Herding in trading by amateur and professional investors.doc

herding in trading by amateur and professional investors byitzhak venezia*, amrut nashikkar* and zur shapira*may 14, 2008rough draft: *the hebrew university, jerusalem and rutgers business school, newark and new brunswick *stern school of business, new york university. we acknowledge the financial support of the sanger family chair for banking and risk management, the galanter fund, the mordecai zagagi fund, the whitcomb center for research in financial services, and the school of accounting, the hebrew university. the helpful comments of participants at the european financial management, (efm), symposium on behavioral finance, 2006, and finance seminar participants at rutgers university and bocconi university are gratefully acknowledged. addresses: itzhak venezia, school of business, hebrew university, e-mail: msvenezmscc.huji.ac.il. zur shapira, stern school of business, new york university, new york, ny 10012, e-mail: , amrut nashikkar. .abstractwe study herding behavior of amateur and professional investors using a unique data-set consisting of all their daily transactions during a four year period, and explore the factors that can explain such behavior. we distinguish between features specific to the stocks, such as their systematic risk, idiosyncratic risk, and size, on the one hand, and market factors such as total stock market volume and returns on the other hand. we find herding tendencies among both types of investors and that this tendency is higher for amateurs. it is shown that herding is affected by both types of factors: it is a decreasing function of the size of the firm, and an increasing function of its risk. idiosyncratic risk tends to positively affect herding but this effect is significantly lower for professionals. systematic risk however influences positively only the herding of professionals. our data also reveal that herding behavior of the two groups is a persistent phenomenon, and that it is closely related to the volatility of market returns. amateurs herding is weakly related to market returns. in general most of the results are consistent with the theory that herding is information based. herding in trading by amateur and professional investorsi. introduction herding behavior by investors has intrigued researchers for a long time because of its potential effects on fluctuations in returns for examples showing how herding and other demand fluctuations affect prices see, e.g., eichengreen et.al. ,1998a,b, furman and stiglitz, 1998, nofsinger and sias, 1999, froot, oconnell, and seasholes, 2001, choe, kho, and stulz, 1997, edelen and warner, 2001, goetzmann and massa, 2002, shapira and venezia, 2006, and sias and starks, 1997. for a comprehensive review of herding literature see bickchandani and sharma, 2000, and hirshleifer and teoh, 2001. in many cases researchers argue that herding is a by product of information availability or the lack of it. different groups of investors receive various types of information and of diverse quality. differences between the group characteristics and different information available to their members may cause such groups of investors to behave differently from each other and yet to exhibit herding within each group. understanding the bases of herding behavior of each of such groups may lead to better understanding of price variations since diverse groups trading may correlate differently with the market. the following examples, although not directly involving herding, but closely related to it, highlight the need to better understand the differences between individuals and institutions trading behavior and their influences on the market. ofek and richardson, 2003, suggest that increased activity of individuals could have contributed to the bubble of the 1990s. on the other hand, griffin, harris and topaloglu, 2003a and 2003b argue, using daily data from nasdaq, that institutions were the main driving force during the crashes of the 1990s. in recent papers kaniel, saar, and titman, 2007, and kaniel, liu, saar, and titman, 2007, show that trading buy imbalances by individuals provide market returns predictive ability. while investors can be classified into groups in various ways for analyzing their behavior, such classification and analysis will be useful only if the groups are sufficiently large for their actions to potentially affect key market features. groups that are too large are counterproductive for the study of herds. if the groups cover the whole market, then the sell of one group would be the buy of the other, hence the behavior of the groups would mirror each other. groups should be somewhat homogeneous and the members of the groups should have some possibility of guessing or observing the actions of other members. whereas we are considering the behavior of amateurs vs. professionals, the herding behavior of some other groups have also been analyzed: offshore investors and onshore investors, (kim and wei, 2001), analysts and newsletters (see, e.g., graham, 1999, welch, 2000, jegadeesh and kim, 2007), futures traders (pritsker and kodres,1995) and banks: (hirofumi and ryuichi, 2007) classifying investors to two general types, amateurs and professionals can help in distinguishing between two large groups whose herding have significant effects on markets, and could also serve as a tool for examining the effect of increasing market savvy on behavior. whereas there exists a considerable empirical research that analyzes herding behavior of institutions, with varying degrees of agreement as we show below, research on how herding differs between professionals and amateurs is scant. exception is nofsinger and sias, 1999. this is partly due to the frequent use of actual market data in the research on herding behavior by institutions, while experimental methods have usually been used in research on herding by