Abstract
We examine analysts’ forecast behavior in a setting in which CEOs are optimistic and analysts react rationally to CEO optimism. We document that the bias in analysts’ consensus forecasts is negatively related to the level of CEO optimism. The negative relation is stronger for small firms, firms with low analyst followings, and firms with high uncertainty. Analysts revise downward their forecasts for next year’s earnings less relative to their revision for current year’s earnings for firms with more optimistic CEOs, a result consistent with optimistic CEOs are subject to self-attribution bias. The stock price reactions to downward forecast revisions and missing analysts’ forecasts are less negative for firms with optimistic CEOs, indicating that investors understand the implications of CEO optimism for analysts’ forecast bias and subsequent revisions.
Introduction
We examine whether CEO optimism explains the systematic pattern in analysts’ forecast bias and investigate the associated consequences. Prior studies show that analysts’ 8-month-ahead forecasts of annual earnings are upwardly biased and the bias is gradually reduced over the course of the year. Several explanations have been advanced and tested. Some prior studies assume that management is rational and that analysts may be irrational (e.g., DeBondt & Thaler, 1990; Easterwood & Nutt, 1999). In this article, we consider the possibility that CEOs are optimistic and analysts rationally follow management guidance by issuing optimistic earnings forecasts to curry favor with management. 1
Under the CEO optimism hypothesis, CEOs express an optimistic view about their companies to equity analysts, as well as other stakeholders. For CEO optimism to have an effect on analysts’ forecasts, it must be costly for analysts not to go along with management guidance. The analyst literature suggests that analysts have incentives to issue upwardly biased forecasts to maintain access to management (Lim, 2001) or to attract investment-banking business (Lin & McNichols, 1998). Hence, analysts make optimistic forecasts, especially early in the year when the benefit of maintaining a good relationship with management is high and the cost of issuing optimistic earnings forecasts is low. Subsequently, analysts revise their forecasts downward as more objective information, such as quarterly earnings reports, arrives throughout the year. As a result, analysts’ forecasts of annual earnings are upwardly biased early in the year, but the bias is gradually reduced over the year.
Optimistic CEOs exist for various reasons. On one hand, companies prefer optimistic CEOs, because they help promote the company’s brand, appeal to customers, motivate employees, and attract investors. On the other hand, CEOs tend to be optimistic because of self-attribution bias developed from rising through the ranks in the corporate world. Anecdotal evidence supports the view that CEOs tend to make brash predictions about earnings and earnings growth (see, for example, Loomis, 2001). Furthermore, CEO optimism is more likely to exist in equilibrium than analyst irrationality, because the executive labor market is less efficient than the analyst labor market. The cost of removing optimistic CEOs is high, especially if they are entrenched. Costly corporate takeovers might be the only way to remove optimistic CEOs (Heaton, 2002). The CEO optimism argument also explains why management would tolerate optimistic forecasts that firms cannot beat, because the CEOs are the source of the optimism bias.
We consider CEOs to be optimistic if they purchase more of their own company stock rather than diversifying their holdings on the basis of standard portfolio theory. We find that consensus analyst forecasts at 8 months before the fiscal year-end are on average biased upward, and the forecast bias (actual earnings minus forecasts scaled by lagged stock price) is negatively related to the level of CEO optimism. These results hold after controlling for other determinants of analysts’ forecast bias and for the possibility that analysts might be irrational, as documented in prior literature. We also find that the relation between analysts’ forecast bias and CEO optimism is stronger for small companies and firms with volatile earnings and stock returns, a result consistent with the notion that analysts have more difficulty collecting data on firms operating in a poor or uncertain information environment and thus have stronger incentives to curry favor with management.
Second, we examine the implications of CEO optimism for analysts’ forecast revision. As more objective information arrives throughout the year, analysts incorporate this information into their forecasts and realistically revise their forecasts downward. Given that earnings forecasts are more optimistic for firms with optimistic CEOs in the first place, it is not surprising to see that analysts’ forecast revision is negatively related to CEO optimism. A more interesting implication of CEO optimism is that analysts would revise next year’s earnings forecasts less downward for firms with optimistic CEOs. The results are consistent with this conjecture, suggesting that CEO optimism is habitual and reflects personal characteristics, so that optimistic CEOs remain upbeat about their companies’ futures. We obtain the aforementioned findings with several firm performance and growth measures included in the regression equations to control for the possibility that recent good performance and growth induce both CEO optimism and analyst optimistic bias.
Finally, we investigate the capital market implications of the analysts’ forecast bias induced by CEO optimism. We find that the stock price reaction to downward forecast revisions and to missing earnings targets is less negative for firms with more optimistic CEOs, indicating that investors understand the implications of CEO optimism for analysts’ forecast bias and subsequent revisions. In other words, analysts’ optimistic forecast bias can exist in an efficient stock market (Baker, Ruback, & Wurgler, 2004).
We consider a number of alternative explanations and conduct several robustness checks. First, we verify that the key findings are not due to reverse causality by showing that non-executives’ trading behavior has no association with analysts’ forecast bias, consistent with analysts having no incentives to curry favor with non-executive insiders. Second, we address endogeneity by examining CEO turnover and find that analysts’ earnings forecasts become more optimistic after an optimistic CEO replaces a non-optimistic CEO, but become less optimistic after a non-optimistic CEO replaces an optimistic CEO. Third, we rule out insider information-driven trades as an explanation, because any inside information-based measurement error in our proxy for CEO optimism actually biases against finding support for our hypothesis. Fourth, we control for the possibility that CEO optimism simply capturing future information that induced analyst optimistic bias by including past sales growth into the regressions. Furthermore, our stock return-based test results are inconsistent with analysts and investors overweighting past growth information. Finally, we consider two alternative proxies for CEO optimism: one based on an indicator variable version of the original proxy and another constructed from management earnings forecasts. We show that the results remain qualitatively unchanged.
