M^i til m mi' m ,./<b5V-35 /9S3 ALFRED P. WORKING PAPER SLOAN SCHOOL OF MANAGEMENT MULTIPERIOD SECURITIES MARKETS WITH DIFFERENTIAL INFORMATION: MARTINGALES AND RESOLUTION TIMES* by Darrell Duffie and Chi-fu Huang WP //1654-85 January 1985 Revised March 1985 MASSACHUSETTS INSTITUTE OF TECHNOLOGY 50 MEMORIAL DRIVE CAMBRIDGE. MASSACHUSETTS 02139 MULTIPERIOD SECURITIES MARKETS WITH DIFFERENTIAL INFORMATION: MARTINGALES AND RESOLUTION TIMES* by Darrell Duffie and Chi-fu Huang WP //1654-85 January 1985 Revised March 1985 *We would like to thank conversations with Michael Harrison and Andreu Mas-Colell. Some technical help from Kai-Ching Lin of the University of Chicago is appreciated. Table of Contents 1 Introduction and summary 2 Differential Information 1 2 3 The formulation 3 4 The advantage 7 of better information 5 Free lunches and martingales 6 The 7 An example resolution times 8 Discussions, generalizations, and concluding remarks 9 12 17 19 Abstract We model multiperiod securities that admits no free lunches tations. We is markets with differential information. related to martingales when agents have A price system rational expec- introduce a concept called resolution time and show that a better informed agent and a worse informed agent must agree on the resolution times of commonly mar- they have rational expectations and keted events if follows that if all there are no free lunches. It then the elementary events are marketed for a worse informed agent then any price system that admits no cally equalize the information free lunches to a better informed agent must dynami- asymmetry between the two. a dynamically fully revealing price system that complete markets. if is We provide an example of arbitrage free and yields elementarily 1. Introduction and summary This paper addresses differential information in a multiperiod model of security markets. Fo- we cusing on asset price processes that preclude "free lunches," We Kreps connection between price processes and martingales. extend the famous Harrison- first go on to study the information that must be revealed by a better informed agent through price processes to a worse informed agent free lunches are to be precluded. Finally, if we formalize the connection between completely revealing price processes and dynamically complete markets for elementary contingent claims. and additional details summarized below. The results are results of Harrison and Kreps The [1979] connecting the behavior of price processes with martingales have already opened up a theory of dynamic equilibrium with symmetric information (Dufiie and Huang [1985]). The results here are a step toward dynamic equilibrium with differential information. The paper is summarized as follows. Section 2 motivates the concepts of differential information The details can easily continuous-time setting. in a less demanding discrete-time be surmized by Section 3 presents the formal model and shows that, in a rational expectations model, readers. agents agree on the joint variation of stochastic price processes. A some contingent claim is marketed if financed by of the advantages of better information: thus a larger space of marketed claims. some security trading strategy. Section 4 shows a larger space of admisible trading strategies and Section 5 shows that the absence of simple free lunches (Kreps [1981]) implies a unique implicit price process for any marketed claim. A better informed agent cannot, in the absence of simple free lunches, attain a consumption claim at a smaller investment than a worse informed agent. initial Kreps [1981] simple free luDcb concept of "no arbitrage" to the free lunch free lunches for a particular agent, we demonstrate the shows a need to strengthen the in infinite-dimensional cases. Barring existence of a martingale measure for that agent: a probability measure, absolutely continuous with respect to the agent's endowed probability measure, under which security price processes are martingales. In Section 6 event is to we introduce the resolution time of an event, the random, and of course may be different for differently An time one learns that an resolution time of an event informed agents. has no later resolution time than a worse informed agent. if The happen or not to happen with probability one. first event B A is better informed agent marketed for there exists a marketed elementary contingent claim for this event, or equivalently, exists a trading strategy paying one unit of assuming the consumption space is consumption in event B and is an agent if there nothing otherwise. By separable and thereby extending a result of Kreps [1981], we are able to demonstrate the existence of an equivalent martingale measure, barring free lunches. This is a martingale measure assigning strictly positive probability to any event assigned non-zero probability by the agent's endowed probability measure. This allows us to show that the absence free lunches for the better informed agent implies that the resolution times of the better and worse of informed agents for commonly marketed events must be equal (almost surely). all events are marketed for the poorly informed agent and if It follows that, if the better informed agent has no free lunches, then the price system has symmetrized the information, neutralizing the better informed agent's informational advantage. Section 7 provides an example of a fully revealing arbitrage-free price system. Uncertainty modeled by a Brownian Motion. Agent a observes the Brownian Motion observe the Brownian Motion as evolves through time, but it Two the final value of the Brownian Motion. The riskless. /3. 