Herding Behavior among Financial Analysts: a literature review. Geert Van Campenhout, Jan-Francies Verhestraeten

Transcription

1 Herding Behavior among Financial Analysts: a literature review Geert Van Campenhout, Jan-Francies Verhestraeten HUB RESEARCH PAPER 2010/39 NOVEMBER 2010

2 HERDING BEHAVIOR AMONG FINANCIAL ANALYSTS: A LITERATURE REVIEW Geert Van Campenhout Assistant Professor of Finance, HUBrussel Voluntary Scientific Researcher, KULeuven Jan-Francies Verhestraeten BOF Doctoral Researcher, HUBrussel Voluntary Scientific Researcher, KULeuven geert.vancampenhout@hubrussel.be ABSTRACT Analysts forecasts are often used as an information source by other investors, and therefore deviations from optimal forecasts are troublesome. Herding, which refers to imitation behavior as a consequence of individual considerations, can lead to such suboptimal forecasts and is therefore widely studied. In this paper we provide a concise literature review of herding behavior among financial analysts. We discuss the concept of herding and review its occurrence & consequences, as well as its motives and determinants. We conclude with some suggestions for further research. Keywords: Financial analysts; herding behavior; earnings forecasts INTRODUCTION Financial analysts are generally considered to be important intermediaries in financial markets. Their earnings forecasts, stock recommendations and research reports are widely used by professional parties and individual investors as inputs for investment strategies and stock valuation models. Analysts might have an edge over other investors if they are better at interpreting public information (e.g. thanks to experience or specialization) and/or if they are better at collecting and analyzing private information

3 (e.g. thanks to better access to (corporate) information). By rationally weighting all their information signals, they should formulate optimal forecasts of the company s future (O Brien 1988, Schipper 1991). Financial research (see e.g. Givoly & Lakonishok 1984, Easterwood & Nutt 1999, Hirshleifer & Teoh 2003, for literature review) has, however, revealed that analysts often deviate from rational decision making. As a result, poor forecasts are submitted with large forecast errors, which in turn mislead investors and impede the extraction of relevant information about the company s future earnings. For instance, Dreman & Berry (1995) and, more recently, Ciccone (2005), show that forecasters are generally optimistic. Ciccone (2005) observes that in the late nineties at least 40 % of all forecasts were too optimistic, which led to a forecast error of about 20% on average. It has been shown that analysts may manipulate their predictions in different ways, either consciously or unconsciously strategic or cognitive bias or even both. This paper focuses on one particular bias, being herding behavior, and discusses its prevalence and measurement, as well as its causes and implications. THE CONCEPT AND CONSEQUENCES OF HERDING Herding is described as imitation behavior resulting from individual factors and often leading to inefficient outcomes for the market as a whole (Bikhchandani et al. 1992). The phenomenon of herding was first studied in psychology. For instance, Asch (1952) studied the impact of an individual s social environment on his decision behavior and observed that within groups individuals often abandon their own private signal to rely predominantly on group opinion. Seminal articles by Shiller (1987), Scharfstein & Stein (1990), Banerjee (1992) and Bikhchandani et al. (1992), among others, introduced herding models into the finance literature and highlighted its possible consequences for the overall functioning of financial markets and information processing by individuals. 1 Concerning the former, copy-cat 1 Herding behavior is found in the decision-making of several market participants (Hirshleifer & Teoh 2003). In fact, at first financial literature was fascinated by herding behavior among investors and its implications for stock prices (e.g. Shiller 1987, Bikhchandani et al. 1992). Moreover, a large part of the literature on herding among financial