This article contributes to two strands of literature in accounting and finance. First, we contribute to the analyst forecast literature by considering CEO optimism as the source of analyst forecast bias. The literature typically attributes analysts’ upward forecast bias to analysts’ behavioral biases (e.g., DeBondt & Thaler, 1990; Easterwood & Nutt, 1999) or to incentives faced by analysts (Jackson, 2005; Lim, 2001; Lin & McNichols, 1998; Mehran & Stulz, 2007). Instead of taking managerial rationality as a given, we consider CEO optimism as the source of the bias in analyst forecasts. The CEO optimism argument explains why management would tolerate optimistic forecasts that firms cannot beat: The CEO acts as the source of the optimism bias. Second, we contribute to the behavioral corporate finance literature (Baker et al., 2004). Researchers have studied CEO optimism to understand corporate investing and financing decisions (Heaton, 2002; Malmendier & Tate, 2005; Malmendier, Tate, & Yan, 2011; Roll, 1986). We contribute to this literature by providing evidence consistent with CEO optimism affecting the forecasts of security analysts. Moreover, we develop and validate an insider-trading-based proxy for CEO optimism. Unlike the proxies used in prior studies, our proxy can be easily computed for a larger sample of CEOs and over a longer sample period.
The article proceeds as follows. “Hypothesis Development” section develops testable hypotheses. “Data and Variable Construction” section describes the sample and variable construction. “Empirical Analysis and Findings” section presents the research design and empirical findings. “Additional Analyses and Discussions” section considers alternative explanations for the key findings. “Conclusion” section concludes.
Hypothesis Development
Top executives are likely to be optimistic for two reasons. 2 First, from the rational perspective, executives are self-selected to be optimistic to promote their companies’ brand, appeal to customers, motivate employees, and attract investors. In this case, optimistic CEOs exist in equilibrium because their optimism is beneficial to shareholders. Second, from the behavioral perspective, senior executives tend to be optimistic because of self-attribution bias acquired from rising through the ranks in a tournament-like setting. As a result, they tend to attribute good outcomes to their own actions and poor results to uncontrollable causes (e.g., see Bettman & Weitz, 1983; Clapham & Schwenk, 1991; Staw, McKechnie, & Puffer, 1983). Even if CEO optimism is suboptimal for a firm, optimistic CEOs exist in equilibrium because the cost of removing them is high. The existence of optimistic CEOs seems to be more likely than that of irrational analysts given that the executive labor market is arguably less efficient than the labor market for security analysts. 3 Hribar and Yang (2006) show that overconfident CEOs are more likely to issue overly optimistic earnings forecasts. Anecdotal evidence also supports the view that CEOs tend to make brash predictions about earnings and earnings growth (see, for example, Loomis, 2001).
The analyst forecast literature documents systematic evidence that analysts’ forecasts of annual earnings are upwardly biased, and that the bias is gradually reduced over the fiscal year (e.g., see O’Brien, 1988). The managerial optimism hypothesis predicts that optimistic CEOs induce the analysts’ forecast bias.
Under the CEO optimism hypothesis, CEOs express an optimistic view of their companies, communicate this view to analysts and, hence, potentially induce biases in analysts’ earnings forecasts. However, knowing that CEOs are optimistic, analysts should discount the information provided by management and issue unbiased forecasts. For CEO optimism to have an effect on analysts’ forecasts, it must be costly for analysts not to go along with management guidance. The risk of issuing an unbiased forecast lies in the possibility of losing access to management and/or the potential loss of future investment-banking business. For most analysts, company management is the most important source of information about a company, its competitors, and the industry. Hence, even if analysts know they are dealing with optimistic CEOs, they will still follow management guidance to maintain a good working relationship with management in the future, especially early in the year when the cost of issuing optimistic forecasts is relatively small. This above reasoning leads to the first testable hypothesis (stated in the alternative form) as follows:
The CEO optimism hypothesis also explains some cross-sectional variations in the relationship between CEO optimism and analyst forecast bias. If a company is operating in a poor or uncertain information environment, analysts are likely to rely more on management and, hence, have a stronger incentive to go along with management guidance. As a result, CEOs tend to exert more influence on analysts’ forecasts, suggesting that the impact of CEO optimism on analyst bias is stronger if the information environment of the followed company is poor or uncertain. This hypothesis is given as follows:
The CEO optimism explanation has implications for subsequent forecast revisions. As more information about the company arrives over the fiscal year (e.g., quarterly earnings announcements), analysts will incorporate the factual evidence into their forecast revisions, resulting in a gradual decline in analysts’ forecast bias over the course of the year. As biases in analysts’ forecasts are stronger for firms with optimistic CEOs, it follows that analysts’ forecast revisions are also greater for these firms. CEOs will also use the same information to update their expectations for the year. As CEO optimism is habitual and reflects personal characteristics, CEOs would remain optimistic about their firms’ longer term performance even when they revise near-term earnings downward (e.g., see Bettman & Weitz, 1983; Clapham & Schwenk, 1991; Staw et al., 1983). As a result, the downward revision of analysts’ forecasts of next year’s earnings will be smaller than that of the current year’s earnings for firms with optimistic CEOs. This discussion leads to the following hypothesis:
Finally, investors understand the implications of CEO optimism for analysts’ forecast bias if it is an optimal equilibrium outcome. We expect the market to incorporate the impact of CEO optimism on analysts’ forecast bias because the capital market is more efficient than the labor market for executives. In other words, optimistic CEOs can exist in an efficient capital market if the cost of arbitraging them away is high (Heaton, 2002). However, investors should know that for companies with optimistic CEOs (a) analysts’ forecasts are upwardly biased early in the year and revised downward gradually over the year and (b) the probability of missing analysts’ forecast at earnings announcement is higher, all else being equal. As a result, the stock price effects of analysts’ downward revisions and missed analysts’ forecasts should be smaller for companies with optimistic CEOs. This line of reasoning leads to our last two testable hypotheses:
Data and Variable Construction
The sample is composed of companies with insider-trading data from Thompson Financial and earnings-per-share (E/P) forecasts from I/B/E/S for the years 1993 through 2006. The sample period begins in 1993 because this is the first year in which both earnings forecasts and 1-year lag CEO trading data are available for a large number of observations. 4 Stock price and return data are from Center for Research in Security Prices (CRSP); financial statement data are from Compustat. We adjust all share and price data for stock splits.
