On the impossibility of weak-form efficient markets Steve L Slezak. Journal of Financial and Quantitative Analysis. Seattle: Sep 2003.Vol.38, Iss. 3; pg. 523 http://proquest.umi.com/pqdweb?did=419627551&sid=10&Fmt=4&clientId=68814&RQT =309&VName=PQD Abstract (Document Summary) Recent theoretical models show that irrational expectations can generate return predictability consistent with apparent violations of weak-form market efficiency documented in the empirical literature. These behavioral models constrain rational investors' ability to exploit inter-temporal predictability by assuming that rational agents face high transactions costs, are myopic, or are non-existent. This paper presents a model in which there are two types of irrational expectations, one that causes momentum and another that creates reversals. I investigate whether these types of predictability will persist in the presence of fully rational agents who face no transactions costs, are long lived, and trade dynamically to optimally exploit any predictability due to irrational mispricings. I show that weak-form market efficiency will be violated under two very weak conditions: rational investors are risk averse and the fundamental value of the asset is risky. The paper also investigates the accumulation of wealth by trader type and shows that irrational agents will survive under a large set of parameters. [PUBLICATION ABSTRACT] Full Text (7699 words) Copyright %Washington% Sep 2003[Headnote] Abstract Recent theoretical models show that irrational expectations can generate return predictability consistent with apparent violations of weak-form market efficiency documented in the empirical literature. These behavioral models constrain rational investors' ability to exploit inter-temporal predictability by assuming that rational agents face high transactions costs, are myopic, or are non-existent. This paper presents a model in which there are two types of irrational expectations, one that causes momentum and another that creates reversals. I investigate whether these types of predictability will persist in the presence of fully rational agents who face no transactions costs, are long lived, and trade dynamically to optimally exploit any predictability due to irrational mispricings. I show that weak-form market efficiency will be violated under two very weak conditions: rational investors are risk averse and the fundamental value of the asset is risky. The paper also investigates the accumulation of wealth by trader type and shows that irrational agents will survive under a large set of parameters. I. Introduction The weak form of the efficient market hypothesis states that the price of a security at any point in time should reflect all of the relevant information about the security's future value that can be gleaned from past market information on price and volume. In risk-neutral markets, this implies that prices must follow a random walk, with future returns being completely unpredictable given past returns. In markets with risk-averse investors who require risk premia, the weak form implies that future returns in excess of predictable risk premia cannot be predicted given past price movements. Recent empirical research, however, has documented dynamic return phenomena that appear to be inconsistent with the weak form of the efficient market hypothesis. For example, studies have documented momentum, whereby stock returns (over intervals of specific length) are positively correlated, and reversals, whereby stock returns are negatively correlated. Both phenomena suggest inter-temporal predictability in stock returns based on past price information. In addition, there are studies that document predictable post-event price behavior, such as post-earnings drift (Bernard and Thomas (1989), (1990)) and long-term predictability after seasoned equity offerings (see Loughran and Ritter (1997), for example). However, there is considerable debate (both in the printed literature and in spoken conversation) regarding whether the documented phenomena constitute profitable trade opportunities (see, for example, De Bondt and Thaler (1985) and Jegadeesh and Titman (1993)), statistical aberrations (see Richardson (1993) and Lo and MacKinlay (1990)), or compensation for risk (see Grundy and Martin (2001)).1 Recent behavioral models have been developed to explain these phenomena by assuming investors make specific types of information processing errors that have been documented in the psychology literature. For