VIEWS: 14 PAGES: 13 CATEGORY: Business POSTED ON: 7/22/2010
What Is a Closing Order in the Stock Market document sample
Expectations of Professionals in Turkish Stock Market: A Study of Monthly Reuters Survey Numan ÜLKÜ* ABSTRACT Reuters-Turkey News Service conducts a monthly survey of forecasts of the senior analysts, portfolio managers, or strategists of major brokerage houses in Istanbul Stock Exchange on the next month’s closing level of the ISE-100 index. This survey data is analysed in this study, in order to assess forecasting performance of market professionals, to test previously documented behavioral biases of analyst expectations and some specific hypotheses of the noisy rational expectations literature. Results indicate that: i) survey participants’ average forecasts have no predictive power, ii) they are too optimistic, iii) they extrapolate from the past despite the fact that the degree of extrapolation is negatively correlated with forecast performance, iv) extrapolation becomes insignificant when previous period’s return is negative, v) the forecast performance is not correlated accross time, vi) participants are heterogenous in terms of their expectations, vii) but no association between the level dispersion in forecasts and market volume and volatility is observed, viii) participants seem to place greater weight on market price than their priors. * Yeditepe University, Istanbul, Turkey; nurolmenkul@superonline.com The author is thankful to Güzin Övünç and Reuters Turkey News Service for sharing the survey data. The views expressed here those of the author and cannot be associated with Reuters. I. INTRODUCTION Reuters-Turkey News Service conducts a monthly survey of forecasts of the senior analysts, portfolio managers, or strategists of major brokerage houses in ISE (Istanbul Stock Exchange) on the next month’s closing level of ISE-100 stock index, since January 2002. This unique data set provides us an opportunity to assess the predictive performance of professional market participants and test behavioral hypotheses of analyst expectations, as well as to examine some of the extant hypotheses in the noisy rational expectations literature. II. LITERATURE REVİEW Efficient markets theory predicts that no forecaster should have the ability to consistently predict future returns, if the market is efficient. Evidence from early empirical studies tended to support this basic prediction of efficient markets theory (Fama, 1970). However, more recent literature of the last two decades contain a great deal of predictability findings. Stock index returns, the focus of this paper, are found to have predictable components from past returns (Fama and French, 1988), or from technical analysis signals (Brock, Lakonishok and LeBaron, 1992). These evidence, inconsistent with market efficiency, are, however, obtained in ex post studies. More direct evidence should come from ex ante forecasts of real-life market participants. So, studies of the performance of market forecasters and portfolio managers have been part of the literature testing efficient markets theory. The earliest example was Cowles (1934) who found that most of the forecasting effort was unsuccessful. See especially Hartzmark (1991) who found that superior performers are less than that would be expected under pure chance and that above- average performance is not correlated over time. The performance of mutual fund managers turns out similar results. Available evidence suggests little systematic predictive ability on the part of market professionals. These studies, taken together, are consistent with market efficiency and imply that predictability found in ex post tests could not have not been utilized by real-life practitioners, at least by an average of those sampled in these studies. Another area of interest in analysts’ forecasts stems from behavioral hypotheses. Lakonishok, Shleifer and Vishny (1994) explain the profitability of value strategies by extrapolation. Dechow and Sloan (1997), on the other hand, find that stock prices appear to reflect analysts’ biased forecasts of future earnings growth rather than naive extrapolation of past trends. These studies open up the need for further tests of biases in analysts’ expectations, especially the tendency to extrapolate from the past. While one way of conducting such tests is experimental settings (see DeBondt, 1993; and on turkish subjects, Muradoğlu, 1996), transferability of findings from experimental designs to real-life pose some problems. Direct surveys with real-life agents may overcome some of such shortcomings. Another behavioral bias documented in the literature is systematic optimism. Dechow and Sloan (1997) report that over their 15-year I/B/E/S sample with 70,000 earnings forecasts per analyst per quarter per company, the mean forecast earnings growth rate is 14% while the mean rate of realized earnings growth is 8%. This suggests that analysts