The Effect of Star Analyst Tournaments on Firms' Information Environment *

Transcription

1 The Effect of Star Analyst Tournaments on Firms' Information Environment * Gil Aharoni University of Melbourne Joshua Shemesh University of Melbourne Fernando Zapatero University of Southern California First draft: April 22, 2014 This draft: August 26, 2016 ABSTRACT Star analysts typically cover 10 to 15 stocks. In half of the stocks they cover, they cross path with other stars (battleground stocks). We show that being the most accurate star in battleground stocks plays a key role in the annual rankings from Institutional Investor magazine. Star analysts respond to this tournament by producing more accurate earnings forecasts in battleground stocks than in other stocks they cover. We explore several alternative explanations and find that they cannot account for the higher accuracy in battleground stocks. Head-to-head competition between star analysts plays a key role in a firm s information environment. * We thank François Derrien, Eti Einhorn, Jordan Neyland, Jianfeng Shen, and conference participants in the 2013 Borsa Istanbul Finance & Economics Conference, the FIRN 2013 Annual Meeting, the 2013 Australian Banking & Finance Conference, the 2014 Auckland Finance Meeting, the 2016 Annual Congress of the European Accounting Association, and seminar participants from Texas A&M University, Claremont McKenna College, IE Business School, BI Norwegian Business School, California State University-Fullerton, Loyola Marymount University, Bar-Ilan University, Hebrew University, The Center for Academic Studies-Or Yehuda, Tel Aviv University, the University of Technology in Sydney, Australian National University, the University of Melbourne, and the University of Southern California for helpful comments. Any existing errors are our sole responsibility. Corresponding author: Marshall School of Business, University of Southern California, 3670 Trousdale Parkway, Bridge Hall 308, Los Angeles, CA , Ph: , Fax: , fzapatero@marshall.usc.edu. 1

2 1 Introduction This paper sheds new light on the relation between analyst coverage and the information environment. We show that the selection process of star analysts amounts to a tournament, and that this tournament incentivizes star analysts to issue more accurate earnings forecasts in the stocks in which they compete directly with each other. Recent literature documents a positive relation between analyst coverage and what can be generally referred to as the transparency of a firm s information environment. In particular, a decrease in analyst coverage is correlated with an increase in analyst optimism (Hong and Kacperczyk (2010)), an increase in earnings management (Yu (2008) and Lindsey and Mola (2014)), an increase in information asymmetry (Chang, Dasgupta, and Hilary (2006) and Kelly and Ljungqvist (2012)) and a decrease in monitoring firm activity (Chen, Harford, and Lin (2013)). Irvine (2003) shows that a firm s stock liquidity improves with the initiation of analyst coverage. Derrien and Kecskés (2013) document that a decrease in analyst coverage leads to a decrease in corporate investment and financing. They argue that a decrease in analyst coverage increases information asymmetry and thus increases the cost of capital. Our paper adds to this literature by examining the effect of direct competition between star analysts on another aspect of a firm s information environment,: the accuracy of its estimated future earnings. In this paper, star analysts is shorthand for analysts selected for the All-American Research Team compiled by Institutional Investor (I/I) magazine, based on a survey sent to hundreds of investment professionals. Arguably, this is the most important professional recognition for analysts: I/I star status has been shown to have a substantial effect on both analyst compensation and the ability of their brokerage houses to attract new clients (e.g., Groysberg, Healy, and Maber (2011); Clarke, Khorana, Patel, and Rau (2007)). In the first part of the paper, we show that a star analyst s chances of keeping her star status depend on 2

3 the tournament that results from the comparison of her earnings forecasts to that of other stars. More precisely, we document that the probability that a star will improve her ranking is positively correlated in a significant way with the number of stocks in which her earnings forecasts are more accurate than those of other star analysts covering the same stock. We call such a feat a win. Also, we refer to stocks that are covered by more than one star analyst as battleground stocks. Therefore, the star analyst whose earnings forecast is closer to the actual earnings than that of all the other star analysts covering the battleground stock, earns a win. The higher the total number of wins in a given year, the higher the probability the analyst will improve her ranking, and in turn get re-elected as a star in the subsequent year. We show that the critical variable is this number of wins, while the accuracy relative to non-star analysts is negligible. That is, star analysts are assessed on a relative basis rather than on their absolute performance, and their performance is evaluated mainly in comparison to other star analysts, rather than to the entire analyst community. Very importantly, we show that this ranking pattern is not driven by the centrality of the stocks the star analysts cover. The concern here is that participants in the I/I survey typically institutional investors are more interested in particular stocks and might reward analysts who cover a large number of these stocks. If so, battleground status would merely proxy for interest from investment professionals. Our results show that wins (as previously defined) are all important, independent from the characteristics of the stock in which they are achieved. These findings mitigate the concern that analysts are ranked based on firm characteristics rather than on comparisons between rival stars. In the second part of this paper, we show that the accuracy of star analysts earnings forecasts is substantially higher in battleground stocks than in other stocks they cover. Previous literature shows that star analysts are more accurate than non-stars (i.e., Stickel (1992)). In this spirit, the literature has mostly focused on comparing the mean accuracy 3