2. Differential which is and one brief concluding remark. model of uncertainty Information central primitive for a An "correct," loosely speaking. any ordered subset of [0,1], allowing Fred receives information termed cr-algebras) on Q. knows whether A CI is assumption which is ^ of C is never forgotten, Vs > 7, F = {7(; an element of the tribe is T correct. For this section the time set an element of A, that is Tt is of "states of the world," one of formalized by specifying a Eltration is or not the correct state that the filtration F Q a set one to cover continuous and discrete time, some subset If is agent, say Fred, receives information through time, refining formalize the notion that such an occurrence A further elementary claims. Section 8 adds for and a Fred's knowledge concerning which of the states in How does not at the beginning of time long-lived securities are traded, one risky Markets are dynamically complete discussion, generalizations, The knows even /? price process of the risky security symmetrizes the original informational assymetry between agents a and some Agent directly. is is, ( is in a single pass. € T} of tribes (also 7t then at time A has "occurred". To whether we say that F is t Fred increasing, or (. often technically convenient when T is a continuum time set is right-continuous, or s>t A different agent, say George, specified by a filtration G= {^,;< might naturally receive € T} at an informational disadvantage at time time t, and perhaps more. If $i C Ti of tribes on C. t, Of course, if ^( C t E T, we write G C then George is George knows at It since Fred "knows" every event that for all through time different information F and say that Fred is better informed than George. The join of 7t and ^j, denoted Jt^ Qi, is the tribe generated by the union of represents the total information held by the two agents at time then at time The t, t. In other words, by pooling their information, Fred and George would know whether joined filtration, denoted F VG is, of course {7t\/ Qt\t E T}. and 7i B \i B €. Qi, and 7t\/ Qt has occurred. The meet (7-algebra), time t of and whether denoted is B $i, on the other hand, and Tt Tt A §f B€ If A 7< the intersection of is In the "rational expectations" genre common it is to is indeed a is and George know independently ^, then Fred has occurred. The meet of the filtrations and §i (which 7t denoted Y AG = {7t A Si]t € T}. assume that agents have private mation and also learn information from "market observables" in particular, , at infor- the market values of traded assets. Market observables can be represented as a vector of stochastic processes, which we S from take to be nothing more than function at time is t t modeled as the is tribe Tf on process If some to 5R^ for integer N. The value B € S up 7t^ then to an agent one will know and including time is at generated by the functions {S{s); s fi S corresponding filtration of tribes generated by if T 5 up denoted S{t), a function on n. The information revealed by observing including time point, x f2 time is denoted F^ whether or not t € filtration is G the the and learns of course Suppose that agents are from a set A, indexed by a, with private information given by process The [0, <]}. t. endowed with private information corresponding to a The market observable to and = {7t^\t G T}. To belabor B has occurred by observing from a vector of market observables S, then the agent's "total" information F'*. S of 5 is fully revea.l'mg GVF . filtrations if yaeA, Yf'^cF'^vf^ a which means that by observing S each agent learns all privately held information not already known. A S stochastic process respect to p/ for all t implicit in G. Indeed, The adapted to a € T. Roughly, S S is adapted to is G G = {get G T} if S(t) is measurable with if 5 only conveys information already to G filtration adapted if and only if F'^ C G. question arises: Are market observables such as prices naturally adapted to the join of private information? This If is is essentially a question about the economic structure of price formation. one believes that market values are a direct consequence of individual strategies, such as bidding, for example, then it would be natural market observables as adapted. to treat One could also imagine, perhaps, that certain market structures add "noise" to price formation. This issue addressed 3. all in Kreps [1977]. is also For our purposes, we are not forced to take a stance. The formulation In this section we consider an intertemporal pure exchange economy under uncertainty with differential information. Taken as primitive u EU denotes a complete description common probability assessment held subset of the interval [0, 1], of the is a complete probability space {Cl,T,P), where each exogenous uncertain environment, and where by agents which includes in the and 1. economy. The time set T is P is the taken to be a assumed that there It is sumption only at time 1. is one consumption good We take V = An u> information structure is (17, V G U. We equip The economy 1 l^a C V; 2 >a is K F" e We F" is an increasing family of subtribes of populated with agents from a set implies that v K v{u!) units of I, consumption r. {7t;t G T}, +k >-a v, where V the positive cone of is interpret that "Va >-a is is = >? E {>:a-,'Vai^°'](^ £ ^}- We shall assumed that there are a degeneracy, assume E V and the strict preference relation derived from >:q, and with the origin deleted; be his preference relation on >:a to we = \ a.s., = number finite 1, 2, . . . , A''. further impose that for Each security n We a normalization. of long-lived securities traded in the and "Va, all n, represented by d„ is shall henceforth P{dn < 1} = economy denote (d„)^_j by d. To rule out 1. {5„(i); n a vector of semimartingales with respect to some information structure F such that F" for all aE in E K. We assume A price system for long-lived securities is an A''-vector of nonnegative processes 5 = 1,2,. .. ,N;t eT} satisfying the following conditions: is a endowed information structure. that X^„_i dn S is, A: strictly increasing in the sense that v Va to be agent a's net trade space, zero net supply indexed by n 1 that F. to be his It is means with the L'-norm topology denoted by a preference relation defined on where 3 EV J, P). Thus v that the following conditions hold for each (>:„, 'Va)^") € available for con- is Let F be a countable collection of information structures. £ltra,tion. k economy which L^{P) to be the commodity space, where L^{P) denotes the space of integrable random variables on at time one in state the in = C F A;^ 2 S„{t) < a.s. I ^5„(0 = and 1 a.s.WtET; (3.1) n=l 3forn = l,2,...,iV, ^(''[5„,5„]i)^ <oo, where {'"[S'„,5„](} is the joint variation process of and Meyer [1982, We assume Requiring loss of generality. S„ and 5„ with respect to F; F is complete in that Tt to be a semimartingale with respect to Any cf. Dellacherie VII.44]. in addition that 5 (3.2) discrete time process 'For the definition of a semimartingale see, for is contains all the P-null sets for some information structure F a semimartingale with respect to example, Jacod [1979, pp.29]. 