systematically overestimate expected earnings. See also a more detailed study by Easterwood and Nutt (1999) who differentiate between systematic misreaction and systematic optimism, and provide results consistent with the latter. In the context of company earnings estimates, systematic optimism bias is often related to agency issues. It will therefore be very interesting to see whether the optimism bias still holds in the context of market index forecasting where no agency issues are present. Finally, noisy rational expectations literature contains several premises and findings that need to be tested in real-life settings in order to be assessed in terms of validity. Market models tend to move from simple structures with uniform information (e.g.; Grossman and Stiglitz (1980) and Kyle (1985)) to complex structures with heterogenous information (e.g.; Foster and Viswanathan (1996)). The shift towards the latter reflects the need for more realistic modeling of financial markets, and results obtained do differ. This suggests that what assumption is valid in contemporaneous financial markets does matter in determining which results will apply. One such issue is heterogeneity of expectations: MacDonald and Marsh (1996), for example, find that “currency forecasters have heterogenous expectations” and “such disagreements are the key variables in determining market trading volume”. More empirical studies are needed to characterize the valid structure of expectations as well as to test the hypothesized relationships between degree of conditional heterogeneity and conditional variances and trading volumes. Another area of interest is the hypothesized role of prices as aggregators of private information (see Cho and Krishnan, 2000). Section III below outlines details of the data and methodology, results are discussed in Section IV, and main conclusions repeated in Section V. III. DATA and METHODOLOGY Forecast of each participant is obtained on the last trading day of the preceding month. The deadline is market closing time, so that informational symmetry among participants is ensured. Survey results are announced on the same day shortly after market closing as a news headline on Reuters screens1. The survey consisted of a single number representing participants’ point forecast of the next month’s closing level of the ISE-100 index. The survey was first applied at the end of December, 2001, so the first survey data available is for January 2002. (As of June 2004, 28 1 The raw data of participants and their forecasts can be obtained both from the author and from Reuters Turkey News Service (contact Ms. Güzin Övünç, who conducted the surveys and reported as a news item, at guzin.ovunc@reuters.com.tr). months of data were accumulated, the study is planned to be finalized when 36 months of data will be accumulated, even though basic results seem to have already appeared.) 2 While participants had been kept anonymous in the first 9 months of the survey, they have been publicized in the name of the brokerage house since October 2002 3. This change of procedure was proposed by the author of this paper to prevent strategic behavior, as participants representing their institutions would be expected to do their best. Though, it has had little impact on mean forecast errors: under anonymity, the average absolute error was slightly larger than median absolute error (12.94% vs. 12.70%), under publicity the average has been slightly smaller than the median (11.77% vs. 11.96%). The difference between the behavior of the mean and median is consistent with some degree of strategic behavior, but having only a small impact on overall results. Average number of participants per survey is around 17-18. There is some degree of discontinuity: Only two out of 26 regular participants has 100% attendance, 15 out of them has an attendance rate of 65% or higher. The remaining 11 with lower attendance are excluded in performance analysis, but included in the computation of average expectations as they are unbiased observations. A few participants have transferred from one institution to another; in such cases we treated the analyst rather than institution as the unit of observation. The average response rate per survey is aoround 80% (i.e.; around 22 participants were asked to reply on average each time), which is high compared to similar surveys done with financial market professionals. This is probably because the survey is conducted with volunteers willing to exhibit or at least challenge forecast ability, their primary job task. Forecast closing levels are first transformed into forecast returns. For each month, the market’s forecast is computed both by taking the simple average and also the median (to avoid the impact of outliers) of forecasts of individual participants. Strategic behavior (i.e.; submitting a forecast, which, if actualized, would benefit the participant) has always been an issue in surveys whose