The variables of interest are computed as follows. We take the consensus analyst forecast 8 months prior to the fiscal year-end to ensure that analysts have access to the prior year’s annual reports and Securities and Exchange Commission (SEC) filings. Specifically, for firms with a December 31 fiscal year-end, we take the analysts’ forecasts as of April 30 of the year. Analysts’ forecast error is computed as the difference between the actual and forecasted earnings, scaled by the closing stock price at the prior year-end [= (Et−F−8)/Pt−1]. 5 Similarly, the forecast revision is computed over the 8-month period from April 30 through December 31 of the year.
We measure CEO optimism by inferring it from CEO trading behavior. Specifically, we consider CEOs to be optimistic if they systematically buy more company shares than they sell, on the premise that, according to standard portfolio theory, risk-averse CEOs should sell their own company’s stocks to diversify away company-specific risk. Hence, we proxy for CEO optimism using net purchase, NB, which is a modified version of the variable Net Buy in Malmendier and Tate (2005). 6 We compute NB as the number of shares purchased minus the number of shares sold in the past 12 months, scaled by the number of shares outstanding at the beginning of the year. 7 The empirical results are robust to calculating NB over the past 3- or 5-year period. NB is measured over the previous fiscal year(s) before we compute the analysts’ forecast error and revisions to ensure that analysts’ forecasts do not induce CEO trading. Prior literature also shows that insider trading captures inside information. Note that any inside information captured in NB works against our hypothesis. This is because, if a CEO buys shares owing to positive inside information and analysts do not fully incorporate such information, analyst forecast error (actual earnings minus forecast) is likely to be higher for firms with higher NB (see “Alternative Explanations for the Main Results” section for more discussion).
CEO purchases and sales are partly driven by stock-based compensation, which is likely to be related to the firm’s growth opportunities. To purge the effect of stock-based compensation from NB, we include only open-market purchases (i.e., we exclude shares acquired from stock option exercises and stock grants). 8 Furthermore, as insider sales are partially driven by insiders unloading their stock-based compensations, we compute adjusted sales by subtracting the amount of stock acquired from stock grants and option exercises from open-market sales. This approach assumes that insiders unload the new shares obtained from stock-based compensation in the same year. Hence, we restrict adjusted sales to be non-negative.
For each CEO trading transaction, we require the availability of the transaction price and the number of shares purchased or sold. As insiders frequently report incorrect or erroneous transaction codes, we require each observation in our sample to be reasonably clean. Thomson Financial cleans the data to some degree on the basis of certain reasonable assumptions, and indicates the level of confidence concerning the accuracy of a particular record in a variable called “Cleanse Indicator.” We exclude observations if the Cleanse Indicator is “S” (no clean attempted) or “A” (numerous data elements were missing or invalid).
Table 1 reports descriptive statistics on our key variables of interest. Panel A shows that the 8-month-ahead forecast error scaled by beginning stock price, FE, has a mean of −0.015 and a median of −0.001, which are consistent with prior evidence of upward biases in analysts’ forecasts. The mean (median) forecast revision of the current year’s earnings over the last 8 months of the fiscal year, REV1, is −0.009 (−0.001). The average (median) CEO net purchase, NB, is −0.516 (0.000), which indicates that on average insiders sell more shares than they purchase. However, at least 25% of insiders are net buyers of company stocks.
Descriptive Statistics.
Note. Pearson coefficients are above the diagonal; Spearman are below, in panel B. The sample includes 16,407 firm-year observations with non-missing FE and NB from 1993 through 2006. Forecast error, FE, is calculated as actual earnings minus analyst median forecast from 8 months before a firm’s fiscal year-end, scaled by prior year-end stock price. NB is CEO net purchase (in %), measured as the number of shares purchased minus the number of shares sold, scaled by year-beginning outstanding shares. We subtract the number of shares received via stock grants and option exercises from the number of shares sold and measure NB over 12 months in the prior year. REV1 is the analyst forecast revision from 8 months before fiscal year-end to fiscal year-end, scaled by the prior year-end stock price. MV is the market value of equity at the end of the previous year. COV is analyst coverage 8 months before a firm’s fiscal year-end. EVOL is earnings volatility, measured as the standard deviation of ROA in the previous 5 years. RETVOL is return volatility, measured as the standard deviation of monthly stock returns over the 12-month period prior to the earnings forecast date. ROA is earnings scaled by average total assets in the previous year. EP is the earnings to price ratio at the end of the prior year. BM is the book to market ratio, measured as the book value of equity divided by the market value of equity at the end of the previous year. SGR is sales growth in the previous year. ΔPPE is changes in net property, plant, and equipment scaled by average total assets in the previous year. RET12 is the 12-month buy-and-hold return up to the analyst forecast date. ACC is accruals at the end of the previous year, measured as net income minus cash flow from operations scaled by lagged assets. All variables except COV are winsorized at the 1st and 99th percentiles.
The market value of equity averages US$3.1 billion, with a range from US$15.2 million to US$68.3 billion. An average of 7.79 analysts follows each sample firm (COV). The 12-month holding period return measured up to 8 months before the fiscal year-end, RET12, is on average 23.6% (median of 11.1%). Statistics not reported in the table indicate that the number of firms ranges from 228 to 1,692 per year over the sample period of 1993-2006, giving a maximum of 17,034 firm-year observations.
Panel B of Table 1 reports the Pearson and Spearman correlation coefficients of the variables. Forecast errors, FE, are positively correlated with forecast revisions, REV1, with a correlation coefficient of 81.1%. NB is negatively correlated with both FE and REV1, with Pearson correlations of −.064 and −.056, respectively.
Empirical Analysis and Findings
Portfolio-Level Univariate Analysis
Table 2, panel A, provides summary statistics on analysts’ forecast errors by forecast horizon. Consistent with the results of O’Brien (1988) and others, the mean forecast errors are negative and monotonically increasing toward 0 over the last 8 months of the fiscal year. More specifically, the mean 8-month-ahead forecast error is −0.0213, which is cut by almost half to −0.0110 3 months before the year-end and is further reduced by another 50% to −0.0053 at fiscal year-end. The negative mean forecast errors at the year-end are somewhat misleading, as they are affected by outliers and skewness of the distribution. In fact, the percentage of observations with upward-biased forecasts decreases from 58.07% at 8 months before the year-end to 37.28% before the earnings announcement, suggesting that more than 60% of firms eventually meet or beat analysts’ forecasts. The median forecast error is positive at the fiscal year-end and earnings announcement.
