 and T{9,z,x\,n) — z : K — > [8,8], f and a = a x + a z we (21) 1 (9 + -±- [8 -$ (^rx [9 - Xl ])]\ : > continuous in To [8,8]. On = 6 - ax e = x - £ with the other hand, for any given function condition (20) determines a threshold x\ G K. see this, note that z 8, Hence, for any given threshold x\ G M, con- 00. M— is An equilibrium solution to (20) and (21). 5 obtain: , i)=g, G M, we have that Y(9,z,x\,n) dition (21) determines a function 8* 8* , ) . and c/b) = ^=(z-9) + &- and any -§(^TX \9*{z) - x\})]\ [8*{z) - a x e, so that z\x 17 ~ M (0, a x + a\a^) is any joint > . Consider now the limit as -> fi Note 0. 6l that, for all (0, z, x\) r(M,s» - G(M) = ^=(z-9) + lOLn ^ 2 ,^0 implies (0) . G is independent of x{ and is the same as in the Morris-Shin benchmark. function 9* R — [0,0] such that, for all 2, Consider Note that now a > : G(9*(z),z)=g. earlier, 9* is As shown unique and only if (22) a z /^/a^ < v2n. if If instead a z /^/a^ > V27T, there are multiple 9* solving (22). Moreover, for any such 9*, (20) admits a unique solution x\ We e R. conclude that, for whenever a z /y/a^ as > OL e ^fOL^ > fi small enough, there are multiple solutions to (20) and (21) = a£ ax But a z ^/2rr. so that multiple equilibria again , emerge as long v2tt. Moreover, the equilibria of this economy approximate the equilibria of our benchmark model £^ in the following sense. Let denote the "dynamic" economy of this section and S the "static" economy of the previous section. Let x* and 9* denote the equilibrium thresholds for £, y the correspondence deQied in (10), and dence. For any x l( )( z) ~~ u > X *(Y( Z )) 0, we can Qid < u f° r an equilibrium of £^y To see only if the composite 9* o Y -> is is is Hence, in equilibrium, and 0^ Y Since y 2 conveys no we pM-p<y(*)) < ( are the thresholds associated with are part of an equilibrium for random variable y 2 deQied $ S if and = $- 1 by (A 2 )+e, is a noisy indicator about the activity unobservable, late agents continue to attack _1 u and a solution to (22). (A2 ) +£ = y 2 (z) if . the same realization as the one in (15). y 2 of late agents. If y 2 changes , note that 9* and y2 where £ a function selected from this correspon- small enough such that au z where x*2 this, Finally, let us introduce a \i Y [x 2 {z) ^JaZ = — y/aZ{x*2 {z) more information than z z] - and y 2 is if and only if < x x 2 {z). a function of z alone: z). and thus no more information than allow late agents to condition their behavior on y 2 in addition to yi y\. , nothing That is, the equilibria of the economy where late agents observe both yi and y 2 coincide with the equilibria of the Now economy where late agents observe only y\ consider again the limit as \i — > 0. Since x*2 , Az) 18 — > x*(Y(z)), we have y 2 (^)(z) — — ^/o^ x*(Y(z)) V2(ii)( z ) ~~> . By Y( z )- That an equilibrium economy z is, in the limit as economy in by a random variable £^ of y-zfji) \x — 0, > is economy variable = t/2(/x) <& _1 — z . that {A{6 i is y)) Hence, part of + e for £ can be approximated This indeed provides a justification for the £( M ). we made which the not a function of z alone, is random the ^JaZ x*(Y(z)) part of an equilibrium for equilibrium selection signal y = solves the Dxed-point condition y any Y(z) that £. Conversely, we have Y(z) definition of Y(z), in Section 3.1. Any equilibrium of the "static" economy in exists, if it can not be approximated by an equilibrium of a "dynamic" economy. Financial Prices 4 The analysis so far has assumed that agents can condition indicator of aggregate behavior. that is We determined Instead, we now on a noisy their decision to attack allow agents to observe a Dnancial price earlier in a competitively asset market. modify the environment as follows. There are two stages and we as the 'asset market'. All individuals begin with the exogenous private signal x, =8+£ { , same endowment ^ where the noise is refer to stage 1 w of wealth and an as before. In the fist stage, agents trade a financial asset that has a dividend that depends on the underlying fundamentals, either directly or indirectly. whether to attack the regime or not as In the second