4 its all is t E T. without natural filtration. time case, almost In the continuous some The assumption filtration). the all that F is known finer processes are semimartingales (with respect to than any of {F";a £ A} allows the possibility that an agent can learn from a price system. There exists discussions Kreps whether F should be at most as [1977], on Our assumption not an issue that concerns us. Condition (3.1) is again Remark assumed that agents denoted by H^iS) is We shall This is general enough to include that as a special case. is is a technical restriction. For a sufficient condition for (3.2) see Dellacherie and Meyer [1982, VII.98]. 3.1: in 5}tra.tion A have it = {hf{S);te generates is denoted by ¥^ = G T}. {7i^;t rationa/ expectations in that they learn from a price system The information to refine their information. S fine as the joint of all agents' information. a normalization and condition (3.2) Given a price system 5, the It is the literature, for example, in structure of agent a after he observes a price system T}, where assume throughout that {{"{S) satisfies the usuai conditions: complete; and 1 Tt = For consistency, we need to require also that d be measurable with respect to h°{S) for all 2 right-continuous: Aux^u' where a E A. That is, u let € be any half open interval contained \r,s) T, then V< e in ['''is), [r,s). at time when 1 be settled, they will know how agents' long and short positions of long-lived securities need to many units of consumption good to which they are entitled or obligated. Before proceeding, Lemma Proof. A 3.1. We we price system first claim that some shall first record is S an A'^-vector of is is an F-semimartingale and [1979] that 5 is is H° {S)-semimartingales By ff"(5)-optional. right continuous, therefore fl^°'(5)-optional; S technical results. cf. S Chung and Williams //''*(5)-optional. an //'"(5)-semimartingale, since construction, It for all a. is [1983, follows then from fl''*(5) is adapted to H°'{S) and Theorem Theorem 3.4]. is Thus 9.19(a) of Jacod a sub-Sltration of F. I If F'^ is system S not coarser than F" and agent will a ignors the information revealed through 5, the price not be an F"-semimartingale, since omy, where trading over time budget constraint for agents must involve stochastic is it is not adapted to F"*. In an intertemporal econ- allowed, before anything interesting can be said, an intertemporal must be formulated. With uncertainty present, integration. Jacod [1979, p. 278-279] has integrals to have desired properties, it is shown that this in budget constraint order for stochastic necessary that the integrators be semimartingales. Thus - when a else, price system carries information with which an agent the agent's intertemporal budget constraint Another lemma Lemma is {^{Sn,S,n]t-te T} and {^"(^^[5„,5„,],; process of Sn and Proof. it. Or not well defined. is < < G T} which e T}, where is a common {^"(^'[5„,5„,],; < version € T} is ^ of the processes the joint variation with respect to H°'{S). Since 5„ and S„i are semimartingales with respect to F and H°'{S), the two processes G T} and {^[5„,5r„],;< VII.42]. 5,,, not endowed, he must learn needed: There exists a process {[5'„,5m](; 3.2. is We first {"°'^^^[Sn,Sm]t;t G T} are well defined; cf. Dellacherie and Meyer [1982, note that and 4 cf. Jacod [1979], Section 2.25. Theorem 9.19(b) of Jacod [1979] shows that there exists a version of {^[5„ + 5m,5„ + 5TO](} {'^"^^'[5„ Sm]t} and - Sm^S,, - {^°'('^'[5„ and + 5m,5'„ + 5^],}, Sm]t}. Therefore there esists a and similarly common for {*"[5n common -5m,5n version of {^[5„,5m](} and {^"(^)[5„,5„.],}. I We shall henceforth use for all a. (This F and H°'{S) is {[5'„,5m]/} to denote a F possible because is a countable since both filtrations are complete. right continuous, {[5n,5TO]/}, {'^[5„, 5„,]t}, Indistinguishable processes are indeed the Lemma will agree common set.) version of {^[5„,5m]/} and {^"^^^[5'n,5m](}| Note that {[5„,5,„](} is adapted to both In addition, since joint variation processes are and {^"^^^[5„,Sm](} are indistinguishable^ processes. same process 3.2 has an important implication. As long on the joint variation processes of security for any pratical purpose. as agents have rational expectations, they prices. In the case that price processes are continuous processes, joint variation processes are covariance processes. Thus rational agents must agree on the variance-covariance matrix process. Now let processes. A Pa{S) be the tribe of subsets of process F : fi x [0, 1] i-h- 5R is Cx [0, 1] generated by said to be left continuous i/" (5) -adapted H" {S)-predictable if it is measurable with respect to Pa{S). Given a price system 5, an admissible trading strategy = {0n{i); n = 1,2,. .. ,N;t GT} for agent a is an A''-vector of processes satisfying the following conditions: ^A process {.Y(()} is a version of another process [Y(t)} if X(t) = Y (t) with 'Two processe {A'(<)} and {V'{0} are indistinguishable if ^(i) = Y{t) for all probability one for all t G T. J £ T with probability one. . 