I N F O R M A T I O N I N S T O C K P R I C E S A N D T R A D I N G V O L U M E b y J A N S C H N E I D E R A T H E S I S S U B M I T T E D I N P A R T I A L F U L F I L L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F D O C T O R O F P H I L O S O P H Y i n T H E F A C U L T Y O F G R A D U A T E S T U D I E S (Bus iness A d m i n i s t r a t i o n ) T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A M a y 2006 © J a n Schneider, 2006 Abst rac t In this thesis I develop two theoretical models to analyze how investors can infer private information from market prices and aggregate trading volume. In the first chapter I provide a closed form solution for a rational expectations equilibrium where all investors infer information about the state of the economy from (1) private sig-nals, (2) the market price and (3) aggregate trading volume. The main result of this model is that trading volume reveals the relative quality of the aggregate private information in the economy. Investors use volume to decide how they should weight the market price relative to their own private signals when they update their be-liefs. In the second chapter, I assume that investors make individual mistakes when they infer information from the price. I show that in a heterogeneous information economy, bounded rationality on the individual level is observationally equivalent to a psychological bias on the aggregate level. If the investors are not able to infer perfectly the true state of the economy from the price, then the aggregate demand corresponds to the demand of a representative agent who is "underconfident". The underconfidence of the representative agent causes the price to adjust to new infor-mation too slowly. ii Table of Contents Abstract ii Table of Contents iii List of Figures iv CHAPTER I Introduction 1 CHAPTER II A Rational Expectations Equilibrium with Informative Trading Volume 18 2.1 Introduction 18 2.2 Setup of the Model 19 2.3 The Equilibrium 22 2.4 Properties of the Equilibrium 31 CHAPTER III Heterogeneous Beliefs and the Underconfident Representative Agent 34 3.1 Introduction 34 3.2 Homogeneous Information 35 3.3 Heterogenous Uncertainty about the Income of the Firm 35 3.4 The Underconfident Representative Agent 47 3.4 Comparative Statics 51 CHAPTER IV Conclusion 61 Appendix Proofs 64 iii List of Figures Figure 2.1 Time Line 20 Figure 2.2 Composition of Trading Volume 28 Figure 2.3 Trading Volume and the Dispersion of Signals 30 Figure 3.1 Price Reaction to New Information 52 Figure 3.2 Momentum Return 53 Figure 3.3 Time Line 54 Figure 3.3 Momentum Return with Reversals during the Formation Period .. 56 Figure 3.3 Momentum Return with Reversals during the Formation Period II 58 iv Chapter 1 Introduction In a financial market every investor is interested in the private information that other investors might possess. There are several ways to learn what other market participants know. In this thesis I examine how investors can infer private infor-mation from market prices and aggregate trading volume. I develop two theoretical models to address the following questions: (1) how can investors fully rationally infer information from prices and trading volume, and (2) how do the results of a world with perfectly rational investors change, if the rationality of investors is bounded, so that they are not fully able to infer all available private information? Grossman (1976) was one of the first authors who showed how investors can extract information about the future payoff of a security from its price. In the first part of this thesis I extend the Grossman model to a case where investors extract information from both the price and the trading volume. This model provides a closed form solution for a rational expectations equilibrium where all investors infer information about the state of the economy from (1) private signals, (2) the market price and (3) aggregate trading volume. 1 Severa l e m p i r i c a l s tudies s u p p o r t the idea t h a t t r a d i n g v o l u m e conta ins infor-m a t i o n a b o u t future re turns . F o r example , L l o r e n t e , M i c h a e l y , Saar , a n d W a n g ( L M S W , 2002) solve a n d test a m o d e l where t r a d i n g v o l u m e pred ic t s changes i n the a u t o c o r r e l a t i o n of r e t u r n s . 1 However , the investors i n L M S W ' s m o d e l r a t i o n a l l y ignore t r a d i n g v o l u m e w h e n t h e y u p d a t e the ir beliefs, s ince v o l u m e does not c o n t a i n a n y i n f o r m a t i o n b e y o n d the ir o w n pr iva te s ignals a n d the m a r k e t pr ice . T h e r e f o r e , i n L M S W , t r a d i n g v o l u m e prov ides i n f o r m a t i o n o n l y to a n outs ide observer of the economy, b u t not to investors w i t h i n the economy. M y m o d e l shows how investors w i t h i n the e c o n o m y c a n l earn f r o m t r a d i n g v o l -u m e , a n d how v o l u m e i n f o r m a t i o n differs f r o m i n f o r m a t i o n c o n t a i n e d i n the pr ice . T h e m a i n result of th is m o d e l is t h a t t r a d i n g v o l u m e reveals the re lat ive q u a l i t y of the aggregate pr iva te i n f o r m a t i o n i n the economy. U n d e r low t r a d i n g v o l u m e , the aggregate i n f o r m a t i o n is m o r e precise c o m p a r e d to p r i v a t e s ignals t h a n u n d e r h i g h t r a d i n g v o l u m e . Investors therefore use v o l u m e t o dec ide h o w t h e y s h o u l d weight the m a r k e t pr ice re lat ive to the ir o w n pr iva te s ignals w h e n t h e y u p d a t e the ir be-liefs. W h e n t r a d i n g v o l u m e is low, investors weight the m a r k e t pr ice m o r e heavi ly . Converse ly , w h e n v o l u m e is h i g h , investors weight the ir pr iva te s ignals m o r e heavi ly . In order to show h o w investors infer i n f o r m a t i o n f r o m t r a d i n g v o l u m e , I deve lop a m o d e l where a large n u m b e r of s m a l l investors observe p r i v a t e no i sy s ignals a b o u t a future d i v i d e n d . In a d d i t i o n to the ir e n d o w m e n t of i n f o r m a t i o n , the investors are also e n d o w e d w i t h pr iva te c la ims to a r i sky future l a b o r income . T h e d i v i d e n d a n d the l a b o r i n c o m e are corre la ted , so t h a t investors have two mot ives for t r a d i n g : ^or other evidence see also Gallant, Rossi, and Tauchen (1992), Campbell, Grossman, and Wang (1993), Conrad, Hameed, and Niden (1994), Lee and Swaminathan (2000), Gervais, Kaniel, and Mingelgrin (2001), and Connolly and Stivers (2003). 2 p r i v a t e i n f o r m a t i o n a n d r i sk shar ing . P r i v a t e s ignals a n d l a b o r e n d o w m e n t s are ident i ca l ly d i s t r i b u t e d for a l l investors , so t h a t a l l investors observe i n f o r m a t i o n of i d e n t i c a l qual i ty . T h e r e f o r e , investors weight the ir s ignals equa l ly w h e n t h e y u p d a t e the ir beliefs. A s a result , the equ i -l i b r i u m pr ice depends o n the average s igna l a n d the average exposure to the l a b o r r isk . S ince investors are u n c e r t a i n a b o u t the average l a b o r r i sk , t h e y are not able to fu l ly infer the average d i v i d e n d s igna l f r o m the pr ice . In a d d i t i o n to the u n c e r t a i n t y a b o u t the aggregate d i v i d e n d i n f o r m a t i o n a n d aggregate l a b o r risk, the investors are also u n c e r t a i n a b o u t the cross- investor cor-r e l a t i o n o f the i n d i v i d u a l errors i n the ir pr iva te d i v i d e n d signals . T h e c o r r e l a t i o n of the i n d i v i d u a l s igna l errors is i m p o r t a n t , s ince th is corre la t ion de termines the q u a l i t y of the aggregate i n f o r m a t i o n re lat ive to the p r i v a t e i n f o r m a t i o n . F o r e x a m -ple , i f the s ignals are perfec t ly corre la ted , t h e n the average s igna l conta ins the same i n f o r m a t i o n as the i n d i v i d u a l signals. However , i f investor specific s igna l errors are u n c o r r e l a t e d , t h e n investors k n o w m o r e i n aggregate t h a n t h e y k n o w i n d i v i d u a l l y . Investors therefore w i s h to k n o w the corre la t ion of the i n d i v i d u a l s ignals , i n order to assess the prec i s ion of the average s igna l i n the pr ice . In a s y m m e t r i c e c o n o m y where pr iva te i n f o r m a t i o n a n d l a b o r risks are ident i ca l ly d i s t r i b u t e d for a l l investors , t r a d i n g v o l u m e reveals the c o r r e l a t i o n of s ignals i n the fo l lowing way: s ince investors weight the ir s ignals a n d e n d o w m e n t s ident i ca l l y w h e n t h e y ca lcu la te the ir d e m a n d s , the n u m b e r of shares t h a t a g iven investor b u y s or sells d e p e n d s o n l y o n the differences between his pr iva te s igna l a n d e n d o w m e n t a n d the s igna l a n d e n d o w m e n t of the average investor i n the economy. T h e r e f o r e , the i n d i v i d u a l t rades are funct ions o n l y of the investor specific c o m p o n e n t s of s ignals a n d endowments . H e n c e , i f these c o m p o n e n t s are i n d e p e n d e n t across investors , a n d 3 if the number of investors in the economy is large, the per capita trading volume depends only on the distribution and not on the realization of these components. As a result, investors can infer the distribution of signals from trading volume. In particular, if investors are uncertain about the correlation of their private signals, then trading volume reveals this correlation. In traditional models of heterogeneous information, such as Grossman and Stiglitz (1980), investors form a weighted average of their private signals and the market price when they update their beliefs. In these models, all investors know the optimal weights for the price and the signals, since these weights are independent of the state of the economy. In my model, investors are uncertain how they should weight the market price relative to their own private information. Observing trading volume removes this uncertainty. Under high trading volume, the quality of the aggregate signal in the price does not exceed the quality of their private signals. Since the price contains additional noise from the aggregate labor income shock, investors weight their own signals more heavily than the price when volume is high. However, under low trading volume, the quality of the aggregate information exceeds the quality of the individual signals. Therefore, investors weight the price more heavily when trading volume is low. The idea that investors have a risk sharing and a private information motive for trading has been previously employed for example by Wang (1994) and LMSW (2002). However, in these models, there are only two agents that trade with each other. Therefore, trading volume does not provide any information for the investors beyond the information that they can infer from their own private signals and the market price. The technical difficulty that arises if investors are allowed to observe trading volume is that volume is a sum of absolute values and therefore not normally 4 distributed. Asset pricing models with heterogeneously informed investors usually rely on the properties of the normal distribution in order to be tractable. My model solves the problem that trading volume is not normally distributed by transforming a non-linear optimization problem into a problem that is linear conditional on the observation of trading volume. Several other authors have ex-amined alternative approaches. For example, Bernardo and Judd (1996) show how to numerically solve a model where investors learn private information from trading volume. Their numerical approach has the advantage that it covers a large set of possible assumptions, however, a numerical approach does not provide the same clean economic intuition as an analytical solution. Blume, Easley, and O'Hara (BEH, 1994) provide a closed-form solution for a model where investors learn from past prices and past trading volume. Similar to my model, the investors in BEH face two dimensions of uncertainty: the realiza-tion and the quality of other investors' signals. However, in order to solve their model, BEH have to assume that investors are not fully rational: even though in-vestors know the price at which they trade, they employ this price in order to update their beliefs only after they have completed their trade. BEH (page 160) comment on the difficulty of solving a rational expectations equilibrium where investors learn from trading volume: "Alternatively, there could be nonrevealing equilibria in which traders condition on price and volume. However, as volume is a sum of absolute values it cannot be normally distributed. So although such an equilibrium might exist there seems to be no hope of constructing it, and hence no hope of using a contemporaneous data approach to study volume." As I show in this paper, the case for a non-revealing equilibrium where investors condition their demands on prices and on volume is not completely hopeless. 