stage, agents decide in the basic model. The return of the asset and the regime outcome are realized at the end of the second stage. Here, in deciding whether to attack agents can no longer condition their choice on a direct signal of the aggregate attack as in Section 3. However, in addition to their private information, they can use information revealed by the equilibrium price from the [1st stage. This framework opens up new modeling choices regarding the specification of the Qiancial asset's payoff and the preferences over risky payoffs. Following Grossman-Stiglitz, our choice with an eye towards tractability. The four specifications we we guide solve below are designed so that they preserve normality of the information structure and are solvable in closed-form. We Qst introduce some general notation and by p the price of the asset or dividend paid by the asset this asset. For all cases his portfolio choice we is state the equilibrium conditions. more generally some measure represented by / and consider we can k{ = of the terms of trade. Let be the investment agent i / the makes in express the indirect utility the agent enjoys from by a function V(k, f,p) so that the Ui We denote V(ki,f,p) 19 + total payoff U(a,i,R), is where the utility Let aggregate from attacking demand U is just as in Section 2. be for the asset One the exogenous net supply of the asset. results it and the private A/^O, k{di. We assume there o~1) This shock 'noisy' traders. and independent Grossman and K=e a shock e to of is is that not observed both the fundamentals which determines an equilibrium price function P(8,e). Stiglitz (1976, 1980), the role of the shock e the information revealed by Qiancial prices about fundamentals. is is noise. Market clearing requires in it is J interpretation of the net supply shock from a shock to the demand of other by individuals and we assume that As K= is to introduce noise in Since the price function a function of both 9 and e the observation of p does not reveal 9 perfectly, leading to a signal-extraction problem. A rational- expectations equilibrium is a price function, p = P(9,e), individual strategies l investment and attacking, k(x ,p) and a(x\p), and aggregate investment and attack for functions, K(9,p) and A(9,e), such that k(x,p) in the Crst stage: = argmaxE[ K(0,p) V(k,f,p) \x,p] (23) = Jk(x,p)d$(Z^\ (24) K(9,P(9,e))=e and (25) in the second stage a(x,p) = axgmaxR[U(a,R)\x,p], a£[0,l] (26) v ; ' A(6,p)= f a (x,p)d<S>(^^-j, where R= 1 if A(8,p) > 8 and R= The information an agent has observed price (27) otherwise. consists of the privately observed signal x and the publicly p. In this sense, the price p takes the place that the signal y observable action model of Section . had in the Condition (25) imposes market clearing in the asset market. Condition (23) and (27) requires that an agent's investment take into account the information contained in their private information and prices. The four speciCcations we consider below generate the following information structure. 20 There is a strictly monotone function Z{y) and a random variable v that and independent = and £ such that Z(P(9,e)) of 9 9 + Thus, the observation of an equilibrium price realization p = observation of a public signal z The is A/"(0, Q{a x a E )~ l , and v for every realization of 9 is ) e. informationally equivalent to the Z(p) on 9 with normal error and precision agent's posterior conditional on his private information x ap = Q(a x ,a e and the observed price ). p is \x,p 9 where 5 = a x /(a x +ap ) and a of equilibrium turns out to ~N(5x + {l-8)z = a x +ap , a' 1 ) , Like in the Morris-Shin benchmark, the determinacy . depend on the ratio ap that is, the ratio of the precision of the public information generated by the price to the square root of the precision of the exogenous private information. the equilibrium is unique. If instead ap j\/a x > ~ If ap /\faZ < V^w, then V2ty, then there are multiple equilibria. Unlike Morris-Shin, however, the precision of public information ap and therefore the ratio , c^p/V^ri are endogenous and affected by a x what In follows, In each case, Q. we