Analysts’ Forecast Errors by Horizon and CEO Optimism.
Note. The sample includes 16,407 firm-year observations with non-missing FE and NB from 1993 through 2006. Forecast error, FE, is calculated as actual earnings minus median analyst forecast, scaled by prior year-end stock price. FE(−t) is analyst forecast error t months before a firm’s fiscal year-end. FE(EA) is analyst forecast error at the earnings announcement. NB is CEO net purchase (in %) to proxy for CEO optimism, measured as the number of shares purchased minus the number of shares sold, scaled by beginning-of-year shares outstanding. We subtract the number of shares received via stock grants and option exercise from the number of shares sold and measure NB over 12 months in the prior year. In panel B, the sample is sorted into five portfolios by NB, where NB1 and NB5 are the bottom and top quintiles, respectively. Forecast errors are winsorized at the 1st and 99th percentiles.
Table 2, panel B, provides descriptive statistics on forecast errors by forecast horizon and the level of CEO optimism, as proxied by NB. We form five NB portfolios. The indicator variable NB1 takes the value of 1 if NB is ranked in the bottom quintile and 0 otherwise. The other indicator variables, NB2, NB3, NB4, and NB5, are defined similarly. Holding NB constant, we find that the mean forecast error increases toward 0 and the percentage of negative forecast errors decreases over the last 8 months of the fiscal year, which closely follows the pattern reported in panel A.
More importantly, holding the forecast horizon constant, we document a negative relation between the proxy for CEO optimism and the bias in analyst forecasts. For example, at 8 months before the year-end, the average forecast error of firms in the bottom NB quintile is −0.0155, compared with −0.0336 for firms in the top NB quintile. At the fiscal year-end, the corresponding figures for the bottom and top quintiles are −0.0038 and −0.0092, respectively. The median forecast errors are virtually identical among the five NB quintiles starting from 2 months before the fiscal year-end. The median firm in each quintile actually beats analysts’ consensus forecast at the earnings announcement. Overall, the univariate statistics suggest that the higher the level of CEO optimism, the larger the bias in analysts’ earnings forecasts.
Implications of CEO Optimism for Analyst Forecast Bias
To test H1 that analysts’ forecasts made early in the fiscal year are influenced by CEO optimism, we estimate the following regression model:
where FE = analyst’s forecast errors, computed as reported E/P minus the median analysts’ E/P forecast made 8 months before fiscal year-end, divided by the closing share price at the prior year-end; NB = insider-trading-based proxy for CEO optimism, computed as the number of shares purchases minus the number of shares sold, scaled by the number of shares outstanding; SIZE = the logarithm of the market value of equity at the end of year t− 1; COV = the logarithm of the number of analysts following the company; EVOL = earnings volatility, measured as the standard deviation of ROA in the previous 5 years; RETVOL = return volatility, measured as the standard deviation of monthly stock returns over the 12-month period prior to the earnings forecast date; ROA = earnings scaled by average total assets in the previous year; EP = the earnings-to-price ratio at the end of year t− 1; DEP = a dummy variable with the value of 1 if EP is negative and 0 otherwise; BM = the logarithm of the book-to-market ratio in year t; SGR = sales growth in the previous year; ΔPPE = the changes in net property, plant, and equipment scaled by average total assets in the previous year; RET12 = trailing 12-month stock return up to analyst forecast date; ACC = total accruals in year t− 1, measured as net income minus cash flow from operations scaled by lagged total assets.
Our main variable of interest is the proxy variable for CEO optimism, NB. A negative coefficient on NB is consistent with the prediction of H1. Following previous research, we control for other determinants of earnings forecast bias. The literature argues that analysts have incentives to issue optimistic forecasts for firms with poor information environments because they want to please management to secure future access (Lim, 2001). We include firm size (SIZE) and analyst coverage (COV) to capture a firm’s information environment (Atiase, 1987; Zhang, 2006). Prior studies also suggest that CEO optimism and analyst optimistic bias are greater when uncertainty is greater. Hence, we control for earnings volatility, EVOL, and return volatility, RETVOL. To account for the possibility that recent good performance and growth induce both CEO optimism and analyst optimistic bias, we include several firm performance/growth measures: ROA, EP, BM, SGR, and ΔPPE. As the dependent variable is price-scaled forecast error, it is correlated with a firm’s E/P ratio. Failure to control for the E/P ratio may lead to spurious correlation between the dependent variable and our proxy for CEO optimism. The dummy variable DEP allows the relation between analysts’ forecast bias and the earnings-to-price ratio to be nonlinear. Finally, prior literature documents evidence consistent with analysts being not fully rational. To allow for this possibility, we include RET12 to control for analysts’ underreaction to the information embedded in the prior period’s stock returns and ACC to control for the possibility that analysts do not fully understand the implication of accruals for future earnings (Bradshaw, Richardson, & Sloan, 2001; Teoh & Wong, 2002). 9
Table 3 summarizes regression results. In panel A, column 1 indicates that analysts’ forecast error at 8 months before the fiscal year-end is negatively related to NB. In other words, the more optimistic the CEO, the greater the upward bias in analysts’ forecasts. Column 2 shows that the qualitative result remains unchanged when we add the control variables to the regression. The estimated coefficients on NB are −0.002 and −0.001 (both significant at less than the 1% level), respectively, in columns 1 and 2. Compared with the mean forecast bias of −0.0213 (panel A in Table 2), the estimated effects are around 5% to 10% of the mean forecast bias.
Regressions of Analysts’ Forecast Errors on CEO Optimism.