results are made publicly available. Even if one can assume that participants are rational enough to consider that their forecasts would have little impact on the equilibrium price in an efficient market and the loss of prestige as a result of bad forecasts would be far more costly, there is another reason to believe that the forecasts are biased toward the trading positions of the participants: Wishful thinking (i.e.; assigning a higher probability to a desirable event than its true probability). While the publicity of the his/her institution and the consequent competition is an incentive for a participant to do his/her best, it cannot perfectly ensure this. Yet, the bias toward already held trading positions is not something to be avoided from the researcher’s point of view, rather it enhances the link between what people say and what 2 There has been an interruption for two months between Dec.2003 and Jan.2004 due to temporary leave of the Reuters News Service reporter conducting the survey. This is unlikely to affext results of the study, nor to alter the treatment of data. 3 This is done with the consent of the participant, expectedly with the approval of the management of participant’s affiliated institution. 3 participants refused this and withdrew from the survey. people do. This link is crucial in making inferences from research results. The chronological order, however, must be taken into account here: To the extent that current forecasts are biased toward already held positions, they will reflect action that must have been already taken in the recent past, rather than to be taken in the future. These ideas will help in interpreting the results. To present some descriptive statistics, the mean realized monthly raw return of the ISE-100 index over the 27-months sample period is 0.009, while the sample standard deviation is 0.127. This suggests a period of negative excess market returns, but remember that the unconditional estimate of the excess market portfolio returns for the ISE since July 1987, the beginning of the available return data, is negative. The mean average forecast return is 0.065 and the sample standard deviation of average forecasts 0.043. IV. RESULTS 4.1. The Forecasting Performance: The forecasting performance of all participants as a group (let’s call “the market”) is assessed by the relation between median forecast returns and realized returns of the ISE-100 index. The results with simple average forecast returns are similar and not reported here. Let’s notate realized return of the ISE-100 index in month t as Rt and median forecast return for month t as FRt . The following simple linear regression is run (so far t = 27): Rt = a + b FRt + et (1) Forecast performance of the market is simply assessed by testing the null hypothesis that b = 0 , and by the R2 of this regression. The null hypothesis that b = 0 (i.e.; that there is no statistical relation between forecast and realized returns) cannot be rejected at 10% level; in fact F = 0.13 and sig (F) = 0.72 is far from any significance. R 2 is 0.005. Put in another way, the correlation between median forecast return and the realized return is insignificant at +0.072. The clear conclusion is that median forecast does not have any significant predictive ability. The average absolute forecast error is 12.16% which is comparable to the standard deviation of monthly returns over the sample period at 12.68%. Additional inspection by subperiods suggests that market’s absolute forecast error and standard deviation of realized returns tend to move together. As a robustness check, the ability of the median forecast to predict the return of the first half of the month and the average index level over the month (rather than the end of the month) is investigated in order to account for the possibility that even though median forecast has some predictive ability under the current information set but unforeseen events later in the month makes them obsolete. Results, though slightly stronger, are still far from significance. The null hypothesis “b = 0” cannot be rejected at 10% level under any of these alternative definitions of the dependent variable (i.e.; forecast month’s return). As to the performance of individual participants, the same regression is run seperately for each of the 15 particiants with attendance rate 65% or higher. Let’s notate the forecast of participant i for month t as FRit . Rt = ai + bi FRti + eti (2) The results are presented in Table 1. Forecast performance is assessed by testing the null hypothesis “b i = 0”. Results suggest that the null hypothesis cannot be rejected at 10% level for none of the 15 participants. The coefficient for 6 participants is negative. The mean absolute forecast error of the “best” participant is 10.5%, and that of other participants range between 12.1% and 16.5%. The overall result is that none of participants has statistically significant predictive ability