4 across different analysts. In this paper, we compare the performance of an analyst across the different stocks in her portfolio, and show that star analysts are about 20% more accurate in battleground stocks than in the other stocks they cover. This is after controlling for variables that have been shown to affect analysts accuracy, including firm, analyst, and forecast characteristics. Our findings are consistent with a higher level of effort and/or resources dedicated to the coverage of battleground stocks. The brokerage house that employs the star analyst also benefits from the star status, and thus the incentives of the star analyst and the brokerage house to preserve the star status are perfectly aligned. We explore several alternative explanations to the higher accuracy in battleground stocks. In particular, we rule out that higher accuracy is the result of stock selection by star analysts who pick stocks with better information environment, which are therefore easier to cover. We also provide evidence inconsistent with the notion that spillover effects in information result from the coverage of other star analysts. Arguably, as more star analysts cover a particular stock, the quality of the available information on that stock increases, which makes it easier to forecast earnings. We show that these alternative explanations cannot account for our results. The starting point of our analysis is the annual survey that I/I sends to thousands of investment professionals. Based on the responses, I/I selects the All-American Research Team whose members are known in the literature as star analysts. This includes analysts ranked from one through three in around 65 different industries and sectors. Importantly, there is no compensation for participating in the survey. Survey respondents are busy professionals and, arguably, they do not have a lot of time to spend on it. A simple rule of thumb, such as choosing the analyst with the best earnings forecast among the most recognizable analysts stars can expedite the process. The prominence of star analysts has made them the focus of many research papers. The literature shows that star analysts are more influential in reducing earnings management 4

5 (Yu (2008)), have a stronger effect on information asymmetry (Kelly and Ljungqvist (2012)), and have more influential recommendations (Loh and Stulz (2011) and Fang and Yasuda (2009)). Given the importance of star analysts in financial markets, the question of whether there are economic reasons behind their selection is of major relevance to finance. Previous literature has largely failed to find economic reasoning behind I/I rankings, thus reinforcing the view that rankings are largely a beauty contest. In this paper, we ask how head-to-head competition between star analysts affects their rankings. We define a ranking improvement as either an improvement in the rankings, or permanence in the top spot, from one year to the next. Also, we call battleground stocks those covered by two or more star analysts. Battleground stocks are larger and more profitable than single-star stocks. However, by including a large number of controls in our analysis, we demonstrate that these characteristics do not drive our results. We show that a higher ranking is crucial in maintaining star status and the benefits that come with it. We study the tournament that results from the comparison of earnings forecasts of star analysts who cover the same stock. We assign a win to the star analyst with the most accurate earnings forecast in a particular stock for a given year. We document that promotions from one year to the next, as we have just characterized them, are positively correlated with the number of wins. It is possible that the important factor in the I/I survey is the centrality of a stock rather than its battleground status. We use a battery of tests to verify that battleground stocks are not merely those that garner more interest from institutional investors. Furthermore, consistent with the conjecture that survey respondents use some type of rule of thumb to speed up the process of completing the survey, we show that a win is important regardless of the distance by which the star analyst beats the other stars covering the stock. Using a sports analogy, each battleground stock presents a game, and the analyst who has accumulated the most wins becomes the winner in the tournament. 5

6 Given the importance of beating other stars for the preservation of star status, the second part of the paper examines whether this incentive affects the strategy of star analysts. We ask whether star analysts put forth extra effort, or receive more resources, on the stocks they cover along with other stars. Such effort or resources may include, for example, field trips to visit the headquarters, factory, or clients of the firm covered, or appointing ad-hoc aides that assist only with the covered firm. Such effort allows star analyst to fine-tune the precision of their earnings forecasts. We find that accuracy in star analysts earnings forecasts is substantially higher in battleground stocks than in stocks that they cover as single stars. The improvement in accuracy is about 20%, which is both statistically and economically significant. The incentives of analysts and their employers to keep the star status are perfectly aligned. Our results stand the inclusion of a large number of controls in our tests. We also find that accuracy does not increase with the number of star analysts beyond two, the threshold at which a stock is classified as a battleground. We next devote a full section of the paper to examine possible endogeneity concerns. As we briefly mentioned and will carefully document later in the paper, we check whether stock characteristics might be behind our results. In addition to the inclusion of many control variables, we also follow stocks that switch between battleground and single-star throughout the sample period, and the effects remain just as strong. That is, the battleground effect is independent of cross-sectional differences between battleground and single-star stocks. In addition to these tests, we explore the possibility that the battleground effect arises from the coverage choice of star analysts. That is, star analysts actively choose stocks that are easier to predict. 1 Analysts covering such stocks are then more likely to be selected as stars, 1 Papers that document the tendency of analysts to choose stocks with better information environment include: Lang and Lundholm (1996); McNichols and O'Brien (1997); Francis, Hanna, and Philbrick (1997); and Bushman, Piotroski, and Smith (2004)). 6