Note. The dependent variable is the forecast error, FE, calculated as the difference between the actual earnings and the median analyst forecast from 8 months before a firm’s fiscal year-end, scaled by the prior year-end stock price. NB is CEO net purchase (in %) to proxy for CEO optimism, measured as the number of shares purchased minus the number of shares sold, scaled by the beginning-of-year shares outstanding. We subtract the number of shares received via stock grants and option exercise from the number of shares sold and measure NB over 12 months in the prior year. SIZE is the logarithm of a firm’s market value. COV is the logarithm of analyst coverage. EVOL is earnings volatility, measured as the standard deviation of ROA in the previous 5 years. RETVOL is return volatility, measured as the standard deviation of monthly stock returns over the 12-month period prior to the earnings forecast date. ROA is earnings scaled by average total assets in the previous year. SGR is sales growth in the previous year. EP is the earnings to price ratio at the end of prior year. DEP is a dummy variable with the value of 1 if EP is negative and 0 otherwise. BM is the book to market ratio, measured as the book value of equity divided by the market value of equity at the end of the previous year. ΔPPE is changes in net property, plant, and equipment scaled by average total assets in the previous year. RET12 is the 12-month buy-and-hold return immediately prior to the analyst forecast date. ACC is accruals, measured as net income minus cash flow from operations scaled by lagged assets. The sample includes 16,407 firm-year observations with non-missing FE and NB from 1993 through 2006. t statistics in parentheses are based on the Fama–MacBeth regression approach with Newey–West adjustment of two lags. All variables except COV are winsorized at the 1st and 99th percentiles.
The control variables exhibit the expected association with analysts’ forecast error. First, analysts issue more optimistic forecasts for small firms and firms with few analysts following them (i.e., firms in poor information environments). Second, both EVOL and RETVOL are negatively related to analysts’ forecast error, indicating that forecast bias is greater for volatile firms. Third, ROA, EP, and ΔPPE exhibit a significant association with analysts’ forecast bias (also included is a dummy variable for observation with negative E/P), suggesting more optimistic analyst forecasts for firms with good operating performance and growth in the past year. However, the statistical negative estimated coefficient on BM is consistent with more optimistic forecasts for low growth firms. Finally, the estimated coefficient on the past 12-month stock return (RET12) is significant and that on ACC is significantly negative, which are consistent with analysts underreacting to the information embedded in the prior stock returns and with analysts not fully understanding the implication of accruals for future earnings, respectively. The negative coefficient on ACC is similar to the results documented in Bradshaw et al. (2001) and Teoh and Wong (2002) that analysts’ earnings forecasts are more optimistic for high-accrual firms.
The results reported in columns 1 and 2 are estimated using the Fama–MacBeth regression method to account for cross-correlation, as well as the Newey–West autocorrelation adjustment of two lags to correct for serial correlation. In columns 3 and 4, we further control for industry-fixed effects by adding industry indicator variables to the regressions. We define industry membership using two-digit Standard Industrial Classification (SIC) codes. The results are almost identical to those reported in columns 1 and 2.
We also conduct a number of robustness checks to ensure that the main results are not sensitive to the construction of the CEO optimism proxy. In particular, CEO stock purchases and sales could be driven by stock-based compensation, which is correlated with growth opportunities and, hence, analyst forecast bias. We attempt to purge such an effect from our proxy for CEO optimism (these adjustments are in addition to those already done, as described above). Panel B in Table 3 reports three sets of robustness checks: (a) we do not adjust open-market sales for stocks acquired through option exercise and stock grants; (b) we capture CEO optimism using only CEO purchases, scaled by the number of shares outstanding; and (c) we include in CEO purchases all kinds of acquisitions (acqdisp = “A”) and in CEO sales all kinds of disposals (acqdisp = “D”). The results are robust to these specification checks, except in column 4 where the estimated coefficient on NB becomes insignificant.
Implications of CEO Optimism and Information Environment
H2 states that the impact of CEO optimism on analysts’ forecast bias is more pronounced for firms operating in a poor or uncertain information environment. Following prior studies (e.g., Zhang, 2006), we use firm size (SIZE) and the number of analysts following the firm (COV) to proxy for a company’s information environment and earnings volatility (EVOL) and analyst forecast dispersion (DISP) to capture uncertainty. Hence, the regression model is given as follows:
where D is an indicator variable that takes the value of 1 for firms operating in a poor or uncertain information environment (i.e., firms in the bottom half of SIZE, bottom half of COV, the top half of EVOL, or the top half of DISP); and 0 otherwise. The rest of the variables are defined as in Equation 1. We predict
Table 4 reports the estimation results from applying the Fama–MacBeth regression method with the Newey–West autocorrelation adjustment of two lags and industry-fixed effects. 10 The estimated coefficients on the interaction terms, NB×D, are significantly negative, expect for that under column 3. The negative coefficient is consistent with a more negative forecast error for small firms, firms with low analyst coverage, and firms with dispersed analyst forecasts, lending support to H2. Prior literature finds that analysts tend to issue more optimistic forecasts for firms with a poor information environment to please management (Lim, 2001). We extend this literature by documenting that a firm’s information environment also interacts with CEO optimism in determining the analyst forecast bias, as suggested by the CEO optimism hypothesis. In particular, firms that operate in a poor or uncertain information environment tend to exhibit more optimistic analyst forecasts, consistent with a higher demand for management access on the part of security analysts, resulting in more pressure to follow the optimistic management guidance.
Regressions of Analysts’ Forecast Errors on CEO Optimism and Information Environment Variables.
Note. The dependent variable is the forecast error, FE, calculated as the difference between the reported earnings and the median analyst forecast from 8 months before a firm’s fiscal year-end, scaled by the prior year-end stock price. NB is CEO net purchase (in %), measured as the number of shares purchased minus the number of shares sold, scaled by beginning-of-year shares outstanding. We subtract the number of shares received via stock grants and option exercise from the number of shares sold and measure NB over 12 months in the prior year. SIZE is the logarithm of a firm’s market value. COV is the logarithm of analyst coverage. EVOL is earnings volatility, measured as the standard deviation of ROA in the previous 5 years. RETVOL is return volatility, measured as the standard deviation of monthly stock returns over the 12-month period prior to the earnings forecast date. D is a dummy variable that takes the value of 1 for firms in the bottom half of SIZE, bottom half of COV, top half of EVOL, or top half of DISP, where DISP is analyst dispersion; and 0 otherwise. ROA is earnings scaled by average total assets in the previous year. EP is the earnings to price ratio at the end of prior year. DEP is a dummy variable with the value of 1 if EP is negative and 0 otherwise. BM is the book to market ratio, measured as the book value of equity divided by the market value of equity at the end of the previous year. SGR is sales growth in the previous year. ΔPPE is changes in net property, plant, and equipment scaled by average total assets in the previous year. RET12 is 12-month buy-and-hold returns up to the analyst forecast date. ACC is accruals, measured as net income minus cash flow from operations scaled by lagged assets. The sample period is from 1993 to 2006, and t statistics in parentheses are based on the Fama–MacBeth regression approach with Newey–West adjustment of two lags. All variables except COV are winsorized at the 1st and 99th percentiles.