over the 1-month horizon. The distribution of the b i coefficients is close to what would be expected under pure chance (i.e.; if all participants were forecasting randomly), as Kolmogorov- Smirnov test cannot reject uniform distribution. Mean Abs. Participant n bi sig (b) Cor sig (cor) For.Err 1 26 -0.255 0.54 -0.126 0.27 0.121 2 22 0.184 0.52 0.146 0.26 0.147 3 24 -0.345 0.21 -0.265 0.11 0.145 4 25 -0.162 0.59 -0.113 0.30 0.165 5 25 0.047 0.87 0.034 0.44 0.141 6 21 0.270 0.45 0.173 0.23 0.121 7 25 -0.081 0.69 -0.084 0.35 0.159 8 25 0.293 0.28 0.130 0.10 0.127 9 18 0.204 0.64 0.117 0.32 0.122 10 24 0.133 0.71 0.079 0.36 0.134 11 17 -0.216 0.63 -0.127 0.31 0.140 12 19 -0.177 0.59 -0.133 0.29 0.139 13 25 0.555 0.29 0.218 0.15 0.105 14 24 0.312 0.44 0.165 0.22 0.125 15 21 0.173 0.54 0.142 0.27 0.132 Median 27 0.213 0.72 0.072 0.36 0.122 Table 1: n is the number of months the participant attended the survey. b i is the coefficient of forecast return in Equaiton 2. Sig (b) is the signicance level of rejecting the null hypothesis “b i = 0”. Cor is the coefficient of correlation between forecast and realized returns. Sig (cor) is the the p-value for the one-tailed test that cor = 0. Avg.For.Err. is the simple average of each participant’s absolute forecast error over the months. Another way of assessing forecast performance is to compare mean absolute forecast errors of participants to random walk model’s mean absolute forecast error. Using a simple random walk model of Rt = μ + et where μ is the mean return over the sample period and et is i.i.d., the mean absolute error is 0.105. This equals to the mean absolute forecast error of the “best” participant. In other words, 14 out of 15 participants have produced larger errors than a simple random walk model, suggesting that participants would have performed better had they been forecasting randomly. As in Hartzmark (1991), the persistence in performance is assessed by dividing the sample into two halves. Ranking participants according to average absolute forecast error, presented in Table 2, suggests that, as in Hartzmark (1991), there is little correlation in average forecast performance of individual participants accross time. This reiterates Hartzmark’s conclusion that luck rather than skill seems to drive forecasting performance. Persistence in Performance First Half Middle Best 20% Second 20% 20% Fourth 20% Worst 20% 13 0.092 1 0.129 14 0.155 2 0.158 4 0.180 9 0.121 6 0.148 3 0.156 5 0.159 7 0.184 10 0.124 8 0.152 11 0.156 15 0.159 12 0.191 Second Half Middle Best 20% Second 20% 20% Fourth 20% Worst 20% 15 0.089 8 0.100 13 0.118 7 0.126 2 0.133 6 0.094 12 0.102 5 0.121 3 0.132 4 0.146 14 0.100 1 0.111 9 0.125 11 0.132 10 0.149 Table 2: Participants are ranked according to their average absolute forecast errors in the first and second halves of the sample period. They are divided into 5 quintiles of 20% or 3. The first coloumn of each quintile provides the participant’s code and the second coloumn his/her average forecast error. 4.2. Behavioral Hypotheses: Extrapolation Bias: To determine whether the participants as a group are extrapolating from the past, the following regression is run: FRt = α + β Rt-1 + et (3) The result indicates that the null hypothesis “β = 0” can be rejected at 10% level (F=3.63 and sig(F)=0.068). If FR is defined to be the average, rather than median, forecast return, the relationship turns out to be stronger: The null hypothesis “β = 0” can be rejected at 5% level (F=6.60 and sig(F)= 0.016). This is quite interesting, because it is difficult to obtain a reliable estimate of the market return from experts’ forecasts, but the opposite is possible: The forecasts of the experts can be estimated from previous month’s market return. Given that the first order autocorrelation in monthly return series over the sample period has been insignificantly negative at − 0.20, there is strong evidence of unwarrented extrapolation from the past. An R 2 of 0.20 suggests that around 1/5 of the variation in average forecasts can be explained by the previous period’s return. 4 These results are consistent with the hypothesis that analysts, as a group, herd on market price. The similar regression ( FRti = α + βi Rt-1 + et, i ) is run for each of the 15 high-attendance- participants to find out individual tendencies to extrapolate from the past. Results, presented in Table 3, indicate that the coefficient of the previous month’s return is positive for 14 of 15 4 Muradoğlu (1996) documents a hedging tendency for forecasts extrapolating past trends with skewed interval estimates. Since Reuters data consists of only point forecasts, it is unfortunately impossible to test this hedging tendency in this study. participants, significant at 10% for 2 and at 5% for additional 3. Interestingly, for the best performer (according to minimum average absolute forecast error criterion), the null hypothesis “βi = 0” cannot be rejected at any