we . consider four alternative speciCcations of the asset market (stage solve for the equilibrium price function We then examine the critical ratio ol ~ p J sJ~oix to and the associated mappings Z 1). and study the determinacy of equilibrium as a function of the exogenous information structure, (a x ,a £ ), or equivalently (a x ,a E ). Risk Aversion — Fundamental Dividend 4.1 We start with an example that maps directly to the CARA-normal framework introduced by Grossman and Stiglitz (1976, 1980) [see also either in a risky asset or a risk-less asset. it costs the agent 1 in the dst stage and Hellwig (1980)]. invest his wealth We normalize the gross return of the risk-less delivers 1 in the second stage. the agent p in the Crst stage and delivers f(8) asset The agent can = 9 in the second. The asset: risky asset costs Here the return of the depends directly on the exogenous fundamental. The agent enjoys utility only from second-stage consumption and his preferences over second-stage consumption exhibit constant absolute risk aversion (CARA). utility from his portfolio choice V(k, is The indirect thus given by f,p)=u(w- pk + fk) , 21 u(c) = - exp(-2 7C )/27, (28) where k c is the amount invested = w — pk + fk We [ x,p] \ = we obtain, then guess and verify that E[ for invested in the risk-less asset, is second-period consumption. is J^E V(k, f,p) Setting w — pk and in the risky asset some 5 G (0, 1) 6 1 | x,p = ] and a > 0, + <fa in (1 -5)p and Var[ x,p (9 | = a" ] 1 , which case the optimal demand reduces to k(x,p) = — (x-p). 7 It follows that the aggregate demand for the asset is K(9,p) = —(e- P ). 1 In equilibrium, By K = J kidi = e, and therefore the equilibrium price implication, the observation of 9 with precision (5a/^) ap = (5a/~f) We now precision 2 aE 2 a e That . p = P(e,e)=e-^-e. p is satis Ces (29) da is, equivalent to the observation of a public signal about in this case we have Z(p) = p, v = —(Sa/j)e, and . determine 5 and a. Note that x and Z(p) a x and ap , E respectively. It follows that Oi X [ a = a x (l + a x a £ /^ 2 ap Recall that in Section 3 | x,p ), 5 = (Sa/^y) 1/(1 } — 2 + 8x — (1 5)p, where ae + ax a = Q(ae ,ax = J-^L we found that the are independent signals of 9 with Q>n a a x + ap a = a x + ap = a x + Solving the above gives 9 =p , £ / 'y 2 ) , and - ) . . (30) precision of endogenous information increased pro- portionally with the precision of private information. Here the precision of public information 22 . ' increases more than proportionally with the precision of private information. This will only and serve to reinforce our conclusions regarding the comparative static To verify this, consider stage by thresholds x*(p) and regime changes A monotone (continuation) equilibrium 6*(p) such that and only if 2. if < 9 9*(p). an agent attacks The threshold if ax limit exercises for and only if x*(p) solves frPr x [9 is . characterized < x*(p) and the < 9*(p)\x,p] = c, or equivalently h * - P<P))) = -^+ 7TT7rS(P) (^+^ (-~-*>) \a + a a +a P V The threshold x x p x *(p)=9*(p) if and only ' '" : if it - (p 31 ) 9*(p), or equivalently + -±=<l>-\9*(p)). (32) = p, we have that 9*(p) Combining the above two conditions and using Z(p) in equilibrium = on the other hand, solves A(9*(p),p) 9*(p), ( b JJ p can be sustained solves 9*(p)) + 3>-\9* (p)) -- J?Z±**-i (1 - c /b) (33) OCt. Similar arguments as those used for Proposition only if aP l ^/ox~ > V2tt, given by (29) and is where ap is 9*{p) if As given by (30). V is given by (28) and f always uniquely determined. and only in the On 9* (p) if and the other hand, the price function is uniquely determined. Proposition 6 Suppose P{9,e) is imply that there are multiple 1 < 7 2 (27r)~ 1/2 if g\<j\ = The equilibrium price function 9. There are multiple equilibrium thresholds x*(p) and . benchmark model, multiple equilibria survive as long as either noise is small enough. Indeed, the common-knowledge outcomes are once again obtained in the limit as ax — > 4.2 for any given a £ . Risk Aversion — Speculative Dividend We now modify the previous example by letting of the asset stage. We is now a earlier / = f(A) = — $ -1 (,4). That is, the dividend function of the equilibrium size of the attack realized in the second showed that, in equilibrium, 23 A = $ /c7x~[x* [y (p) - 9}) . It follows that = / — y/a^[9 x* and therefore (p)] o^E[ = v / k 9 | x,p } ja x Ven[ - y/a^x*(p) 9 \x,p] -p Let 1 = P zp + X*(p), (34) >OL n and note demand that, for every p, the above deQies a unique as = E[0 k | where 7 = r )y/a~x~. The rest x,p \ can then rewrite the optimal -p x,p] 7Var[ 9 We p. ] then as in the previous example, provided we replace 7 with is we have 7. In particular, (35) da so that Z(p) = p, v = —(5a/^)e, and ap = a £ a x2 j^( precision of the information revealed by the price ap Once . is = Using 7 now = Q(a £ ,a x = 7^/0^, we conclude that the given by (36) ) /y2. again, the precision of public information increases more than proportionally with the precision of private information, which only reinforces our results. Indeed, consider stage solve (31) p = -7==P and The (32). + x*(p). As 2. in the previous difference Hence, (??) is now is given by (36). What may be mined. Using (34), (35), where A = $ _1 (37), =$ 1 + ap (1 - now we have ax + p — >• is p , > as p (37) -P and x*(p) are uniquely deter the price function. a7 +A + a. lot.. ' — +00 = that p must solve (-^) ^foTx /\/a x~+~ap and z —00, and F(p) given by Z(p) a x + ap ' = 9 — — > ax + (Sa/j)e. equation (14) in the benchmark model. Note that F(p) as is ar c/b) follows that the threshold 9*(p) indeterminate and F{p) It signal replaced by z®- y/a x where ap now the endogenous 'OL, =$ 9*{p) that is example, the thresholds x*(p) and 9*{p) is p+A (38) a. This equation is analogous to continuous in p, and F{p) +00, which implies that a solution always 24 — —00 > exists. . Moreover, = **(p) +a -^[^-«M--^-p a +a a \ a +a x so the solution is unique \ p for all z if x p av j ^fal < p ap /^/a^ > y/2n. If instead v2tt, there are thresholds z and z such that there exist multiple equilibrium prices whenever z G V Proposition 7 Suppose olds x*(p) and = — <& -1 (A). given by (28) and f 9*{p) are always uniquely determined. functions P{6, e) The is if and only if (z, z) The equilibrium thresh- There are multiple equilibrium price a1a x < 7 2 j\phx. results here are reminiscent of the ones in the benchmark model. In equilibrium, the we found price plays the role of an anticipatory signal of the size of the attack. Indeed, as there, the strategies of the agents are uniquely determined, although the equilibrium price may function not be. In contrast, in the previous example the price played the role of a signal for the exogenous fundamental and not 9. Here indeterminacy arises for individual strategies for the price function. In both cases endogeneity of public information implies that with the precision of private information. In both cases its precision this implies that multiplicity survives with small noise and the common-knowledge outcomes obtain in the limit as a x given a £ — > for any . Risk Neutrality - Fundamental Dividend 4.3 We increasing is modify the asset market and the preferences as unit of the asset costs utility in the Crst period 1 from the portfolio choice is follows. and pays / — p There in the is no risk-less bond. One second period. The indirect thus given by V(kJ,p)=u 1 (w-k)+u 2 ((f-p)k) where k utility is the amount invested, u\ is the utility from Crst-period consumption, u 2 from second-period consumption, and Consider an agent who (39) U is is the the payoff from attacking. receives a private signal x and observes a price p. His optimal investment k solves u l (w-k)=E[(f-p)u'2 ((f-p)k) \x,p}. / We assume that U\(c) linear relation, ki = is quadratic and k {E [/|x,p] 112(c) is linear, in — p] + A, for which case some constants k > 25 (40) (40) reduces to 0, AeM. a simple With out any we normalize A loss of generality, = we 0. Finally, = / let = f(9) That 9. the return of is, the asset depends only on the exogenous fundamental. The analysis here is similar to that in the Crst example. k = The optimal individual demand for the asset is We k {E / [ | x,p - p} = ] k {E [ 9 x,p | ] - p} . conjecture B[d\x,p]=5x + (l-6)p some