5 In the first part of this thesis I assume that all investors are fully rational, and that they are able to optimally analyze prices and volume. However, in reality prices aggregate information in a complicated way, and extracting this information is a difficult task. In the second part of this thesis I assume that the computational skills of the investors are limited: they make individual mistakes when they infer information from the price. I show that in a heterogeneous information economy where investors' rationality is bounded prices react too slowly to new information. A large number of empirical studies document that prices potentially underreact to new information. For example, Ball and Brown (1968), Bernard and Thomas (1990) and others find that firms reporting unexpectedly high earnings outperform firms reporting unexpectedly low earnings. Givoly and Lakonishok (1979), and Stickel (1991) document similar drifts after analyst forecast revisions. Jegadeesh and Titman (1993) rank stocks according to past returns and find that past winners outperform past losers. The debate whether these findings violate market efficiency is ongoing. Some authors argue that return continuations can be explained with changes in firm's risks. Other authors argue that prices underreact to new information because investors are psychologically biased. For example, Barberis, Shleifer, and Vishny (BSV, 1998) link underreaction to the conservatism bias. The conservatism bias has been identified in experiments by Edwards (1968). Individuals who are subject to this bias tend to underweight new information when they update their priors. BSV show that, if the representative agent in the economy suffers from the conservatism bias, then prices will adjust to new information slowly.2 2Other psychologically motivated explanations include over/confidence as for example in Daniel, 6 In this paper I show that even if the representative agents appears to be psy-chologically biased, this is not necessarily true for the individual investors. In a heterogeneous information economy where the computational skills of the investors are limited, so that they are not perfectly able to infer each other's information from the price, the aggregate demand is equivalent to the demand of a representa-tive agent who underestimates the quality of his information. This underconfidence of the representative agent arises from the fact that the investors know more in aggregate than they know individually, and that they fail to infer perfectly the ag-gregate knowledge from the price. In that way, bounded rationality on the individual level is equivalent to a psychological bias on the aggregate level. Several empirical studies support the notion that heterogeneous beliefs play an important role in the momentum phenomenon. For example, Verardo (2002) finds that profits from momentum strategies increase with the dispersion in analyst fore-casts. Hong, Lim, and Stein (2000) find that momentum strategies work better among small stocks with low analyst coverage. Zhang (2005) finds that the post earnings announcement drift, the drift after analyst forecast revisions, and the profits from price momentum strategies all increase with various proxies for heterogeneous beliefs. To see how heterogenous information causes return continuations consider the following example. There is a firm that will pay out an uncertain future dividend. The investors receive a signal about this dividend of the form "dividend + noise". Assume for the moment that this signal is public, and that all investors interpret the signal in the same way. If this signal becomes more precise as the dividend payout Hirshleifer, Subrahmanyam (1998), and the disposition effect as for example in Frazzini (2005) and Grinblatt and Han (2005). 7 date approaches, the investors will increase the weight of the signal, as they update their beliefs. This updating process has two effects on the serial correlation of the returns. First, the "true" component of the signal produces a positive effect, since this value will be slowly incorporated into the price. Second, the noise component produces a negative effect, since, as the signal becomes more precise, the investors will reverse that part of their initial reaction to the signal that was due to the noise. If the economy is in a steady state, and the risk that the average investors has to bear does not change over time, these two effects will offset each other exactly. Hence, uncertain information does not produce serially correlated returns, if the investors are homogenously informed. Assume now that the signal contains an investor specific noise component, so that the investors axe heterogeneously informed. Assume that the rationality of the investors is bounded, and that the investors are not able to infer perfectly all the private information from the price. Since the individual signals are noisy, the investors will assign only small weights to these signals when they update their priors. However, the individual noise components will at least partially cancel each other out, when one aggregates the demands. Therefore, the aggregation of demands will reduce the negative noise effect on the correlation of returns, without reducing sufficiently the positive effect resulting from the true information component of the signals. The aggregate demand will be equivalent to the demand of a representative agent who is underconfident. As a result, prices adjust to new information too slowly. Several other authors have developed heterogeneous information models to ex-plain momentum. In these models, returns are positively autocorrelated because investors receive information sequentially. For example, in Hong and Stein (1999) in-8 formation about a liquidating dividend spreads slowly through a group of "newswatch-ers". Since information spreads slowly, and since the newswatchers are not able to infer information from the price, the price adjusts slowly to each piece of new infor-mation. In Holden and Subrahmanyam (2002) some investors receive certain pieces of information before other investors. In their model, noise traders prevent the price from revealing all the information. Holden and Subrahmanyam state on page 4: "Thus, our consideration of the sequential nature of information acquisition is the key to generating positive serial correlation within a rational expectations model." In this paper, I show that the sequential information flow is not a necessary condition for return continuations. Instead, the key to momentum is simply the disagreement about future payoffs. I show in this paper how heterogeneous beliefs combined with bounded rational-ity leads to aggregate underreaction to new information. If the price underreacts to new information, momentum traders will rationally chase trends, as for example in Hong and Stein (1999). I do not examine the effect of momentum traders on prices and returns in this paper. In the real world, we would expect that the actions of momentum traders diminish the abnormal profits from momentum strategies. It is therefore surprising that, for example, Grundy and Martin (2001) find that mo-mentum strategies can produce risk adjusted returns of more than one percent per month. However, these momentum profits are profits before trading costs. The ev-idence for the profitability of momentum strategies after trading costs is less clear. For example, Lesmond, Schill, and Zhou (2004) find that momentum strategies are not profitable after trading costs. Korajczyk and Sadka (2004) construct a liquid-ity weighted momentum strategy in order to minimize trading costs. Taking price impact trading costs into account, they estimate that this strategy earns positive 9 abnormal profits for an investment of up to $5 billion.3 In this paper, I do not ad-dress the question to which degree momentum traders should arbitrage momentum profits away. Instead, my focus is to show a compelling source of momentum. 3 As of December 1999. At that time, the total market capitalization of the NYSE was $11.7 trillion. 10 Bibliography [1] Admati, Anat R., 1985, A noisy rational expectations equilibrium for multi-asset securities markets, Econometrica, 53, 629-658. [2] Admati, Anat R., and Paul Pfleiderer, 1988, A theory of intraday patterns: volume and price variability, Review of Financial Studies, 1, 3-40. [3] Back, Kerry, Cao, C. Henry, and Gregory A. Willard, 2000, Imperfect compe-tition among informed traders, Journal of Finance, 55, 2117-2155. [4] Ball, Ray, and Philip Brown, 1968, An empirical evaluation of accounting in-come numbers, Journal of Accounting Research, 6, 159-178. [5] Barberis, Nicholas, Shleifer, Andrei, and Robert Vishny, 1998, A model of investor sentiment, Journal of Financial Economics, 49, 307-343. [6] Berk, Jonathan B., Green, Richard C., and Vasant Naik, 1999, Optimal invest-ment, growth options, and security returns, Journal of Finance, 54, 1553-1607. [7] Bernard, Victor L., and Jacob K. Thomas, 1990, Evidence that stock prices do not fully reflect the implications of current earnings on future earnings, Journal of Accounting and Economics, 13, 305-340. 11 [8] Bernardo, Antonio E., and Kenneth L. Judd, 1996, Volume and price formation in an asset trading model with assymetric information, working paper, UCLA. [9] Blume, Lawrence, Easley, David, and Maureen O'Hara, 1994, Market statistics and technical analysis: the role of volume, Journal of Finance, 49, 153-181. [10] Brown, David P., and Robert H. Jennings, 1989, On technical analysis, Review of Financial Studies, 2, 527-551. [11] Campbell, John Y., Grossman, Sanford J., and Jiang Wang, 1993, Trading volume and serial correlation in stock returns, Quarterly Journal of Econmics, 108, 905-939. [12] Chan, Louis K.C., and Jegadeesh, Narasimhan, and Josef Lakonishok, 1996, Momentum strategies, Journal of Finance, 51, 1681-1713. [13] Chordia, Tarun, and Lakshmanan Shivakumar, 2002, Momentum, business cy-cle, and time-varying expected returns, Journal of Finance, 57, 985-1019. [14] Conolly, Robert, and Chris Stivers, 2003, Momentum and reversals in equity-index returns during periods of abnormal turnover and return dispersion, Jour-nal of Finance, 58, 1521 - 1555. [15] Conrad, Jennifer S., Hameed, Allaudeen, and Cathy Niden, 1994, Volume and autocovariances in short-horizon individual security returns, Journal of Finance, 49, 1305-1329. [16] Daniel, Kent, Hirshleifer, David, and Avanidhar Subrahmanyam, 1998, Investor psychology and security market under- and overreaction, Journal of Finance, 53, 1839-1885. 12 [17] Easley, David, and Maureen O'Hara, Time and the process of security price adjustment, Journal of Finance, 47, 577-605. [18] Edwards, Ward, 1968, Conservatism in human information processing, in: For-mal representation of human judgement, B. Kleinmutz (ed.), John Willey and Sons, New York, 17-52. [19] Fama, Eugene F., and Kenneth R. French, 1996, Multifactor explanations of asset pricing anomalies, Journal of Finance 51, 55-84. [20] Foster, F. Douglas, and S. Viswanathan, 1990, A theory of interday variations in volume, variance, and trading costs in securities markets, Review of Financial Studies, 3, 593-624 . [21] Foster, F. Douglas, and S. Viswanathan, 1993, The effect of public information and competition on trading volume and price volatility, Review of Financial Studies, 6, 23-56. [22] Foster, F. Douglas, and S. Viswanathan, 1996, Strategic trading when agents forecast the forecasts of others, Journal of Finance, 51, 1437-1478. [23] Frazzini, Andrea, 2005, The disposition effect and underreaction to news, Jour-nal of Finance, forthcoming. [24] Gallant, A. Ronald, Rossi, Peter E., and George Tauchen, 1992, Stock prices and volume, Review of Financial Studies, 5, 199-242. [25] Gallmeyer, Michael, Hollifield, Burton, and Duane J. Seppi, 2005, Demand Discovery and Asset Pricing, working paper. 