Analyst Coverage, Analyst Optimism, and Stock Price Crash Risk: Evidence from a Transitional Economy

Transcription

1 Analyst Coverage, Analyst Optimism, and Stock Price Crash Risk: Evidence from a Transitional Economy Kam C. Chan Gordon Ford College of Business Western Kentucky University Johnny.chan@wku.edu Xuanyu Jiang School of Business Renmin University of China jxynini@sina.com Nianhang Xu School of Business Renmin University of China nhxu@ruc.edu.cn Zhihong Yi School of Business Renmin University of China zhyi7620@163.com This version: January

2 Analyst Coverage, Analyst Optimism, and Stock Price Crash Risk: Evidence from a Transitional Economy Abstract We examine the relation among analyst coverage, analyst optimism, and firm-specific future stock price crash risk. If analysts are overly optimistic in their recommendations and earnings forecasts, the negative information of the firms they cover cannot be timely revealed to the outside investors. When the accumulated negative information reaches a tipping point, it will then be revealed to market, resulting in bubble bursting and a stock price crash. Using samples of 3,416 (analyst recommendations) and 3,239 (earnings forecasts) firm-years, our findings suggest that analyst coverage and analyst optimism contributes to future stock price crash risk in China. We find that an increase in analyst coverage for a firm leads to an increase in the future stock price crash risk of the same firm and this positive relation is more pronounced when firms are covered by more optimistic analysts. In addition, the impact of analyst optimism on future stock price crash risk is more pronounced when analysts are from investment banks with underwriting services and from employers with small brokerage revenue. In contrast, the impact of analyst optimism on stock price crash risk is less pronounced when analysts have high personal reputations and are affiliated with reputable investment banks and brokerage firms. Our results are robust to alternative variable measures and different regression model specifications. Keywords: Analyst optimism; Crash risk; Conflict of interest; Reputation JEL classification: G00, G24-2 -

3 Analyst Coverage, Analyst Optimism, and Stock Price Crash Risk: Evidence from a Transitional Economy 1. Introduction The determinants of aggregate and firm-level stock price crash risk draw the attention of investors, regulators, and policy makers as they all have a vested interest in lowering such a risk. It is natural that there is a high level of interest on the subject of forecasting future stock price crash risk. For instance, Jin and Myers (2006) provide a model and country-level empirical evidence that a positive correlation between stock market opaqueness and stock price crash risk in 40 countries. In contrast to the country-level evidence, Hutton, Marcus, and Tehranian (2009) find a positive association between future stock price crash risk and the opaqueness of financial reports using firm-level data in the US. Kim, Li, and Zhang (2011a, b) suggest that future stock price crash risk for a firm is positively correlated with its CFO s option portfolio value and corporate tax avoidance. Kim and Zhang (2011) document that accounting conservatism reduces the likelihood of a firm experiencing stock price crashes. The basic argument of these studies is that managers have a tendency to withhold bad news for an extended period of time. Thus, bad news of a firm stockpiles. When the accumulation of bad news passes a threshold, it is revealed to the market at once, leading to a large negative drop in price for the stock. Stock price crash risk studies, however, primarily focus on the impact of accounting characteristics to such a risk. Few studies consider determinants other than a firm s accounting characteristics. A separate voluminous number of studies on analyst coverage suggest that there is indeed analyst optimism in recommendations and earnings forecasts. In the context of analyst coverage, prior literature considers several incentives for analyst optimism including: (1) to maximize trading commissions (Hayes, 1998; Irvine, 2001; Jackson, 2005; Cowen, Groysberg, and Healy, 2006; - 3 -

4 Beyer and Guttman, 2011); (2) to curry favor with management (Lim, 2001); and (3) to gain or maintain investment banking business (Dugar and Nathan, 1995; Lin and McNichols, 1998; Michaely and Womack, 1999). Despite the existing literature on analyst coverage, analyst optimism, and stock price crash risk, few studies link the major research areas. The role of analyst coverage and analyst optimism bias related to individual future stock price crash risk is unclear. The objective of this paper is to examine the relation among analyst coverage, analyst optimism, and firm-specific future stock price crash risk. We contend that analyst coverage, through their optimistic forecasts and recommendations, can increase the future stock price crash risk of the firms they covered. If analysts are overly optimistic in recommendations and earnings forecasts, the negative information of the firms they cover cannot be timely revealed to outside investors. When the accumulated negative information reaches a tipping point, it will then be revealed to the market, resulting in bubble bursting and a stock price crash. While our argument of future stock price crash is similar to the literature, the cause is different. The explanation of crash risk in the literature stems from a firm s internal factors, such as accounting conservatism, but our focus is to relate such a risk with a firm s external factors (analyst coverage and analyst optimism). Hence, our study offers a new perspective on the determinant of future stock price crash risk. Specifically, we examine several hypotheses using a rich database in the emerging Chinese stock market. The literature, with the exception of Jin and Myers (2006), all study US data. A detailed study using the Chinese stock market allows us to understand stock price crash risk in an emerging market and the findings would be particularly useful to investors, regulators, and - 4 -