example, Daniel, Hirshleifer, and Subrahmanyam (1998) show that overconfidence and biased self-attribution (the phenomenon that agents excessively attribute success to ability while excessively discounting low ability as the possible source of failure) can generate short horizon momentum followed by reversals at longer horizons. In Barberis, Shleifer, and Vishny (1998), investors behave in a manner consistent with representativeness, in which agents view events as typical or representative, in violation of the probabilistic laws governing the event process. Hong and Stein (1999) also generate momentum followed by reversals by considering the interaction of two types of boundedly rational agents, one type that conditions on information about the future but ignores current and past price and another type that forms overly simple forecasts based solely on lagged information. Kyle and Wang (1997), Odean (1998), and Gervais and Odean (2001) consider models in which the agents are overconfident about the quality of the information they trade on, while De Long, Shleifer, Summers, and Waldmann (1990a), (1990b) consider a model in which some investors make forecast errors that are correlated with their information sets. In all of the behavioral models discussed above, rational traders are limited in their ability to exploit the mispricings generated by irrational investors, either by their absence or via their myopia.2 In either case, it is a priori unclear whether the types of dynamic phenomena that obtain in these models will persist if long-lived dynamically rational agents are active in the market. Furthermore, it is also unclear whether it would ever pay an investor to learn to become rational. That is, even if agents start with these psychological biases, might they learn to overcome their natural tendencies when faced with the prospect of speculative profits? It seems natural to inquire whether an agent who learns to be rational will be rewarded and whether the trades he/she uses to exploit mispricings will make the irrational phenomena disappear in equilibrium. If so, then these behavioral explanations may not be very robust or convincing. Rather than investigate the robustness of the above-cited behavioral models by examining the equilibria that obtain when fully rational agents are added to each model, I chose to generally examine robustness by taking the common feature of their equilibria, namely inter-temporal predictability, and to see if it survives the presence of rational investors.3 In particular, I examine two types of irrational expectations that can potentially generate momentum or reversals. The main result of the paper is that intertemporal predictability generated by irrational expectations is robust to the inclusion of dynamic rational agents under very weak conditions. I purposefully define an economy in which it is most likely that rational trade will correct any mispricings and then show that, even in this economy, irrational trades affect the equilibrium price process. By showing that behavioral phenomena persist in my setting, it is clear that they will also persist in less frictionless settings (as in Shleifer and Vishny (1997), who examine limits to arbitrage due to institutional constraints). As frictions lessen over time with the advent of better institutional features and order processing, my results indicate that irrational behavior will continue to matter. To define an economy in which rational traders are most able to correct irrational behavior, I employ the following assumptions. First, I consider an infinite agent economy in which rational agents are competitive price takers. In this setting, competing rational agents have no incentive to let deviations from fundamental value persist (perhaps to generate a continued flow of profitable trade opportunities). Second, I assume that the rational agents are long lived and optimally choose demands to maximize expected utility of terminal wealth, exploiting any predictable variation in returns that the irrational agents generate. That is, I assume that rational agents solve a dynamic programming problem to maximize their expected utility of terminal wealth given all the profitable trade opportunities provided by irrational agents. I provide closed-form solutions for optimal dynamic demand in the presence of correlated normally distributed returns without relying on approximations. Third, I abstract away from the specific predictable patterns documented in the empirical literature and consider only a