13 [26] Gervais, Simon, Kaniel, Ron, and Dan H. Mingelgrin, 2001, The high-volume return premium, Journal of Finance, 56, 877-919. [27] Givoly, Dan, and Josef Lakonishok, 1979, The information content of finan-cial analysts' forecasts of earnings: some evidence on semi-strong inefficiency, Journal of Accounting and Economics, 1, 165-185. [28] Glosten, Lawrence R., 1994, Is the electronic open limit order book inevitable?, Journal of Finance, 49, 1127-1161. [29] Glosten, Lawrence R., and Paul R. Milgrom, 1985, Bid, ask, and transaction prices in a specialist market with heterogeneously informed traders, Journal of Financial Economics, 13, 71-100. [30] Griffin, John M., Xiuqing Ji, and J. Spencer Martin, 2003, Momentum investing and business cycle risk: evidence from pole to pole, Journal of Finance 58, 2515-2547. [31] Grinblatt, Mark, and Bing Han, 2005, Prospect theory, mental accounting, and momentum, Journal of Financial Economics, forthcoming. [32] Grossman, Sanford J., 1976, On the efficiency of competitive stock markets when traders have diverse information, Journal of Finance 31, 573-585. [33] Grossman, Sanford, and Joseph Stiglitz, 1980, On the impossibility of informa-tionally efficient markets, American Economic Review, 70, 393-408. [34] Grundy, Bruce D., and Maureen McNichols, 1989, Trade and the revelation of information through prices and direct disclosure, Review of Financial Studies, 2, 495-526. [35] Grundy, Bruce D., and Spencer J. Martin, 2001, Understanding the nature of risks and the source and rewards to momentum investing, Review of Financial Studies 14, 29-78. [36] Harris, Milton, and Arthur Raviv, 1993, Differences of opinion make a horse race, Review of Financial Studies, 6, 473-506. [37] He, Hua, and Jiang Wang 1995, Differential information and dynamic behaviour of stock trading volume, Review of Financial Studies, 8,919-972. [38] Holden, Craig W., and Avinidhar Subrahmanyam, 2002, News events, informa-tion acquisition, and serial correlation, Journal of Business, 75, 1-32. [39] Hong, Harrison, Lim, Terence, and Jeremy C. Stein, 2000, Bad news travels slowly: size, analyst coverage, and the profitability of momentum strategies, Journal of Finance, 55, 265-295. [40] Hong, Harrison, and Jeremy C. Stein, 1999, A unified theory of underreaction, momentum trading, and overreaction in asset markets, Journal of Finance, 54, 2143-2184. [41] Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying winners and selling losers: implications for stock market efficiency, Journal of Finance, 48, 65-91. [42] Johnson, Timothy C , 2002, Rational momentum effects, Journal of Finance 57, 585-607. 15 [43] Kandel, Eugene, and Neil Pearson, 1995, Differential interpretation of public signals and trade in speculative markets, Journal of Political Economy, 103, 831-872. [44] KarpofF, Jonathan M., 1987, The relation between price changes and trading volume: a survey, Journal of Financial and Quantitative Analysis, 22, 109-126. [45] Kim, Oliver, and Robert Verrecchia, 1991, Trading volume and price reactions to public announcements, Journal of Accounting Research, 29, 302-321. [46] Kim, Oliver, and Robert Verrecchia, 2001, The relation among disclosure, re-turns, and trading volume information, Accounting Review, 76, 633-654. [47] Korajczyk, Robert A., and Ronnie Sadka, 2004, Are momentum profits robust to trading costs?, Journal of Finance, 59, 1039-1081. [48] Kyle, Albert S., 1985, Continuous auctions and insider trading, Econometrica, 6, 1315-1336. [49] Lee, Charles M.C., and Bhaskaran Swaminathan, 2000, Price momentum and trading volume, Journal of Finance, 55, 2017-2069. [50] Lesmond, David A., Schill, Michael J., and Chuncheng Zhou, 2004, The illusion-ary nature of momentum profits, Journal of Fiancial Economics, 71, 349-380. [51] Llorente, Guillermo, Michaely, Roni, Saar, Gideon, and Jiang Wang, 2002, Dy-namic volum-return relation of individual stocks, Review of Financial Studies, 15, 1005-1047. [52] Lo, Andrew, and Jiang Wang, 2000, Trading volume: data analysis, and impli-cations of portfolio theory, Review of Financial Studies, 13, 257-300. 16 [53] Milgrom, Paul, and Nancy L. Stockey, 1982, Information, trade, and common knowledge, Journal of Economic Theory, 26, 17-27. [54] Pfleiderer, Paul, 1984, The volume of trade and the variability of prices: a framework for analysis in noisy rational expectations equilibria, Manuscript, Graduate School of Business Stanford University. [55] Rudin, Walter, 1976, Principles of Mathematical Analysis, McGraw-Hill, Inc. [56] Shalen, Catherine, 1993, Volume, volatility, and the dispersion of beliefs, Review of Financial Studies, 6, 405-434. [57] Stickel, Scott, 1991, Common stock returns surrounding earnings forecast revi-sions: more puzzling evidence, The Accounting Review, 66, 402-416. [58] Verardo, Michela, 2002, Heterogeneous beliefs and momentum profits, working paper [59] Verrecchia , Robert E., 1982, Information acquisition in a noisy rational expec-tations economy, Econometrica, 50, 1425-1430. [60] Wang, Jiang, 1993, A model of intertemporal asset prices under asymmetric information, Review of Economic Studies, 60, 249-282. [61] Wang, Jiang, 1994, A model of competitive stock trading volume, Journal of Political Economy, 102, 127-168. [62] Zhang, X. Frank, 2005, Information uncertainty and stock returns, Journal of Finance, forthcoming. 17 Chapter 2 A Rational Expectations Equil ibr ium with Informative Trading Volume 2.1 Introduction In this chapter I develop a rational expectations equilibrium where investors infer private information from the market price and from trading volume. A l l investors in this model observe private noisy signals about a future dividend and they are also endowed with claims to a risky future labor income. Investors have therefore two motives for trading: private information and risk sharing. Since investors are uncertain about the economy wide exposure to the labor risk, they are not able to fully infer the aggregate private information from the price. In addition to the uncertainty about the aggregate dividend information and aggregate labor risk, the investors are also uncertain about the cross-investor corre-18 lation of the individual errors in their private dividend signals. The correlation of the individual signal errors is important, since this correlation determines the quality of the aggregate information relative to the private information. In the equilibrium, investors learn the correlation of signals from trading volume, and they learn the average signal from the price. 2.2 Setup of the model The economy is populated by a countable set of investors. I will refer to an individual investor as investor i, i = 1,2, There are two time periods, t = 0 and t = 1. Figure 2.1 shows a picture of the time line. At time t = 0, investor i is endowed with Nt = N + 7ii units of a non-traded asset. At time t = 1 the investors receive a payoff of Y for each unit Ni they are endowed with at time t — 0. The total non-traded income of investor i at time t = 1 is therefore given by NY. I will refer to NY as the labor income of investor i, even though other interpretations are possible. In addition to the labor income, the investors also receive income from their investments in the financial market. The financial market consists of two assets: a risk free bond and a risky firm. One dollar invested in the bond at time t = 0 pays one dollar at time t = 1. Investors can buy or sell an unlimited amount of the bond. Investors can trade shares of the firm at time t = 0 at the equilibrium price P. At time t = 1 the firm pays a liquidating dividend D for each share the investors hold at time t = 0 after trading. At time t = 0 the investors observe private noisy signals 19 l a b o r e n d o w m e n t s payoffs Ni = N + m NiY r i s k y asset d e m a n d s payoffs XiD p r i v a t e i n f o r m a t i o n a b o u t r i s k y asset p u b l i c i n f o r m a t i o n a b o u t r i s k y asset s ignals bi = D + r} + ei pr ice P v o l u m e V F i g u r e 2.1: Time Line. T h i s figure shows endowments , d e m a n d s , payoffs, p r i v a t e i n f o r m a t i o n a n d p u b l i c i n f o r m a t i o n . a b o u t th i s d i v i d e n d . T h e s igna l of investor i is g iven b y Di = D + rj + €i where r\ a n d e$ are error t erms . T h e corre la t ion of the across investors de termines h o w m u c h the investors disagree a b o u t the future payoff. T h e r e are two poss ible states of the w o r l d r e g a r d i n g this corre la t ion . In state L , the c o r r e l a t i o n of s igna l errors is low. In th is state, the are i n d e p e n d e n t across investors . In state H the c o r r e l a t i o n o f errors is h i g h . In this state, a l l i n d i v i d u a l s ignals c o n t a i n the same error t e r m = e, so t h a t the errors are perfect ly corre la ted across investors a n d a l l investors observe the same s igna l D + 77 + e. T h e investors m i g h t have some i n f o r m a t i o n a b o u t th is corre la t ion , however t h e y d o not k n o w the r e a l i z a t i o n o f the c o r r e l a t i o n w i t h certainty . F o r a n y g iven rea l i za t ion of the c o r r e l a t i o n state the 20 r a n d o m var iables D, N, Y, rj, {ei}^0, { n j g 0 are j o i n t l y n o r m a l l y d i s t r i b u t e d w i t h m e a n zero a n d var iances a2D, a%, aY, c 2 , , a\ a n d CT2.1 A l l var iables are u n c o r r e l a t e d except for the corre la t ion of the e, i n state H, a n d except for D a n d Y, w h i c h are corre la ted w i t h Cov[D, Y] = ODY > 0. L e t Xi be the t i m e t = 0 d e m a n d for the r i sky asset of investor i. L e t 1 h X= l i m T y X i i=l be the per c a p i t a d e m a n d , p r o v i d e d t h a t this l i m i t exists. A n e q u i l i b r i u m is g iven b y a p r i c e P t h a t satisfies X = s u p p l y per c a p i t a (2.1) w i t h p r o b a b i l i t y one. T o s impl i fy the n o t a t i o n I w i l l set the s u p p l y e q u a l to zero, a n d I assume t h a t a l l investors o w n zero shares p r i o r to the t r a d i n g date t = 0. T h e a s s u m p t i o n of zero s u p p l y means t h a t a l l d i v i d e n d s t h a t investors w h o h o l d l o n g pos i t ions of the f i r m receive are p a i d b y investors w h o h o l d short pos i t ions . T h i s a s s u m p t i o n w i l l r e m o v e a constant from the e q u i l i b r i u m pr ice , b u t it w i l l not affect a n y of the results . S ince investors h o l d zero shares before t r a d i n g , the n u m b e r of shares t h a t investor i t rades is g iven b y \Xi\. L e t 1 h V= l i m TY\Xi\ i=l 1 Since all variables are normally distributed, dividends and labor income can be negative. It is possible to choose means and variances so that the probability for negative payoffs wi l l be arbitrarily small. However, since all investors know these distributions, the choice of the mean does not affect the trading volume. In order to simplify the notation I set therefore all the means equal to zero. 21 be the (double counted) per capita trading volume, provided that this limit exists. Investor i chooses his demand Xt by maximizing E e-Pw, T where p is the coefficient of absolute risk aversion, Wi is the future wealth, and T is the information set of investor i. This information set is given by Ti = {Di,Ni,P,V}. Investors can therefore condition their demand on their private dividend signals, their private labor endowments, the equilibrium price, and the equilibrium trading volume. Note that investors know Ni, their own exposure to the labor risk, but they do not observe N, the economy wide exposure to this risk. This assumption will prevent the equilibrium price from completely revealing all private information. 2.3 The equilibrium Let 1 h D= lim - V D. h-too h Z—' i = l be the aggregate information about the future dividend. Then we have D + 77 in state L D + n + e in state H If the correlation of signal errors ti is low (state L), then the individual signal errors cancel each other out in the average signal. Only in this case investors are able to learn information about the future dividend from the price. If the correlation of 22 signals is high (state H), all investors observe the aggregate signal JJ = D + r\ + e directly, so that the price does not provide any additional information about D. Investors are not only uncertain about the realization of the average signal D, but they are also uncertain wether the world is in state L or in state H. In order to find an equilibrium for this economy, I will first analyze the case where investors know the correlation of signals. Lemma 1 (Correlation of signals is high). Assume the world is in state H, and assume all investors know that the world is in state H. Then there exists an equi-librium price of the form P = $£(D + V + e) -$%N, (2.2) where and are constants. If the price is given by (2.2), then the coefficients are given by ®Z = 2 , 2 ^ 2 a T l d ^" = P 2 2 • 2 ' and the demand of investor i is given by and trading volume is given by XH = _^DYn. V 7T Of D Lemma (1) shows that, if all investors know that the world is in state H, the price P is a linear function of the aggregate signal D + 77 + e and the exposure to the aggregate labor risk N. The price decreases with N, since the labor payoff Y and the dividend D are positively correlated. 