5 policy makers in China and other emerging markets in terms of understanding the contributing factors of stock price crashes. In addition, the Chinese stock market offers several unique features to test the several hypotheses. First, China s financial market and listed firms are associated with a poor information environment (Piotroski, Wong, and Zhang, 2011). For instance, Ball, Robin, and Wu (2000) document that despite the introduction and adoption of international accounting standards among listed firms in China, the timely loss recognition practices are still behind the practices of common law countries. The opaque nature of the Chinese stock market makes the role of analyst coverage particularly important in terms of providing information to investors. Second, there have been several reports raising the concern of optimism bias among the analysts in China. Li (2008), Liu and Zhang (2008), and Wang (2009) report that displeasing institutional clients in China can hurt an analyst s future career. 1 Similar to Piotroski et al (2011), we argue that it is a Chinese culture not to release negative information unless it is absolutely unavoidable. Therefore, analysts, to the extent possible, would value social conformity and maintain a good relationship with the firm they covered. Overall, analyst optimism is a concern in China. Third, the database in China permits us to identify analyst characteristics such as whether an analyst is any of the following: (1) from a top investment bank; (2) from a top brokerage firm (high brokerage revenue) 2 ; and (3) a star analyst. The rich database allows us to disentangle the potential conflict of interest among different analyst affiliations and the impact of analysts personal and investment bank/brokerage firm reputations on the relation between analyst optimism and stock price crash risk. 1 Several news reports in China say that analysts are under pressure to provide optimism bias opinions. A detailed survey was presented at Sina.com ( assessed on November 27, 2011). 2 Data on brokerage revenue is unique in China. In a US study, Ljungqvist et al. (2007) proxy for the size of the brokerage business using the annual number of registered representatives because the brokerage revenue is absent in the US

6 For our empirical analysis, we follow Chen, Hong, and Stein (2001) and Kim, Li and Zhang (2011a, b) and use the negative coefficient of skewness (NCSKEW) and down-to-up volatility (DUVOL) to measure the stock price crash risk. Using samples of 3,416 (analyst recommendations) and 3,239 (earnings forecasts) firm-years, our findings suggest that analyst coverage and analyst optimism contributes to future stock price crash risk in China. We find that an increase in analyst coverage for a firm leads to an increase in the future stock price crash risk of the same firm and this positive relation is more pronounced when firms are covered by more optimistic analysts. In addition, the impact of analyst optimism on future stock price crash risk is more pronounced when analysts are from investment banks with underwriting services and from employers with small brokerage revenue. In contrast, the impact of analyst optimism on stock price crash risk is less pronounced when analysts have high personal reputation and they are affiliated with reputable investment banks and brokerage firms. Our results are robust to alternative variable measures and different regression models specifications. We make several contributions to the literature. First, our research extends the emerging literature of forecasting future stock price crash risk. This paper identifies a new negative effect of analyst coverage and analyst optimism: they can increase future stock price crash risk. Our evidence suggests that, at the firm level, there exists a determinant beyond a firm s internal characteristics (opaqueness of financial reports, CFOs option portfolio value, corporate tax avoidance, and accounting conservatism). Second, we provide additional insights into the large body of literature on the role of conflict of interest (Ljungqvist et al., 2007; Mehran and Stulz, 2007) and reputation (Hong, Kubik, and Solomon, 2000; Cowen, Groysberg, and Healy, 2006; - 6 -

7 Fang and Yasuda, 2010) in sell-side analyst research. 3 We find that, on one hand, conflict of interest can bias analyst research upward and aggravate future stock price crash risk. On the other hand, both personal reputation and investment bank or brokerage firm reputation can work as a disciplinary mechanism against analysts tendency to issue optimistic research and finally attenuate stock price crash risk. Third, our findings offer a new bad news releasing view of stock price crashes from information production agents (analysts). While our bad news releasing view is similar to the bad news hoarding theory of stock price crashes (Jin and Myers, 2006; Bleck and Liu, 2007), we focus on analysts behavior rather than corporate insiders. Our results suggest that analysts, through their optimism bias during information processing, contribute to the slow releasing the bad news for the firms they covered. By not releasing the bad news in a timely fashion, analyst coverage and analyst optimism make firms prone to future stock price crashes. 2. Hypotheses development A sell-side analyst s supposed role is to act as an information intermediary, channeling information from companies to investors in the form of investment recommendations, earnings forecasts, and detailed reports. If analysts really play such a role as revealing firm-specific information (especially bad news) to investors, then a firm s information transparency would increase. Recent studies shows that the lack of information transparency increases a stock s future crash risk by enabling managers to hide and accumulate bad news (e.g., Jin and Myers, 2006; Hutton, Marcus, and Tehranian, 2009; Kim, Li and Zhang, 2011a, b). Thus, our first hypothesis is that analyst coverage can reduce future stock price crash risk, that is, 3 Mehran and Stulz (2007) make an excellent review on the conflicts of interest in financial institutions