9 1 /f"(5)-predictable; is 2 the stochastic integral d{s)'^dS{s) / Jo is 3 5 well defined with respect to H°'{S) for self-financing, that is = e{o)^s{o)+ f with respect to H'^iS), and n= all 61(1)"^^ yteT e{s)^ds{s) a.s. By {er,{t)fd[Sn,S„]y <00. (3.4) a natural budget constraint and condition (3.4) (3.3) is The advantage 4. technical. is a when the price system is the linearity of the stochastic integral and an application of the Kunita-Watanabe inequality Meyer Dellacherie and (cf. (3.3) € V; Let 0"[5] denote the space of admissible trading strategies for agent 5. S; 6 l,2,...,iV e(J Condition G T, henceforth denoted hy t is, e{t)'^s{t) 4for all We we know 0'*[S] is a linear space. of better information show shall [1982, pp.277]), in this section that a better informed agent will be better the sense that off in he or she has access to more admissible trading strategies and therefore can enjoy a larger feasible net trade space. if agent a has The direct converse of the above statement is not true, however. access to a bigger space of admissible strategies than agent information as refined by a price system must be finer than that of agent /9, say. We show that Then agent a's except possibly at date 13 1. Following Harrison and Kreps [1979] agent 6 is a given a price system S if there exists 6 a and said to generate v for agent we say that a consumption claim v E 0"[S] such ^(0)''^5(0) is that d'^{l)'^d = €V is marketed v a.s. In such event, an implicit price for v at time zero. Let Ala denote the space of marketed consumption claims for agent a. By the fact that ©"[5] space, A/q We is, T,^ is a linear subspace of shall say that agent c Theorem 7," Vf G 4.1. If a is is a linear V better informed than agent /3 if F^ is a sub-filtration of F", that T. agent a is better informed than agent /3, then 0^[5] C ©"[S] and Mp c Ma. Proof. Let 9 for € ©^[S]. We want to show that satisfies 7 the defining properties of 0'*[5]. therefore H°(S) Firstly, the fact that the stochastic integral 9 assertion of Theorem 9.26 of Jacod [1979]. Also, ^(l)'^d Secondly, the fact that £ 5 is V since 6 well defined follows from the G e^[5]. a self-financing strategy with respect to is first follows from the H^lS) second assertion of Theorem 9.26 of Jacod [1979]. Lastly, the fact that 6 satisfies (3.4) As long M^ C M^. € Q^lS] and thus obvious. Therefore 6 is as a better informed agent has rational expectations, whatever strategy a worse in- formed agent can employ can be employed by him. incorrect, however. The converse of the above proposition Consider the following example. Let a and /9 be two agents, whose informa- tion structures cannot be ordered. Suppose that the information generated by the price system is H^{S) c i7"(5) and thus e^[5] C e"[5]. identical to F^. Therefore The is following proposition shows that the statement that 9^ [5] C ©"[5] implies H^[S] C ff'*[5] is also incorrect. C hf{S) Proposition 4.1. Suppose that h^ {S) V< € T/{0}. Then Pp{S) C Pa(S) and e^[5] C G"[S]. Proof. of the 2 of The predictable tribe P^(5) form {0} x Bo and (s, Chung and Williams The B x is 6 h^[S) and B G with Bq h^s{S) for By the hypothesis that 7f C [1983]. P^(5)-predictable rectangle of the assertion follows t] generated by the collection of H^{S)-predictable rectangles is J," V« s,teT,s <t; G T/{0}, we know a ;'„(S) -predictable rectangle. Therefore, Pff{S) from similar arguments as those of Theorem intuition behind Proposition 4.1 is clear. C Chapter cf. that every Pa{S). The 4.1. In a finite horizon single commodity economy where admissible strategies are predictable ones, agents are not able to take advantage better information revealed to the final date! A Theorem Suppose that Proof. and 4.2. correct statement Suppose that there B^ them /if exists hf{S). The rectangle Chung and Williams (5) < at the final date of the /if (5). will be no trading at Then 6^ [5] C e"[S] implies H^{S) C H"{S). G T/{l} such {t, l] economy. There of their as follows. is C rest x B is that hf{S) ^ hf{S). Then an element of Pd{S) but is there = 1(,,i]xb('^,s) (1 d„{oj,s) = -1(,^,]xb(w,s)5,(cj,0 for 8 - Si{ojj)) for all n a set BG hf{S) not an element of Pa{S); [1983, pp. 28]. Define a self-financing trading strategy by ei{u,s) is some yt i. i cf. It is To quickly checked that 6 defined above see this c e^[S] we note that, d~^ (5R+/{0}) = an element of ©''[5]. But 6 is (t,l]x e°\S]. Thus we must have H^{S) B^ is not PQ(S)-predictable. Pa{S). This contradicts the hypothesis that c H"{S). I A corollary available. is = Corollary 4.1. Suppose that h^{S) Proof. Using Theorem /if (S) and that e"[5] = e^[5]. Then H''{S) = H^{S). 4.2 twice gives the result. I It is clear from the above discussion that whether a converse statement of Theorem true will depend upon whether the information conveyed by a price system is 4.1 is a substitute or a complemeDt to agents' information. 5. Free lunches and martingales In this section we shall show that a price system that admits no free lunches is intimately connected with martingales when agents have rational expectations. The definitions of a simple free lunch and a free lunch that we employ are those of Kreps Definition 5.1. A a simple free lunch for agent is [1981]. a strategy 6 € 0°'[S] such that ^(0)''^5(0) < and d{0)^d€K. Definition 5.2. A free lunch for agent a is a net {{d^, f-^); A € A} C e"[5] x V and a bundle k e K such that e^{l)'^d-v^eKu{0}ioia\lX, v^^k, and liminf^^ (0)^5(0) Implicit in the definition of a free lunch ences. We is < 0. obviously a sense of continuity of agent q's prefer- take this definition as a primitive in the analyses to follow and refer interested readers to Kreps [1981] for a host of related issues. Here are some direct consequences of the no simple Proposition 5.1. Suppose that a price system Then for all n = Proof. Obvious. 