simple predictable deviation from fundamental value. Thus, there are no informational issues, such as structural/parameter uncertainty (as in Brav and Heaton (2002)), that prevent rational agents from recognizing deviations from fundamental value. Furthermore, I assume no transactions costs, short-sale constraints, borrowing constraints, or asymmetric information and abstract away from consumption/production effects that might generate time-varying risk premia. That is, I allow rational agents to trade in a perfectly frictionless market in which the source of predictability is well known to be due to irrational mispricings. I show that even in this frictionless market, optimal dynamic rational trade will not drive prices back to fundamental value. While there are no exogenous barriers that prevent competition from rational agents in the model I develop, I assume that rational agents are risk averse and face fundamental risk. However, I prove that only in the limit as either fundamental risk or rational risk aversion goes to 0 will irrational behavior have no effect on equilibrium prices and markets be weak-form efficient.4 This result is in contrast to the result in De Long, Shleifer, Summers, and Waldmann (DSSW) (1990b) that irrational traders cause deviations from fundamental value even when there is no fundamental risk. Unlike my model, DSSW considers an economy in which rational investors are myopic. Given myopia, unpredictable future irrational expectations impose risk, which prevents the risk-averse myopic rational investors from forcing price to equal fundamental value in equilibrium. Although DSSW state that the impact of irrational trade will diminish as the horizon of the rational agents increases, they do not consider situations in which the rational investors can trade multiple times during that longer horizon. Rather, they consider only rational agents that must maintain a given position over their investment horizon. Many of the empirical phenomena on inter-temporal predictability, however, concern behavior over fairly short horizons.5 Thus, the fixed-position strategies of DSSW will not be effective at exploiting the types of inter-temporal predictability in the data. In my model, when rational investors can trade frequently, are not myopic, and have investment horizons that extend beyond the predictability, rational trades eliminate any predictability in equilibrium when there is no fundamental risk. If fundamental value is risky, however, then inter-temporal predictability caused by irrational expectations will exist in equilibrium and markets will not be weak-form efficient. The paper also examines the wealth accumulation of rational and irrational investors and shows that although the rational agents have greater returns on average, irrational agents can still enjoy positive growth in their wealth. Numerical analysis shows that both momentum and reversals are consistent with the positive wealth growth for irrational agents in equilibrium. The paper also shows that the return to being rational is bounded and falls as the proportion of rational investors increases. In the limit as all investors become rational, the return to being rational falls to 0. Thus, if irrational agents can choose to become rational on the basis of expected return, a mass of irrational investors will persist in equilibrium as long as there is a positive cost associated with overcoming an irrational tendency. The paper is organized as follows. Section II describes the economy: fundamental value is defined and the structure of irrational expectations and irrational demands are specified. In Section III, the solution to the dynamic programming problem of the dynamically rational agents is specified and discussed. In Section IV, the equilibrium price function is derived from the market-clearing condition. In Section V, various properties of the equilibrium price function under specific special cases are investigated. This section also derives the conditions under which irrational traders affect price. Section VI describes the accumulation of wealth by rational and irrational traders and discusses survivorship and the incentive to be rational. Section VII provides concluding comments. II. The Model A. Fundamental Value and the Arrival of Information A single risky security and riskless money are traded by rational and irrational investors in a market that is open at dates t - 1, 2, . . . , T - 1. Without a loss of generality, the riskless asset generates a zero net return and, as