8 H1a: All else being equal, analyst coverage is negatively associated with future stock price crash risk. Analysts, however, are not bound to fully and truthfully report their private information (Beyer, et al., 2010). Generally, sell-side analysts tend to issue optimistic forecasts and recommendations. 4 Thus, if analysts tend to be optimistic in forecasts and recommendations, the negative information of the firms they cover cannot be timely revealed to outside investors. When the accumulated negative information reaches a tipping point, it will be suddenly released to the stock market, resulting in bubble bursting and a stock price crash (Jin and Myers, 2006; Hutton, Marcus, and Tehranian, 2009). Thus, analyst coverage, in contrast to H1a, may increase future stock price crash risk. We have alternative Hypothesis 1: H1b: All else being equal, analyst coverage is positively associated with future stock price crash risk. Following H1b, we argue that the positive impact of analyst coverage on crash risk should be more pronounced for firms covered by more optimistic analysts. This leads to the second hypothesis: H2: All else being equal, the positive association between analyst coverage and crash risk is more pronounced when firms are covered by more optimistic analysts. 4 For example, from 1995 through 2001, only 4% of all recommendations on seasoned stocks are rated underperform or sell. Most recommendations issued during that period are favorable, up to the rating of strong buy. In addition, after 2002, analyst tendency toward optimism persists and stock recommendations are still biased upward (see Mola and Guidolin, 2009)

9 Empirical results consistent with H2 corroborate the analyst optimism explanation of H1b for the positive relation between analyst coverage and crash risk. That is, if the positive relation between analyst coverage and crash risk is not caused by analyst optimism, we will not observe evidence consistent with H2. In the next two hypotheses, we propose that the factors that influence analyst optimism also impact the relation between analyst coverage and crash risk. Specifically, we consider these factors from two perspectives: the presence of conflict of interest in sell-side research (i.e., Conflict of Interest Hypothesis) and the role of reputation (i.e., the Reputation Hypothesis). For the Conflict of Interest Hypothesis, we argue that sell-side analysts who work for investment banking houses could come under pressure to publish more favorable research about their employers current or potential clients to help boost investment banking fee revenue (Dugar and Nathan, 1995; Lin and McNichols, 1998; Michaely and Womack, 1999; Agrawal and Chen, 2008). If analysts provide optimistic research to attract lucrative underwriting business to their firms, then analysts working at investment banks that provide underwriting services should issue more optimistic forecasts and recommendations, compared with those employed in other places that do not provide any underwriting services. Therefore, under the conflict of interest argument, firms covered by analysts from investment banks face greater crash risk, as shown in the following hypothesis: H3: All else being equal, the positive association between analyst coverage and crash risk is more pronounced when firms are covered by analysts from investment banks

10 Similarly, analysts can also come under pressure to help generate brokerage commissions. Irvine (2004), Jackson (2005), and Cowen, Groysberg, and Healy (2006) argue that bullish research stimulates trading and thus generates additional brokerage revenue. This brokerage revenue pressure also provides incentives for analysts to present positively biased opinions. For example, Irvine (2004) uses a unique dataset of trades of the largest 100 companies from the Toronto Stock Exchange over 1993 to 1994 and finds that favorable stock recommendations are associated with larger increases in the brokerage house s share of trading volume of covered firms. Consistent with these results, Jackson (2005) finds evidence suggesting that analysts who issue optimistic stock recommendations generate more trade for their brokerage firms based on a sample from the Australian equity market over the period Ljungqvist et al. (2007) also find that analysts issue more optimistic recommendations when they work for banks with larger brokerage business. Therefore, under the Conflict of Interest Hypothesis, firms covered by analysts from with large brokerage firms face greater crash risk, as shown in the following hypothesis: H4a: All else being equal, the positive association between analyst coverage and crash risk is more pronounced when firms are covered by analysts from employers with larger brokerage revenue. We argue that due to, fierce competition among brokerage firms, analysts at small brokerage firms are under more pressure to generate brokerage revenue for their firm survival relative to big brokerage firms. For instance, Li (2008) reported a case on an analyst s negative comments

11 on Kweichow Moutai, a major liquor producer in China. His employer immediately received substantially less trading commissions as institutional investors direct less trading to the brokerage firm. After a few weeks, the analyst changed the tone of his comments from negative to uncertain regarding Kweichow Moutai s future stock price. The anecdotal evidence suggests that analysts are likely to inflate their recommendations and earnings forecasts in hope of attracting more trading business. An alternative hypothesis to H4a is: H4b: All else being equal, the positive association between analyst coverage and crash risk is more pronounced when firms are covered by analysts from employers with smaller brokerage revenue. For the Reputation Hypothesis, prior studies show that reputation might have some countervailing effect on analyst optimism. At the personal level, analysts face a trade-off between generating revenues for their employers brokerage and investment banking businesses and their own future career concerns. In the short run, analysts may gain substantial underwriting or trading-related compensation by publishing optimism research. However, in the long run, their biased research can damage their reputation and long-term career prospects (Hong, Kubik, and Solomon, 2000; Hong and Kubick, 2003). Because analysts with a better reputation have greater long-term benefits to lose, they are less likely to succumb to investment banking or brokerage pressure in the short run (Fang and Yasuda, 2010; Ljungqvist et al., 2007). Similarly, at the employer level, an investment bank or a brokerage firm s reputation capital is important for its long-term success. Investment banks and brokerage firms have an incentive to build and preserve their reputations, which helps reputable investment banks and brokerage firms better supervise

