استمرار عملکرد نسبی در توصیه های بورسی تحلیل گران مالی طرف فروشنده ها

Analysts with above-median risk-adjusted performance in the estimation period persistently outperform those with below-median performance in the subsequent holdout period. The annualized risk-adjusted returns of trading strategies based on performance persistence are statistically and economically significant, with a magnitude around 10% even after adjusting for transaction costs and trading delays. This stems mostly from past above-median performers and is not simply a decomposition of previously documented post-event return drift. The results support the hypotheses that more information is contained in above-median performers’ recommendations and that investor reaction to these recommendations is incomplete during the event periods.

In this paper I examine the performance persistence of the so-called “sell-side” financial analysts who work for a brokerage house and provide research for the house and its clients. These analysts are prominent in the investment process; indeed, the financial press has called the decade of the 1990s the “Age of the Analysts.” While analysts have traditionally influenced institutional investors, they have increasingly influenced individual investors as well. I measure analyst performance using risk-adjusted returns on portfolios created with analyst recommendations. Performance persistence, or persistence in the relative performance differences among individual analysts, is measured by the correlation between the differences in individual analyst performance in the estimation period to the one-period-ahead holdout sample. I find significant persistence in relative performance differences among individual analysts. Specifically, analysts whose recommended portfolios generate stronger abnormal returns in the estimation period continue to produce recommended portfolios with stronger abnormal returns in the subsequent holdout periods. The annualized risk-adjusted returns of trading strategies based on performance persistence are statistically and economically significant, with a magnitude around 10% even after adjusting for transaction costs and trading delays. When I separately investigate analysts with above-median and below-median performance, denoted as past winners and losers, respectively, I find performance persistence in buy recommendations for past winners; analyst performance for sell recommendations also exhibits persistence, but much less consistently. Given little post-event return drift in the overall sample, the performance persistence in buy recommendations is not simply a decomposition of the previously documented post-event return drift (Elton et al., 1986; Stickel, 1995; Womack, 1996). Given that past winners’ recommendations outperform those of past losers, both within the event period and in the post-event period, the results support the hypotheses that there is more information contained in past winners’ recommendations and that investor reaction to these recommendations is incomplete during the event periods, with a substantial portion of the incomplete reactions realized around subsequent earnings announcements, stock recommendation dates, and earnings forecast dates. My paper is related to the literature that examines differential analyst performance and subsequently explains this difference. Richards (1976), Brown and Rozeff (1980), O’Brien (1987), Coggin and Hunter (1989), O’Brien (1990), Butler and Lang (1991), Sinha et al. (1997), and Cooper et al. (2001) investigate differences in the earnings forecast accuracy across analysts; Sinha et al. (1997) is the closest in spirit to my paper. They find that superior earnings forecasters in one period tend to be superior in subsequent holdout periods, which suggests persistence in earnings forecast accuracy. Prior literature also attempts to explain within-sample differences in both the forecast accuracy (Mikhail et al., 1997; Clement, 1999; Jacob et al., 1999) and the abnormal returns associated with stock recommendations and earnings forecast revisions (Stickel, 1995; Gleason and Lee, 2003). My paper is also related to the ongoing debate concerning the efficiency with which the market incorporates the information in research reports (see, e.g., Kothari, 2001; Lee, 2001). For earnings forecast revisions, early research by Griffin (1976), Givoly and Lakonishok, 1979 and Givoly and Lakonishok, 1980, and Imhoff and Lobo (1984) documents abnormal event-period returns. Subsequent research demonstrates an incomplete return reaction during the event period (Givoly and Lakonishok, 1980; Stickel, 1991). For stock recommendations, early studies focus on the aggregate performance of a recommendation or recommendation revision category. For example, Elton et al. (1986), Stickel (1995), and Womack (1996) show that positive (negative) abnormal returns accompany favorable (unfavorable) changes in recommendations, with a post-event return drift of up to six months.1 Similarly, Barber et al. (2001) find that abnormal returns increase with a stock's average recommendation. More recent studies investigate cross-sectional performance differences in stock recommendations. For example, Barber et al. (2000) find no systematic differences in stock recommendations made by brokerage houses of different sizes. My most important contribution is to provide clear evidence of a violation of the semi-strong form of the efficient markets hypothesis. The statistically significant abnormal returns found in extant research are eliminated by transaction costs and trading delays and do not constitute violations. In contrast, I document abnormal returns that are many times greater than those reported in prior studies of analysts and fund managers. Even after adjusting for transaction costs and trading delays, these returns are statistically significant and economically meaningful. Compared to prior research on earnings forecasts, I examine stock recommendations that provide independent information about share value (see, e.g., Francis and Soffer, 1997). Compared to prior research on recommendations, I employ analyst-level analysis. First, the value of a sell-side research department is its analysts, and thus focusing on performance differences at the analyst level is likely to yield a more profitable trading strategy. Second, while it is impossible even for institutional investors to hold portfolios of the stocks recommended by all analysts or by analysts from a single brokerage house, any investor can trade on the recommended portfolios of individual analysts. Since many buy-side fund managers rely on research from sell-side analysts (Cheng et al., 2005), it is interesting to compare the performance persistence of sell-side analysts with that of buy-side fund managers. The evidence on fund managers is mixed. For example, Brown and Goetzmann (1995) and Malkiel (1995) find evidence of performance persistence in mutual funds, but Carhart (1997) shows that controlling for momentum eliminates such persistence. A recent study by Bollen and Busse (2005) finds performance persistence for mutual funds even after adjusting for momentum, but the persistence is only for quarterly or shorter intervals and the average quarterly abnormal return for the top 10% of fund managers is about 1.8%. While analysts and fund managers function in substantially different working environments—for example, funds incur transaction costs, are required to hold some short-term securities if they are open-ended, are subject to stricter regulations, and generally cannot short securities—the magnitude of differences in performance persistence between fund managers and analysts is unlikely to be explained by most variation in working environments. In contemporaneous work, Mikhail et al. (2004) (MMW, hereafter) provide evidence of analyst performance persistence in stock recommendations. They find that the market does recognize the performance persistence in the five days surrounding recommendation revisions. They also find that the market reaction is incomplete. The profitability of their suggested trading strategies, however, disappears after accounting for transaction costs and trading delays. To understand the greater persistence and trading profit in my results, I modify my sample selection and test design methodologies separately for each of the following major ways in which our two studies differ: (1) MMW combine all recommendations except for hold, whereas I examine buy and sell recommendations separately, and I focus on the more informative strong buy recommendations and exclude the less informative buy recommendations; (2) MMW use three-factor matching to measure analyst performance, whereas I use a much wider range of risk adjustments; (3) MMW's database does not include brokerage houses such Goldman Sachs, Merrill Lynch, and Donaldson, Lufkin & Jenrette, whereas the recommendations from these large houses represent close to 10% of my sample; (4) MMW use fixed holding periods, whereas I base holding periods on recommendation revision dates, which could add precision to performance measurement; (5) MMW examine performance persistence at longer horizons, whereas I focus on shorter horizons; and, (6) MMW use the direction of revisions and an event-time approach, whereas I use recommendation levels and a calendar-time approach that directly reflects the experience of potential investors. Appreciating the impact of these methodological differences is very important to our understanding of market efficiency. Moreover, each of these differences implies a significant change in the implementation of trading strategies in practice. The effects of the first two methodological differences are the strongest, and are in opposite directions. After combining strong buy recommendations with either sell recommendations or less informative buy recommendations, performance persistence and trading profit are both considerably reduced although still statistically significant at conventional levels; the reduction when including the less informative buy recommendations is particularly great. Also, performance persistence and trading profit when using only MMW's three-factor matching to measure analyst performance are much greater than those when using my main risk adjustment, a six-factor model combined with the three-factor matching. The third and fourth aspects above, namely, the exclusion of recommendations from the three large brokerage houses and the use of fixed holding period, strengthen the performance persistence in my sample. The last two differences do not seem to have material impact. An interesting finding is that performance persistence appears to be the strongest at semiannual intervals. Complementing MMW, I examine the potential explanations for performance persistence in detail. MMW conjecture that the performance persistence they document may be a decomposition of the post-event return drift found in the prior literature. I show that performance persistence is a phenomenon of market inefficiency, independent of the existence of post-event return drift. Also, I find that performance persistence is mainly due to past winners. Further, I provide substantial evidence that persistence is likely to be due to market inefficiency, rather than inadequate risk adjustments or particular testing methodologies. For example, I use a range of risk adjustments to measure analyst performance and many testing methodologies to gauge performance persistence. I find that performance persistence cannot be explained by industry momentum. A sizeable portion of the incomplete reactions that follows the stock recommendations of the best analysts are realized around subsequent information releases, further supporting a market inefficiency explanation. The rest of my paper is organized as follows. Section 2 describes the data and performance measures. Section 3 presents the empirical results and Section 4 concludes.