for K(9,p) 6 5 = k6 (0, 1) to be determined. It follows that k — p) (9 . In equilibrium, P By p implication, Z(p) =p ensured is is for It ) a x high enough. To complete a — ¥ oo. = see this, let 5 e Using these results, ' a = ax j (a e K 2 ) — 5) < sufficiently \ — 5) — as either > is V is always uniquely determined. and only if ax is a and 5, in this is, example with 5 — * a = as as S 5 /(1 a — and > — 5 5). — > 1 r2 {Ky/a £ ) / Oirr \ / LXn- Kyfire )y/5(l-8). ax — ax or > — > oo then implies that, given a £ therefore the equilibrium in unique ae < sufficient for multiplicity. Proposition 8 Suppose is ( 1/4 necessarily and therefore high . and rewrite the above = On either sufficiently small or sufficiently high. 5 (1 a E That an increasing function of a u and a decreasing K 1x2 CXr Otrp. we have that av j\fotx < v2tt and is . we Old = fact that 5(1 is ax + ae 52 K 2 + ap (0, 1) as ap The therefore the analysis, note that Obviously, this gives a monotonic relation between as — p) and bounded above by K 2 a e and therefore we immediately have that unique- The above uniquely determines . k8(x Hence, the equilibrium price = P (M = *_-L e ol. a £ To = remains to pin down 8 and the function Q. d function of e. k(x,p) a public signal about 9 with precision k 2 5 = — ^e. and v Note that ap ness is K= = Sn/K, We if and only the other hand, for given 2 is if ax a x we have , whereas ae The equilibrium price function P sufficient for uniqueness, conclude: given by (39) and f = 9. There are multiple equilibrium thresholds x*(p) and 8*(p) either sufficiently small or sufficiently high relative to 26 , a£ . if $ Thus, the Morris-Shin limit result we have examined, the is preserved in this example. Like in 4.4 when a x is is not strong enough to restore sufficiently small. Risk Neutrality — Speculative Dividend We modify the previous is previous cases precision of public information increases with the precision of private information. Unlike the previous cases, however, this effect multiplicity all example by letting the return of the asset the aggregate size of the attack occurring in the second stage. to the second example. In particular, but the price function in stage 1 we will show that the be / The =— _1 analysis (A), is where now A similar strategies in stage 2 are unique can be indeterminate. Let x*(p) denote the threshold agents use in stage 2 in deciding whether to attack. In equilibrium, so that the asset return k = k{E[ = / is f y/a^[9 x,p | ] — x* (p)]. The demand -p] = k{^/c^E[ 9 \ for the asset is thus x,p]-p- y/a^x*(p)} . Let + x*(p) P= —p=P JOL* and note that, for every p, the above deQies a unique k where k = Ky/a^. We now follows that We can thus write the demand as = H{E[9 x,p]-p} | conjecture E[9 It p. K = k5(9 —p) | x,p] =6x + (l -5)p. and therefore p = 9-±s. ko Hence, the observation of p 9 with precision ap = 2 equivalent to the observation of p, which is 2 K 5 ae E[0 It . | follows that x,p] =E[0 | x,p] 27 =8x + (l-8)p, is a public signal for where This is ct x = r o ax -^3oa x + aE b k = a x + ap the same as in the previous example, with k replacing k. Using k = n^/a^, is proportional to we infer i 5 = 1 so that 5 ax is like in , decreasing in a e but independent increasing in both and is We conclude: Proposition 9 Suppose and 9* functions P(6,e) Hence, the Final 5 of ' a x This means ~2 c-2 b ae a £ and ax The rest . V , 22 and only if o~ x and/or a £ common- knowledge outcomes now ap given by — of the analysis given by (39) and f is is N is = -$ _1 are always uniquely determined. (p) if k that . the benchmark model. Indeed, the critical ratio av olds x* (p) + a s 5 2 K? similar to the second example. ( The equilibrium 4). J thresh- There are multiple equilibrium price are sufficiently small. are once again obtained as a x —> 0. Remarks Building on Morris and Shin (1998) this paper introduced instruments that endogenized the sources of public information in models where coordination public information by either: (i) is important. a noisy signal of aggregate activity; or price that reveals information in equilibrium. cases is An that the precision of public