23 Since all investors hold zero shares before they start trading, the demand Xf is equivalent to the number of shares that investor i buys or sells. Note that these equilibrium trades can also be written as *>'--%(»<-») In state i / , investor i sells shares if his private labor risk exposure N is higher than the average risk exposure N. The equilibrium demands are independent of the dividend signals Dt since all investors observe the same signal Di = D + r) + e. Since the equilibrium demands depend only on investor specific components nit and since these components are independent across investors, the per capita trading volume is given by the unconditional expectation of the absolute number of shares that any given investor trades. Since the nt are normally distributed, we have V« = nXi\ = ^ [ X i ] = ^ - ^ a n As a result, volume is constant. Hence, if all investors know that the world is in state H, then investors ignore trading volume when they choose their demands. Lemma 2 (Correlation of signals is low). Assume the world is in state L, and as-sume all investors know that the world is in state L. Then there exists an equilibrium price of the form P = $ LD{D + V)-$LNN, (2.3) where $£> and are constants. If the price is given by (2.3), then we have § L D < $g < 1 and > 0 and the demand of investor i is given by Xi = tyD€i - ^NUi, 24 where typ > 0 and > 0, and trading volume is give by Lemma (2) shows that, if all investors know that the world is in state L, the price P is a linear function of the aggregate signal D + n and the exposure to the aggregate labor risk N. The investor specific signal errors are not part of the price, since they cancel each other out in the aggregate demand. Note that the equilibrium trades X± can also be written as Xl = tfD (Di - (D + r,)) - * N (Nt - JV) . As opposed to state H, investors in state L do not only trade to share risk, but also because they are heterogeneously informed. In this case, the equilibrium demands depend on the differences between the individual signals Di and the average signal D+r] and the individual labor endowments N and the average endowment N. Since investors are uncertain about the aggregate endowment N, the price does not fully reveal the aggregate dividend signal. In both correlation states, L and H, the demands Xi depend only on the differ-ences between individual and average signals and endowments, since information and labor risks are identically distributed across investors. If signals and endowments are identically distributed, then investors weight their signals and endowments iden-tically when they form their demands. As a result, the price depends on the averages whereas the equilibrium demands depend on the differences of signals and endow-ments. Therefore, the price is a function only of the common components D and N, and the demands are functions only of the investor specific components e* (in state H) and n». As a result, for a given state of signal dispersion, equilibrium demands and the price are independent. 25 T h e o r e m 1 (Informative trading volume). If the equilibrium price is given by (2.2) in state L, and by (2.3) in state H, then we have VL < VH. Hence, if investors are uncertain about the correlation of signal errors, then there exists an equilibrium where trading volume reveals this correlation. Note that investors cannot learn the correlation of signals by comparing their private signals A to the price P, since the price depends partially on the unknown aggregate labor risk exposure N. However, as Theorem 1 shows, investors can infer the dispersion of beliefs in the economy from trading volume. Given this infor-mation, the investors use their private signals and the price to estimate the future dividend. In that way, observing trading volume helps the investors to separate two sources of uncertainty: uncertainty about the realization of the aggregate informa-tion, and uncertainty about the quality of the aggregate information. Theorem 1 shows that trading volume in state H is higher than trading volume in state L. This result might seem surprising, since in state L investors have two reasons for trading, risk sharing and private information, whereas in state H investors only trade to share risk. To understand why volume increases with the correlation of signals, note that, given CARA utility and normal distribution, the general demand functions of the investors are given by v E[D - P\Tt\ - pCov[D - P, NiYlft] The demand of investor % depends on the conditional expectation of the future investment payoff, the variance of this payoff, and the covariance of the investment 26 payoff with the labor payoff. As Appendix 4 shows, in state L, the conditional expectation of the future dividend is a linear combination of the private dividend signal, the private labor risk, and the price: E[D\Fi) = ipDbi + i>NNi + ippP, (2.5) where tpo > 0, tpN > 0, and ipp > 0. Investors increase their expectations of D with their private labor risks JVj, since they use Nt to estimate the aggregate labor risk N in the price, and since the price decreases with this risk. As Appendix 4 shows, plugging (2.5) into (2.4) gives the state L equilibrium demands X i ~ pVar[D|^] a l Ui ^ Hence, the equilibrium demands in state L can be written as the sum of two com-ponents: the first component is due to the fact that investors observe heterogeneous signals, and the second component is given by the equilibrium demand Xf1 in Lemma (1). By (2.6) we can write trading volume in state L as 7T VpVar[£>|^]; Ge i r \ a l pVarpI^] As appendix 4 shows, we have a 2D pVar[D|^]" Hence, private information affects trading volume in two opposing ways. On the one hand, private information increases volume since investors trade based on the differences of their dividend signals. On the other hand, private information reduces volume, since the fact that investors use their private labor risks to estimate the aggregate risk reduces the trading that is due to differences in labor risk endowments. 27 trading due to differences in dividend signals reduction of risk sharing trading due to differences due to private information in labor endowments A = v 7 ^ -J Y total trading volume in state H Figure 2.2: Composition of Trading Volume. If the world is in state H, so that investors only trade to share risk, then trading volume is given by VH. The private signals in state L have two effects on VH: (1) private information reduces VH by A, since investors use their private labor endowments to estimate the aggregate endowment, and (2) the private signals induce the information trading C. The total trading volume under heterogeneous information is given by VL = \ A B 2 + C 2 . VL is less than the sum of B and C, since these two trading motives partially cancel each other out for the average investor in the economy. 28 As Theorem 1 shows, the total effect of private information on trading volume is always negative. The reason for this negative effect is that for some investors in the economy the demand due to private signals will partially offset the demand due to risk sharing. Therefore, the number of shares that the average investor in the economy trades is less than the sum of his risk sharing demand and his private information demand. Figure 2.2 shows a geometric interpretation of the two compo-nents of trading volume volume under heterogeneous information and the relation of these components to trading volume under homogeneous information. I have shown so far how trading volume depends on two extreme cases: the error terms are perfectly correlated or they are not correlated at all. In order to examine how trading volume depends on the correlation of error terms in general, assume the dividend signal of investor i is given by Di = D + r)-rVuel + y / l - u e2i, C T 2 = a\2. = a 2 (2.7) where 0 < u < 1, t\ is a common error term and e^i are investor specific error terms, which are independent and identically distributed across investors. The case u> = 0 corresponds to state L, and the case u — 1 corresponds to state H in the theorem. Since o\x = of2. the total variance of the signal does not depend on u. Hence, the specification of the dividend signal in (2.7) allows to examine the effect of a change in the correlation of signal errors, independent of the effect of a change in the total error variance. Figure 2.3 shows the equilibrium trading volume, assuming that all investor know u. As Figure 2.3 shows, trading volume decreases with the dispersion of signals. 29 Volume (low correlation) (perfect correlation) Figure 2.3: T r a d i n g V o l u m e and the Dispers ion of Signals. This figure shows how trading volume in a symmetric economy with a large number of small investors depends on the correlation of the investor specific signal errors, holding the total variance of the error terms constant. The private signal of investor i is given by Di = D+n+i/u) ei + y/1 — uje^i, where w and €\ are common error terms and ej is an investor specific error term, and <r2x = cr 2 2 i = a\. The total variance of the signal errors is given by a^+wa^ + ^—^a^. = a^+a2. The remaining parameters are given by p = a2D = UDY = = an — !• 30 2.4 Properties of the Equilibrium 2.4.1 Updat ing of beliefs Coro l la ry 1. If the equilibrium is given by the Theorem 1, the expected future dividend conditional on the information of investor i is given by E[D\Ft] = if trading volume is high, and E[D\Ti} = + ipNNt + ibpP, if trading volume is low, and we have tpp > 0, ipx > 0, and 0 < ipp < ipp. Corollary 1 shows that investors use the price to update their beliefs only when trading volume is low. Under high trading volume, the aggregate signal is identical to the individual signals. Since the uncertain aggregate labor risk iV prevents the price from revealing the aggregate signal, investors completely ignore the price when they update their beliefs. However, under low trading volume, the quality of the aggregate information exceeds the quality of the private information. Therefore, if trading volume is low, investors reduce the weights on their private signals and weight the price more heavily. In this case, the investors also use their own labor risk Ni in order to estimate the aggregate risk N in the price. 31 2.4.2 Trading strategies Corollary 2. // the equilibrium is given by the Theorem 1, the demand of investor i is given by X? = VZDi-V%Ni-*$P where all coefficients are greater than zero and we have typ > $p. Corollary 2 shows that, after controlling for the signals A and the labor en-dowment N, investors always trade against the price. However, investors trade less aggressively against the price when trading volume is low than when trading vol-ume is high. The reason for this behavior is that, under low trading volume, the price does not fully adjust to the aggregate information in the economy. Investors interpret therefore high prices partially as good news and low prices partially as bad news. 2.4.3 Volume predicts the future risk premium I have so far assumed that the risky asset has a zero net supply. This assumption removes a risk premium from the return, since the average investor does not require a risk premium if he does not hold any shares. The following result shows the relation between volume and the risk premium, if the asset has a positive supply. Corollary 3. Assume the per capita supply of the risky asset is given by S > 0, and assume that all investors hold S shares prior to trading. Let P§=o> s e {A H} be the equilibrium price in Theorem 1. Then we have PL = P | = 0 - $LSS and PH = PjLo - <&f S, 32 and E[D - P\VS] = $SSS and E[D - P\VH] > E[D - P\VL] > 0. Corollary 3 shows that the price decreases with the supply of the risky asset. The expected future return is given by the risk premium that investors require in order to hold the per capita supply S. Since the quality of the aggregate information decreases with trading volume, investors require a higher risk premium when trading volume is high. Hence, the expected future return increases with trading volume. 33 Chapter 3 Heterogeneous Beliefs and the Underconfident Representative Agent 3.1 Introduction In the first part of this thesis I have assumed that all investors are fully rational, and that they are able to optimally analyze prices and volume. However, in reality prices aggregate information in a complicated way, and extracting this information is a difficult task. In this chapter I assume that the computational skills of the investors are limited: they make individual mistakes when they infer information from the price. I show that in a heterogeneous information economy where investors' rationality is bounded prices react too slowly to new information. The remainder of this chapter is organized as follows. Section 3.2 describes the model for the base case where all investors are homogeneously informed. Section 34 3.3 adds heterogeneous beliefs. In section 3.4 I compare the aggregate demand to the demand of a representative agent. In section 3.5 I examine how the size of the momentum profits depends on the parameters of the model. 