numeraire, has a price equal to 1 at each trade date t. The price of the risky asset, however, fluctuates given the arrival of information. At trade date t, the price of the risky security is P^sub t^. The information that affects the price is with respect to a single liquidating dividend that is paid at the terminal date T. The firm does not pay dividends in any other periods and the per-share liquidating dividend is denoted by v. The per-capita supply of the risky security is X. Enlarge 200% Enlarge 400% B. Rational and Irrational Expectations Here I make a distinction between a publicly announced objective value implication and publicly available information that must first be processed to yield a subjective value implication. I assume that the public announcement of [eta]^sub t^ at t is the first type of information; once [eta]^sub t^ is publicly announced, all investors agree that it is objectively correct. I will henceforth refer to the public announcement of [eta]^sub t^ as objective information at t. An example of the second type of information includes newspaper articles on market conditions in a particular industry or on the discovery of a new technology. Because this type of information must be processed by an individual to yield a value implication, I refer to this information as subjective information. Even though investors may possess common subjective information (they all read the same newspapers), they may have different expectations on the objective value implication that will be publicly announced in the future. At date t, all agents (rational and irrational) possess the same information set [Omega]^sub t^, where [Omega]^sub t^ includes all past and current objective information (including past prices, past public announcements of [eta]^sub t^ (for [tau] < t), and the current objective announcement of [eta]^sub t^), and current subjective information.6 That is, [Omega]^sub t^ = {O^sub t^, S^sub t^}, where O^sub t^ = {[eta]^sub 1^,[eta]^sub 2^,...,[eta]^sub 1^, P^sub 1^, P^sub 2^,..., P^sub t^} denotes the objective information at t, and S^sub t^ denotes the subjective information at t. For concreteness, I assume that the subjective information received ar t allows investors to predict only the next period's objective public announcement of [eta]^sub t^+1. Although the various types of investors have the same information set [Omega]^sub t^, they differ in the way in which they process the subjective information in forming their forecasts of [eta]^sub t^+1.7 I denote the conditional forecast of an investor of type i in period t by E^sup i^[[eta]^sub t+1^/[Omega]^sub t^]. Enlarge 200% Enlarge 400% I assume that there are two different types of irrational investors. Their expectations of [eta]^sub t+1^ given [Omega]^sub t^, differ from the forecast of rational investors, with the differences having a specific form for each type. So that I can assess the robustness of behavioral explanations of both momentum and reversals, I allow irrationality to take two forms, one form that, if dominant, will generate positive serial correlation (as in momentum), and another form that, if dominant, will generate negative serial correlation (as in reversals). One type of irrational investor (denoted C) makes a mistake that is correlated with the subjective information contained in his/her information set [Omega]^sub t^. Specifically, the expectation of the next period news announcement for type C investors takes the following specific form, (3) E^sup C^[[eta]^sub t+1^/[Omega]^sub t^] = [eta]^sub t+1^ + [epsilon]^sub t^, where [epsilon]^sub t^ is normally distributed with mean 0 and variance [sigma]^sup 2^^sub [epsilon]^sub t^^ and is independent of [eta]^sub t+1^. As I will show below, this type of non-rational forecast creates price behavior that makes a contrarian trade strategy profitable. Thus, I denote investors who have this type of irrational expectation by C.8 The expectations of the type C agents are irrational since the forecast error, [eta]^sub t+1^ - E^sup C^[[eta]^sub t+1^\[Omega]^sub t^] = [epsilon]^sub t^, is correlated with the forecast, E^sup C^[[eta]^sub t+1^\[Omega]^sub t^] = [eta]^sub t+1^ + [epsilon]^sub t^. As a result of this correlation, type C forecasts are biased. For ease of exposition, I will refer to [epsilon]^sub t^ as the bias shock at t. Enlarge 200% Enlarge 400% In contrast to some of the behavioral models, I assume that there is no asymmetric information across the different investor types. By adopting this assumption, however, I