12 the actions of individual analysts (Fang and Yasuda, 2010). Cowen, Groysberg, and Healy (2006) find that analysts at the six largest Wall Street investment banks that dominate the underwriting market have less optimistic forecasts and recommendations than analysts working at other investment banks, syndicates, or brokerage firms. Ljungqvist et al. (2006, 2007) also find that analysts employed by high reputation investment banks issue less aggressive recommendations. Therefore, under the Reputation Hypothesis, we predict that the reputation of both the analyst and the investment bank or brokerage can curb the analysts optimism, and firms covered by star analysts and analysts from high reputation investment banks or brokerage firms face less crash risk, as shown in the following hypotheses: H5a: All else being equal, the positive association between analyst coverage and crash risk is less pronounced when firms are covered by star analysts. H5b: All else being equal, the positive association between analyst coverage and crash risk is less pronounced when firms are covered by analysts from high reputation investment banks and brokerage firms. 3. Sample development, variable measurement, and research design This section presents the empirical methods, sample selection, and variable definitions The sample We use two sources to construct our variables: China Stock Market and Accounting Research (CSMAR) database and the Wind Financial Database (WindDB). For analyst coverage, we use analyst recommendations and analyst earnings forecasts as the metrics. Although CSMAR has tracked analyst recommendations and earnings forecasts since 2001, its coverage in the first two

13 years is sparse and the number of analysts contributing recommendations or earnings per share (EPS) forecasts to the database significantly increases in Consequently, we start collecting data from 2003 and continue through the end of The stock return data and financial data for the covered firms are extracted from CSMAR and WindDB. We have 10,227 firm-year observations in the period from 2003 to We then exclude: (1) firms with fewer than 30 weeks of stock-return data; (2) B-share stocks 5 ; (3) financial services firms; (4) firm-year observations with insufficient financial data to calculate control variables; and (5) firm-year observations without analyst coverage. We are left with a final sample of 3,416 (using analyst recommendations) and 3,239 firm-year observations (using analyst earnings forecasts). We present the details of the sample development and the fiscal years of the sample in Table 1. <Table 1> 3.2. Measuring firm-specific crash risk We construct two measures of crash risk following Chen, Hong, and Stein (2001) and Kim, Li and Zhang (2011a, b). We first estimate firm-specific weekly returns, denoted by W, as the natural log of one plus the residual return from the expanded market model regression for each firm and year: ri, t= α + β1 irm, t 2 + β2 irm, t 1+ β3 irm, t+ β4 irm, t+ 1+ β5 ir m, t+ 2 +ε i, t (1) where r i,t is the return on stock i in week t, and r m,t is the value-weighted A-share market return in week t. The firm-specific weekly returns for firm i in week t is represented by W i,t = Ln(1+ε i,t ), where ε i,t is the residual in Equation (1). 5 China trades both A-share (denominated in local currency) and B-share (denominated in Hong Kong dollars or US dollars) stocks. The trading volume for B-share stocks has been low since

14 The first measure of crash risk is the negative coefficient of skewness, NCSKEW, is calculated by taking the negative of the third moment of firm-specific weekly returns for each sample year and dividing it by the standard deviation of firm-specific weekly returns raised to the third power. Specifically, for each firm i in year t, the NCSKEW is ( 1) 32 3 ( 1)( 2)( 2 ) 32 (2) NCSKEWit, = n n Wit, n n W it, where n is the number of observations on firm-specific weekly returns of firm i during year t. In addition to NCSKEW, the second measure of crash risk is down-to-up volatility, DUVOL, which is computed as follows. For any stock i in year t, we separate all the weeks with firm-specific weekly returns below the annual mean ( down weeks) from those with firm-specific weekly returns above the period mean ( up weeks), and compute separately the standard deviation for each of these subsamples. We then take the log of the ratio of the standard deviation on the down weeks to the standard deviation on the up weeks. Thus we have: DUVOL = Ln ( n 1) W 2 ( n 1) W 2 it, u it, d it, Down Up (3) where n u and n d are the number of up and down weeks, respectively. Both NCSKEW and DUVOL are used in the crash risk literature Measuring analyst optimism Recommendations and earnings forecasts are two key elements of analysts research reports. Prior literature documents that analysts tend to be optimistic in recommendations and forecasts. Following prior literature, we measure analyst optimism in two ways: stock recommendations

15 and earnings forecasts. Because analyst recommendations and earnings forecasts data come from different databases in CSMAR, the number of records of analyst recommendations and earnings forecasts are different. We exclude from our two datasets any records with anonymous analysts employers, records without issuance date of recommendation or forecast, and records without a standardized rating or forecasted EPS. If an analyst issues more than one recommendation or earnings forecast in a calendar year, we keep the most recent record before the end of the fiscal year. Following Lin and McNichols (1998), we first change the recommendations data numerically with values from -2 to 2, where Strong sell is coded as -2, Sell is coded as -1, Hold is coded as 0, Buy is coded as 1, Strong buy is coded as 2. We denote Rankscore i,j,t as the numerical recommendations issue by Analyst j for firm i in year t. Then for each firm and year, we calculate the average value of Rankscore i,j,t for different analysts, denoted by Rankscore i,t. If Rankscore i,t locates above the median value in year t, the first variable to measure analyst optimism, Optimism_Recommend, is equal to one, and zero otherwise. Following Jackson (2005), we construct the second analyst optimism variable using earnings forecasts issued by analysts. Specifically, this analyst optimism variable is ( ) Opti, jt, = Fi, jt, Ait, Pit, 1 (4) where F i, j,t is Analyst j s earnings forecast for firm i in year t; A i,t is the actual earnings per share for firm i; and P i,t 1 is the stock price at the end of the last fiscal year. Then for each firm and year, we calculate the average value of Opt i,j,t for different analysts, denoted by Opt i,t. The