individuals. see for example: cipriani and guarino, 2005, drehmann, oechssler, and roider, 2005, gilo-shalom, levy and venezia, 2007, to name just a few. in a recent paper alevy, haigh and list, 2007, experimentally compare the herding behavior of students with that of cbot traders. lakonishok, shleifer, and vishny, 1992, study the trading behavior of tax exempt mutual funds and they only find little evidence of herding in their sample. grinblatt, titman, and wermers, 1995, however, find weak evidence of herding in the mutual funds they analyzed. they find that when herding exists its impact is important; stocks purchased by funds that herd outperform stocks they sell by four percent during the following six months and this return difference is especially pronounced for small stocks. at the same time, they do not have any evidence of herding behavior by individual investors. wermers, 1999, using quarterly holding data finds that mutual funds herd when trading small stocks and mainly in trading by growth-oriented funds, but he observes little herding by mutual funds on average. nofsinger and sias, 1999, examine herding by both institutional and individual investors. they report evidence on herding behavior in institutions, using quarterly holdings data. using a smaller data-set than the one they used for their analysis of institutions they also find herding by individuals and that herding by institutions impacts prices more than herding by individual investors. in addition, they note that institutional herding into a stock results in better performance relative to the market than does herding by individuals. moreover, they demonstrate that institutional herding shows a strong contemporaneous relationship with daily stock returns. in a later study, sias, 2004, finds positive auto-correlation between institutional investors demand for a security from one quarter to another and argues that most of this apparent momentum trading by institutional investors can be explained by herding rather than by the claim that institutions follow a momentum strategy based on past winners and losers. griffin, harris, and topaloglu, 2003, document strong evidence of feedback trading (a behavior closely related to herding) by institutional investors at the daily level. they find that on average, the top performing securities based on the previous days return is more likely to be bought in by institutions (and sold by individuals) than poorly performing securities. however, they do not explore herding behavior.further insights on herding were obtained from studies on non-us markets. walter and weber, 2006, investigate herding behavior in german mutual funds and find some evidence that german mutual funds managers tend to herd and exhibit positive feedback trading patterns. they also find that a large proportion of apparent herding behavior can be attributed to changes in the benchmark index composition. wylie, 2005, uses the portfolio holdings of equity mutual funds in the u.k. using quarterly data to test for herding. he finds a modest amount of herding in the largest as well as in the smallest individual u. k. stocks but little herding in average size stocks. wylie also finds that mutual fund managers tend to herd out of large stocks after high excess returns. kim and nofsinger, 2005, study institutional herding in japan and find evidence of a lower herding than in the us, but a higher price impact of herding on japanese stocks. while herding by institutional investors has been extensively studied, the question of this behavior among individuals and its effect on prices has received less attention (see, however feng and seasholes, 2002). other features of behavior of uninformed retail investors were extensively explored; for example, barber and odean, 2000, study the performance of individual investors who hold common stocks and show they under-perform the market on average. however, they do not explore whether investors herd or not. a similar conclusion is also reached by shapira and venezia, 2001, using an israeli data set. they too do not study herding nor do they document the kind of trading strategies that are used by individual investors. odean, 1998, shapira and venezia, 2001, and dhar and zhu, 2006, study the disposition effect and other phenomena in individual investors trading but did not analyze herding behavior. in the present study we examine herding behavior of individuals and professionally managed investors in israel by observing their daily transactions over a four year period. we analyze the factors that contribute to this behavior; factors specific to the herded stocks as well as market wide factors. our study adds to the literature in various ways. first, we provide new evidence on herding and its correlation with stock characteristics such as firm size, its systematic and its idiosyncratic risks. second, we study herding in a market which was not studied before. in addition to the differences in market settings, size, and investment cultures between markets, which potentially make the study of any new market constructive, there exists an important feature in the market we study distinguishing it from the us that makes its analysis important. the compensation of the professional investors in our study, unlike professional investors in the us, does not include a percentage of the profits they make; instead, their fees are based only on the volume of assets under management, leading to flat incentives with respect to performance. there are long run effects of performance on the future demand for the professional investors services in our sample, and hence their performance is important for them but presumably to a lesser extent than in the us. also when future demand is the main motivation for performance, the benchmarking effect that may lead to herding is weaker than in the us where benchmarking is an official standard for compensation. the