subject myself to the following critique in the event that inter-temporal predictability does not survive rational trade. If inter-temporal predictability does not survive rational trade, then my result may be erroneous due to an internal inconsistency. For example, in some of the behavioral models, the irrational agents possess asymmetric (superior) information. If my model ignores asymmetric information (by focusing on the intertemporal predictability of returns and not its source), then the rational agents in my model may trade more aggressively than they would if they were trading against (albeit irrational) better informed agents; they may trade so much more aggressively that they eliminate any inter-temporal predictability. However, if my model incorporated informational asymmetry, then rational agents may trade less aggressively, allowing predictability to persist. Thus, if inter-temporal predictability did not survive rational trade, the result implied by my approach would be misleading. My approach is not misleading, however, if I find that inter-temporal predictability persists. If rational investors cannot eliminate the behavior in a frictionless world without asymmetric information, then they will not be able to eliminate it in markets with frictions, whatever the source. Alternatively, if the rational agents would trade less in the face of asymmetric information, then the results from my approach are biased in favor of finding no predictability. As I show below, however, my approach shows that predictability persists and, as a consequence, my results are informative. C. The Objective Functions of Investors Enlarge 200% Enlarge 400% Enlarge 200% Enlarge 400% I assume that even if [Delta][pi]^sup t+1^^sub 0^ changes through time, the econometrician knows its deterministic path. Given this knowledge, the covariance of adjacent price changes is simply cov([Delta]P^sub t+1^, [Delta]P^sub t^) = (1 - [pi]^sup t^^sub 1^)[pi]^sup t^^sub 1^[sigma]^sup 2^^sub [eta]t+1^ - ([pi]^sup t^^sub 2^)^sup 2^[sigma]^sup 2^^sub [epsilon]t^ while the covariance of non-adjacent price changes is 0 (i.e., cov([Delta]P^sub t+1+k^, [Delta]P^sub t^) = 0, for k > 0). I define any non-zero covariance between excess price changes as a violation of weak-form market efficiency. A non-zero covariance implies that, once the predictable part of the price change that is due to risk premia is taken out, the past excess price change can (at least partially) predict the subsequent excess price change. The first part of the covariance of adjacent price changes is due to the correction of nonrationally reflected subjective information on [eta]^sub t+1^ at t getting rationally reflected once [eta]^sub t+1^ is objectively announced at t + 1. If [pi]^sup t^^sub 1^ < 1, then a component of price displays momentum since a component of adjacent price changes is positively correlated. The more current prices reflect the rational expectation of [eta]^sub t+1^ (i.e., the closer [pi]^sup t^^sub 1^ is to 1), the less the amount of momentum or positive correlation that exists. The second part of the covariance is negative as the price in period t + 1 bounces back from the temporary deviation from fundamental value caused by the irrational component of irrational expectations at t (the bias shock [epsilon]^sub t^) once the price reflects the objective announcement of [eta]^sub t+1^ at t + 1. Thus, when the effect of the bias shock at t is reversed at t + 1, a portion of the price change is negatively correlated. Whether the serial covariance is positive or negative depends upon the relative strength of the momentum and reversion effects discussed above. The covariance is 0 only when i) the effects completely offset one another or ii) [pi]^sup t^^sub 1^ = 1 and [pi]^sup t^^sub 2^ = 0. This latter condition holds when only the rational expectation is accurately reflected and the irrational component of irrational expectations has no effect on equilibrium price. In all other cases, there is unconditional short horizon predictability in returns. The issue is whether the trades of rational agents, designed to exploit this predictability, will, in equilibrium, make the predictability disappear. III. Optimal Demands Enlarge 200% Enlarge 400% Enlarge 200% Enlarge 400% In contrast to irrational investors, the rational investors are long lived and solve a dynamic programming problem that maximizes their expected utility of terminal wealth given expected future profitable trading opportunities created by irrational