1,2, . . . , TV, 5„(1) = d„ a.s. free lunch definition. S admits no simple free lunches for some agent a. S admits no Proposition 5.2. Suppose that functional Ra ' Mq "-^ simple free lunches for agent a. There exists a linear that gives implicit prices at time zero for every marketed consumption 3? claim (for agent a). Proof. clear that It is if 5 admits no simple free lunches for agent a, then there exists a unique implicit price for every marketed claim (for agent a). Let 6 m e Ma. Define tt^ : Mo, h^ by 5R ndm) = e 0"[5] be a strategy that generates e{0)'^S{0). I Proposition 5.3. Let m £ Ma he marketed for agent a. If there exist no simple free lunches, then there exists an implicit price process {S^{t);t price process Proof. This is, m for agent a. In addition, the implicit a direct consequence of no simple free lunch and the fact that there exists a portfolio of long-lived securities, available for all agents, (That for uniquely determined up to indistinguishability. is is G T} whose implicit price process hold one share of each long-lived security.) In fact, the assertion where there exist ways is is unit throughout. valid in any economy to transfer strictly positive values across time. I A corollary is immediate. Corollary 5.1. Suppose that agent a is better informed than agent admits no simple free lunches for agent a. Let m € M^. Then (3 and that the price system {5,^,(0} ^"'^ {^m{t)} are indistin- guishable processes. Proof. This a consequence of Theorem 4.1 and Proposition 5.3. I When a price system admits no simple free lunches for a better informed agent, he cannot dynamically manufacture a consumption claim corollary no simple a complement to is free lunches, in the sense that When is Theorem 4.1. less costly than a worse informed agent can. This The advantage some consumption claims can be manufactured $ By the probability measure Qq following proposition *Here we remark that since A on the measurable space {Cl,T) that such that under which the price system The and not Let <E> denote Riesz representative theorem,"* we with the positive cone of L°°{P) with the origin deleted, where L°°{P) denotes the space of essentially bounded random variables on (n, T,P). P there are less costly. a price system admits no free lunches we are able to say a bit more. can identify when strictly in the sense that the space of feasible net trades is bigger the set of nontrivial positive linear functionals on V. to of better information, V is is well 5 is is martingaie measure for agent a a absolutely continuous with respect a vector of //°'(5)-martingales. known. a Banach lattice, any positive linear functional 10 is is continuous. Proposition 5.4. Suppose that extends ttq, Proof. See The Lemma of 1 4>a \ admits do free lunches for agent a, then there exists Ma = Kreps 4>a ^ ^ that ttq. [1981]. following theorem connects the no free lunch condition with martingales. Theorem Qa, and denoted by S 5.1. There exists a one-to-one correspondence between martingale measures linear functionals 4>a €^ with (/>a Qa{B) = ^a[lB) wiere E^{) is Ma = \ e r and >^B The correspondence tTq. <l>^{v) = lies in $. is for agent a, given by VuGV, E*Jv) the expectation under Qa- be an extension of Proof. Let (pa essentially bounded random variable to all of tTq j/a V that on (n,7) such that Thus there <l>a{v) = exists a nonzero positive E{vya) Vv e V. Define a set function Be Qa{B)^ f ya[oj)P[d^) Jb easy to see that It is Qa is 7. Next we want to show that 5 a probability measure on (0,7). is a vector of /^^(Sj-martingales under Qa. Fix an integer i with I <i < N. < Let <i < <2 < and {<i,«2} 1 C T. For any BG /i",(5) consider the following trading strategy: ei{u,t) 0„{u!,t) = 1 = Si{u,t2) claim that d since cf. B G G 6" [5]. Meyer = -S,(w,<i) for = 5,(w,<2) 1, - e (<i,<2] Si{u,ti) for otherwise, t e te all n ^ is and t e tj (<2,1] and (<i,<2] Si{u,ti) for otherwise, for € B anduiE [1982, wG B (<2,1] and w€ B i. 5 is well-defined with respect to H''{S) Chapter VIII]. Secondly, the fact that d is self-financing follows in Theorem 3.1 of Huang [1985a]. we have ^{lo B a left-continuous (therefore predictable) simple trading strategy; from the reasoning similar to that given bounded by - t Firstly, the stochastic integral ^ hf^{S) and since ^ Dellacherie and 5,(w,<i) for = = We - i^"{t)f d{S„,S„]ty < E {[S„,S„],)^ < 11 oo, Thirdly, since | ^ j is . where the second inequality follows from 6{iyd G V. Note that element of = that 9(1^(1 Is 3.3 (5,(<2) and Equation - (3.2). Finally, which 5',(<i)), we want to show bounded and naturally an is V The consumption claim zero. Since Lemma and <i, ^2 are arbitrary, i — 15(5,(^2) it 5',(ii)) is marketed and has a implicit price of zero at time follows that 5 is an i7"(S)-martingales under A'^-vector of Qa. Conversely, positive, and be a martingale measure for agent a. Putting y Qa let = dQo/dP, y is nonzero, bounded. As long as we can show that essentially k(J {On{t)?d[S„,S„],y <<:x., the assertion then follows from arguments similar to those in Theorem 3.1 of Huang [1985a]. Now observe k(J {en{t))^ d[S„, Sr^jtV <(esssup y)E( < where we have used that (^„(0)' c?[5„,5„],y J CX), under a substitution of an fact that joint variation processes are invariant absolutely continuius probability measure. I A corollary Theorem Huang 5.1 immediate: is a generalization of the martingale results of Harrison and Kreps [1979] and [1985a]. If a price 5 then is is system S admits no free lunches for a vector /r"(5)-martingale under some probability respect P. The converse We shall say is an agent a with Qa is that rational expectations, absolutely continuous with also true. that a price system A. In that case, for any agent a S is arbitrage-free if it admits no free lunches for there exists a martingale measure Qa- We all agents in can go on to talk about completeness of the securities markets. That being straightforward generalization of Harrision and Kreps [1979] we leave Remark portfolio < £ T rate, is The whose value for interested readers. driving force behind the martingale result is unit throughout. certainly not necessary. For example, available The some martingale measure. upon We is Thus the assumption that if the fact that there exists a X!)n=i '^n(0 — 1 '^^- f°r ^^' there exists a riskless asset with a zero interest then a price system that admits no free lunches respect to 6. 