3.2 Homogeneous Information There are two assets in the economy, a risk-free asset and a risky asset. Investors can trade these assets at the trading dates t = 0,1,2, One dollar invested in the risk-free asset at time t pays 1 + r dollar at time t + 1. Let R = ^ be the time t price of one dollar at time t + Investors can buy or sell an unlimited amount of the risk-free asset. Assumpt ion 1 (Risky asset). Investors can trade shares of a single risky asset at the equilibrium price Pt. At each trading date t, investors receive a dividend Dt for each share they are holding at time t — 1. The value Dt is given by the process where Do = do, dt ~ ./V~(0, cr2,), and E[dtds] = 0 for t ^ s. The dividends are normally distributed with an unconditional mean of zero. The assumption of normal distribution allows to solve for the equilibrium price explicitly. Setting the unconditional expectation equal to zero simplifies the notation, but does not affect the results of the model. Assume for the moment that the investors know the realizations of future divi-dends with certainty. Then the price of the asset is given by the present value of all future dividends Dt = A - i + <k 35 I will call Pf the full information value of the risky asset. Let [ii,^] be the time interval starting on trading day t\ and ending on trading day t2. Assume an investor buys one share of the asset at time ti and sell it at t2. Let T\ = t2 — h be the length of this interval. Then Ti A t l t 2 = YlRSL>t+s + RTlPt* -Pt, (3-2) s=l is the cumulative dollar excess return that this investor receives. Plugging the full information value of (3.1) into (3.2), we get 3=1 Hence, under full information returns are serially uncorrected.1 Next, assume investors do not know the realizations of future dividends. At each trading date, investors know only the realizations of all past dividends up to the current dividend Dt. In addition, the investors receive noisy public signals about each future dividend innovation. The time t signal about dt+s is given by dt+a,t — dt+s + Vt+s,t s Vt+s,t = y ~ ] £ t + a j ) where all e(+Sij are independent normally distributed random variables with mean zero and variance of. The first subscript of dt+s,t and nt+ait refers to the dividend 1Returns axe only serially uncorrelated, if one calculates returns as in (3.2). Without taking dividends and the discount rate into account, we have, for example, Pf+\ — Pf = 2~Is=i (l+r^*-1' and Cov[Pf+1 - PtF, PtF+2 - PtF+1] = 7,(1 + r ) d ( 2 + r ) -36 innovation dt+s, and the second subscript refers to the trading date at which the investor observes the signal. The time t information about dt+s is given by the true dividend innovation dt+s plus s independent noise terms. At each trading date, one noise term is removed from the signals, so that the signals become more precise, as the payout date of the corresponding dividend approaches. In that way, the investors know more about dividends that the firm will pay in the near future than dividends in the far future. For example, the time t = 0 signal about the dividend innovation at time t = 3 is given by ^3,0 = d3 + r]3t0, 773,0 = e 3 i i + e3]2 + e3)3. At time t = 1, this signal becomes ^3,1 = cfo + % , i = £ 3 , 1 + £ 3 , 2 -The information of the investors about d3 is therefore more precise at t = 1 than at t = 0. Note that the signal tZ 3 i i encompasses the information contained in the signal ci3]o. An investor who observes ci3]i does not need ci3]o to forecast future dividends, since d3fl is distributed as d3ii plus noise. The time t information about all future dividends is therefore given by the vector It = { A , dt+i,t, dt+2tt, dt+3,t, • • •}• (3.4) I omit past dividends in (3.4), since past dividends are irrelevant information for an investor who knows the current dividend. Given the information in (3.4), each investor has to decide how many shares to buy of the risky asset. The economy is populated by a countable set of investors. An individual investor is indexed by i € {1,2,3,...}. Let Xit be the number of shares that investor i holds at time t. 37 Assumption 2 (Investors). Investor i chooses Xit by maximizing Eit -exp{-pWit+1} , where Eit [•} is the expectations operator conditional on the information of investor i at time t, p > 0 is the parameter of risk-aversion, and Wu is the wealth of investor i at time t. For simplicity I assume that investors are myopic. This assumption allows to derive analytical results. It is straight forward to extend the model to the case where all investors maximize the utility of life-time consumption. I have analyzed this case numerically and found that the qualitative results match the results under a myopic investment policy. If the time t + 1 wealth of investors i is normally distributed, conditional on his information set at time t, the demand for the risky asset of investor i is given by In order to find an equilibrium, we have to aggregate the demand of all investors. Let be the average demand per investor. Definition 1 (Equilibrium). An equilibrium price is given by a real-valued price process Pt such that (a) Xt = average supply, Xit — Eit[Pt+1 + Dt+1]-{l + r)Pt PVarit[Pt+1 + A + i ] (3.5) (3.6) 38 (b) there exists real numbers M, <&t, $Dt, ®dts such that |$(| < M, \$Dt\ < M, \$dts\ < M, and In general there will exist more than one price process that satisfies condition (a).2 Condition (b) rules out non-linear equilibria and equilibria that contain ra-tional "bubble-components". The requirement that the equilibrium price depends only on the current dividend Dt and not on past dividends is not restrictive, since the investors do not need past dividends to forecast future dividends. The supply of the risky asset will determine the risk premium. Since, in this model, the riskiness of the asset does not change over time, the risk premium will be a constant. Hence, the risk premium will not affect the correlation of returns. Therefore, in the remainder of the paper I will set the average supply equal to zero. This assumption will simplify the notation by removing the risk premium from the returns, but it will not affect any result of the model. Lemma 3 (Equilibrium with noisy public signals). If the information of the in-vestors about future dividends is given by (3.4), then there exists a unique equilibrium price given by P t = $t + $DtDt + ® d t s d t + s . s=l Pt = Et P t F -. oo dt+s + Vt+s,t where A, -s Cov[dt+g,dt+s] Var[dt+S] 1 and Pf is given by (3.1). 2Whenever applicable, equations hold with probability one. 39 Lemma 3 shows that the equilibrium price equals the expectation of the full information value of the risky asset, conditional on the public information available at time t. The equilibrium price does not depend on the parameter of risk aversion, since all investors are homogenous, and the supply of the risky asset equals zero. Therefore, the demand of each investor equals zero in the equilibrium. The full information value of the asset is the present value of all future dividends. Since investors do not bear any risk in the equilibrium, the price must be given by the expected future payoff of the asset. If there would a positive supply of the risky asset, the price in Lemma 3 would be given by the expected fundamental value of the asset plus a constant depending on the risk aversion. If the price is given by Lemma 3, we can write the price as Pt. Pf + - OO " - OO -}~2(As-l)Rs-'dt+s + - ^ A S J R S - V s , t 3 = 1 S= l , (3.7) full information value uncertainty noise The price consists of three parts: the full information value of the asset given by (3.1), an uncertainty part, and the signal noise. From Lemma 3 we have 0 < As < 1. The uncertainty part decreases therefore the effect of the dividends on the price, rel-ative to the full information value. The weights of the dividends for the price in Lemma 3 are smaller than the weights for the full information value, since the in-vestors receive uncertain signals about future dividends. The uncertainty increases 2 with the ratio of signal noise variance to dividend variance s%. If the investors d observe undisturbed signals, the uncertainty effect in (3.7) disappears. Lemma 3 shows that As decreases monotonically with s. The uncertainty effect is therefore stronger for dividends in the far future than dividends in the near future. The un-certainty effect is stronger for dividends in the far future, since the investors receive more information about the dividends as the dividend payout date approaches. 40 Let PF be the uncertainty part and PtN be the noise part of the price in (3.7), so that we have Pt = PtF + PF + P». (3.8) Recall from (3.2) that A t l t 2 is the cumulative excess return of the asset during the time interval [£i,£2]- If the price is given by the full information value Pf, this return is given by the fundamental return A£ t 2 . If the price is given by (3.8), the uncertainty part PF and the noise part P^ will cause the return to deviate from the full information return. These deviations are given by = ( T T ^ ^ - ^ - ( 3- 9 b) The return A^ t 2 is the dollar excess return that is due to the dividend uncertainty, and A^ t 2 is the dollar excess return that is due to the noise. Plugging the price in (3.8) into the definition of the return in (3.2), we get A M 2 = A£ t 2 + A£ t 2 + A £ 2 . The total return is the sum of the full information return, the uncertainty return, and the noise return. The full information return A£ ( 2 equals zero by (3.3). Since the uncertainty return is not correlated with the noise return, we have Cov[Aht2, At2t3] = Cov [ A g ] +Cov . (3.10) V v ' * ' uncertainty effect noise effect Hence the covariance of the returns during the two time intervals [ti, £2] and [£2, h] is the sum of two effects: the uncertainty effect and the noise effect. 41 L e m m a 4 (Correlation of returns with noisy public information). / / the price is given by Lemma 3, then we have Cov[A»t2,A»t3] < 0, Cov [A M 2 , At2t3] = Gov [ A J k , + Cov [A^2, = 0. Lemma 4 shows that the uncertainty effect is positive, the noise effect is negative, and the two effects offset each other exactly. The uncertainty effect is positive for the following reason. Assume the firm will pay a positive dividend at some trading date in the future. Since the price reacts to this information slowly, the positive future payoff will increase the return during the first period [ti,t2] and during the second period [t2, £3]. In this way, the slow reaction of the price to news has a positive effect on the correlation of returns. To see why the noise effect is negative, assume a signal associated with a partic-ular future dividend contains a positive noise component. On each trading date, the investors increase the weight that they put on this signals when they update their beliefs. Therefore, the positive noise component will initially increase the price and produce a positive return, since the investors can not distinguish between noise and true information. However, at some future trading date, the investors will learn that this particular part of the signal was just noise. A positive noise component today will therefore decrease the price at some point in the future. In that way, the signal noise has a negative effect on the correlation of returns. Lemma 4 shows that the uncertainty effect and the noise effect offset each other exactly. This result makes intuitively sense. Since the riskiness of the firm stays con-stant over time, changes of the price can only be due to changes in the information, 42 and not to changes of the risk premium. But the price at any given time reflects all publicly available information. Future price changes are therefore caused only by the arrival of new information. Since new information is not correlated with today's information, past returns cannot have any predictive power for future returns. 3.3 Heterogenous Uncertainty about the Income of the Firm I will now compare the economy with homogeneous beliefs of the previous section to an economy with heterogeneous beliefs. Assume Investor i receives at time t the information lit = {Dt, dit+ij, dit+2,ti dit+3,t, • • •}, (3-11) where s dit+s,t — dt+s + Y^ ^•it+sj > and the noise terms eit+Sij are independent across the investors. The only differ-ence between the homogeneous information structure in (3.11) and the information given in (3.4) is that now the investors observe signals with individual noise terms instead of common noise terms. There are two possible interpretations for the in-dividual noise terms. One is that the investors receive private information about future dividends. The other interpretation is that the investors observe the same public information, but understand this information in different ways. In this case, dt+s can be interpreted as the true information contained in the public signal, and the individual noise terms can be interpreted as the mistakes that the individual investors make when they analyze the public signal. 