16 Optimism_ Forecast is an indicator variable that equals one if Opt i,t is above zero, and zero otherwise. 3.4 Measuring analyst coverage We measure the intensity of analyst activity as the number of analysts who issued recommendations or earnings forecasts for a firm during a given calendar year, denoted by Analyst_Recommend and Analyst_Forecast respectively. We further classify different types of analyst coverage according to analysts different affiliations or characteristics, as shown below: (1) Investment bank and non-investment bank analyst coverage: If an investment bank has underwriting business during year t with the covered firm, then the number of analysts from the investment bank that cover the firm are denoted as investment bank analyst coverage (Analyst_IB) from year t on. The difference between analyst coverage and investment bank analyst coverage is the number of non-investment bank analysts (Analyst_NIB). (2) Top 10 and non-top 10 brokerage firms analyst coverage: WindDB has collected commission revenue for each brokerage firm. For each year, we sort brokerage firms in descending order according to their commission revenues. If Analyst j s employer belongs to a top 10 brokerage firm in year t, we regard Analyst j as a top 10 brokerage firm analyst (Analyst_TopBRG). The difference between analyst coverage and top 10 brokerage firm analyst coverage is the number of non-top 10 brokerage firm analysts (Analyst_NTopBRG) in year t

17 (3) Star and non-star analyst coverage: If an analyst is selected by The New Fortune magazine as the best analyst in year t, we regard it as star analyst coverage (Analyst_Star) since year t. 6 The difference between analyst coverage and star analyst coverage is the number of non-star analysts (Analyst_NStar). (4) Top 10 investment bank and non-top 10 investment bank analyst coverage: WindDB also has collected underwriting income for each investment bank. For each year, we sort investment banks in descending order according to their underwriting income. If Analyst j s employer belongs to a top 10 investment bank in year t, we regard Analyst j as a top 10 investment bank analyst (Analyst_TopIB) in year t. The difference between investment bank analyst coverage and top 10 investment bank analyst coverage is the number of non-top 10 investment bank analysts (Analyst_NTopIB). Then we add the suffix _Recommend and _Forecast to the different types of analyst coverage variables to measure the number of analysts that issue recommendations and earnings forecasts, respectively. 3.5 Control variables Following Chen, Hong, and Stein (2001), Hutton, Marcus, and Tehranian (2009), and Kim, Li, and Zhang (2011a, b), we include a set of control variables that are deemed to be potential 6 The selection process is similar to the All American Research Team. To select star analysts in the current year, The New Fortune sent questionnaires covering all industries to institutional investors. The questionnaire does not pre-list any analysts names. Each respondent writes in the names of analysts for whom they wish to vote. If the respondent votes for more than one analyst, the names are ranked. The New Fortune then adds up the scores and identifies the star analyst selections

18 predictors of crash risk. The variable DTURN is the detrended stock trading volume, which is a proxy for investor heterogeneity, or the difference of opinions among investors. The lag NCSKEW variable is the negative skewness of past firm-specific stock returns, which is included to capture the potential persistence of the third moment of stock returns. The variable SIGMA is the standard deviation of past firm-specific stock returns. RET is the average firm-specific weekly return over the past year. We also include the standard control variables such as SIZE, defined as the firm s log of total assets; MB, defined as the ratio of the market value of equity to the book value of equity; LEV, defined as the book value of all liabilities scaled by market value of assets; and ROA, defined as income before extraordinary items divided by lagged total assets. We further control one information transparency variable (ABACC) in our analysis, which is defined as discretionary accruals that are estimated from the modified Jones model (Dechow, Sloan, and Sweeney, 1995). Prior studies find that DTURN, lag NCSKEW, SIGMA, RET, SIZE, MB, and ABACC are positively related to future crash risk, while LEV and ROA are both negatively related to future stock crash risk. We also include industry and year dummies to control for industry and time fixed effects. The detailed variable definitions are in the Appendix Empirical models Analyst coverage and crash risk To investigate the effect of analyst coverage on stock crash risk, we first estimate the following regression: CrashRisk = α + β Analyst + γ ControlVariables +ε, it, 1 it, 1 it (5) In Equation (5), the stock crash risk is proxied by NCSKEW or DUVOL, and the analyst coverage is proxied by Analyst_Recommend or Analyst_Forecast. The dependent variable is measured in