5.1: it for, say, agent a is a //°'(5)-martingale with prepared detailed derivations of this case. They are request. resolution times In the last section we have demonstrated that an arbitrage free price system is intimately connected with martingales. In this section we shall show that no free lunches necessitates that a 12 better informed agent and a worse informed agent agree on the resolution times of a particular set of events. That We equalized. through an arbitrage free price system some endowed asymmetry of information is, is go on to formalize the linkage between dynamically elementary market completeness, Under some to be defined, and a dynamically fully revealing price system. regularity conditions a price system that admits no free lunches for a better informed agent and yields elementary completeness of markets for a worse informed agent must convey better informed agent BG Let is endowed with to the worse informed agents over time. h^{S). Define an optional time r| = inf {« € [0, 1] where as usual when the infimum does not we have E{\b h") \ = Ijg a.s. T^ : : we Tq is the first >-^ Tg exist [0,oo] by h^{S)) \ B mapping from time that agent for agent a. q its to fl is range is [0, l], B following proposition formalizes this intuition. Let 6.1. Proof. Define Be f^ : fi h^{S). Then ^ [0,oo] BG Some h"{S), lies in [0,1]. We however. Thus without loss of if We agent a will not is to happen shall thus call is or not to Tg the resolution better informed than agent be later than those for agent technical happen lemmas /?. /?, The are first recorded. h^a{S). by = It is B sure that event € h^[S) Tq B€ [0, 1]. Intuition suggests that then agent a's resolution times for every Lemma or 0}, 1 takes the value oo. Note that since with P-probability one, after observing the price system. time of event = Therefore, except on a set of P-measure zero, shall treat TJi as a Literally, Q E{Ib can always redefine Tg on a P-nuU set such that generality the information that the all Vw ^ B. oo clear that f§ = M{te[0,l]:E{lB\hf{S)) = Thus Tg is and Meyer an if" (S)-optional random variable. [1982] that BG It l}. then follows from Theorem IV.53 of Dellacherie h^^{S). I Remark For the definitions of optional random variables and hj-a{S), see Chung and 6.1: Williams [1983, Section Lemma 6.2. If Furthermore, Proof. Let T 1.7]. H^{S) C H°'{S), then any H^{S)-optional random let T be H^[S)-optional. Then h^{S) C h^{S). be /r''(S)-optional. Then we have {T<t}ehf{S) 13 V<GT variable is H^iSyoptional. C hf{S) by the definition of optionality. Since hf{S) {T<t}eh^{S) T This, by the definition of optionality, implies that Next we note that h^{S) contains yt e B e Let T. C h^{S). Since H'^S) VieT. /f"(5)-optional. is B e the sets all e T, we know Vi hf(S) such that we know H'^iS), B E h^{S) and Bf]{T < t} e hf {S) t} e hf{S) BfllT < WteT. Thus Beh^iS). I C Proposition 6.1. Suppose that F^ Proof. From the hypothesis, it is BE F". Let claer that /if Then T^ < T| (5). P-a.s. H^{S) C /f"(5). From Lemma 6.1 we know that BEh^.{S). Thus E{1b \ ^ From Lemma 6.2 E [\b /ij,(5)) I /i°'^(5)) lies in [0,1] iterative expectation we get B =e{e [ib = | P-a.s. 1b B we know that h^{S) C h"g{S). Therefore by ^B Since £'(1b = h^,{S)) /i;:,(5)) 1 /ij,(5)) I F- a.s. 1b with P-probability one, we must have P-a.s., E(\B\h''j.,J,S)\^lB which implies that T^{S) < T^{S) P-a.s. I The following proposition shows that the definition of resolution times invariant under a is substitution of an equivalent probability measure. Proposition 6.2. Let Tg be B E h"{S), resolution times for B let Q be a probability measure equivalent to P, and with respect to P and Q, respectively. Then Tg = Tg let T^ and P-a.s. and therefore Q-a.s. Proof. Putting ^ = dQ/dP and fixing a right continuous version ^{t) absolutely continuous with respect to F, {^(0} P-null set. We is have = ^ 1b 14 \ h"[S)). Since Q is a strictly positive process except possibly on a E[\Bi\h''MS)) EQ{lB\h"T^)= = E[^ P- a.s. This implies that Tq < Tg and therefore Q-a.s. P-a.s. Q since absolutely continuous with is respect to P. By the hypothesis that Q and P are in fact equivalent, P we know = Tg P-a.s. absolutely continuous Tg > Tg with respect to Q. Thus we can reverse the above argument and have Hence we must have Tg is F-a.s. and Q-a.s. and Q-a.s. I Before proceeding to the main theorems of this section, we give first some strengthings of the results of Section 5. We assmue henceforth that the probability space (n,7,P) shall the space of strictly positive linear functionals on a is a martingale measure for him that Let follows from an extension of Theorem ^ denote equivaient martingale measure for agent . equivalent to P. V'a I ^^a = whose proof 3 of kreps [1981], S admits do Proposition 6.3. Suppose that Proof. Let denote the space of strictly positive linear functionals on V. The following proposition '9 such that is V An separable. is is given in the Appendix. i/jq G ^ a separable normed space. The Then there free lunches for agent a. exists ^a- Given that {U,7,P) a separable probability space, is V is assertion then follows from the theorem in the Appendix. I A direct consequence of the above proposition and Theorem 5.1 is: Proposition 6.4. There exists a one-to-one correspondence between equivalent martingale measures for agent a, Qa, and linear functionals V^q € ^ with V'o | Ma = The correspondence ttq. is given by Qa{B) = yBeT rPa{lB) and4>a{v) = E;{v) Vt; € F, wiere Ea{) denotes the expectation under Qa- When the probability space a strictly positive extension of is tt^ separable, no free lunches for agent to all of V . This in a implies the existence of turn implies the existence of an equivalent martingale measure for agent a. Since all the probability measures to