Implications of CEO Optimism for Analysts’ Forecast Revisions
We use the following regression model to test H3a that analysts’ forecast revisions for the current year’s earnings are negatively related to the level of CEO optimism:
where REV1 is the analysts’ forecast revision for the current year’s earnings from 8 months before the fiscal year-end to the year-end, scaled by the closing stock price at the prior year-end. The rest of the variables are as defined in Equation 1.
Table 5 summaries the estimation of Equation 3. In particular, it documents a significantly negative association between forecast revisions and NB, thereby supporting H3a. Moreover, the estimated coefficients on the control variables are largely as expected and in line with those reported in Table 3. Taken together, these results suggest that the more optimistic the CEOs, the more pronounced the downward revisions in analyst forecasts over the course of the year.
Regressions of Analysts’ Forecast Revisions for Current Year’s Earnings on CEO Optimism and Other Control Variables.
Note. The dependent variable is the analysts’ forecast revision, REV1, from 8 months before the fiscal year-end to fiscal year-end, scaled by the prior year-end stock price. NB is CEO net purchase (in %), measured as the number of shares purchased minus the number of shares sold, scaled by beginning-of-year shares outstanding. We subtract the number of shares received via stock grants and option exercise from the number of shares sold and measure NB over 12 months in the prior year. SIZE is the logarithm of a firm’s market value. COV is the logarithm of analyst coverage. EVOL is earnings volatility, measured as the standard deviation of ROA in the previous 5 years. RETVOL is return volatility, measured as the standard deviation of monthly stock returns over the 12-month period prior to the earnings forecast date. ROA is earnings scaled by average total assets in the previous year. EP is the earnings to price ratio at the end of prior year. DEP is a dummy variable with the value of 1 if EP is negative and 0 otherwise. BM is the book to market ratio, measured as the book value of equity divided by the market value of equity at the end of the previous year. SGR is sales growth in the previous year. ΔPPE is changes in net property, plant, and equipment scaled by average total assets in the previous year. RET12 is the 12-month buy-and-hold return immediately prior to the analyst forecast date. ACC is accruals, measured as net income minus cash flow from operations scaled by lagged assets. The sample period is from 1993 to 2006, and t statistics in parentheses are based on the Fama–MacBeth regression approach with Newey–West adjustment of two lags. All variables except COV are winsorized at the 1st and 99th percentiles.
H3b predicts that analysts would revise their forecasts for next year’s earnings less downward relative to their forecast revision on current year’s earnings for companies with more optimistic CEOs. The model for testing H3b is
where REV2 (REV1) is the analysts’ forecast revision for next (this) year’s earnings made over the last 8 months of this year, scaled by last year’s closing stock price. The rest of the variables are defined in Equation 1. The test is based on the premise that forecast revisions for current and future years are positively related, except when the reason for this year’s revision is transitory in nature. Because CEO optimism is habitual and reflects personal characteristics, optimistic CEOs would remain optimistic about future performance even when current-year earnings are revised downward. Therefore, we expect the positive relation between forecast revisions for current-year and next-year earnings to be lower for firms with optimistic CEOs (i.e., a negative
Table 6 summarizes the estimation of Equation 4. As expected, both columns 1 and 2 indicate that analyst forecast revisions of current-year and next-year earnings are significantly and positively correlated (estimated coefficients on REV1 of .644 and .624, respectively), suggesting that analysts revise the next-year forecasts in the same direction as their current-year forecasts. However, the significant and negative coefficients on REV1×NB indicate that the higher the level of CEO optimism, the lower the correlation between REV2 and REV1. These results are consistent with the notion that CEOs remain optimistic about their companies’ futures and communicate their belief to analysts, resulting in smaller analyst revisions in longer term earnings forecasts. 11
Regressions of Analysts’ Forecast Revisions for Next Year’s Earnings on Current Year’s Forecast Revisions, CEO Optimism, and Other Control Variables.
Note. The dependent variable, REV2, is analysts’ forecast revision for next year’s earnings from 8 months before this fiscal year-end to this fiscal year-end, scaled by prior year-end stock price. REV1 is forecast revision for this year’s earnings from 8 months before the fiscal year-end to the fiscal year-end, scaled by the prior year-end stock price. NB is CEO net purchase (in %), measured as the number of shares purchased minus the number of shares sold, scaled by the beginning-of-year shares outstanding. We subtract the number of shares received via stock grants and option exercise from the number of shares sold and measure NB over 12 months in the prior year. SIZE is the logarithm of a firm’s market value. COV is the logarithm of analyst coverage. EVOL is earnings volatility, measured as the standard deviation of ROA in the previous 5 years. RETVOL is return volatility, measured as the standard deviation of monthly stock returns over the 12-month period prior to the earnings forecast date. ROA is earnings scaled by average total assets in the previous year. EP is the earnings to price ratio at the end of prior year. DEP is a dummy variable with the value of 1 if EP is negative and 0 otherwise. BM is the book to market ratio, measured as the book value of equity divided by the market value of equity at the end of the previous year. SGR is sales growth in the previous year. ΔPPE is changes in net property, plant, and equipment scaled by average total assets in the previous year. RET12 is the 12-month buy-and-hold return immediately prior to the analyst forecast date. ACC is accruals, measured as net income minus cash flow from operations scaled by lagged assets. The sample period is from 1993 to 2006, and t statistics in parentheses are based on the Fama–MacBeth regression approach with Newey–West adjustment of two lags. All variables are winsorized at the 1st and 99th percentiles.