information (ii) We modeled a Qiancial asset's important feature of the equilibrium in is endogenous and rises all with the precision of private information. We showed that in all but one of the six models considered this effect is strong enough to reverse the limiting uniqueness result obtained with exogenous public information. Thus, typically, with endogenous public information multiplicity serve fundamentals with small if idiosyncratic noise We is enough idiosyncratic is ensured when individuals ob- noise. Conversely, uniqueness is ensured large enough. view the main theme in Morris-Shin as emphasizing the importance of the details of the information structures for the multiplicity or uniqueness of equilibria. contributes to this same theme by studying the importance aggregation. 28 This paper of endogeneous information References [1] Angeletos, George-Marios, Christian Hellwig, and Alessandro Pavan (2003) "Coordination [2] and Policy Traps," MIT/UCLA/Northwestern working paper. Angeletos, George-Marios, Christian Hellwig, and Alessandro Pavan (2004) "On the Dy- namics of Information, Coordination, and Regime Change," MIT/UCLA/Northwestern working paper. Andrew on Morris and Shin," [3] Atkeson, [4] Calvo, G. (1988), "Servicing the Public Debt: (2000), "Discussion Economic Review, [5] [6] Econometrica Cole, Harold, Bank Damme the Role of Expectations." American Games and Equilibrium (1993), "Global Selec- 61, 5, 989-1018. and Timothy Kehoe (1996), "Self-fulQling Debt Crises," Federal Reserve of Minneapolis staff report 211. "Coordination, Learning and Delay," [7] Dasgupta, Amil (2002) [8] Diamond, Douglas, and Philip Dybvig , Liquidity," Journal of Political [9] Macro Annual 78(4), 647-661. Carlsson, Hans, and Eric van tion," NBER paper. "Bank Runs, Deposit Insurance, and (1983), Economy LSE working 91, 401-19. Green, Jerry "Information, Efficiency and Equilibrium," Harvard Institute of Economic Research Discussion Paper No. 284, Harvard University. [10] Grossman, Sanford (1977). "The Existence of Futures Markets, Noisy Rational Expectations [11] and Informational Review of Economic Studies Externalities," 44, 431-449. Grossman, Sanford (1981) "An Introduction to the Theory of Rational Expectations under Asymmetric Information," Review of Economic Studies, 48, 541-559. [12] Grossman, Sanford, and Joseph Efficient Markets," [13] Stiglitz (1980), "On the American Economic Review Grossman, Sanford, and Joseph Stiglitz (1976), Impossibility of Informationally 70, 393-408. "Information and Competitive Price Systems," American Economic Review 66, 246-253. [14] Hellwig, Martin (1980) "On the Aggregation Journal of Economic Theory, 22, 477-498. 29 of Information in Competitive Markets" [15] Hellwig, Christian (2002), "Public Information, Private Information, ity of Equilibria in [16] Coordination Games," Journal of Economic Theory, 107, 191-222. Morris, Stephen, and Hyun Song Shin (1998), "Unique Equilibrium in a Model of FulQling Currency Attacks", American Economic Review, 88, [17] Morris, Stephen, and Macroeconomics," [18] Hyun Song Shin NBER Morris, Stephen, and tions," in Self- 587-597. 3, "Rethinking Multiple Equilibria in (2000), Macro Annual. Hyun Song Shin (2001), "Global Games - Theory and Applica- Advances in Economics and Econometrics, 8th World Congress of the Econo- metric Society (M. Dewatripont, L. Hansen, and versity Press, Cambridge, [19] and the Multiplic- Morris, Stephen, and S. Turnovsky, eds.), Cambridge Uni- UK. Hyun Song Shin (2003), "Coordination Risk and the Price of Debt," forthcoming in European Economic Review. [20] Morris, Stephen, and Hyun Song Shin (2004), "Liquidity Black Holes," forthcoming in Review of Finance. [21] Obstfeld, Maurice (1986), "Rational American Economic Review [22] Crises", 76, 1, 72-81. Obstfeld, Maurice (1996), "Models of Currency Crises with Self-FulQling Features," European Economic Review [23] and Self-FulQling Balance-of-Payments 40, 3-5, 1037-47. Velasco, Andres (1996), "Fixed ity," Exchange Rates: European Economic Review Credibility, Flexibility 40, 3-5, 1023-36. 3837 027 30 and Multiplic- MIT LIBRARIES 3 9080 02617 8373