7 and hence they become battleground stocks. We find, however, that coverage is very sticky and analysts rarely change their portfolios from year to year. We also report that the battleground effect persists even when we exclude all coverage initiations made by star analysts. Our results indicate that a reigning star is more accurate when competing with another reigning star(s) than when competing with a former or future star(s). This way, we focus on changes in battleground status that stem strictly from the ranking of other analysts and not from their coverage choice. We next study whether the higher accuracy could be the result of spillover effects of higher quality analysts. According to this argument, star analysts are selected based on their ability, as perceived by survey participants. Given their skills, star analysts improve the information environment of the stocks they cover and help other analysts including other star analysts covering the same stocks. As a result, star analysts are able to issue more accurate forecasts for battleground stocks than for the stocks in which they are the only single star. If this explanation is correct, we should find that the accuracy of the forecasts increases with the number of star analysts covering the stock. However, we find no relation between star coverage and star analyst accuracy when we examine changes in star coverage in which competition is always present (between 2 and 3), or in which competition plays no role (between 0 and 1). Furthermore, we study the case of battleground stocks that become singlestar stocks strictly because one of the analysts is not re-selected as a star. We find that the accuracy of the remaining star drops but only after the demotion of the other star becomes public. As we explain in the paper, given that accuracy is revealed only after actual earnings are announced, the drop in accuracy of the remaining star cannot be driven by the deterioration in the quality of the former star. We contribute to the literature in several ways. We provide further evidence of outstanding ability on the part of star analysts. As a point of departure with the mainstream 7

8 literature on analysts in general, and star analysts in particular, we disaggregate the portfolio of a star analyst and analyze differences across the stocks that they cover, instead of differences in the average accuracy at the portfolio level across analysts. We split the portfolio of stocks that a star analyst covers into two groups, depending on whether the same stocks are covered by other star analysts. We document that the presence of head-to-head competition, or a match between star analysts, is critical to determine their subsequent ranking. We conjecture that this is the result of the rule of thumb used by respondents to the I/I survey on which the selection of star analysts is based. As a result, the earnings forecasts of star analysts in stocks where there is head-to-head competition are more accurate than those in stocks where they are the only star. This is consistent with star analysts exerting more effort on stocks covered by other star analysts or with their employers allocating more resources to their coverage of these stocks. By looking at different types of stocks within the portfolio of a star analyst, our paper identifies a novel strategic consideration that influences the forecasts issued by star analysts. Most existing papers report that analysts mainly affect relatively small stocks that are covered by five analysts or less. The existing evidence is consistent with the notion that analyst coverage predominantly affects the information environment of smaller stocks while large firms are hardly affected. We show that star analyst accuracy among the largest firms is strongly dependent on the competition between star analysts. Our findings also highlight the importance of star analysts in financial markets. Over 20 ordinary analysts cover the average battleground stock, yet the presence of one other star analyst reduces the forecast error by more than 20%. This study documents the effect of competition among the most influential analysts covering the largest firms. While the existing literature indicates that a firm s information environment may affect forecast accuracy, we find that head-to-head competition between stars presents a case 8

9 in which the forecast accuracy affects the information environment, but not the other way around. We use a drop in head-to-head competition between stars, merely due to the demotion of rival stars, as an instrumental variable. Unlike other instrumental variables, such as brokerage house mergers (Hong and Kacperczyk (2010)) and brokerage house closures (Kelly and Ljungqvist (2012)), our approach ensures that the demoted star continues to cover the firm. We show that the forecast error of the remaining stars in stocks that experience a drop in star coverage only rises if these stocks happen to lose battleground status. The demotion of rival stars is therefore correlated with head-to-head competition between star analysts but not with the information environment of the firm. This novel instrumental variable allows us to distinguish between our tournament argument and alternative explanations. The rest of the paper is organized as follows. In Section 2, we discuss the data and methodology. In Section 3, we study the effect of forecast accuracy relative to other stars on the prospects of star analysts to keep their status. In Section 4, we analyze the difference in forecast accuracy of star analysts across the stocks they cover, in particular between battleground stocks and the rest. Section 5 examines potential alternative explanations and endogeneity concerns. We close the paper with some conclusions in Section 6. 2 Data, Methodology, and Summary Statistics 2.1 Data and methodology Our data is drawn from two main sources. The data on analysts earnings forecasts comes from the Institutional Brokers Estimate System (I/B/E/S) files. We limit the sample period to because many papers show that the nature of analyst estimates changed materially after the Regulation Fair Disclosure was adopted. Throughout the paper, we 9