19 year t, while most of the independent variables are measured in year t-1 (except for ROA, which is measured in year t). We also include a series of control variables (discussed in the prior section) Analyst coverage, analyst optimism and stock crash risk To test whether analyst optimism is the main mechanism that leads to the relation between analyst coverage and crash risk, we run the following regression: CrashRisk = α + β Analyst + β Analyst Optimism + β Optimism + it, 1 it, 1 2 it, 1 t 1 3 t 1 γ ControlVariables + ε it, (6) In Equation (6), Optimism is an indicator variable proxied by Optimism_ Recommend or Optimism_ Forecast. The control variables are the same as in Equation (5). If H2 holds, β 2 in Equation (6) will be positive and significant as analyst optimism is more pronounced when firms are covered by more optimistic analysts Conflict of Interest Hypothesis, analyst coverage, and stock crash risk Investment bank analyst coverage, non-investment bank analyst coverage, and stock crash risk To test how investment bank analyst optimism relates to stock crash risk (H3a), we classify analyst coverage into investment bank analyst coverage (Analyst_IB) and non-investment bank analyst coverage (Analyst_NIB). Then we run the following regression: CrashRisk = α + β Analyst _ IB + β Analyst _ NIB + γ ControlVariables +ε it, 1 it, 1 2 it, 1 it, (7)

20 In Equation (7), Analyst_IB and Analyst_NIB are proxied by Analyst_IB_Recommend (Analyst_IB_Forecast) and Analyst_NIB_Recommend (Analyst_NIB_Forecast), respectively Top 10 brokerage firm analyst coverage, non-top 10 brokerage firm analyst coverage, and crash risk To test whether analyst optimism due to brokerage pressure relates to stock crash risk, we classify analyst coverage into analyst coverage from top 10 brokerage firms (Analyst_TopBRG) and analyst coverage from non-top 10 brokerage firms (Analyst_NTopBRG). Then we run the following regression: CrashRisk = α + β Analyst _ TopBRG + β Analyst _ NTopBRG + it, 1 it, 1 2 it, 1 γ ControlVariables + ε it, (8) In Equation (8), Analyst_TopBRG and Analyst_NTopBRG are proxied by Analyst_TopBRG_Recommend (Analyst_TopBRG_Forecast) and Analyst_NTobBRG_Recommend (Analyst_NTopBRG_ Forecast), respectively Reputation Hypothesis, analyst coverage, and stock crash risk Star analyst coverage, non-star analyst coverage and stock crash risk To test whether an analyst s personal reputation works as a disciplinary mechanism against analyst optimism (H4a), we classify analyst coverage into star analyst coverage (Analyst_Star) and non-star analyst coverage (Analyst_NStar) and then run the following regression: CrashRisk = α + β Analyst _ Star + β Analyst _ NStar + it, 1 it, 1 2 it, 1 γ ControlVariables + ε it, (9) In Equation (9), Analyst_Star and Analyst_NStar are proxied by Analyst_Star_Recommend

21 (Analyst_Star_Forecast) and Analyst_NStar_Recommend (Analyst_NStar_Forecast), respectively Top 10 investment bank analyst coverage, non-top 10 investment bank analyst coverage, and crash risk To further investigate whether investment bank reputation affects the relation between analyst coverage and crash risk, we classify analyst coverage into top 10 investment bank analyst coverage (Analyst_TopIB) and non-top 10 investment bank analyst coverage (Analyst_NTopIB), and then run the following regression: CrashRisk = α + β Analyst _ TopIB + β Analyst _ NTopIB + it, 1 it, 1 2 it, 1 β Analyst _ NIB + γ ControlVariables +ε 3 it, 1 it, (10) In Equation (10), Analyst_TopIB and Analyst_NTopIB are proxied by Analyst_TopIB_Recommend (Analyst_TopIB_Forecast) and Analyst_NTopIB_Recommend (Analyst_NTopIB _Forecast), respectively. 4. Empirical results 4.1. Descriptive statistics Table 2 presents the descriptive statistics of the sample. Both of the two stock price crash risk measures, MCSKEW and DUVOL, share similar profiles in terms of mean, standard deviation, and median. At the analyst level, the mean for analyst optimism measures, Rankscore and Opt, are and 0.011, which suggests that the average analyst recommendation is close to 1 ( buy recommendation) and the analyst earnings forecast is higher than actual earnings. These statistics are intuitively consistent with notion that analysts, in general, have optimism bias about

22 their recommendations and earnings forecasts. At the firm level, for analyst optimism measures (Optimism_Recommend and Optimism_Forecast), 51.6% of the analysts recommendations are above the median, and 55.7% of the analysts forecasts are larger than zero. < Table 2> 4.2. Analyst coverage and crash risk We present the results of the impact of analyst coverage on stock price crash risk in Table 3. The combinations of NCSKEW and DUVOL, analyst recommendations, and earnings forecasts offer four different regression models for a robust testing of Hypothesis 1. The coefficients associated with Analyst_Recommend and Analyst_Forecast are positive and significant at 1% levels, suggesting stock price crash risk is positively correlated with an increase in analyst coverage in China. The findings offer support for H1b, not H1a. The result of not supporting H1a (regarding the information role of analysts) is consistent with Chan and Hameed (2006), who find that analysts collect more market-wide information rather than firm-specific information in an emerging market. Thus, analyst coverage does not reduce stock price crash risk. Given the general optimism bias of analyst recommendations and earnings forecasts, an increase in analyst coverage is associated with an increase in stock price crash risk at each firm. The signs of control variables in Table 3 are consistent with prior literature on stock price crash risk. < Table 3> 4.3. The effect of analyst optimism on the relation between analyst coverage and crash risk The results in Table 3 show a more general finding. To be precise about the association between analyst optimism and stock price crash risk, we examine the joint impact of analyst coverage (recommendation or earnings forecast) and optimism bias. We incorporate two additional variables: Optimism_Recommend and Analyst_Recommend*Optimism_Recommend (also the