Capital Market Implications of CEO Optimism
We use the following regression model to investigate H4a that the stock price reaction to analysts’ downward forecast revisions is lower for firms with optimistic CEOs:
where RET = raw return computed over the last 8 months of the fiscal year; REV1 = analysts’ forecast revision for the current year’s earnings from 8 months before the fiscal year-end to the year-end, scaled by the prior year’s closing stock price; NB = insider-trading-based proxy for CEO optimism, computed as shares purchases minus shares sold, scaled by shares outstanding; BETA = stock’s equity beta; SIZE = the logarithm of the market value of equity in year t; BM = the logarithm of the book-to-market ratio in year t; RET12 = past 12-month stock returns up to analysts’ forecast date.
Table 7 documents the regression results. Column 1 shows that forecast revision and contemporaneous stock return are positively correlated, indicating a strong market reaction to analyst forecast revision. Column 2 reports that this result remains unchanged after controlling for beta, size, book-to-market, and return momentum effects. Columns 3 and 4 document test results for whether the valuation effect of forecast revisions varies with the level of CEO optimism. The estimated coefficients on the interaction terms (REV1×NB) are significantly negative under both columns 3 and 4, suggesting that investors understand the implications of CEO optimism for analyst forecast revisions. While the coefficient estimates on BETA, SIZE, BM, and RET12 exhibit the expected signs, only those on SIZE and RET12 are distinguishable from 0. 12
Regressions of Stock Return on Analysts’ Forecast Revisions and CEO Optimism.
Note. The dependent variable, RET, is stock returns from 8 months before the fiscal year-end to the fiscal year-end. REV1 is analysts’ forecast revision for this year’s earnings from 8 months before the fiscal year-end to the fiscal year-end, scaled by the prior year-end stock price. NB is CEO net purchase (in %), measured as the number of shares purchased minus the number of shares sold, scaled by beginning-of-year shares outstanding. We subtract the number of shares received via stock grants and option exercise from the number of shares sold and measure NB over 12 months in the prior year. BETA is the coefficient estimate calculated by regressing monthly excess stock returns on excess market returns over the past 60 months. SIZE is the logarithm of the market value of equity. BM is the logarithm of the book-to-market ratio. RET12 is the trailing 12-month stock return immediately prior to 9 months before the fiscal year-end. The sample period is from 1993 to 2006, and t statistics in parentheses are based on the Fama–MacBeth regression approach. All variables are winsorized at the 1st and 99th percentiles.
Finally, we test H4b that the market reaction to companies missing analysts’ forecasts is less negative for companies with more optimistic CEOs. The regression model is specified as follows:
where ARET = 3-day [−1, +1] market-adjusted abnormal return around earnings announcement, where day 0 is the announcement date; UE = earnings surprise, computed as actual earnings minus the last consensus forecast, scaled by prior fiscal year’s closing stock price; MISS = dummy variable indicating reported earnings lower than analysts’ consensus forecasts (i.e., missed forecasts); ACC = total accrual in year t− 1, measured as net income minus cash flow from operations scaled by lagged total assets.
Table 8 summarizes the estimation of regression Equation 6. As expected, abnormal returns around earnings announcements are positively associated with earnings surprises, UE. Column 2 indicates that the estimated coefficient on MISS is statistically negative, consistent with companies that missed analysts’ forecasts experiencing an additional reduction in stock price. Most important, column 3 shows that the negative effect on stock return for missing analysts’ forecasts is mitigated for companies with optimistic CEOs. In columns 4 through 6, we further control for other variables that are associated with expected stock return. While the results are qualitatively similar to those reported under columns 1 through 3, the estimated coefficient on the interaction term (MISS×NB) is no longer distinguishable from 0.
Regressions of Market Reactions Around Earnings Announcements.
Note. The dependent variable, ARET, is a 3-day [−1, +1] market-adjusted stock return, where day 0 is the earnings announcement date. UE is earnings surprise, computed using the last analysts’ consensus forecast of the fiscal year, scaled by prior year’s closing stock price. MISS is a dummy variable indicating that a company’s reported earnings is less than analysts’ forecast; that is, missed analysts’ earnings expectation. NB is CEO net purchase (in %), measured as the number of shares purchased minus the number of shares sold, scaled by beginning-of-year shares outstanding. We subtract the number of shares received via stock grants and option exercise from the number of shares sold and measure NB over 12 months in the prior year. SIZE is the logarithm of the market value of equity. BM is the logarithm of the book-to-market ratio. RET12 is the trailing 12-month stock return immediately prior to 9 months before the fiscal year-end. ACC is accruals, measured as net income minus cash flow from operations scaled by lagged assets. The sample period is from 1993 to 2006, and t statistics in parentheses are based on the Fama–MacBeth regression approach. All variables, except MISS, are winsorized at the 1st and 99th percentiles.
In summary, we find that the magnitude of analyst forecast bias is positively related to CEO optimism and that the positive correlation is stronger for firms with a poor or uncertain information environment. We also find that the more optimistic the CEOs, the greater the reduction in analysts’ forecasts for the current year’s earnings, but the smaller the reduction in analysts’ forecasts for the next year’s earnings. Finally, we show that the stock price reaction to downward forecast revisions and to missing the earnings target is smaller for firms with more optimistic CEOs, indicating that investors understand the implications of CEO optimism for analysts’ forecast bias and subsequent revisions.
Additional Analyses and Discussions
Alternative Explanations for the Main Results
Reverse causality
Optimistic analysts induce rational CEO buying, because CEOs want to take advantage of the medium-term momentum in stock returns after good-news forecasts. Under this alternative explanation, the causality is in the opposite direction from that hypothesized in the article. We note that insiders cannot take advantage of medium-term momentum, if any such momentum exists, because they are restricted by the SEC short-swing rule (Rule 16b of the Securities Exchange Act of 1934). This rule requires insiders to return any trading profits to the company when a purchase and a sale are made within a 6-month period. Furthermore, our net purchase variable is measured before analysts issue forecasts. Finally, this interpretation takes analyst forecast optimism as a given and, hence, does not explain the origin of the bias in analyst forecasts. While prior studies have examined various behavioral- or incentive-based explanations for analyst optimism, most are silent as to why CEOs would tolerate optimistic forecasts that CEOs cannot meet or beat. The CEO optimism argument provides a plausible explanation for this phenomenon: CEOs act as the source of the optimism bias.