43 In either case, if the investors receive heterogenous signals they will have an incentive to use the price as an additional source of information. If the investors are perfectly rational, the result of Grossman (1976) will apply, and the price will fully reveal all information. In this case, the only possible equilibrium price is the full information value of the risky asset. The assumption of perfect rationality is strong, since it requires that the investors completely understand the underlying model of the economy, and are able to infer information from the price without an error. Instead, I will now assume that the rationality of the investors is bounded. When the investors use the price to infer information, they make individual mistakes. In addition, I will for simplicity assume that the investors only use the current price and not past prices as a source of information. This assumption is not as restrictive as it might seem, since, if all investors are rational, past prices are redundant information as the result in Lemma 5 will show. Note that, since investors are myopic, their demands depend only on their beliefs about the next periods' payoff of the asset. Let Vit = Pt-E[Pt\Iit}. (3.12) We can interpret as the unexpected part of the price Pt, after accounting for the information contained in 1^ . If In and Pt are jointly normally distributed, then it follows form the properties of the multivariate normal distribution that I# and are independent, and that E[Dt+1 + Pt+1\Iit, Pt] = E[Dt+1 + Pt+1|I,] + C W [ ^ r { P ^ ] P t | I < t 1 ^ - (3-13) Therefore, we can interpret I# and in the following way: lu contains the non-price related information that investor i at time t uses to forecast the future payoff, and ^u contains the additional information that the investor obtains when he observes the 44 price. Since 1^ and ^ are independent, the additional information in is strictly incremented to the non-price related information in Iit. I will now assume that the investors make individual mistakes when they infer this additional information from the price. The information that investor i infers from the price at time t is given by *wt = tfat + e*« (3-14) where the individual noise terms emt are independent normally distributed with mean zero and variance cr2 . The demand for the risky asset in (3.5) becomes then X a = E[Pt+1 + Dt+1\Iit,y2it}-(l + r)Pt pVar[Pt+1 + Dt+1\Iit,^2it] Note that by (3.13) the demand function in (3.15) is fully rational, if cr2^ = 0. Lemma 5 (Equilibrium with heterogeneous beliefs). If the demand of the investors is given by (3.15), then there exists an equilibrium price of the form r Dt + f^^R'-'dt+s s=l (3.16) If the price is given by (3.16), then we must have and As < < 1 for As given by Lemma 3, and lim^^o Pt = Pf. Lemma 5 shows that the coefficients $ds have the same properties as the coeffi-cients As in the case of homogenous information in Lemma 3. However, as opposed to Lemma 3, the price in Lemma 5 does not contain any signal noise. The price does not contain any signal noise, because the price depends on the aggregate de-mand and therefore on the aggregate information in the economy. Since the investor 45 specific noise terms in (3.11) cancel each other out in the aggregate demand, the equilibrium price cannot depend on these noise terms. We can write the price in Lemma 5 as Pt + 1 0 0 s=l full information value uncertainty The price consists of two parts: the full information value of the asset given by (3.1) and a part that is due to the uncertainty about future dividends. By Lemma 5, the uncertainty part decreases the effect of the dividends on the price, relative to the full information value. Since > Aa, the uncertainty effect in the heterogeneous information economy is smaller than in the comparable economy with homogeneous information. The uncertainty under heterogeneous information is small, since the investors know more on aggregate in the heterogeneous information economy than in the homogeneous information economy, and since the individual investors can access the aggregate information by observing the price. If the individual mistakes that the investors make when they learn from the price are small, the price will be close to the full information value by Lemma 5. Lemma 6 (Correlation of returns with heterogenous beliefs). / / the price is given by Lemma 5, then we have Cov [Atlt2, At2t3] = Cov [A?lta, A£t3] > 0, where A^ t is defined according to (3.9a). Lemma 6 shows that the returns of the time periods [ii,^] and are posi-tively correlated. Recall from equation (3.10) that, if the investors receive homoge-nous information, the covariance of the returns is the sum of two effects: a positive 46 uncertainty effect and a negative noise effect. In the case of heterogenous infor-mation, there is no noise component in the price, since the investor specific noise components cancel each other out in the aggregate demand. Therefore, the only remaining effect on the covariance is the positive effect generated by the uncertainty return A¥,t.. This uncertainty return is due to the fact that the price reacts slower to future dividends than the full information price. In that way, the slow adjustment of the price to future dividends produces positively correlated returns. Chordia and Shivakumar (2002) find that winners outperform losers not only during the post-formation period of a momentum trading strategy, but also during the pre-formation period. Chordia and Shivakumar interpret this finding as evidence that time varying expected returns cause momentum. Note, however, that the covariance in Lemma 6 works in both directions. If the returns during the formation period are positively correlated with the returns during the post-formation period, then the same will be true for the correlation between formation and pre-formation period. The findings of Chordia and Shivakumar are therefore consistent with the idea that aggregate underreaction to new information causes momentum. 3.4 The Underconfident Representative Agent I will now construct a representative agent for the economy of the previous section. Recall that the heterogeneously informed investors of Lemma 5 observe the signals lit = {Dt, dit+i,t, du+2t, • • •}, dit+Sit = dt+s + Vit+s,t, (3-17) where the r]it+3,t are individual noise terms. If the information of the investors is given by (3.17), then the aggregate demand does not contain any signal noise, since the individual noise terms cancel each other out. Since the aggregate demand does 47 not depend on the signal noise, a representative agent must be able to observe the undisturbed signal It = {Dt,dt+1,dt+2,---}. (3.18) As I will show in the following, the representative agent suffers from a psychological bias. He updates his priors too slowly, given the quality of his signals. Specifically, he calculates the distributions of future prices and dividends, as if he observed the noisy signal It = {A, d~t+\,u <k+2u •••}, dt+S}t = dt+s + VZ • f]t+s,u (3.19) where Z is a positive constant, and the noise terms wt+s,t in (3.19) have the same distribution as the r]it+a^ in (3.17). I will assume that the psychological bias of the representative agent is restricted to the way he uses his information in (3.18). When he calculates the distributions of future prices, he fully rationally takes the correct functional form of the equilibrium price into account. The representative agent is therefore aware of the fact that the equilibrium price does not contain any of the noise terms nst. However, he does not use the price as a source of information, since he is the only agent in the economy, and since the price can only contain information that he already knows. Even though he is aware that he has access to the full information in (3.18), his psychological bias drives him to update his priors according to (3.19). Lemma 7 (Underconfident representative agent). Let Z - l - 1 1 +r Cov[Dt+1 + Pt+i,Pt\Iit} Var[Pt\Iit} + ol Then the equilibrium price in the representative agent economy is equivalent to the price in the economy populated by individual investors of Lemma 5, and we have Z e [0,11, lim„2 ^,0Z = 0, and lima2 Z = 1. 48 Note that the variable Z in Lemma 7 does not depend on i, the index for the individual investor, since covariances of normally distributed random variables only depend on the distribution of the signal and not on the realization of the signal. Part (a) of Lemma 7 shows that the aggregate demand in the heterogeneous information economy of the previous section is equivalent to the demand of a representative agent, who observes the undisturbed signal It = {Dt, dt+i, dt+2, • • • }, but updates his priors as if he observed the noisy signal It from equation (3.19). In that sense, the representative agent is underconfident. He underestimates the quality of his information. The coefficient Z € [0,1] measures the degree of underconfidence. Recall from the previous section that Iit is the non-price related information of investor i at time t, and that $ 2 i t is the additional information that this investor obtains when he observes the price. Similarly to (3.13), it follows from the properties of the normal distribution that E[Dt+1 + Pt+1\Iit, $«] = E[Dt+1 + Pt+1\Iit] + (1 + r)(l - Z)#it (3.20) (see Appendix C ) . The function Z determines therefore the degree to which the beliefs of the investors depend on the information obtained from the price. If —> 0, the investors are fully rational, and the price related information will affect the expected future payoff with the factor 1 + r. If a\ —> oo, the investors are not able to infer any information from the price, and the price related information tyit will not effect their beliefs at all. By equation (3.19) the representative agent is underconfident, if Var[\/Z -nt+sj} = Z -sal > 0, for s = 1,2,3,... . The representative agent is therefore underconfident if and only if both of the following two conditions are satisfied: the individual investors have 49 heterogeneously beliefs (a2 > 0), and the investors make individual mistakes when they learn from the price (Z > 0). Comparison of equations (3.19) and (3.20) shows that the measure Z ties the ability of the investors to learn from the price to the underconfidence of the representative agent. The better the investors are able to learn from the price, the higher is the influence of the information coming from the price on the private beliefs, and the lower is the aggregate underconfidence. The relation of individual mistakes and aggregate underconfidence shows that bounded rationality on the individual level is equivalent to a psychological bias on the aggregate level. As a result of his underconfidence, the representative agent does not put enough weight on his signals, when he updates his beliefs. Thus, it seems as if the representative agent suffers from a conservatism bias, identified in experiments by Edwards (1968). Individuals who are subject to this bias tend to underweight new information when they update their priors. Barberis, Shleifer, and Vishny (1998) assume that a representative agent suffers from the conservatism bias to model the underreaction of the price to new information. Lemma 5 and Lemma 7 show that even if the representative agent is psychologically biased, this does not need to be the case for the individual investors. If the investors make individual mistakes when they infer information from the price, then the representative agent will be underconfident because the investors rationally reduce the weights that they put on the price related information when they update their beliefs. The aggregate conservatism bias and the resulting underreaction occurs therefore naturally in a world with heterogeneous information and bounded rationality. 50 3.5 Comparative Statics Figure 3.1 shows how the dividend coefficients of the price in Lemma 5 depend on the parameters of the economy. The top curve shows the dividend coefficients for the case, where all investors know the realizations of all future dividends (of = 0). Since investors discount dividends in the far future more than they discount dividends in the near future, the dividend coefficients are higher for dividends in the near future than for dividends in the far future. The remaining curves of Figure 3.1 show how the the dispersion of beliefs and the ability to learn from the price affects the discounting. The fact that the dividend coefficients under heterogeneous information are smaller than under full information shows that the price underreacts to new information about future dividends. The price adjusts to news about future dividends slower if the dispersion of beliefs among investors is higher (high value of CT2) or if the investors are less able to infer information from the price (high value of of ). I will now demonstrate how the expected return of a momentum strategy depends on these parameters. 