10 concentrate on the earnings per share (EPS) forecast for the next fiscal year. Analysts rankings are drawn from the files of I/I. We determine star status according to Institutional Investor s annual All-American Research Team. Each year, Institutional Investor proclaims the top three analysts in various industries and sectors. We therefore define stars as those in the first, second, or third place. Although additional runner-up(s) may be nominated, they are excluded from the All- American Research Team and accordingly not considered stars in our analysis. The vast majority of analysts included in the team are selected as stars in a single industry. However, close to 10% of star analysts are selected in more than one industry. In such cases, in order to avoid double counting, we consider the higher ranking as the star analyst s ranking for that year. We manually match the remaining analyst-years from I/I rankings with the I/B/E/S files. Throughout our sample period, we have 1,184 unique star analyst-year observations. In this paper, we want to study the effects of head-to-head competition between star analysts and thus we divide all stocks in the I/B/E/S universe according to the number of star analysts who cover each stock. Of the 20,293 firm-years in our sample, 63.5% are not covered by any star analyst, while the rest are split almost evenly between stocks that are covered by a single star analyst and those that are covered by two or more star analysts (3,749 and 3,663, respectively). In order to compare analysts performance across different stocks, we need a measure of accuracy. For this purpose we introduce the metric of forecast error, which we define as the absolute difference between the forecast and realized earnings scaled by the realized earnings. We note that realized earnings (and accuracy) are only announced one year after the forecast was made (after the forecast s fiscal-year s end). In order to reduce the potential influence of outliers, most of which are driven by obvious data errors, we exclude from our sample all forecasts with an error larger than 4. We follow Clement and Tse (2005) in defining the control variables and we normalize them relative to all analysts 10

11 covering the firm (whether stars or non-stars) using the following formula, in which analyst i covers firm j at year t: Characteristic ijt RawCharacteristic ijt RawCharacteristic max RawCharacteristic min jt RawCharacteristic min jt jt. As noted by Clement and Tse (2005), the normalization of all variables to a value between 0 and 1 allows us to examine their relative importance by directly comparing the coefficients. Our focus is on the competition between existing stars and thus we require that the analyst achieve star status one year prior to her forecast s fiscal year end. For every firm that an analyst covers, we include in our study only the earliest announcement in each year. We focus on the earliest announcement rather than on the last announcement for two reasons. First, the earliest forecast is arguably the most challenging because the forecast horizon is the longest. Second, previous literature shows that analysts forecasts are subject to herding and hence later announcements carry less information about analysts quality. Therefore, for each firm an analyst covers, we maintain only the earliest EPS forecast for the next fiscal year as long as it is made before the fiscal year s end Summary statistics We start our empirical investigation by dividing the I/B/E/S universe into three types of stocks: (1) stocks that are not covered by any star analyst, (2) stocks that are covered by a 2 We note that I/I respondents are required to send back their questionnaires between March and May, thus for firms with a fiscal year end other than December, the actual EPS may not yet available when the survey closes. This suggests that in some cases the respondents need to estimate the winner (possibly based on revisions in analysts forecasts and quarterly earnings). Such measurement errors in our dependent variables are generally expected to bias the regression coefficients toward 0 and thus they work against finding a relation between the dependent and independent variables. Nevertheless, we can report that when we exclude these firms (roughly 20% of all forecasts) our main results hold. 11

12 single star analyst, and (3) stocks that are covered by more than one star analyst (henceforth called battleground stocks). In Table 1 we present summary statistics of accounting variables, market performance, and analyst coverage for each category. (Insert Table 1 about here.) The first row presents our sample size (in firm-years) under each category. Almost two-thirds of the firms in the I/B/E/S universe are not covered by any star analyst and the rest are divided almost equally between firms that are covered by a single star analyst and those covered by more than one star analyst (i.e., battleground stocks). The second row presents the number of large firms. Throughout the paper, we refer to firms that are larger than the median size of the New York Stock Exchange (NYSE) as large firms. We find that close to 80% of all large firms are covered by at least one star analyst and that over 50% of all large firms are battleground stocks. Focusing on firm size, our results show that battleground stocks are larger than those covered by a single star analyst, which are in turn larger than those not covered by any star analyst. Importantly, the proportion of firms smaller than the lowest NYSE-size quintile in battleground stocks is only 3%. Battleground stocks are also more profitable than other stocks. The proportion of firms with a negative net income among battleground stocks is only one-fifth of the proportion of firms covered by a single star analyst. Analyst coverage increases with the number of star analysts that cover the stock. The average number of analysts that cover stocks with no star coverage is less than 7 but this number increases to 11.5 for stocks with singlestar coverage and further increases to 18.5 for battleground stocks. Furthermore, while analyst coverage increases throughout our sample period, as evidenced by the change in the number of analysts (compared with the previous year), battleground stocks experience the 12