expectations. Note that the price function in equation (8) above admits a wide variety of price dynamics depending upon the series of values for [pi]^sup t^^sub 0^, [pi]^sup t^^sub 1^, and [pi]^sup t^^sub 2^. Since I do not want to a priori restrict the dynamic behavior of returns that might result, I must derive optimal demands for rational agents for arbitrary price processes (with arbitrary dynamics). In my model, the only restriction that is imposed (for tractability) is that current and future prices are jointly normally distributed. The following theorem (which is proven in Slezak (1994)) provides the optimal demands for a long-lived dynamically rational investor in that setting. Enlarge 200% Enlarge 400% Enlarge 200% Enlarge 400% IV. The Equilibrium Price Process This section derives the market-clearing equilibrium price process and examines whether the irrational component of irrational investors' expectations impacts price in equilibrium. More specifically, I next examine whether [pi]^sup t^^sub 2^ is non-zero and whether [pi]^sup t^^sub 1^, differs from 1. If [pi]^sup t^^sub 2^ is non-zero, then type C investors have an effect on prices and if [pi]^sup t^^sub 1^, is not equal to 1, type M investors affect price. In addition, irrational trade can have an effect on prices via the risk premium or discount [pi]^sup t^^sub 0^, which is also examined below. Enlarge 200% Enlarge 400% Enlarge 200% Enlarge 400% For the remainder of the analysis, I assume that the shock variances are constant for all periods. That is, [sigma]^sup 2^^sub [eta]t^ = [sigma]^sup 2^^sub [eta]^ and [sigma]^sup 2^^sub [epsilon]t^ = [sigma]^sup 2^^sub [epsilon]t^. Given constant variances [sigma]^sup 2^^sub [eta]^ and [sigma]^sup 2^^sub [epsilon]t^, I examine the resulting stationary equilibria in which the price coefficients are constant across periods (i.e., [pi]^sup t^^sub 1^ = [pi]^sup t+1^^sub 1^, [pi]^sup t^^sub 2^ = [pi]^sup t+1^^sub 2^, and [Delta][pi]^sup t+1^^sub 0^ = [Delta][pi]^sup t+2^^sub 0^). That is, I consider equilibria in which the values of the future coefficients imply (via the relationships in Lemma 1) the same values for the current price coefficients. As in Slezak (1994), I refer to these equilibria as steady response equilibria (SRE) since, in equilibrium, the responsiveness to a particular type of shock is steady (or the same) across periods. Lemma 2 uses the results in Lemma 1 to characterize the equilibrium price function coefficients in an SRE. Enlarge 200% Enlarge 400% In the next section, I use this system of equations to characterize when irrational investors affect equilibrium prices. From the system of equations (18)-(25), the price coefficient at t depends upon the whole future sequence of price coefficients. By imposing a transversality condition that the price coefficients equal a particular set of values at a particular future date (say T), one can determine the coefficients at t by recursively working backward. (The transversality condition acts as a seed to the dynamic system (18)-(25).) An SRE price function (when it exists) may obtain in the limit as T goes to infinity and the parameters are constant. Alternatively, without the imposition of a transversality condition, SRE price functions will obtain as long as the conditions in Lemma 2 hold. That is, as long as the future price functions are an SRE price function, then the current price function will be that SRE function and an SRE will have been achieved. Thus, SREs are self-sustaining. Furthermore, an SRE does not require transversality conditions. As a consequence, since all of the properties of price I analyze derive from the conditions stated in Lemma 2, the results do not rely on price equaling fundamental value at some future point.10 V. The Effect of Irrational Expectations on Equilibrium Price A. Conditions under Which Irrational Expectations Do Not Affect Equilibrium Price Given the characterization of SRE from Lemma 2 in the previous section, I now investigate the conditions under which irrational investors will have no effect on price and market will be weak-form efficient. Theorem 2 is the primary result of this section. Theorem. 2. Under each of the following sufficient conditions, the market is weak-form efficient. If any one of these conditions holds, then irrational expectations have no effect on equilibrium prices (i.e., [pi]^sub 1^ = 1 and [pi]^sub 2^ = 0 in equilibrium). 