3.5.1 Momentum profits and length of the holding period For the two consecutive time intervals £2] and [t2, t3], I will refer to the first interval as the formation period and to the second interval as the holding period. Consider the regression coefficient _ Cov[Atlt2,At2t3] Var[Aht2] ' { 6 - Z i ) where Atitj is the cumulative excess return for the period [U, tj], as denned in (3.2). The coefficient \I> measures the ability to forecast future price differences based on 51 lit*** 100 Figure 3.1: Price Reaction to New Information. This figure shows the dividends coefficients of the price given by Lemma 5. The remaining parameters are given by p = a\ = \, r = 0.004. the observations of past price differences. Since the unconditional expected returns are zero in this model, we have E H2t3 tit2- (3.22) Therefore, we can interpret \t as the cumulative expected dollar return for a buy-and-hold strategy with time horizon T2 = £3 — t2, if the asset has increased by one dollar during the period [ti,^ ]- Figure 3.2 shows how this momentum return increases monotonically with the length of the holding period. The dependence of the momentum return on the remaining parameters is less clear. Even though, as Figure 3.1 shows, the parameters of the economy affect the underreaction of the price to new information about future dividends in an intuitively obvious way, this effect is not as obvious for the momentum return. The dependence of the momentum return on the parameters is complicated, because the underreaction affects current and future prices, and the momentum return depends on the difference between 52 0.3 0.1-1 36 Figure 3.2: Momentum Return. This figure shows how the momentum return = ^^j^^f^ depends on length of the holding period T 2. The momentum return \I/ • Ti is the expected dollar return for a buy-and-hold strategy with time horizon T2, if the asset has increased by one dollar during the formation period. The remaining parameters are given by Ti = 6, a\ = 0.1, ol = 105, o\ = p = 1, r = 0.004. these prices. Therefore, instead of focussing on the dependence of the momentum return on the parameter values, I will now show how the behavior of the asset during the formation period affects the holding period return. 3.5.2 Momentum profits and reversals during the formation period I will now divide the formation period [ti, t2\ into two separate time intervals [t\a, t\b] and [ti6,i2]- Figure 3.3 shows a picture of the time line. In order to be able to 53 formation per iod holding per iod H 1 1 h-ha hb £2 £3 T\a Tib T2 Figure 3.3: T i m e L ine . compare momentum returns for formation periods with different lengths, I multiply the expected holding period return ^ with the length of the corresponding formation period. If the length of the formation period is Ti, then the resulting return \I> - Ti is the expected dollar return for a buy-and-hold strategy with time horizon T 2, if the asset has increased by an average of one dollar per trading date during the formation period. The top part of Figure 3.4 shows the expected returns from three separate regres-sions, where the dependant variable is either the return during the first part of the formation period, or the second part of the formation period, or the total formation period. For example, ^aTia in the top part of the Figure is the expected cumulative return for the holding period [ii,^], if the asset has had an average return of one dollar per trading date during the period [tia, Similarly, ^bTu is the expected holding period return for an average return of one dollar during the period [t\a, tu,]. The return \&Ti is the expected return, if the average return was one dollar during the period [ti, ^2] • The bottom part of Figure 3.4 shows coefficients \I>a and ^b from E A t l b t 2 =t f a A t l o f u + Vl>bAtl6t2. (3.23) The sum tyaTiaA + ^bTuB can be interpreted as the expected dollar return for the holding period [£2 ,^3 ] , if the asset has increased by an average of A dollar per" trading date during the first part of the formation period [£i05 l^b]? and an average 54 of B dollar per trading date during the second part of the formation period [t\a, t\b]. The top part of Figure 3.4 shows that taken separately, the returns during both parts of the formation period are positively correlated with the return during the holding period. However, if one calculates expected returns as in (3.23), the bottom part of Figure 3.4 shows that the return during the first part of the formation period is negatively correlated with the holding period return, and that the return during the second part of the formation period is positively correlated with the holding period return. Assume the return of the asset during the second part of the formation period [tu,, t2] is positive. Then the expected holding period return will be higher, if the return at the beginning of the formation period [tia, is negative, than if the return at the beginning of the formation period is positive. In other words, the asset exhibits stronger momentum if the asset has experienced a recent reversal. In addition, it is also possible to predict future reversals. Assume, that the returns A t l o t u and A t l f c t 2 are both positive, but that the return at the beginning of the formation period A t l a t l 6 is sufficiently higher than the return at the end of the formation period A t l b t 2 . In that case, as Figure 3.4 shows, the expected holding period return will be negative, even though the total return for the formation period [£1,^ 2] is positive. As Figure 3.5 shows, for certain parameter values it is possible to predict the timing of the reversal during the holding period. To understand the source of the predictive power of reversals note that future dividends produce trends in the price, if the price reacts to news about dividends slowly. Assume now that the price of the risky asset has increased in the recent past. Then it is not clear whether this trend has just started or whether it is already at the end. Dividing the formation period into two subperiods helps to distinguish between these two situations. If the price has decreased during the first part of the 55 F i g u r e 3.4: M o m e n t u m R e t u r n wi th Reversals dur ing the Format ion P e r i o d . In this figure, the formation period [ti,t2] is divided into two time intervals [hajtib] and [t\b,t2\- The return \f/Tj in the top and in the bottom part of the figure is the expected dollar return for a for the holding period \t2,ta\, if the asset has increased by an average of one dollar per trading date during the formation period \t\,t^. The return tyaTia (^bT\b) in the top part of the figure is the expected dollar return, if the asset has increased by an average of one dollar per trading date during the formation period [tia,tib] ([tib,^])- In the bottom part of the figure, the sum 9aTiaA + ^bTuB is the expected dollar return, if the asset has increased by an average of A dollar per trading date during the first part of the formation period [tia,tib], and an average of B dollar per trading date during the second part of the formation period [t\a,tib\- The remaining parameters are given by T l a = Tib = 3, T i = T i a + Tu, of = 0.1, a\ = 105, a\ = p=\, r = 0.004. 56 m-m-m u mm m mm • • • > • • • • > • • > • • • • • • • • 36 57 F i g u r e 3.5: M o m e n t u m re turn wi th reversals dur ing the formation pe-r iod I I . In this figure, the formation period [t\,t2] is divided into two time intervals [tia,tib] and [ii&jtz]. The return i t T i in the top and in the bottom part of the figure is the expected dollar return for a for the holding period [ £ 2 , ^ 3 ] , if the asset has increased by an average of one dollar per trading date during the formation period [ti, £2]- The return ^a.T\a (&bT\b) in the top part of the figure is the expected dollar return, if the asset has increased by an average of one dollar per trading date during the formation period [t\a, t\b] ([hb,^])- In the bottom part of the figure, the sum aTiaA + bTibB is the expected dol-lar return, if the asset has increased by an average of A dollar per trading date during the first part of the formation period [tia,tib], and an average of B dollar per trading date during the second part of the formation period [t\a,t\b\. The remaining parameters are given by Tla = Tlb = 3, r x = Tla + Tlb, a2t = 0.01, 0% = 105, a\ = p = 1, r = 0.004. 58 59 formation period and increased during the second part, then it is more likely that the recent price trend is a new, and therefore the trend is more likely to continue. 60 Chapter 4 Conclusion In this thesis I examine how investors can infer private information from market prices and aggregate trading volume. I develop two theoretical models to address the following questions: (1) how can investors fully rationally infer information from prices and trading volume, and (2) how do the results of a world with perfectly rational investors change, if the rationality of investors is bounded, so that they are not fully able to infer all available private information? In the first part of this thesis, I provide a closed form solution for a rational expectations equilibrium where all investors infer information about the state of the economy from (1) private signals, (2) the market price and (3) aggregate trading volume. My model shows how investors can learn from trading volume, and how volume information differs from information contained in the price. The main result of this model is that trading volume reveals the relative quality of the aggregate private information in the economy. Under low trading volume, the aggregate infor-mation is more precise compared to private signals than under high trading volume. Investors therefore use volume to decide how they should weight the market price 61 relative to their own private signals when they update their beliefs. When trading volume is low, investors weight the market price more heavily. Conversely, when trading volume is high, investors weight their private signals more heavily. In this model, the price does not fully reveal all private information, since the price also depends on the unknown aggregate exposure to the labor risk. This combination of private information and unknown labor risk has two effects on the autocorrelation of returns. First, private information causes return continuations, since the price incorporates this information slowly. Second, the aggregate labor endowment causes return reversals, since this endowment influences the price tem-porarily without affecting the future payoff of the risky asset. The total effect in this model is always negative, so that returns are negatively autocorrelated, if investors are fully rational. In the simplified world of the model in the first chapter of this thesis, the price can only be partially revealing, if the price contains an additional source of noise, such as the uncertain aggregate labor risk exposure. Without this additional noise, fully rational investors can infer all private information from the price. In real financial markets, however, prices aggregate information in a complicated way, and extracting this information is a difficult task. In the second part of this thesis I assume therefore that the computational skills of the investors are limited: they make individual mistakes when they infer information from the price. If investors make individual mistakes when they analyze prices, then they are not able to infer all private information from the price, even if there is no additional source of noise that disturbs the price. Therefore, in a world with bounded rationality, private information can have a positive effect on the autocorrelation of returns, without the negative noise effect that is required if all investors are fully rational. Hence, if the 62 computational skills of the investors are limited, prices react to new information too slowly. 63 Appendix Proof of Lemma 1 Assume the price is given by P = §l{D + ri + e) -$%N (4.1) Let Ti = {D, N, P,V} be the information set of investor i. Then the demand of investor i is given by v _ E[D - P\T] - pCov[D - P, NY\Ti} pVax[D-P\Ti] ' [4- Z) where Cov[D-P,NiY\Fi] = T l ^ t ^ ^ - ( 4 3 c) Plugging the demand into the equilibrium condition in (2.1) and solving for the price and comparing coefficients with (4.1), we get a n d $N i Q Lemma 1. Plugging 64 these coefficients into (4.1), and (4.1) into (4.2), we get the equilibrium demand v aDY A , = ^ - n j . Since the are independent across investors, we have Proof of Lemma 2 Assume the price is given by P = &j>{D + ri)-^N (4.5) Then the demand of investor i is given by (4.2). Let T = {£>i,Ni,V} and T = {Di, Nh P, V}. To simplify the notation, I will write § D for and $ N for in the following. Then we have E[D\Ti\ = ^_+a}l + ( J A (4.6a) v r n i - r i + g « ) (A K M V»V>\n = o J T a J T a l ( 4 6 b ) Cov[D,NiY\ti] = T I V / ' ^ (4-6c) E [ P |^ ] = $ v t ^ + ^ A - ^ ^ T ^ ^ ( 4 6 d ) 65 Let Then we have Cov[£>, P\Tj) Vax[P|^ i] E[D-P\T} = E[D\^ + z(p-E[P\^ -P Var[D|7i] = Var[D|^i]-ZCov[D,P|^;] Cov[D,Y\Fi) = C T 2 + C T 2 ( l - Z $ C ) " (4.7) (4.8a) (4.8b) (4.8c) ' D ' " JJ Plugging (4.8) into the demand of investor i in (4.2) and plugging these demands into the equilibrium condition in (2.1) and solving for the price we get P = 1 1 a% + tf+ o-t Comparing coefficients with (4.5), we get o-l al + al + al{l-Z) (4.9a) PO~DY — ol + a\{\-Z) °N + ol (4.9b) o% + <T* + o*(l-Z)o*N + a*(l-Z) Note that an equilibrium exists, if there exists a real number Z, such that the denominators in (4.9) are not equal to zero, and such that Z satisfies (4.7), where $ u and $JV are given by (4.9). Let g(Z) be the right hand side of (4.7). Then it 66 follows from (4.9) that 5 (0) > 0, and we have m = <>b<>?