4 behavior of investors might feed speculative investment bubbles, which would lead to substantial welfare losses when these bubbles burst. With respect to information processing, herding leads to the suboptimal use of valuable information leading to inefficient decision making. For instance, Trueman (1994) investigates these insights into the decision processes of financial analysts and observes herding behavior when they submit their earnings forecasts. Furthermore, the extent of herding is higher among inexperienced analysts. Olsen (1996) documents that analysts herded in 52 to 72 % of their submitted forecasts between Moreover, this form of behavior increases with the unpredictability of earnings. De Bondt & Forbes (1999) confirm these results by investigating herding behavior based on more recent UK data (period ). They find that 50 to 60 % of herding forecasts are observed within 12 to 9 months of an earnings announcement. Herding, however, decreases as we approach the actual announcement date. Among the same lines, Guedj & Bouchaud (2008) consider forecast error to be much larger than the extent of forecast dispersion and see this as strong evidence of herding behavior among analyst forecasts from 1987 until Recently, there has been some debate on the importance of herding. For instance, Chen & Jiang (2006) and Bernhardt et al. (2006) argue that overconfidence (i.e. overweighting of private information) - resulting in estimates that are biased away from the consensus - is more important than herding. This issue remains an open question. To gain a better understanding of analysts herding behavior, we have to analyze their information dissemination in detail. In particular, analysts constructing their earnings forecasts essentially rely on two kinds of information, namely a public and a private signal (Ramnath et al. 2008). The private signal consists of private information about the company, thanks to good access to management, sophisticated models, distinctive interpretation, etc. The public signal comprises all publicly available analysts was based on observations among investors in general. Similarly, extensive literature exists on herding among other financial intermediaries, such as portfolio managers, and their incentives (e.g. Maug & Naik 1996, Chevalier & Ellison 1997). Even within firms managers are often prone to herding in their investment, financing and reporting decisions (Hirshleifer & Teoh 2003).

5 information, namely past earnings, industry and macro-economic information, and also forecasts of other analysts as reflected in the prevalent consensus forecast. It is exactly this last source of information that often induces herding behavior. Analysts might be prone to deviate from their true belief and overvalue consensus forecast information in their earnings prediction to ensure that their forecast is closer to the consensus forecast. 2 In doing so, they reduce their own private signal which in turn reduces the information value of their individual forecast and the resulting consensus forecast. According to Olsen (1996) this has two main implications. Firstly, the dispersion in opinions typically decreases. Secondly, he argues that analysts are inclined to be rather optimistic about a company s earnings, with the result that they will prefer to conduct herding when their private signal is pessimistic. Hence, these effects will result in a more centered forecast distribution with a higher mean earnings forecast. 3 By means of an illustration, we provide a stylized example of the forecasting process of 10 analysts issuing earnings estimates with and without herding for a fictive company. The details are provided in Table 1. When analysts herd, they deviate from their unbiased estimate by overweighting. Unbiased estimates are obtained by Bayesian weighting of public and private information. We assume a weighting scheme of here for unbiased estimates, and in the event of herding. Since analysts tend to herd their pessimistic forecasts, we assume that they will only deviate from their unbiased estimate if their private signal is lower than the public signal. Consequently, analysts A,B, F and I will revise their initial 2 Given rational expectations, it is possible to determine the optimal weights of public and private information that ensure that the expected value of earnings forecasts will be equal to the actual earnings. Deviations from these weights are labeled under- or overweighting depending on the direction of deviation. 3 Obviously, other inefficiencies such as overconfidence, for instance may influence the forecast distribution as well. However, with a view to brevity, this paper focuses on analysts herding behavior and ignores these inefficiencies.

6 forecast upwards (by overweighting their public information (weight of 90% instead of 10%)). As postulated, this herding behavior results in a higher consensus forecast and lower dispersion. Note that the forecast error has also increased. As a result, inefficient behavior occurs as the individual s conduct is not consistent with group interest. Panel A: Assumptions about the weighting of public and private information by analysts Public information Private information No herding Herding Panel B: Consensus Forecast, forecast error and dispersion with and without herding (i) Analyst Public signal Private signal Forecast without herding Forecast with herding A B C D E F G H I J Signal Average Consensus ( ) Forecast Error ( ) Dispersion ( ) i) Actual earnings of this quarter and previous quarter are assumed to be equal to 10. Table 1: Stylized example of impact of herding on forecast distribution. This table considers the forecasting process of 10 analysts issuing earnings estimates for a fictive company in a certain quarter. For convenience the actual quarterly earnings, ex-post, are assumed to be equal to 10. Unbiased estimates are given by Bayesian weighting of public and private information (assumed to be fifty-fifty here).herding results in biased estimates by overweighting public information (we assume that a weighting of for the public and private signal is used ) The effect of herding also becomes apparent if we relate it to the wisdom-of-crowds effect, which states that the best attainable forecast can be obtained by taking the average opinion of a group of independent individuals rather than the opinion of an individual expert (Surowiecki 2004). This principle