I examine the persistence in the relative performance of stock recommendations made by individual analysts and find significant evidence of a violation of the semi-strong form of the efficient markets hypothesis. Analysts with above-median risk-adjusted performance in the estimation period (past winners) consistently outperform those with below-median performance (past losers) in the subsequent holdout periods, even with alternative adjustments for cross-sectional correlation, risk, testing methodologies, portfolio weighting schemes, sample period selection, return measurement intervals, and various analyst and firm characteristics. The annualized risk-adjusted returns of trading strategies based on performance persistence are statistically and economically significant, with a magnitude around 10% even after adjusting for transaction costs and trading delays. This magnitude is many times greater than what is typically found in the existing literature on analysts and fund managers. The overall persistence stems mostly from past winners and is not a decomposition of post-event return drift as documented by the prior literature. A significant portion of the abnormal returns following analyst recommendations are realized around subsequent announcements of earnings, stock recommendations, and earnings forecasts, especially for the 10th decile analysts, further supporting a market inefficiency explanation. In addition, these abnormal returns that are much greater than those for fund managers do not seem to be due to differences in working environments or data availability between analysts and mutual fund managers, which point to an interesting research question given many fund managers use analyst research. Analyst buy recommendations exhibit much more consistent performance persistence than sell recommendations. In summary, the evidence is consistent with the hypotheses that (1) past winners’ recommendations are more informative, and (2) investors underreact to past winners’ recommendations and overreact to past losers’ recommendations. My results also point toward a trading strategy that can potentially be exploited by investors. Because an analyst covers only a limited number of stocks and does not frequently revise recommendations, even individual investors can realistically mimic a stock portfolio recommended by a top analyst. Raw return persistence offers a strategy for investors such as hedge fund managers to follow.