1. [gamma]^sub R^ = 0 2. [sigma]^sup 2^^sub [eta]^ = 0 3. N^sub R^ = 1 When all three of the above conditions are not satisfied, irrational trades affect price and prices will not be weak-form efficient. In particular: If 0 < N^sub C^ < or = 1 and 0 < [sigma]^sup 2^^sub [epsilon]^ < [infinity], then [pi]2 > 0. If 0 < N^sub M^ < or = 1 and 0 < or = [delta] < 1, then [pi]^sub 1^ < 1. Proof. See Appendix B. The first part of Theorem 2 states that as long as rational investors are either risk neutral or face no fundamental risk, then the presence of irrational investors (regardless of their mass) will have no effect upon the equilibrium price. When rational agents are risk neutral, they are willing to take infinite positions to exploit even tiny deviations of price from expected fundamental value generated by irrational traders. As a result, for market clearing to hold, the equilibrium price must fully reflect the rational expectation of [eta]^sub t+1^ (and current and past objective news via [theta]^sub t^) and nothing else (i.e., [pi]^sub 1^ = 1 and [pi]^sub 2^ = 0). Similarly, when there is no fundamental risk (i.e., when [sigma]^sup 2^^sub [eta]^ = 0), even risk-averse rational investors are willing to take infinite positions on the basis of an irrational deviation. Notice, however, that the future random irrational errors made by type C investors are unpredictable (to rational investors) and, in principle, might cause future prices to fluctuate, generating risk to the rational investors (as in DSSW (1990b)). The above theorem states, however, that this is not true. Given that rational agents are long lived and dynamic, potential fluctuations due to type C irrational errors do not generate risk when the underlying fundamental value of the firm is constant over time. Thus, rational agents are not averse to taking large positions on the basis of deviations created by [epsilon]^sub t^ while facing future volatility created by these shocks, knowing that the future fluctuations do not represent risk but rather generate profitable future trading opportunities. Competition among the rational agents for these trade opportunities, however, assures that these deviations go to 0 in equilibrium. The second part of Theorem 2 states that as long as there is a non-negligible mass of at least one type of irrational trader, and rational agents are risk averse and face fundamental risk, the price function will differ from the fully rational price function. When there are M type traders present, the price will only partially reflect the rational expectation [eta]^sub t+1^; when there are C type traders present, the price will have fluctuations due to the irrational component of their expectations, [epsilon]^sub t^. Furthermore, it is never the case that there exists a combination of parameters for which the effects of the two types of irrational traders cancel, yielding a rational price function in equilibrium. Given risk aversion and fundamental risk, rational agents are unwilling to take positions that are large enough to make the irrational deviations disappear in equilibrium. In this case, when a rational investor trades to exploit a deviation from fundamental value created by [epsilon]^sub t^, he exposes himself to fundamental risk. B. How Irrationality Affects Equilibrium Price In this section, I examine the features of the equilibrium price function. Because analytical solutions for the non-linear system in Lemma 2 do not exist, I provide numerical solutions for the price function under a variety of parameters to show how the parameters affect equilibrium. Although I provide solutions for a limited set of parameters here, the solutions I provide are representative of the behavior that I obtained under numerous other parameter assumptions. Below, I summarize my findings with respect to the price coefficients and the intertemporal predictability of returns. 1. Price Function Coefficients Figures 1-3 show how the price function coefficients ([pi]^sub 1^2, [pi]^sub 2^ and [Delta][pi]^sub 0^, respectively) vary with the variance of news ([sigma]^sup 2^^sub [eta]^) and the variance of type C forecast errors ([sigma]^sup 2^^sub [epsilon]^) for various values of [delta] and N^sub R^. Each individual graph shows how the coefficients vary with [sigma]^sup 2^^sub [eta]^ and [sigma]^sup 2^^sub [epsilon]^. By comparing across the rows of individual graphs in each figure, one can see how the coefficients vary with [delta] while holding the percentage of rational investors fixed. By comparing down the columns, one can see how the coefficients vary with the percent of rational traders in the market while