7 is based on the observation that the aggregation of independent opinions results in a reduction in idiosyncratic variance. 4 Clearly, herding violates the wisdom-of-crowds -principle. HERDING MEASURES With respect to the wisdom-of-crowds effect, a test statistic based on deviations of the actual forecast distribution from the theoretical benchmark distribution of independent forecasts could be an interesting measure to quantify the degree of herding, but is unfortunately difficult to develop in practice. More generally, the measurement of herding is an elusive topic, and various proxies have been developed in the literature. An exhaustive review of these measures goes beyond the scope of this paper. Hence, Table 2 only gives a brief overview of the most important approaches. Other approaches exist, but are mostly a combination of two or more of the measures in Table 2. Concept References Herding as relative tightness of forecast distribution E.g. Olsen (1996) Herding based on dispersion & forecast errors (cross-sectional) E.g. De Bondt & Forbes (1999), Kim & Pantzalis (2003) Herding in terms of ex-post probability of deviation from consensus & actual earnings Herding based on comparison of forecast with deviation from consensus & previous forecast (bold vs. herding forecast) E.g. Bernhardt et al. (2006) E.g. Clement & Tse (2005) Herding as underweighting of private information E.g. Zitzewitz (2001), Chen & Jiang (2006) Table 2: Overview of different categories of herding measures in the literature This table gives an outline of the approaches used in the literature for measuring herding behavior. The alternatives are grouped with respect to the concepts used to construct the proxy. As we notice from Table 2, the various concepts all apply some kind of deviation measure to quantify herding, but they differ in the manner in which this is applied. The first three approaches yield cross- 4 Besides the fact that the opinions of the individuals need to be formed independently, diversity and decentralization of the group of individuals and the method used to summarize opinions are also of importance for the wisdom-of-crowds to hold (Surowiecki 2004).

8 sectional measures, whereas the last two tend to be analyst-specific. Moreover, the first and the third approach produce probabilities of herding, while the other measures do not. More specifically, the second approach relates the company forecasts dispersion to the sector dispersion. Approach four relies on a binary variable, which indicates whether herding is present or not, and approach five uses the regression coefficient of the forecast error on the deviation from the consensus in order to measure herding. Unfortunately, no universal herding measure exists and the degree of herding that is documented is to some extent dependent on the herding proxy that is applied (see also supra on overconfidence). Hence, a generally accepted herding measure is needed to arrive at more definite conclusions on the degree of herding that is present in financial analysts earnings forecasts. MOTIVES & DETERMINANTS OF HERDING What drives herding behavior? In the literature, three main herding motives are put forward, namely information-based herding, reputation-based herding and compensation-based herding (Bikhchandani & Sharma 2000). 5 The first explanation is supported by the theory on information cascades, introduced by Banerjee (1992) and Bikhchandani et al. (1992). Information-based herding occurs when analysts lack confidence about their private information and there exists (a lot of) uncertainty about the quality of public information. As a consequence, analysts abandon their private signal (which is needed to optimally update the available information), and follow the herd that maintains an inefficient consensus. As such, the analysts actions are uninformative to later observers and a cascade arises that blocks the inflow of new information (Hirshleifer & Teoh 2003). 5 Explaining herding behavior based on these three motives is suggested by Bikhchandani & Sharma (2000). For an overview of the different theories developed to explain herding, we refer to Welch (2000) and Hirshleifer & Teoh (2003).