(°D + tf)°N 9 { ) o*Do*(o*D + o*)o% + f?o*{o% + a\ + o*)o*(o% + o * ) ' hence 0 < 5(1 ) < 1. Define f(Z) = Z-g'Z). Then we have / ( 0 ) < 0 and /(l) > 0. Hence, since f(Z) is continuous for Z e [0,1], it follows from the intermediate value theorem that there exists an equilibrium for some Z e [0,1]. Next I will show that in any (linear) equilibrium we must have 0 < Z < 1. Assume the economy is in an equilibrium. From (4.6f) and (4.9a) we have C o » [ A P | i - , ] = ^ ( i - ^ | ^ Using (4.7) we get Using (4.6e) and (4.7) again we get M. So, since <&D =^ 0 by (4.9a), we have 0 < Z < 1. Assume Z = 1. Then we have 2 = <£+o'1 by (4.9a), and we must have <&JV = 0 by (4.10). But we can only have $ J V = 0 if Z > 1 by (4.9b). Hence Z e (0,1). For the demand we have from (4.2) and (4.8) where o% + l* + o* ifD ~ ^ + a*)Z*D) (412a) = P °DY, Ll + al(l-Z®D))-Z$N^— (4.12b) *p = 1 - Z (4.12c) 67 Plugging (4.5) into (4.11) we get 1 ($z> - VP<bD)D + %D€i - (9N - yP®N)N - VNm (4.13) pWs,i[D\Ti} From (4.9a) and (4.12) we have * r j - * p * D = 2 , \ , 2 ( 4 ~ ( 4 + o$)Z$D) - (1 - Z)$ c al{al + al)Z an + ai + ai 2\°~D ol + aZ + a^l-Z) (1 - Z)$D = (1-Z)$D-(1-Z)$D = 0 Similarly, from (4.9b) and (4.12) we have (4.14) ol + al + al (al + a£2(l - W)) - - (1 - Z)*N \ \ / / aN -r an *Nr»+f^:z)-z - - ( 1 - Z ) 'JV ^ u n = 0 (4.15) From (4.8b) and (4.9a) we have VarfDl^ 7,] = a2D(l - $D). Hence we get from (4.13), (4.14), and (4.15) 1 - Z Since the and rij are independent across investors, we have (4.16) (4.17) 68 Proof of Theorem 1 B y (4.4) a n d (4.17) we have VH > VL i f ^ > * 2 D ( 7 2 + * > 2 . (4.18) °~D F r o m (4.8), (4 .12b) , a n d (4.16) we have Xi = ^Dei-^Nni 'z$N-& p V a r ^ l ^ j a2D H e n c e we have f r o m (4.18) VH > VL i f ,2 \ 2 ® p V a r [ D | ^ J I cD / ) V a i [ D | ^ ] I W e have pVar[D\Fi]2M = p V a r ^ ] 2 * 2 ^ 2 + Z^-g^ (z$>N^—2 - 2 ^ 1 p V a r [ D | ^ ] > ) F r o m (4.8b) a n d (4.9b) we have 2 2 Z $ " - T ? - 2 - ^fpV*i[D\Fi] = Z<S>N^-^-paDY{l-<S>D) <TN + an °~D °~N + 0~n = Z * J V - O — ; — o - h~;—o -(1 - ^ ) $ J V H e n c e we get 17JV "t" JV CTn CTJV "r ° n 69 From (4.14) and (4.10) we have Vjjai - Z ( l - Z)$ N N „2 (1 - Z)2&Dal - Z ( l - Z)$ = 0. 'N N~2 il-Z) (1-Z) s2 JI 'N <N „2 a N Hence so VH > VL. M = Z& N aWn a% + al(l-Z) < 0 , Proof of Corollary 1 2 ^ If signals are perfectly correlated, we have -E[-D|.Fj] = g 2 _3+0.2 A by (4.3a). In this case, investors do not use the price to update their beliefs since they cannot directly observe the aggregate labor risk shock N. If the correlation of signals is low, we have E [ D | ^ i ] = E[D|£] + z ( p - E [ P | £ ] ) = a2 • I2 • a2 { ? » - ( 4 + < f lS*£) A + + ^ (1 - Z)$£A + ^ ' A T rJVi + Z P by (4.8a) and (4.14). So since $^ > 0, 0 < $ g < and Z € (0,1), we have V $ > 0, > 0, and 0 < ^ < if>%. 70 Proof of Corollary 2 Let Vaxx, = Varf/}^], if the correlation of signals is low, and let Var# be the corresponding variance if the correlation of signals is high. Then we have \l/p = ^ i ^ - by (4.2). From (4.3b) and (4.8b) we have VarL = Vai„g3+<ffi~,f"Z). So = Z)a^%%Dz)^P- So> s i n c e *D e (0,1) and Z € (0,1), we have Proof of Corollary 3 Assume the price is given by P = + 77 + e) - $jfN -when the dispersion of signals is low, and by P = $Z(D + ri)-§5N-$%S when the dispersion of signals is high. Then it follows directly from Appendix A and B that CT2 + C T 2 ( 1 - Z $ D ) 1-Z hence 0 < < $f. 71 Proof of Lemma 3 If an equilibrium exists, the price must be of the form CO P t = $t + + Y $dtsdt+s by Definition 1. Then we have 3=1 A + i + Pt+i = $t+i + (1 + $Dt) A + i + E ^ t + i s d t+l+3 ^t+l+SJ 3=1 The variance of the signals contained in the vector It is given by Var dt+s,t Var dt+s + E 6 t + s j 3=1 2 , 2 ad + sa€. Let Since all the elements of It are independent, we have E H+l+s A a + i d t + i + S t t , E y~i et+i+s,j .3=1 It s a. j~ A s + i d t + i + s j -So E i t [ D t + i + Pt+i] CO = $ t + i + (1 + § D t + i ) { D t + AA+^t) + E ®dt+is s=l A Sa* A As+i -\ 2~ A s + i (4.19) (4.20) (4.21) (4.22) (4.23) d, t+l+s,t $ t + i + (1 + $rj t+ i ) (A + Axdt+U) + E " W s - ^ d 3=1 72 Plugging this conditional expectation into (3.5), and (3.5) into the equilibrium con-dition in Definition 1, and solving for the price, we get Pt = i f 00 A \ dt+ls—Z dt+l+s,t I • Comparison of coefficients with (4.19) gives 1 -(- r = Y ^ ( i + $ D t + i ) , $«tt+n = r 7 - ( l + ^+i)> 1 + r . l A , 1 + r A a_i If an equilibrium exists, then there exists a real number M such that |$ t| < M, \®Dts\ < M , and \$dts\ < M for all t, s, by condition (b) of Definition 1. So $ t = 0, ®Dt = and $ d f s = Proof of Lemma 4 I will prove the Lemma for Cov A t ) t + i , A t + i ) t + 2 . The case for general time intervals follows by induction. From (3.7) and (3.8) we have So r ^ (1 + r) s=l v ' - i + r ^ + i - r 2_. ( i + r ) , - i s dt+s-73 So Cov \u Au _ ad Aa — Aa+1 As_i — Aa ^ n s ince A s > A a + i . W e have 1 ^ A r 2 ^ ( i + r ) - i ^ + - . ' ' 5=1 v 7 SO t^i*2 - (1 + r ) T x / * a ^ t i _ r (1 + r ) s s=l v ' So we get Cov 0 0 / A A \ oo . E l A s - 1 — As ^ \ _ <72 ^ i , " A + l 8=1 * ' H e n c e Cov At,t+i, A t + i ) t + 2 A t+l,t+2 AN AN *-%t+l> ^t+l,t+2 8=1 * ' 8=1 ^ ' _ _1_ sr^ Aa — As+i ~ ^2 2-^1 (\ _|_ r ] 2 s - l s=l ^ ' = 0. A a - l A s - , A A a . So Cov AN AN < 0. 74 Proof of Lemma 5 Assume the price is given by oo Pt = a>DDt + Y,*d.dt+a. (4.24) 3=1 Since all elements of lu are independent and since Cov[dt, ds] = 0 for t ^ s, we have Var[dt+s\Iu] = sa2As, (4.25) where A„ is given by (4.21), and we have Cov[dt, ds\Iu] = 0 for t ^ s. So we get oo E[Pt\liuDt] = $DDt + J2®dsE[dt+s\Iit], 3=1 Var[Pt + €Pit\Iit, Dt] = £ $2sVar[dt+s\Iit] + a2p, 3=1 O O E[Dt+1 + Pt+i\Iu,Dt} = {l + $D)(Dt + E[dt+1\Iu}) +Y®d3-iE[dt+s\Iit], 3=2 oo Cov[Dt+1 + Pt+i,Pt + ePit\Iu, Dt] = (1 + $D)<f>diVar[dt+1\Iu\ + ^ §ds-i$dsVar[dt+s\lu 3=2 Note that all of these conditional moments are finite, since \<tds\ < M. Let v _ Cov[Dt+i-rPt+i,Pt+epu\Iit} ( . Var[Pt + ePu\ltt] " ^ 75 We have E[Dt+1 + Pt+1\Iit,Dt,Pt + ePit] = E[Dt+1 + Pt+i\Iiu Dt] +Y(Pt + ePit - E[Pt\Iit, A ] ) = (1 + $ D ( 1 - Y))Dt + (1 + $D - Y^dl)E[dt+1\Iit, A ] O O + £(4>*-i - Y$ds)E[dt+s\Iit} + Y(Pt + ePit). (4.27) s=2 Plugging the conditional expectation into (3.15), using (3.6), (4.22), and the equi-librium condition in Definition 1, we get «- 1 i + r - y (1 + - Y))Dt + (1 + $D - YQ^AKk+i + J ^ - i - Y$ds) 8=2 Comparison of coefficients with (4.24) gives $ D = - (4.28a) r *- " T r H - | ! t , - r ) ( 4 2 8 b ) Next I will show that 0 < Y < 1 + r, if a2^ > 0. From (4.28a) and (4.28b) we have 1 + $ D = $ d l [y + A71(l + r- Y)]. So, using (4.28c) we get oo Cov[Dt+1 + Pt+1,Pt + ePit\liuDt) = + + r ~ 5=1 76 Using (4.26) we get ( l ± r ) E ~ x A-^lsVar[dt+s\lit] Cov[Dt+i+Pt+i,Pt+£pit\Iit, Dt] - ——— ^ n i 2 ^ . r j IT I 1 + Var[Pt + ePit\\iu Dt\ 2wS=i(A» ~ 1 ) < J ) d s v / a r l . £ f t + « l i ^ J So, applying (4.26) again, we get Y _ (l + r ) S Z i ^ ' ^ H ^ M ( 4 2 9 ) So 0 < Y < 1 + r, since As > 0. Let R 1 + r ^ s y + ^ -Hi + r - r ) ' and s From (4.28c) we have **=—n T # - = ( 4 - 3 ° ) 1 + r A B , 1. Since 0 < 7 < 1 + r we have As < Bs < 1. So Aa < $ d s < 1, and > $ds+i> and l ims^oo = 0. From (4.29) we have lima2 Y = 1 + r. Hence limCT2 ^ 0 B s liixw _ 0 $ds = 1-I will now show that an equilibrium exists. Let L = K r u, - 1 r(l + r)s~1' Then we have from (4.30) L < < Us. Let B be the set of all infinite sequences of real numbers x = (xo,x\,...), such that xs € [L, Us]. For $ = ($/0) $71) • • •) £ # let /,(*) be the right hand side of (4.30). Let f ( * ) = (/ 0,/i,.. .)• Then f € B. 77 With each element x € B associate the norm ||x|| = siux,. 6 x |a;s|. Then f($) is continuous. Since lims_00(l7s - L) = 0, it is easy to show that B is compact.1 So B is complete. Also B is clearly convex. Hence it follows from Schauder's fix point theorem that there exist a <&* € B, such that f ($*) = <&*. Proof of Lemma 6 If the price is given by Lemma 5, then we have P t = Pf + P/7, where 1 0 0 P " = -Y(^s-Rs-1)dt+s. So So 3=1 1 1 ^ t , t + i - 1 + r / t + i rt - r v^v A , _ i - A , (1 + r ) 3 - 1 since As > As+\. AU A U 1 _ ad s ^ t , t + l » ^ t + l , t + 2 j - r? / „ ~7J A s - As+1 As-! - A ^ (l + r)s (l + r)» f >0, Proof of Lemma 7 Assume the price is given by o o P t = $ D D t + J2$dsdt+s. (4.31) 8=1 l rThe proof that B is compact is a simple extension of the proof that every k-cell is compact. See for example Rudin, 1976, page 39, Theorem 2.40. 78 For Z given by Lemma 7, Y given by (4.29), and Aa given by (4.21), we have Var 3=s = sa2£Z = sa l-Y 1 + r / y +4-1(1+ r - y ) 1 + r Let Then we have 1 + r y + 4 1 ( 1 + r - y ) ' ( 4 - 3 2 ) Under the assumptions of Lemma 7, the conditional expectation for the representa-tive agent is given by CO E[Dt+1 + Pt+1|Ift] = (1 + ®D){Dt + Aidt+i) + ^ * d . _ i 4 < W s=2 So we get Pt 1 + r (1 + + Aidj+i) + ^ ^-lAadj+a s=2 Comparison of coefficients gives $£> = and 7 A , r 1 + r (4.33a) (4.33b) Now the result follows from (4.28), (4.32), and (4.33). 79
- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Theses and Dissertations /
- Information in stock prices and trading volume
Open Collections
UBC Theses and Dissertations
Featured Collection
UBC Theses and Dissertations
Information in stock prices and trading volume Schneider, Jan 2006
pdf
Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.
Page Metadata
Item Metadata
Title | Information in stock prices and trading volume |
Creator |
Schneider, Jan |
Publisher | University of British Columbia |
Date Issued | 2006 |
Description | In this thesis I develop two theoretical models to analyze how investors can infer private information from market prices and aggregate trading volume. In the first chapter I provide a closed form solution for a rational expectations equilibrium where all investors infer information about the state of the economy from (1) private signals, (2) the market price and (3) aggregate trading volume. The main result of this model is that trading volume reveals the relative quality of the aggregate private information in the economy. Investors use volume to decide how they should weight the market price relative to their own private signals when they update their beliefs. In the second chapter, I assume that investors make individual mistakes when they infer information from the price. I show that in a heterogeneous information economy, bounded rationality on the individual level is observationally equivalent to a psychological bias on the aggregate level. If the investors are not able to infer perfectly the true state of the economy from the price, then the aggregate demand corresponds to the demand of a representative agent who is "underconfident". The underconfidence of the representative agent causes the price to adjust to new information too slowly. |
Genre |
Thesis/Dissertation |
Type |
Text |
Language | eng |
Date Available | 2010-01-18 |
Provider | Vancouver : University of British Columbia Library |
Rights | For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use. |
DOI | 10.14288/1.0092960 |
URI | http://hdl.handle.net/2429/18558 |
Degree |
Doctor of Philosophy - PhD |
Program |
Business Administration |
Affiliation |
Business, Sauder School of |
Degree Grantor | University of British Columbia |
Campus |
UBCV |
Scholarly Level | Graduate |
AggregatedSourceRepository | DSpace |
Download
- Media
- 831-ubc_2006-200834.pdf [ 2.76MB ]
- Metadata
- JSON: 831-1.0092960.json
- JSON-LD: 831-1.0092960-ld.json
- RDF/XML (Pretty): 831-1.0092960-rdf.xml
- RDF/JSON: 831-1.0092960-rdf.json
- Turtle: 831-1.0092960-turtle.txt
- N-Triples: 831-1.0092960-rdf-ntriples.txt
- Original Record: 831-1.0092960-source.json
- Full Text
- 831-1.0092960-fulltext.txt
- Citation
- 831-1.0092960.ris
Full Text
Cite
Citation Scheme:
Usage Statistics
Share
Embed
Customize your widget with the following options, then copy and paste the code below into the HTML
of your page to embed this item in your website.
<div id="ubcOpenCollectionsWidgetDisplay">
<script id="ubcOpenCollectionsWidget"
src="{[{embed.src}]}"
data-item="{[{embed.item}]}"
data-collection="{[{embed.collection}]}"
data-metadata="{[{embed.showMetadata}]}"
data-width="{[{embed.width}]}"
data-media="{[{embed.selectedMedia}]}"
async >
</script>
</div>
Our image viewer uses the IIIF 2.0 standard.
To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0092960/manifest