appear will be equivalent probability measures, we shall henceforth simply use a.s. to denote almost surely under any probability measure involved. Here Theorem is our 6.1. first main result of this section. Suppose that agent a is admits do free lunches for agent a. Let better informed than agent BG /if (5) be such tiat 1b r^ = T^ P-a.s. 15 /3, € and tiat the price system Afg. Tien Let {S'^(i)} ^^^ {^'bIOI denote the implicit price processes for Is for agents Proof. From Corollary respectively. Therefore, from now on we From Propositions 5.1 we know that {S%[i)) and {5^(i)} 5.1, 6.3, and for agents Q^ B El{lB\h^i{S)) V<€T a.s. Vi€T a.s., and E*J) denote the expectations under equivalent martingale measures a and /?, respectively. under {H^{S),Qfi) are equal almost times of are indistinguishable processes. and 6.4 we know that = £'„(•) /9, use {5b(<)} to denote both. SB{t)^E*,[lB\h1[S)) where as usual a and Thus the resolution times of B Qa under [H'^{S),Qa) and then follows from Proposition 6.2 that the resolution surely. It under {H°'{S),P) and under {H^{S),P) are equal almost surely. I Commonly marketed If the resolution time of events will be revealed to differentially informed agents in the B under (F°,F) strictly earlier is theorem implies that the price system that admits no We shall say A price system 5 Vf than that under (F^,F), the above free lunches for agent B information to equalize the assymetry of information about a must between agents a and that markets are elementarily complete ioi agent is same time. a if 1b G said to be dynamically fully revealing between agents Ma a and carry enougii jS. B G /if (5). if hf{S) = h^ (5) for all /? G T. Here Theorem our second main result of this section. is 6.2. Suppose that agent a better informed than agent is markets are eiementariiy complete for agent agent a must that T{' = and that J" if implies that pleteness hypothesis implies that any event in h'^{S) (5) /if a and is to exists t Now we G T claim that hf{S) such that hf [S) know P{B) > since all = - h^{S). marketed. h^ {S) ^ hf{S). Let the filtrations are completed. Also, E{lB\hf{S))==lB and on a set of strictly positive It I hf{S)) 16 ^ 1b. B G we have a.s. F-measure E{1b • If the The elementary market com- hf [S) V< G T. Suppose this C hf{S) and ^i 13. then follows from Theorem 6.1 that the resolution times for all the events in h^{S) under {H°'{S),P) are equal. = then a price system that admits no free lunches for /?, be dynamically fully revealing between The hypothesis Proof. /3 is and under {H^{S),P) not the case. Then there hf{S) and B ^ hfiS). We By the right-continuity of the filtrations, we know the resolution times for B under H'^{S) and under H^{S) are not equal, a contradiction. I A corollary immediate: is Corollary 6.1. Suppose that agent a in is A, and that niarkets are elementarily complete for agent system must be dynamically This section, in = the least informed agents in A, that 7^ T{ for Then any arbitrage (3. all agent free price fully revealing in the sense that h^{S) = our view, is e T and V( hf{S) agents in A. all the most important contribution of this paper. recognized by probability theorists that the way information revealed over time is is It has been closely related to the behaviors of martingales and optional times. In an economic context, the former has been extensively studied by proved in Huang [1985a]; while the latter is introduced in this section. the Appendix and used to demonstrate Proposition 6.3 conditions in Theorem of independent interest. is 7. we have The 3 of Kreps [1981] to insure that there exists a strictly positive continuous extension of a linear functional are sometimes hard to verify in applications. along as The theorem It turns out that as separability matters are simple. An example In this section we present an example along the lines of securities traded, one risky and one standard Brownian motion as it riskless informed and there is its 5.1. There are two long-lived with a zero interest rate. Agent comes along. Agent ^ knows motion at time one but cannot observe Remark a gets to observe a at time zero the value of the sample paths over time. Two Brownian agents are differentially no ordering between the two endowed information structures. Markets are elementarily complete for both agents and the price system is dynamically fully revealing between them. Formally, there denoted by W . The We W = is defined on the basic probability space (n, /, P) a standard Brownian motion, {W{t);t e set of trading dates of the assume that F" = We [0,1]}. denote by F^^ economy F^^ and that, for T all is t if 1 at the very beginning agent he could observe W directly, There are two long-lived di{u)) j3 is {5;"';< taken to be e T, 7f Note that although consumption only occurs at time at time = 1, = G [0,1]} the natural filtration of [0, 1]. (t{W{1)}, the tribe generated by W{1). knowing the value of the Brownian motion not better informed. The optimal might well be path-dependent. securities traded. Their payoff structures are = In(o^) 17 net trade for agent /9, where 7 The a real number. is price system Si{uj,t) = ln(w) 52(0;, = W{uj, + 1) is W{u,t) - (^7 J^ + [_s ) Corollary 1.1 of Jeulin and Yor [1979] implies that w{t) Since 7 - t)w{i) = + {i-t)(f j^^dS2{s) + + (1 is a constant, this implies that {W{t) iy(i) + - (1 generate the same filtration, denoted henceforth by F^ J ^ YT's'^^) ^ [0' 1] t)\V{l)]t € = € T}. We assume {7t^:,t P- and {S2{t);t G [0,1]} ^•*- [0,1]} that F'^ has been completed. Corollary 1.1(a) of Jeulin and Yor implies that £>0 Thus ¥^ H^{S) right-continuous and is = ¥^ That . is, is finer the price system Next we claim that 5 admits is than either F" or F^. dynamically no free lunches. It fully revealing It suffices follows then that between a and H"{S) = /?. to demonstrate the existence of a martingale measure. Put Jo Theorem 1 of Jeulin to F'^ under P. and Yor shows that {W{t);t E T} Now = s a standard Brownian motion with respect exp{7'M^(l)-i7'}, set function Q{B) = [ Jb Theorem on is - define e and a 1 (r2, 6.1 and lemma 6.5 Tf) equivalent to P. of Liptser ^{co)P{doj) and Shiryayev B G 7f [1977] ensures that then follows from the Girsanov theorem It Q is (cf. a probability measure Liptser and Shiryayev [1977, pp. 225]) that W*{t) is = W{t)--it = W{1) - f (1+ imi^I^^ ds teT a standard Brownian motion adapted to T^ under Q. Thus S2 S2{t) = W*{t) + a F'^-martingale under Q, since W{l). Finally, the fact that 1.1(e) of Jeulin is and Yor markets are elementarily complete for both agents follows from Corollary [1979]. 