9 Second, analysts are found to conduct reputation-based herding. Scharfstein & Stein (1990), Trueman (1994) and Prendergast & Stole (1996) argue that analysts manipulate their forecasts to get closer to the consensus in order to signal that their information is correlated with their peers. The principal idea here is that to fail as a group will not harm one s reputation as much as being mistaken on one s own. Collective failure may be attributable to uncertainty in the environment (and not due to a lack of skill). Hence, analysts herd to avoid their own forecasts becoming too distinct from the collective (consensus) forecast. Cote & Sanders (1997) and Graham (1999) find empirical evidence for this relationship. Finally, compensation-based herding builds on the work of Maug & Naik (1996) and Chevalier & Ellison (1997) in order to illustrate that herding can also arise as a consequence of payoff externalities. As with reputation-based herding, analysts will also herd to avoid deviant bad forecasts, but in this instance it is the penalization that analysts face (e.g. job loss) when making such a forecast that triggers the herding behavior. Since the pay-off of dissident forecasts is asymmetric (i.e. a larger negative pay-off in case of a deviant negative forecast compared to the benefits of a bold positive forecast), analysts will play safe and herd. Results in Hong et al. (2000) corroborate this hypothesis. They document that inexperienced analysts herd more than experienced analysts, because being wrong for an inexperienced analyst, who is at the start of his career, is more costly than for an older analyst, who has built a reputation throughout years of being in the business. In the literature, these three basic motives are also often explained in terms of behavioral considerations and strategic motives. In the first instance, herding arises because of a psychological need to comply with the consensus, such as a lack of self-confidence (e.g. Bikhchandani et al. 1992). In the second instance, personal incentives like reputation and payoff considerations lead to herding (e.g. Scharfstein & Stein 1990, Trueman 1994, Friesen & Weller 2006).

10 What determines (the propensity of) herding? Extensive research has been conducted to discern determinants of herding behavior. Analyst experience, analyst ability, complexity of forecasting task, brokerage house size and forecast horizon are characteristics that are most frequently related to herding. Firstly, research shows evidence of a negative relationship between the degree of herding and analyst experience (e.g. Hong et al. 2000; Clement and Tse 2005; Krishnan et al. 2006). These findings are commensurable with reputation- and compensation-based herding: younger analysts still have to earn their reputation and have a higher risk of being dismissed. Secondly, the analyst s ability, usually measured by his past performance, has a negative impact on the propensity to herd (Stickel 1990, Cote & Sanders 1997, Graham 1999). Thirdly, the complexity of the forecasting job is found to be positively related to herding, because complexity is representative of uncertainty about the company s performance (Olsen 1996). This is consistent with the first motive of imperfect information. Fourthly, and in line with the first characteristic, the size of the brokerage house is negatively related to herding, because it is assumed that larger brokerage houses are more experienced and have better access to information (Clement & Tse 2005). Finally, the forecast horizon has been shown to have a positive impact on herding (Krishnan et al. 2005). As the forecast horizon decreases, more information becomes available, thereby reducing information uncertainty and herding behavior. CONCLUSION Financial analysts forecasts are widely used by other investors as inputs for investment strategies and other applications. Hence, deviations from optimal forecasts are not only problematic from an academic viewpoint (inefficient information dissemination in financial markets), but also have important practical consequences for investors that rely on their information. Herding (i.e. excessive agreement among analysts due to suboptimal use of private and public information), leads to much less informative forecasts and is therefore an important research subject. In this paper, we reviewed the existing

11 literature by discussing its occurrence, importance, and implications. Subsequently, a brief overview was given both of analysts incentives that induce herding behavior, and of analyst characteristics that enforce its occurrence. In general, evidence of herding was documented, but difficulties in measuring herding made it difficult to give a decisive answer about the degree of herding. Moreover, research widely focused on individual analyst characteristics to explain the herding. Besides a few studies (e.g. Kim & Pantzalis (2003) on the complexity of the company), the role of stock characteristics is largely ignored. Hence, a profound analysis of this relationship, together with the construction of a widely accepted herding measure, appear to be promising topics for further research. REFERENCES Asch, S., Social Psychology. Englewood Cliffs, NJ: Prentice Hall. Banerjee, A., A simple model of herd behavior. The Quarterly Journal of Economics, 57(3), pp Bernhardt, D., Campello, M. & Kutsoati, E., Who herds? Journal of Financial Economics, 80(3), pp Bikhchandani, S., Hirshleifer, D. & Welch, I., A theory of fads, fashion, custom and cultural change as informational cascades. Journal of Political Economy, 100, pp Bikhchandani, S. & Sharma, S., Herd Behavior in Financial Markets: A Review. IMF Staff Papers, 47(3), pp Chen, Q. & Jiang, W., Analysts Weighting of Private and Public Information. Review of Financial Studies, 19(1), pp