significance level, implying that he is not extrapolating from the past. Moreover, 4 of the 5 participants who has exhibited significant tendency to extrapolate from the past entered Equation 2 with negative signs. These findings are consistent with the hypothesis that extrapolation and poor performance coincide. A regression of the b i coefficients on βi coefficients turned out significant inverse relationship (n=15, F=3.77, sig(F)=0.074, the coefficent on β i = − 0.97, R2= 0.22 implying that 22% of the variation in performance could be attributed to extrapolation bias). Participant n Βi sig (F) R-sqd 1 27 0.208 0.02 0.20 2 23 0.206 0.25 0.07 3 25 0.249 0.09 0.12 4 26 0.022 0.88 0.00 5 26 0.083 0.58 0.01 6 22 0.244 0.10 0.13 7 26 0.446 0.02 0.22 8 26 0.074 0.64 0.01 9 19 0.173 0.23 0.09 10 25 0.049 0.69 0.01 11 18 0.186 0.21 0.10 12 20 0.335 0.04 0.22 13 26 0.042 0.60 0.01 14 25 -0.014 0.90 0.00 15 22 0.152 0.39 0.04 Median 28 0.213 0.07 0.07 Table 3: Extrapolation bias tested by the significance of rejecting the null hypothesis “βi = 0” in the regression equation FRti = α + β i Rt-1 + et, i. n is the number of months. β i is participant i’s coefficient on previous month’s return. Sig(F) is the significance level of rejecting the null hypothesis “β i = 0”. R-sqd is the ratio of variation in participant i’s forecasts explained by R t-1 to total variation. An interesting question arises here: Can these findings of extrapolation be in fact a reflection of wishful thinking? Most, if not all, of the survey participants are active traders in the market. It is likely that when the market has moved in a particular direction in the recent past most active traders have positions in that direction. Hence, wishful thinking may be as well a potential explanation of these findings as extrapolation. Unfortunately, however, the coverage of Reuters’ survey does not enable us to differentiate between the two alternative explanations. To answer this question, participants’ transactions data is needed which is unlikely to be available. My opinion based on available information is that extrapolation and wishful thinking act together hand-in-hand. Systematic Optimism Bias: The mean forecast error of the market is + 5.6 %, which is significantly positive at 1% level. This indicates that forecasters participating in the Reuters survey have, as a group, overestimated the ISE-100 index return, consistent with the systematic optimism bias. Again, the degree of the optimism bias is negatively correlated with the forecast performance. 32% of absolute forecast errors of individual participants can be explained by optimism bias. This finding is quite important: It suggests that the systematic optimism bias is not confined to only analysts’ earnings forecasts, but is a more general phenomenon. Moreover, as the forecasts are made for the market index, which consists of 100 stocks, it is unlikely that agency issues are involved here. Hence, the agency issues referred to as the likely reason underlying optimism bias in the extant literature fail to account completely for it. Allthough the fact that all participants are employed in the investments sector desiring the market to be attractive may suggest another type of agency issue, this is clearly different and far less direct in effect from that mentioned in the extant literature. A more appropriate explanation may be wishful thinking. Easterwood and Nutt (1999), differentiating between reaction to positive and negative information, document that analysts underreact to negative information but overreact to positive information (which they summarize as “systematic optimism”). In a similar manner, our data set allows us to compare the extent of the extrapolation bias with respect to nature of the information (previous month’s return). Employing a simple approach, data is grouped into two according to the sign of the realized return in previous month, and Equation 3 is estimated seperately for subsamples of positive and negative monthly returns, respectively. Results suggest that extrapolation bias is significant when previous month’s return is positive (n=13, β=+0.2925, F=3.18, sig(F)=0.102, R2=0.22; note that reduced significance levels are due to halved sample size), but insignificant when previous month’s return is negative (n=15, β=+0.1096, F=0.42, sig(F)=0.527, R2=0.03). Similarly, a comparison of mean forecast errors on subsamples conditioned by sign of the previous month’s return, indicates significant differences in the optimism bias: When the ISE-100 index rose in month t-1, the mean forecast error for month t is +0.075, when it fell in month t-1, the mean forecast error for month t is +0.034 (the difference being significant at 10% level according to Wilcoxon’s nonparametric test). These findings reiterate some of the asymmetry, noticed by Easterwood and Nutt (1999), in analysts’ reactions due to the nature of information and enhance their conclusion of “systematic optimism rather than systematic misreaction”. More general conclusion is that part of the extrapolation and optimism biases is confined to extrapolating positive information. 