23 Optimism_Forecast and Analyst_Forecast*Optimism_Forecast) into the base models in Table 3 to examine how the interaction of analyst recommendation and optimism bias relate to stock price crash risk. Among the four models, the interaction terms (Analyst_Recommend*Optimism_Recommend and Analyst_Forecast*Optimism_Forecast) are positive and three of them are statistically significant at 1% or 5% levels. While the Analyst_Recommend and Analyst_Forecast variables continue to be positive and significant in three out of the four models, the magnitude of the coefficients are smaller than the interaction terms. For instance, the Analyst_Forecast variable in Model 2 has a coefficient of while the Analyst_Forecast*Optimism_Forecast variable has a coefficient of The findings in Table 4 support Hypothesis 2, i.e., the positive association between analyst coverage and crash risk is more pronounced when firms are covered by more optimistic analysts. The signs of control variables in Table 4 are the same as those in Table 3. < Table 4> 4.4. Conflict of interest, analyst coverage, and crash risk To show potential conflict of interest among analyst coverage, we present some summary statistics for analysts in investment banks vis-à-vis non-investment banks as well as top-10 brokerage firms vis-à-vis non-top-10 brokerage firms. The results are included in Table 5, Panels A and B. In both Panels, the optimism bias is present in the proportion of average recommendations greater than 1 and the proportion of average forecast errors greater than zero. In Panel A of Table 5, there are 33.2% optimism recommendations made by investment bank analysts while only 14.7% of non-investment bank analysts made optimism recommendations. For positive forecasting errors, investment bank analysts made them 55.8% of the time, while non-investment bank analysts only made them 52.1% of the time. The differences between the

24 two groups of analysts are significant at 1% or 5% levels. In Panel B of Table 5, the findings for the top-10 brokerage firms and non-top brokerage firms are similar; non-top brokerage firms show more optimism bias than those in top-10 brokerage firms. < Table 5> The effect of investment bank affiliation We examine the Conflict of Interest Hypothesis with respect to investment bank affiliation of analysts in more detail. The findings are shown in Table 6, Panel A, Models 1 to 4. The investment bank analyst variables (Analyst_IB_Recommend and Analyst_IB_Forecast) in all four models are positive and significant at the 1% level while the non-investment bank analyst variables (Analyst_NIB_Recommend and Analyst_NIB_Forecast) are not significant, suggesting the stock price crash risk is positively associated with investment bank analyst coverage. Our findings support Hypothesis The effect of brokerage affiliation Panel B of Table 6 presents the Conflict of Interest Hypothesis with respect to brokerage revenue. Models 5 to 8 show the results for four different combinations of top-10 vis-à-vis non-top 10 brokerage revenue analysts. The findings in Models 5 to 8 consistently show that analyst coverage by non-top-10 brokerage revenue is associated with a higher stock price crash risk. The results support Hypothesis 4b. Analysts from brokerage firms with low revenue face greater pressure to generate more brokerage revenue for their employers, which results in more analyst optimism and greater crash risk in the end. < Table 6> Panel B of Table 6 seems to be in contrast with the finding in Ljungqvist et al. (2007), who find

25 that analysts issue more optimistic recommendations when they work for banks with larger brokerage business. We attribute the difference to two reasons. First, Ljungqvist et al. proxy for the size of the brokerage business using the annual number of registered representatives rather than the brokerage revenue. We argue that brokerage revenue is a better measure of brokerage business and thus brokerage pressure. Second, the emerging stock market in China attracts many smaller firms enter the brokerage business. The strong competition is no doubt putting up pressure for their analysts to issue optimism bias recommendations and earnings forecasts The Reputation Hypothesis, analyst coverage, and stock price crash risk The effect of personal reputation Panels C and D of Table 5 show possible analyst optimism bias due to personal reputation. In terms of the recommendations made, there are no significant differences between star vis-à-vis non-star analysts and between top-10 investment bank vis-à-vis non-top-10 investment bank analysts. However, in terms of average earnings forecast errors, both star and top-10 investment bank analysts show significantly less bias relative to their counterparts. An analyst s personal reputation (in terms of star status or his affiliation to a reputable investment bank) does matter. We present the impact of an analyst s personal reputation on the stock price crash risk in Table 7, Panel A, Models 1 to 4. All four models show positive and 1% significant coefficients in non-star analysts for their recommendations and earnings forecasts. In contrast, all four coefficients associated with star analysts are not significantly related to stock price crash risk