13 sharpest increase. The average increase in coverage for battleground stocks is almost twice the increase for stocks covered by a single star analyst (1.05 and 0.66, respectively). The average forecast error across all analysts also decreases with coverage by star analysts from 0.55 for stocks with no star coverage to 0.41 for stocks with single-star coverage and 0.31 for battleground stocks. Panel B presents data regarding the coverage choice of star analysts. A star analyst covers an average of 12.3 stocks and close to 60% are battleground stocks. The fact that star analysts typically cover many battleground stocks is of utmost importance in studying the relation between star status and forecast error. In particular, it is likely that changes in information environment will offset each other at the portfolio level. Most importantly, it is extremely unlikely that changes in the information environment of one particular firm out of many in the star analyst s portfolio will affect star status. A star analyst initiates coverage of 1.8 stocks per year on average and our unreported results show that roughly three-quarters of initiations represent large firms. Most initiations take place within two years after the analyst becomes a star. The last row reports the number of firms that are abandoned. Of the 12.3 firms that star analysts cover, only 0.6 firms are dropped each year on average. This suggests that star analysts coverage portfolios are very sticky, with over 95% of the firms carried over from year to year. Specifically, star analysts are unlikely to drop stocks after they initiate coverage. The fact that star analysts hardly ever modify their coverage portfolio suggests that star analysts do not self-select into stocks with a better information environment. More specifically, star analysts do not constantly change their stock coverage in response to changes in individual firms information environment. 13

14 3 I/I Rankings as an Incentive for Incumbent Star Analysts The question of whether there are economic reasons behind analyst rankings is of major importance to finance. Previous literature has largely failed to find economic reasoning behind I/I rankings, thus reinforcing the view that rankings are mostly a beauty contest. A possible reason for this failure is the fact that previous literature treats all stocks with equal importance, while in reality I/I survey respondents seem to use rules of thumb, which allow them to respond to the survey in a limited amount of time while providing adequate answers. We suggest that success in battleground stocks can serve as one such rule of thumb as it allows the I/I respondents to focus only on existing stars, and to compare apples to apples, as battleground stocks naturally control for heterogeneity in information environment and earnings surprises. While analyst tournaments likely incorporate a long set of complex rules, our goal in this section is merely to establish that the competition between stars has a strong economic effect on analyst rankings. With this in mind, the focus of the rest of paper is to study the consequences of this effect on star analyst accuracy and, in turn, on the firm s information environment. Previous literature has found that star status is associated with higher analyst pay as well as higher deal flow to the brokerage house. In particular, Groysberg, Healy, and Maber (2011) find that analysts selected to the I/I All-American Team earn much higher salaries than other analysts. In addition, they show a large increase in analyst compensation of roughly 25% when they become stars. Clarke, Khorana, Patel, and Rau (2007) find that when stars switch brokerage houses, the deal flow of the new (old) brokerage house increases (decreases). Arguably, the most important challenge a star analyst faces is to maintain star status. While star status is sticky, roughly 25% of all stars fail to retain their star status in the consecutive year (e.g., Emery and Li (2009) and our own findings). Maintaining star status 14

15 becomes even more difficult with age, as the I/I survey respondents seem to prefer younger blood (as evident from the negative relation between experience and star status). Arguably, the compensation for a star analyst and her ability to attract new deals are likely to improve with her place in the rankings. Most importantly, a higher ranking substantially decreases the probability that the analyst will lose star status i.e., will no longer be ranked in any of the top three places. Table 2 illustrates this point using a simple transition matrix. The rows represent the ranking of the analyst in year t and the columns represent the ranking at year t + 1. Table 2 shows that during our sample period, the probability of an analyst ranked in the highest position to be demoted out of the first three places is close to 12%. The probability of demotion more than doubles for an analyst ranked in second place, and an analyst ranked third faces a probability of more than 40% of not being selected into the top three places in the subsequent year. Results of Table 2 demonstrate that current ranking is crucial in maintaining star status and the benefits that come with it. Star analysts who manage to improve their ranking considerably reduce the risk of demotion. (Insert Table 2 about here.) According to Hong and Kubik (2003), analysts are evaluated according to their relative accuracy and not their absolute accuracy. We conjecture that star analysts are evaluated according to their relative accuracy as well, but that their reference group is other star analysts rather than the entire analyst community. Taking into account the importance of the I/I rankings, we ask whether performing well in battleground stocks is rewarded with a better chance to remain a star. The I/I rankings are based on a questionnaire sent to thousands of professionals in hundreds of institutions on an annual basis. Importantly, the survey respondents do not receive any type of compensation and thus it seems reasonable to assume 15