incentive of the professionals in our sample to beat any specific index is weaker, and thereby their motivation to herd is lower than in the us. if herding behavior is nevertheless found among the professional investors in our sample, it suggests that there are other behavioral or informational explanations for their conduct. third, our study is based on more detailed data than previous studies. nofsinger and sias, 1999, for example, base their conclusions on quarterly reports and therefore their herding measures depend only on holdings in the first and last days of the quarter. our study on the other hand, employs daily data from which monthly herding measures were formed.the paper is structured as follows. in section ii we describe the data. section iii discusses the methodology used for constructing the herding measures and for designing the tests. in section iv we explore the herding tendencies of amateurs and professionals and study stock specific characteristics (size, idiosyncratic risk, and systematic risk) that affect herding into them. in section v we examine market wide factors that lead to herding, and section vi concludes. ii. data the data consist of records of all investment transactions of 2428 managed and 7429 independent clients of one of the largest banks in israel (banks in israel also act as brokerage houses) during the period january 1, 1994 through december 31, 1997. we count as clients in any given year only those who transacted at least once during that year. the number of amateurs is higher than that of the professionals. however, since the professionals traded almost 5 times more frequently than the amateurs, there were no significant differences between the groups in terms of total volume and total number of transactions. independent clients manage their own portfolios, but process their transactions through the bank. managed clients solicit the assistance of professional portfolio and money managers (pmms) who also act as brokers. most of these pmms are not members of the tel aviv stock exchange (tase), so they execute their transactions through an exchange member, (usually a large bank or another financial institution). when a client chooses to have her portfolio managed by a pmm, she opens an account at the bank and authorizes the pmm to manage it. our database consists of all the transactions of clients, both independent and managed that had accounts in 1994. iii. methodologyherding occurs when investors imitate the behavior of other investors and in doing so partially disregard their own information and beliefs. however, not all herding can be described as non-rational. in the literature we find three types of herding that are classified as rational: first, information based herding; investors observing other investors who have invested in a stock may assume the latter have done so in a bayesian updating manner, and therefore, the former may conclude that there is no point in obtaining further signals about the stock because it is unlikely to affect the investors priors sufficiently to change their mind. this argument is in line with work carried by researchers who study cascades (cf., banerjee, 1992, bickchandani, hirshleifer and welch, 1992, and welch, 1992, and a later paper by avery and zemsky, 1998, that critiques their models). second, reputation based herding; scharfstein and stein, 1990, and froot, scharfstein, and stein, 1992, developed a model where there are two investment managers and an employer, and none of the three is certain of the two managers ability. the manager not wanting to take a risk that her decision will reveal her to be of a lower quality, finds it useful to imitate the other manager, and third: compensation based herding; since compensation of investment managers is often linked to some market benchmark it pays for them to imitate the actions taken by other investors (cf., maug and naik, 1996). it is usually impossible to directly test which of these types of herding exists and whether a herd-like behavior of investors is true herding, or it may just seem so since all investors received similar signals and hence behaved alike. for practical purposes however, one needs to construct some proxies for herding behavior. in what follows, we employ techniques and herding measures similar to those used by lakonishok, shleifer, and vishny, 1992, (lsv), and grinblatt, titman and wermers, 1995, (gtw). the main variable used in this analysis is the proportion of buy transactions out of all trades (buy and sell) of some stock during a given period of time relative to the long run proportion of buy transactions. since the long run proportion of buy transactions of any stock is 50%, the above definition only applies when restricted to a particular class of investors, and for a limited period of time. herding is considered to be the case where the proportion of buys significantly differs from its long run average. this measure does not take into account the volumes of trades. for example buyers and sellers could be of same numbers but each of the buyers demands a large amount and each of the sellers a small amount. in such a case herding actually occurs but the measure will not pick it up. wermers, 1999, developed the portfolio change measure, pcm, to correct for that. this measure however provides larger investors greater weight, and the measure itself has other statistical deficiencies.we examine two types of herding: herding into specific stocks, which we call micro herding, and concentrating on either buying or selling stocks in general which we call macro herding. the micro herding measures we construct assess to what extent there is a concentration of buy trades or sell trades on a specific stock. we use this herding measure to analyze the characteristics of the stocks that lead to herding. the macro herding measure evaluates how much are investors buy decisions concentrate