holding [delta] fixed. For all of these graphs, I divide the irrational population equally between M and C type investors. That is, I assume N^sub C^/(N^sub C^ + N^sub M^) = 0.5, or, equivalently, N^sub C^ = N^sub M^ = (1 N^sub R^)/2. Enlarge 200% Enlarge 400% FIGURE 1 Equilibrium [pi]^sub 1^ as a Function of the Exogenous Parameters Figure 1 shows that [pi]^sub 1^ has the following properties. Property 1. [pi]^sub 1^ is increasing in [delta] and N^sub R^. For lower values of [delta], price reflects less of the unbiased forecast of the future public announcement [eta]^sub t^. Recall that [pi]^sub 1^ is the weighted average of 1 and [delta] and, as a consequence, the lower [delta], the lower [pi]^sub 1^. In addition, as the percentage of rational investors increases (or the percentage of agents that are type M decreases), greater weight is placed on 1, thus increasing [pi]^sub 1^. Property 2. [pi]^sub 1^ is both increasing and decreasing in [sigma]^sup 2^^sub [epsilon]^ and [sigma]^sup 2^^sub [eta]^. As [sigma]^sup 2^^sub [eta]^ goes to 0, [pi]^sub 1^ goes to 1. As [sigma]^sup 2^^sub [eta]^ gets larger, [pi]^sub 1^ falls toward [delta]. For very low values of [sigma]^sup 2^^sub [epsilon]^ however, [pi]^sub 1^ will fall and then rise as [sigma]^sup 2^^sub [eta]^ increases. [pi]^sub 1^ is increasing in [sigma]^sup 2^^sub [epsilon]^ for low values of [sigma]^sup 2^^sub [eta]^ but can be decreasing for higher values of [sigma]^sup 2^^sub [eta]^. Thus, the [pi]^sub 1^ surface twists as [sigma]^sup 2^^sub [eta]^ increases. Whether [pi]^sub 1^ increases or decreases in [sigma]^sup 2^^sub [epsilon]^ depends upon which of two effects dominates. The first effect is due to the effect of increasing [sigma]^sup 2^^sub [epsilon]^ on type C trades. For low levels of [sigma]^sup 2^^sub [epsilon]^, type C agents are less irrational and trade more like rational agents. That is, the variance of their bias falls and, as a result, their expectations get closer to rational expectations. As a result, there is a larger mass of traders with expectations that are close to fully reflecting [eta]^sub t+1^, which pushes [pi]^sub 1^ closer to 1. The second effect is from the effect of [sigma]^sup 2^^sub [epsilon]^ on type M demand. For smaller values of [sigma]^sup 2^^sub [epsilon]^, the M type traders trade more aggressively since the variance of future price, the appropriate measure of risk for type M traders, is smaller. As [sigma]^sup 2^^sub [epsilon]^ increases, M type investors trade less aggressively and, thus, do not push [pi]^sub 1^ as far down toward [delta]. When [sigma]^sup 2^^sub [epsilon]^ is small, the first effect (due to C type investors) dominates and [pi]^sub 1^ increases in [sigma]^sup 2^^sub [epsilon]^. When [sigma]^sup 2^^sub [epsilon]^ is large, then the second effect (due to M type investors) dominates and [pi]^sub 1^ decreases in [sigma]^sup 2^^sub [epsilon]^. Figure 2 provides the following properties for [pi]^sub 2^. Property 3. [pi]^sub 2^ is decreasing in N^sub R^. As N^sub R^ increases, there is more competition to exploit the mispricing generated by type C investors and, as a result, the type C irrational errors have less impact on price. Property 4. [pi]2 is increasing in [sigma]^sup 2^^sub [eta]^ and decreasing in [sigma]^sup 2^^sub [epsilon]^. As [sigma]^sup 2^^sub [eta]^ increases, fundamental risk increases and rational agents are willing to take less extreme positions to exploit the mispricings caused by type C irrational forecast errors. As such, less of the mispricing is traded away and [pi]^sub 2^ is larger. As [sigma]^sup 2^^sub [epsilon]^ increases, the return per unit of risk of trading on mispricings increases and competition among rational investors makes the bias shock affect price less. Figure 3 provides the following relationships for mean returns. Property 5. The change in the risk discount is increasing in both [sigma]^sup 2^^sub [eta]^ and [sigma]^sup 2^^sub [epsilon]^. That the risk premium is increasing in [sigma]^sup 2^^sub [eta]^ is not surprising; as the variance of news increases, there is more fundamental risk and both rational and irrational agents require larger mean returns if they are to want to hold the fixed supply of the risky asset. Why the risk premium is increasing in [sigma]^sup 2^^sub [epsilon]^ is less obvious since, for rational investors, [sigma]^sup 2^^sub [epsilon]^ does not generate risk, but rather represents trading opportunities. However, the irrational agents are myopic and as such they care about the variance of the