16 that they use rules of thumb, which allow them to respond to the survey in a limited amount of time while providing adequate answers. We suggest that success in battleground stocks can serve as one such rule of thumb as it allows the I/I respondents to directly compare the performance of star analysts without the need to take into account the heterogeneity in information environment across firms. Since three-quarters of incumbent star analysts retain their star status, it seems that the I/I rankings are mainly affected by this exclusive tournament between star analysts. Several papers examine whether star analysts have better predictive ability, both before and after they become stars, and typically report a positive relation between accuracy and star status. For example, Stickel (1992) reports that star analysts are more accurate than non-star analysts. Leone and Wu (2007) find a positive relation between pre-selection accuracy and star status. Emery and Li (2009) use a logistic regression to examine the variables that affect the probability of being a star. They report that overall accuracy (in all stocks in the analyst s portfolio) is not a significant determinant of becoming an I/I star and only plays a modest role compared with that of other determinants for I/I stars trying to repeat. Our tests are distinct from the previous literature in several important regards. First, we differentiate between battleground and non-battleground stocks, which allows us to evaluate star analyst accuracy relative to other star analysts and not just relative to the entire analyst community. We show that the performance of star analysts is not equally important across all of the stocks they cover. Second, we are only interested in the determinants of ranking improvements of existing star analysts. Since existing star analysts are already highly recognized, variables related to recognition are likely to play a minor role. Third, our sample period begins after the Regulation Fair Disclosure (FD) was introduced. Earlier papers using pre-regulation FD data document that analysts' accuracy is not only influenced by their intellectual ability but also by their relationship with the firm management. For example, 16

17 Cohen, Frazzini, and Malloy (2010) find that prior to Regulation FD, equity analysts outperform on their stock recommendations when they have educational ties with senior officers of firms that they cover. This school-tie premium disappears after the introduction of Regulation FD, which mandates that all publicly traded companies must disclose material information to all investors at the same time. 3.1 Success in battleground stocks and ranking improvements We next aim to establish that success in battleground stocks has a material effect on analyst rankings. In particular, we study whether being the most accurate star analyst in a battleground stock has an economically significant effect on the likelihood of ranking improvements. For this purpose, we create a binary variable win, to which we assign the value of 1 if the star analyst is closer to the actual earnings than all other star analysts that is, her forecast error is the smallest among all star analysts covering the stock. We then count the total number of wins that an analyst has accumulated in a given year, which we refer to as No. of wins. To measure promotion in the I/I rankings, we define a ranking improvement, which takes the value of 1 if the star analyst improves her ranking in the subsequent year. A ranking improvement takes place when an analyst ranked in the second or third place moves up or when an analyst ranked in the highest position remains in that position in the subsequent year. 3 To illustrate the relationship between the accuracy of star analysts in battleground stocks and their promotion in the I/I rankings, Figure 1 plots the frequency of ranking improvement by the number of wins. The probability of improving one s ranking 3 Our notion of improvement includes analysts ranked in the highest position who manage to remain in the top place, which is consistent with the incentive to maximize the probability of retaining star status. Alternatively, we use a more restrictive definition of actual improvement by dropping analysts ranked first in year t since they technically cannot improve their ranking. After dropping all analysts ranked first, our sample decreases by roughly one-third, and the proportion of ranking improvement decreases to 21.6%. The results remain qualitatively the same as in Table 3 although with lower significance. 17

18 substantially increases with every additional win, and more than doubles going from 0 to 5 wins. Importantly, stars that accumulated more wins than the median number of wins (2 in our sample) are 50% more likely to be promoted in the rankings than those who did not. In Table 3, we use logit regressions to formalize the statistical significance of this big economic effect. We define a ranking improvement and the number of wins as in the previous paragraph. We follow Clement and Tse (2005) in defining the control variables. Control variables include the market value of equity (firm size); the total number of analysts (whether stars or non-stars) that cover the stock (No. of analyst); the number of days elapsed since the previous forecast on the same stock by any analyst (days elapsed); the number of days remaining until the end of the fiscal year (forecast horizon); the order in which analysts submitted their forecast (order); the number of analysts employed by the analyst s brokerage house (brokerage size); the general experience of the analyst (general experience) as measured by the number of years that the analyst has been in the I/B/E/S database; and the specific experience of the analyst in covering the firm (firm experience) as measured by the 18

19 number of years that the analyst has covered the firm. All control variables are normalized to take a value between 0 and 1, relative to all analysts covering the firm (whether stars or nonstars) as described in the methodology section. Given the annual frequency of I/I rankings, we aggregate all forecasts made by each analyst in each year by using a simple mean. We exclude from the analysis firms below the NYSE size median. Our sample includes 1,184 analyst-years, of which 35% experience a ranking improvement in the subsequent year. (Insert Table 3 about here.) as follows: In Model 1, the main independent variable is mean relative accuracy, which we define Relative accuracy i, j, t 1 Error MAX Error i, j, t MIN Error jt, MIN Error j, t j, t, where i is the analyst, j is the firm, and t is the year. We then calculate the mean (simple average) across all stocks the analyst covers in each year. Note that the accuracy is normalized relative to all analysts covering the firm (whether stars or non-stars) and thus it inherently controls for firm-specific differences. The results of Model 1 show that the coefficient of mean relative accuracy is positive (0.89) and significant at the 5% level. Due to the loss of information in aggregation, the insignificant results for most of our control variables are to be expected. The only control variable that is significant is firm experience. The negative coefficient suggests that a tendency exists to promote relatively young analysts. This finding is consistent with Emery and Li (2009), who suggest that the assessment of older analysts is less likely to change. Surprisingly, there is no significant relation between 19