18 Discussions, generalizations, and concluding remarks 8. The assumption that the price system has been normalized cannot be taken lightly. Walrasian economy, a normalization of prices economy such a procedure of a price system may be will not economically neutral. In a rational expectations is be economically neutral Take 5 to be a vector regularity conditions except that they sum to a process normalize prices to sum to one, denoted by 5*, but generated by S. Then The assumption all The information in general. Our altered by changing numeraires. the following direction, however. analysis can be generalized in bounded away from zero. We is made the general case without further assumption. condition needed Other than the criticism arbitrage is a can now for the ease of exposition. economy where the commodity space [1985b]. Results on resolution times that need an equivalent martingale No the the characterizations in this paper apply to 5*. that consumption only occurs at date one The all agents have access to the information the space of bounded variation processes representing agents' accumulated net trades; longer normable. content F-semimartingales satisfying of still let All the results before Section 6 are readily extended to an in In a is in the first weak requirement ment has pervasive implications. The reason commodity space that the cf. paragraph, the results presented common We Huang measure cannot be treated that a price system be viable; of a price system. In a is cf. is Kreps in this is no [1981]. paper are robust. have demonstrated that this require- information economy, it is the connection between martingales and an arbitrage free price system observed by Harrison and Kreps [1979] that makes a dynamic equilibrium theory possible; context, no arbitrage must is the satisfy. It necessitates minimum cf. DufRe and Huang [1985]. In a differential information condition a rational expectations equilibrium price system that a price system be a martingale under some probability. Readers should convince themselves that putting this papar and Duffie and Huang [1985] together hard to prove the exietence of a dynamic rational expectations equilibrium with a price system. is Whether there exists a partially revealing an open question. 19 it is not fully revealing dynamic rational expectations equilibrium Appendix Let A' be a separable normed space and and X2 be elements is We of A'. M a pair (Af,7r), a subspace markets model is shall write xi of A' be a cone let A' > X2 xi if - X2 == 1,2, 'f^n '^ . . .} x„ ^f k Xn E that ip \ Proof. M= A.l. Suppose that n, where From Lemma ^ is 5 of (M, tt) n. It follows that F weak* topology and E^=i A„ = 1. Now let on K 0. Then there exists a. i) E the space of strictly positive continuous linear functionals on — A: T {tpk'ik = {A„; n = 1,2, . . '^ such X. exists a collection of equicontinuous G K} such that i'kik) > a separable metric space in the relative weak* topology; Let {V')t„;" 5.4.7 of Schaefer [1980]. A securities markets model M. A free lunch in the securities A" such that Kreps [1981] we know that there is with the origin deleted. Let xi U{0}- a.dmits no free lunches. positive continuous linear functionals on M= A' tt c A/ x and lim infn7r(mn) < Theorem G and a linear functional a sequence {(m„,a:n);n in A' and i/'Jb | cf. Theorem 1,2,...} be a countable dense subset of F in the relative .} be a sequence of strictly positive real numbers such that define oo n=l It is quickly verified that V' is a strictly positive continuous linear functional on A" with 20 xIj \ M= tt. References 1 K. Chung and R. Williams, An Introduction to Stochastic Integration, Birkhauser Boston Inc., 1983. 2 C. Dellacherie and P. Meyer, Probabilities and Potential A: General Theory of Process, North- Holland Publishing Company, 3 C. Dellacherie and New York, 1978. Meyer, Probabilities and Potential B: Theory of Martingales, North- P. Holland Publishing Company, New York, 1982. 4 D. Duffie and C. Huang, Implementing Arrow-Debreu equilibria by continuous trading of few long-linved securities, forthcoming in Econometrica, 1985. 5 D. Duffie, Stochastic equilibria: existence, spanning number, and the 'no expected gains from Graduate School trade' hypothesis. Research Paper No. 762, of Business, Stanford University, 1984. 6 M. Harrison and D. Kreps, Martingales and arbitrage in multiperiod securities markets, J. Econ. Theor. 20 (1979), 381-408. 7 C. Huang, Information structure and equilibrium asset prices, J. Econ. Theor. forthcoming 1985. 8 C. 9 J. Huang, Information structure and viable price systems, Jacod, Calcul Stochastique Springer- Verlag, 10 T. Jeulin and abilites New 11 D. Kreps, A in note on mimeo, 1985. Problemes de Martingales, Lecture Notes in Mathematics 714, York, 1979. M. Yor, XHI, L.N. et MIT Inegalites de Hardy, semi-martingales et faux amis, Seminaire de Prob- Math. 721, 332-359, 1979. fulfilled expectations equilibrium, Journal of Economic Theory, 1977. 12 D. Kreps, Arbitrage and equilibrium in economies with infinitely many commodities, J. Math. Econ. 8 (1981), 15-35. 13 R. Liptser and A. Shiryayev, Statistics of Verlag, New Random Processes I: General Theory, Springer- York, 1977. 14 H. Schaefer, Topological Vector Spaces, Springer- Verlag, 21 New York, 1980. l*5^lv 52 NovToia NAY S 1 WSI Lib-26-67 MIT LIBRARIES 3 SDflO DDM Sfll flED