12 Chevalier, J. & Ellison, G., Career Concerns of Mutual Fund Managers. Quarterly Journal of Economics, 114(2), pp Ciccone, S., Trends in analyst earnings forecast properties. International Review of Financial Analysis, 14(1), pp Clement, M.B. & Tse, S.Y., Financial Analyst Characteristics and Herding Behavior in Forecasting. The Journal of Finance, 60(1), pp Cote, J. & Sanders, D., Herding behavior: Explanations and implications. Behavioral Research in Accounting, 9, pp De Bondt, W.F.M. & Forbes, W.P., Herding in analyst earnings forecasts: evidence from the United Kingdom. European Financial Management, 5(2), pp Dreman, D.N. & Berry, M.A., Analyst forecasting errors and their implications for security analysis. Financial Analysts Journal, 51(3), pp Easterwood, J.C. & Nutt, S.R., Inefficiency in analysts earnings forecasts: systematic misreaction or systematic optimism? The Journal of Finance, 54(5), pp Friesen, G. & Weller, P., Quantifying cognitive biases in analyst earnings forecasts. Journal of Financial Markets, 9(4), pp Givoly, D. & Lakonishok, J., The quality of analysts forecast of earnings. Financial Analysts Journal, 40, pp Graham, J.R., Herding among investment newsletters: theory and evidence. Journal of Finance, 54, pp

13 Guedj, O. & Bouchaud, J., Experts earning forecasts: Bias, herding and gossamer information. Working paper. Hirshleifer, D. & Hong Teoh, S., Herd Behaviour and Cascading in Capital Markets: a Review and Synthesis. European Financial Management, 9(1), pp Hong, H., Kubik, J.D. & Solomon, A., Security Analysts Career Concerns and Herding of Earnings Forecasts. The RAND Journal of Economics, 31(1), p.121. Kim, C. (Francis) & Pantzalis, C., Global/Industrial Diversification and Analyst Herding. Financial Analysts Journal, 59(2), pp Krishnan, M., Lim, S. & Zhou, P., Who Herds? Who Doesn t? Working paper Olsen, R., Implications of herding behavior for earnings estimation, risk assessment, and stock returns. Financial Analysts Journal, 52(4), pp O Brien, P.C., Analysts forecasts as earnings expectations. Journal of Accounting and Economics, 10(1), p Maug, E. & Naik, N., Herding and delegated portfolio management. Working paper (London Business School). Prendergast, C. & Stole, L., Impetuous youngsters and jaded old-timers: Acquiring a reputation for learning. Journal of Political Economy, 104(6), pp Ramnath, S., Rock, S. & Shane, P., The financial analyst forecasting literature: A taxonomy with suggestions for further research. International Journal of Forecasting, 24(1), pp

14 Scharfstein, D. & Stein, J., Herd behavior and investment. The American Economic Review, 80(3), pp Shiller, R., Investor behavior in the October 1987 stock market crash: survey evidence. NBER working paper, (October) Schipper, K., Analysts forecasts. Accounting Horizons, 5, pp Stickel, S.E., Predicting Individual Analyst Earnings Forecasts. Journal of Accounting Research, 28(2), pp Surowiecki, J., The wisdom of the crowds: Why the many are smarter than the few and how collective wisdom shapes business, economics and nations. US, Anchor Books. Trueman, B., Analyst forecasts and herding behavior. Review of Financial Studies, 7(1), pp Welch, I, Herding among security analysts. Journal of Financial Economics, 58(3), pp Zitzewitz, E., Measuring Herding and Exaggeration by Equity Analysts and Other Opinion Sellers. SSRN Electronic Journal, (1802).