4.3. Hypotheses in the Noisy Rational Expectations Literature Heterogenous Expectations: MacDonald and Marsh (1996) measure the significance of individual differences as follows: FRti = Xt + gi + uti (4) where Xt is the component based on public information common to all participants, g i is the individual (idiosyncratic) effect, and u ti is an individual random disturbance term, which could occur due to measurement errors (such as strategic behavior). The average forecast is: FRtA = Xt + gA + utA (5) . Normalizing such that gA equals zero and subtracting (5) from (4), we obtain: FRti − FRtA = gi + ( uti − utA ) (6) The individual effects (biases) gi may then be measured by the significance of a nonzero g i , without specifying the common information set X t . It is possible that heterogeneity follows both from individual effects and idiosyncratic coefficients on information I t (i.e., each forecaster placing different weights on some element of public information set). Then, FR ti = Xt + gi + βi It + uti where βi is the weight that participant i places on information I. The equivalent for Equation 6 is then: FRti − FRtA = gi + (βi − βA ) It + ( uti − utA ) (7) Under this formulation, individual biases (g ≠ 0) and idiosyncratic interpretation of public i information (βi − βA ≠ 0) can be jointly tested. Equation 7 is estimated by OLS for each of the 15 high-attendance participants. Lagged return Rt-1 is used as the information term. Results indicate 5 significant individual biases and 2 significant idiosyncratic interpretation of the common information (Rt-1). Compared to the results reported by MacDonald and Marsh (1996)5, these imply a slightly lower degree of heterogeneity. The Relation between Dispersion in Expectations and Volatility and Trading Volume: The dispersion of expectations is measured by the standard deviation of forecast returns on a month accross participants. Two regressions of volatility and trading volume, respectively, are run on dispersion of expectations, to test hypotheses in the literature that link trading volume and volatility with disagreement of prior opinions. Volatility in month t is measured in two ways: i) absolute value of monthly index return ii) standard deviation of daily returns in month t. The results suggest that there is no significant relationship between the dispersion of forecasts and volatility (F=0.02, sig(F): 0.88 with monthly absolute returns; and F=0.93, sig(F)=0.34 with standard deviation of daily returns). When the volatility is measured by the standard deviation of daily returns, a slightly positive relationship is observed, but it is far from being significant. There is no significant relationship between dispersion of forecasts and the monthly trading volume, adjusted both for trend and volatility, either (F=1.29, sig(F)=0.27). These findings contradict with those of MacDonald and Marsh (1996) on currency forecasters, who concluded that heterogeneity of forecasts was a factor determining trading volume. Although reading too much of this survey data should be avoided as it represents only a small proportion of market participants, our results support new information arrivals, rather than heterogeneity of prior expectations, as the factor driving volume and volatility6. The Role of Market Price as Aggregator of Private Information: A direct test of the hypothesized role requires the use of multiple forecasts taken at different points in time for the same terminal value. The mere possible test with the limited coverage of this survey data is to check the relationship between forecast errors in month (FRt-1 − Rt-1) and forecasts (FRt). A positive 5 See Tables 1 and 2 on pp.670-671. 6 Note that over the sample period there has been a strong positive relationship between monthly volume and absolute monthly return, and not between monthly volume and standard deviation of daily returns.. relationship would be inconsistent with the role of market price as aggregator of private information, as it would suggest that participants insist on their private beliefs (expecting the market to eventually reverse) rather than revise them in line with the market price. No-relation would be consistent with an assumption that participants, as a group, conform to the information delivered by market price, whereas a negative relationship would suggest that analysts employ some form of an error correction model. The analysis is clouded, however, by the fact that average forecast erorrs and realized returns, which we already know to be significantly positively associated with next month’s forecasts, are strongly positively correlated (the coefficient of Pearson correlation is +0.97 significant at p=0.0000). To avoid multicollinearity, the following regression