26 The findings support H5a, i.e., an analyst s personal reputation does matter. Non-star analyst recommendations and earnings forecasts contribute to higher crash risk. < Table 7> The effect of investment bank reputation The results for the impact of investment bank reputation on stock price crash risk are shown in Table 7, Panel B, Models 5 to 8. Because there are investment bank and non-investment bank analysts, we include top-10 investment bank, non-top-10 investment bank, and non-investment bank analysts in these models. For investment recommendations (in Models 5 and 7), the impact of analyst coverage at top-10 investment banks and non-investment banks are not significant while those of non-top-10 investment banks is positive and significant. Hence, the findings in Models 5 and 7 support H5b. In terms of analyst earnings forecasts, only the impact of analyst coverage in non-investment banks shows no significant coefficients. Both top-10 and non-top-10 investment bank analyst forecasts show 1%, 5%, or 10% significant coefficients. Nonetheless, the magnitude of coefficients for non-top-10 investment bank coefficients is larger than those of top-10 investment bank coefficients. The extent of the impact for non-top-10 investment bank analyst coverage on crash risk is more pronounced. Thus, the investment bank s reputation does matter, i.e., analyst earnings forecasts at reputable investment banks contribute a smaller magnitude of crash risk relative to those of less reputable investment banks. 5. Robustness checks with sub-sample analysis In this section, we use sub-sample regressions to examine whether the impacts of analyst

27 optimism on the relation between analyst coverage and crash risk are different for firms with different characteristics. The sub-sample analyses also disentangle the possible masking effect of analyst optimism levels and (1) an analyst s investment bank affiliation; (2) an analyst s brokerage firm reputation (in terms of revenue); (3) an analyst s personal reputation; and (4) an analyst s investment bank reputation. To save space, we only report the results that are based on analyst optimism in their recommendations (Optimism_Recommend). The results based on an alternative optimism measure (Optimism_Forecast) are qualitatively similar. Tables 8 and 9 present the results using NCSKEW and DUVOL as the dependent variables, respectively. 5.1 The effect of investment bank business We classify firms into two sub-samples: Yes group and No group, according to whether all analysts come from investment banks, and we then re-estimate Equation (6). Our logic is that if analysts that covered firm i are from investment banks with underwriting services, the competition for underwriting business will make analysts issue more optimistic recommendations or forecasts. However, if at least one of the analysts comes from an employer that does not provide any underwriting services, these non-investment-bank analysts will come under less or no pressure to publish more optimistic research to attract underwriting business. Therefore, we predict that β 2 in Equation (6) will be more pronounced in the Yes group where all the analysts come from investment banks. The results in Panel A of Tables 8 and 9 show the Yes group with positive and significant coefficients and the No group with insignificant coefficients. The results support our prediction and are consistent with the results in Table The effect of brokerage business

28 We classify firms into two sub-samples: the Low (below median) group and the High group (above median), according to whether the proportion of analysts from top 10 brokerage firms is above the median. We then re-estimate Equation (6). If the pressure for analysts to publish optimistic research to stimulate brokerage business is greater at banks with smaller brokerage business, then we predict that the β 2 in Equation (6) will be more pronounced when the proportion of analysts from non-top 10 brokerage firms is high. Consistent with this prediction, the results in Panel B of Tables 8 and 9 show that β 2 in Equation (6) is more pronounced in the Low group, which is consistent with the result in Table 6. This result means that analysts from banks with low brokerage revenue face greater pressure to generate more brokerage revenue for their employers, which results in more analyst optimism and greater crash risk in the end. 5.3 The effect of personal reputation We classify firms into two sub-samples: the Low group (below median) and the High group (above median), according to whether the proportion of star analysts is high or low. We then re-estimate Equation (6). For each firm-year observation that is covered by at least one star analyst, if the proportion of star analysts is above the first quartile value, we classify it into the High group. Otherwise, we classify those observations that are below the first quartile value and the observations that are covered by all non-star analysts into the Low group. We predict that the β 2 in Equation (6) will be more significant when the proportion of star analysts is low if personal reputation can play as a disciplinary mechanism. The results in Panel C of Tables 8 and 9 support this prediction and are consistent with the results in Table The effect of investment bank reputation

29 We classify firms each year into two sub-samples: the Low group (below median) and the High group (above median), according to whether the proportion of analysts from top-10 investment banks is above the median. We then rerun Equation (6). Similarly, we predict that the β 2 in Equation (6) will be more significant when the proportion of analysts from the top 10 investment banks is low if bank reputation can play as a disciplinary mechanism. The results in Panel D of Tables 8 and 9 support this prediction and are consistent with the results in Table Summary We examine the relation among analyst coverage, analyst optimism, and firm-specific future stock price crash risk. We contend that analyst coverage, through their optimistic forecasts and recommendations, can increase the future stock price crash risk of the firms they covered. If analysts are overly optimistic in recommendations and earnings forecasts, the negative information of the firms they cover cannot be timely revealed to the outside investors. When the accumulated negative information reaches a tipping point, it will then be revealed to the market, resulting in the bubble bursting and a stock price crash. Using samples of 3,416 (analyst recommendations) and 3,239 (earnings forecasts) firm-years, our findings suggest that analyst coverage and analyst optimism contributes to future stock price crash risk in China. We find that an increase in analyst coverage for a firm leads to an increase in the future stock price crash risk of the same firm and this positive relation is more pronounced when firms are covered by more optimistic analysts. In addition, the impact of analyst optimism on future stock price crash risk is more pronounced when analysts are from investment banks with underwriting services and employers with small brokerage revenue. In contrast, the impact of analyst optimism on stock price crash risk is less pronounced when analysts have high personal