20 brokerage size and the probability of promotion. A possible explanation of this finding is that star analysts are already recognized, thus making recognition variables such as the brokerage house less important. Model 2 adds the number of battleground stocks that the analyst covers during the year to the regression. The coefficient of this variable is positive (0.06) and significant at the 1% level. This suggests that the higher the number of battleground stocks an analyst covers, the more likely she is to be promoted in the I/I rankings. One possible explanation is that battleground stocks are more important to institutional investors, and institutions vote for analysts that cover more of these important stocks. However, it is also possible that institutional investors rank star analysts by their accuracy relative to other star analysts. In this case, the number of battleground stocks will mostly capture a larger number of opportunities to beat other stars. To further explore whether I/I rankings are solely determined by how many important stocks a star analyst covers, we study whether being the most accurate star analyst in a battleground stock particularly affects the likelihood of ranking improvements. Adding No. of wins to the regression significantly changes the results. The coefficient of No. of wins is positive (0.10) and highly significant. The coefficient of mean relative accuracy is reduced by roughly one-half and becomes statistically insignificant. Most importantly, the coefficient of No. battleground is reduced by more than half (0.02) and becomes statistically insignificant. This stands to show that I/I rankings are not determined merely by how many important stocks a star analyst covers, but rather by success relative to other stars. Therefore, our results suggest that wins in battleground stocks are pivotal to one s chance to be promoted in the I/I rankings. Using a sports analogy to understand this ranking pattern, each battleground stock presents a game, and the analyst who has accumulated the most wins becomes the winner in the grand tournament. 20

21 In Model 4, we examine the importance of the number of wins in comparison to that of relative accuracy. We do so by normalizing the value of No. of wins relative to all star analysts covering the firm, so that winsit MIN ( wins) t Normalized wins MAX ( wins) MIN ( wins) t t, where wins is the number of wins of analyst i in year t, and MAX (MIN) wins is the maximum (minimum) number of wins of all star analysts in the same year. Since both Normalized wins and mean relative accuracy are normalized as are all of the other variables between 0 and 1, the magnitude of the coefficient indicates their relative importance. Our results show that the coefficient of Normalized wins is almost four times greater than that of mean relative accuracy (0.15 and significant, compared with 0.04 and insignificant). In Panel B of Table 3, we ask whether battleground stocks merely represent those stocks that garner more interest from institutional investors. Our concern is that stock centrality affects analyst rankings rather than the other way around. According to our competition argument, institutional investors pay closer attention to stocks covered by multiple stars. Conversely, institutional investors may simply tend to vote for analysts who have high-interest stocks in their portfolio, in which case high-interest stocks will likely become battleground stocks. If so, then ranking improvement would be correlated with the number of battleground stocks rather than the success in them. In Table 3 Panel A, we already show that, once we control for the success in battleground stocks (No. of wins), the coefficient of the number of battleground stocks becomes small and insignificant. It is also possible, however, that institutional investors pay closer attention to the performance of star analysts in stocks that currently garner more interest. According to this argument, 21

22 institutional investors favor analysts who not only have many high-interest stocks in their portfolio, but also managed to perform well in these stocks. In our next set of tests, we thus examine whether wins in more central stocks carry more weight in ranking improvements. For this purpose, we classify wins based on different proxies for the centrality of the stock they were achieved in. Size is a natural candidate to be correlated with centrality. Previous literature shows that firm size is associated with institutional ownership, and so larger firms are likely to garner more interest from institutional investors. We thus ask whether performing well in large battleground stocks is more important than in smaller battleground stocks. In Model 1, we split wins by the median firm size among all battleground stocks. Results show that performing well is important regardless of firm size. Stock centrality is also likely correlated with the number of stars covering the stock. If institutions vote for analysts that cover high-interest stocks, then more analysts who cover these stocks are likely to be selected as stars. In Model 2, we split wins by the level of star coverage. That is, we distinguish between battleground stocks covered by two stars and those covered by three or more stars. The number of stars covering a battleground stock does not make the win more important. 4 This stands to show that I/I rankings are not determined merely by how many high-interest stocks a star analyst covers, but rather by success relative to other stars. Our next exercise directly tests whether central stocks become battleground or the other way around. We do so by focusing on stocks that recently became battleground i.e., in the year during which they switch from single-star to battleground. We test whether the success in these stocks carries more weight than that in stocks that were already battleground 4 For robustness we also examine whether wins in battleground stocks in which the number of star analysts increases carry more weight than in stocks in which the number of star analysts decreases or remains the same. We find no difference in the importance of wins in the two types of stocks. 22