is estimated: FRt = a + b Res.FEt + et (8) where Res.FEt is the residual of a regression of average forecast error on realized return. Result from the estimation of Equation 8 suggests no significant relation between the forecast error for month t-1 and forecasts for month t (b= 0.053, sig(b)=0.799, F=0.06). This is consistent with the assertion that analysts place greater weight on the information they derive from the market price, so that they conform to the information implied by the market price innovation. V. CONCLUSIONS The main conclusions from the results presented in Section IV are as follows: Participants of the survey conducted by Reuters Turkey news service exhibit no significant ability to predict monthly returns of the ISE-100 index. The distribution of performance is similar to what would be obtained under pure chance. There is little evidence of performance correlated accross time. Almost all of the participants forecasts performs worse than a simple random walk model. These results are more than consistent with the implications of efficient markets theory. Participants of the survey analysed in this study, as a group, exhibit significant biases of optimism and extrapolation, unjustified by the behavior of realized returns. The degree of being vulnerable to these biases is positively correlated with forecast errors. Individual differences in forecast performance can partly be explained by the degree of these biases. Optimism bias, which has been traced to agency issues, is documented in a context of index return forecasting where no agency issue is present. A potential explanation could be wishful thinking. The asymmetry observed in extrapolating positive vs.negative past returns prefers optimism bias over systematic extrapolation (misreaction) as the valid explanation. The fact that short positions are very costly to build in ISE, however, supports “wishful thinking” as the most plausible explanation. Participants’ forecasts are heterogenous, supporting the use of heterogenous expectation models in market equilibrium dynamics. However, the degree of the heterogeneity in forecasts did not seem to drive market volume and volatility. Another possible reason for this finding, however, could be that survey participants constituted only a small proportion of the market. Finally, participants seem to place greater weight on the information they derive from market price than their priors. One contribution of this study might be to confirm the significance of previously documented behavioral biases in analysts’ expectations, in a different real-life context of stock index forecasting in an emerging market. It seems that little predictive ability of forecasters is more than wiped out by behavioral biases to turn out worse performance than a simple random walk model. REFERENCES Brock, William, Josef Lakonishok and Blake LeBaron, 1992, Simple Technical Trading Rules and the Stochastic Properties of Stock Returns, Journal of Finance 47, 1731-1762 Cho, Jin-Wan and Murugappa Krishnan, 2000, Prices as Aggregators of Private Information: Evidence from S&P500 Futures Data, Journal of Financial and Quantitative Analysis 35, 111-126 Cowles, Alfred, 1934, Can Stock Market Forecasters Forecast?, Econometrica 1, 309-324 Dechow, Patricia M. and Richard G. Sloan, 1997, Returns to contrarian investment strategies: Tests of naïve expectations hypothesis, Journal of Financial Economics 43, 3-27 DeBondt, W.F.M., 1993, Betting on trends: intuitive forecasts of financial risk and return, International Journal of Forecasting 9, 355-371 Easterwood, John C. And Stacey R. Nutt, 1999, Inefficiency in Analysts’ Earning Forecasts: Systematic Misreaction or Systematic Optimism?, Journal of Finance 54, 1777-1794 Fama, Eugene F., 1970, Efficient Capital Markets: A Review of Theory and Empirical Work, Journal of Finance, 383-420 Fama E.F. and K.R. French, 1988, Permanent and Temporary Components of Stock Prices, Journal of Political Economy 96 (2), 246-273 Foster, F. Douglas and S. Viswanathan, 1996, Strategic Trading When Agents Forecast the Forecasts of Others, Journal of Finance 51, 1437-1470 Grossman S. And J.E. Stiglitz, 1980, On the Impossibility of Informationally Efficient Markets, American Economic Review 70, 393-408 Hartzmark, Michael L., 1991, Luck versus Forecast Ability: Determinants of Trader Performance in Futures Markets, Journal of Business 64, 49-73 Kyle, Albert.S., 1985, Continuous Auctions and Insider Trading, Econometrica 53, 1315-1335 Lakonishok, Josef, Andrei Shleifer and Robert W. Vishny, 1994, Contrarian Investment, Extrapolation and Risk, Journal of Finance 49, 1541-1578 MacDonald, Ronald and Ian W. Marsh, 1996, Currency forecasters are heterogenous: confirmation and consequences, Journal of International Money and Finance 15, 665-685 Muradoğlu, Gülnur, 1996, Portfolio Managers’ Forecasts of Risk and Return: Are there predictable forecast errors?, paper presented at ISF 1996