23 in the prior year. According to the centrality argument, ranking improvements will be strongly correlated with success in stocks that recently started garnering more interest by institutional investors (new battleground stocks) and much less affected by stocks for which the degree of interest remains unchanged (incumbent battleground stocks). While all new battleground stocks represent high interest, some incumbent battleground stocks represent stocks that have fallen out of interest (and, according to the centrality argument, shrink in star coverage) and yet remain battleground stocks (e.g., dropping from three to two stars). In contrast, our tournament argument suggests that ranking improvements will be strongly correlated with success in incumbent battleground stocks as what counts is beating other existing stars and not any recent increase in interest. We reiterate that analysts issue their forecast earlier in the year, before the I/I rankings are released in October. In many cases, new battleground stocks are not yet revealed when the forecast is made. Therefore, if star analysts are rewarded for beating other stars, as our tournament argument suggests, we would expect them to focus their effort on performing well in incumbent battleground stocks. We thus split the total number of wins into wins in incumbent battleground stocks and wins in new battleground stocks. The results in Model 3 show that the coefficient of wins in incumbent battleground stocks is much larger than that of wins in new battleground stocks. Furthermore, we note that only 38% of all wins are in new battleground stocks, and thus having more of such wins cannot compensate for the weaker effect of a win in a new battleground stock. In contrast with the centrality argument, which suggests that wins in new battleground stocks should have a much stronger effect on ranking improvements, our results show that only a win in an incumbent battleground stock has a significant effect. These findings suggest that the ranking of star analysts is uncorrelated with changes in battleground status among the different stocks in their portfolio. To provide further evidence, we also add to the regression the change in the number of battleground stocks from 23

24 the prior year. Our unreported results show that, while the number of wins remains strongly significant, the change in the number of battleground stocks is not. These findings are in contrast to the stock centrality argument, according to which high-interest stocks become battleground stocks simply because institutional investors vote for analysts who have highinterest stocks in their portfolio. Our results emphasize the value of being the most accurate star among all star analysts covering the stock. In Model 4, we explore the possibility that I/I respondents use more complex rules and distinguish between different types of wins. We split wins by the median gap (the difference in normalized error) between the first- and second-most accurate stars. The difference between the coefficients of the two types of wins is relatively small. This is consistent with a winner-take-all tournament, in which star analysts are being reelected based on being the most accurate star in battleground stocks. We perform several additional tests that examine whether some wins weigh more than others. For example, we split wins by whether the star was the earliest of all stars to announce, and find that both types of wins carry similar weight. Our results show that wins are all important, independent from the characteristics of the stock in which they were achieved. These findings mitigate the concern that there exists an omitted or unobservable time varying firm characteristic correlated with analyst rankings. 3.2 Success in battleground vs. single-star stocks We have shown that accuracy among star analysts in battleground stocks is highly correlated with an improvement in I/I rankings. However, we have yet to examine the importance of a win in battleground stocks compared with performing well in single-star stocks. To address this issue, we create another binary variable, ibes win, to which we assign the value of 1 if the analyst is the most accurate in the entire I/B/E/S universe, including both 24

25 stars and non-stars. We then count the total number of I/B/E/S wins in both battleground and single-star stocks that an analyst has accumulated in a given year, which we refer to as No. of top1. Our main goal is to compare the importance of performing well in battleground stocks (No. of wins) to that of performing well in single-star stocks (No. of top1). (Insert Table 4 about here.) Table 4 shows the relation between star analyst accuracy in both battleground and single-star stocks and their promotion in the I/I rankings. As in Table 3, the dependent variable indicates whether the star analyst improves the ranking or remains in the highest position. In Model 1, we estimate the effect of the total number of I/B/E/S wins that an analyst has accumulated in a particular year on the probability of ranking improvement, and our results show that the coefficient of No. of top1 is positive and highly significant. In Model 2, we add back to the regression the variable No. of wins (the number of battleground stocks in which the analyst is the most accurate relative to other stars) and we find that the coefficient of No. of top1 drops by almost half and is significant only at the 10% level. In comparison, the coefficient of No. of wins is significant at the 1% level. Interestingly, both coefficients seem to be of the same magnitude. Using our sports analogy, beating all analysts covering a stock appears to have the same value as beating other stars in a battleground stock. Both types of wins count just as much for the tournament. Beating all analysts, however, is much harder than focusing only on other stars because while the unconditional probability of a star analyst being the most accurate in the entire I/B/E/S universe is less than 10%, the probability of a star analyst being the most accurate relative to other stars is 35%. This 25