Nevertheless, we cannot completely rule out the possibility of such a reverse causality. To further address the concerns of reverse causality and other potential alternative explanations, we examine the role of trades by non-executive insiders, such as directors and beneficiary shareholders. If CEOs indeed follow analysts’ optimistic views, we expect non-executive insiders to follow analysts’ views as well, given that non-executive insiders are not in a better position to forecast earnings than CEOs. Therefore, the reverse causality argument predicts similar results if we replace insider trades by CEOs with insider trades by non-executives. However, our story predicts an insignificant association between non-executives’ trading behavior and analysts’ optimism because analysts have no incentives to maintain a good relationship with non-executive insiders.
To conduct this sensitivity test, we construct the NB variable for non-executives, who are defined as insiders other than the chairman and company officers. Table 9 reports that the estimated coefficients on non-executive insiders’NB are insignificant using three different estimation methods, lending no support for the reverse causality argument.
Regressions of Analysts’ Forecast Errors on Non-Executive Optimism.
Note. The dependent variable is the forecast error, FE, calculated as the difference between the actual earnings and the median analyst forecast from 8 months before a firm’s fiscal year-end, scaled by the prior year-end stock price. NB is non-executives net buy to proxy for optimism, measured as the number of shares purchased minus the number of shares sold, scaled by the beginning-of-year shares outstanding. Non-executives include those with role code 1 not in (“CEO,”“CFO,”“CB,”“P,”“AV,”“CI,”“CO,”“CT,”“EVP,”“O,”“OB,”“OP,”“OS,”“OT,”“OX,”“S,”“SVP,”“VP”). We subtract the number of shares received via stock grants and option exercise from the number of shares sold and measure NB over 12 months in the prior year. SIZE is the logarithm of a firm’s market value. COV is the logarithm of analyst coverage. EVOL is earnings volatility, measured as the standard deviation of ROA in the previous 5 years. RETVOL is return volatility, measured as the standard deviation of monthly stock returns over the 12-month period prior to the earnings forecast date. ROA is earnings scaled by average total assets in the previous year. EP is the earnings to price ratio at the end of prior year. DEP is a dummy variable with the value of 1 if EP is negative and 0 otherwise. BM is the book to market ratio, measured as the book value of equity divided by the market value of equity at the end of the previous year. SGR is sales growth in the previous year. ΔPPE is changes in net property, plant, and equipment scaled by average total assets in the previous year. RET12 is the 12-month buy-and-hold return immediately prior to the analyst forecast date. ACC is accruals, measured as net income minus cash flow from operations scaled by lagged assets. The sample period is from 1993 to 2006. t statistics in parentheses are based on the Fama–MacBeth regression approach with Newey–West adjustment of two lags. All variables except COV are winsorized at the 1st and 99th percentiles.
Endogeneity issue: Change in analysts’ forecast bias around CEO turnover
We use CEO turnover to document further evidence that optimistic CEOs induce analyst forecast bias. In particular, we test whether analysts’ forecast bias changes after an optimistic CEO replaces a non-optimistic CEO and vice versa. We restrict the sample to new and old CEOs who have insider-trading data at t− 2, where t is the year of the turnover; 154 observations meet this restriction. We compute NB for both new and old CEOs using their trading data at t− 2 and we independently sort them into three groups by NB. We call the CEOs in the top group optimistic CEOs and those in the bottom group non-optimistic CEOs.
In untabulated results, we find that eight companies replaced an optimistic CEO with a non-optimistic CEO. We compare analysts’ forecast bias at t− 1 and t+ 1 and find that analysts’ forecasts become less biased for these eight companies (change = 0.006 and t value = 0.36). More importantly, for nine companies that replaced a non-optimistic CEO with an optimistic CEO, analysts’ forecasts become more optimistic from t− 1 to t+ 1 (change = −0.023; t value = −1.75, which is significant at the 10% level using a one-sided test). As for companies that replaced a non-optimistic CEO with a non-optimistic CEO (n = 29) and for those that replaced an optimistic CEO with an optimistic CEO (n = 30), we find insignificant changes in their analysts’ forecast biases around CEO turnovers. Taken together, these results provide some support for the idea that CEO optimism induces upward biases in analysts’ earnings forecasts, although the sample size is too small to draw any definitive conclusion.
CEO optimism versus inside information
The inside-trading literature suggests that insiders are trading on inside information. Information-driven buying suggests that actual earnings will be higher than the market’s expectation and insiders want to take advantage of such inside information. As a result, insider buying and analysts’ forecast bias (actual earnings minus forecast) should be positively correlated, as opposed to a negative correlation predicted by our CEO optimism story. Therefore, any inside information-based measurement error in our CEO optimism proxy is likely to work against our hypothesis.
CEO optimism capturing future information
Our main results could be attributed to our CEO optimism proxy capturing future information and analysts overweighting this information, leading to a negative relation between the CEO optimism proxy and analysts’ forecast bias. 13 We rule out this alternative explanation for two reasons. First, we include past sales growth (SGR) into regression models (Equations 1-4) to capture the notion that investors extrapolate past growth too far into the future (e.g., Lakonishok, Shleifer, & Vishny, 1994). Our results show that the estimated coefficients on SGR are typically insignificant. Second, our return results are not consistent with the analyst-overweighting argument. Specifically, if analysts overweight growth information, investors will do the same (La Porta, 1996). However, the results documented in Tables 7 and 8 show that investors understand the implications of CEO optimism for analysts’ forecast bias and subsequent revisions.
Alternative CEO Optimism Proxy
CEO optimism proxy as an indicator variable
The construction of the NB (net buy) variable is very similar in spirit to the Net Buyer variable in Malmendier and Tate (2005). We prefer NB over Net Buyer, because it is an interval variable and Net Buyer is an indicator variable. However, an indicator variable version of NB may be appropriate if NB is a noisy measure of CEO optimism. Hence, we repeat our empirical analyses using an indicator variable version of NB. In particular, we create I(NB), where I(NB) = 1 for NB > 0 and I(NB) = 0 otherwise. The results (not tabulated) using I(NB) as our proxy for CEO optimism are qualitatively similar to those using NB.
CEO optimism proxy based on management earnings forecasts
In the main analyses, we use insider-trading behavior to construct the proxy for CEO optimism. While management earnings forecasts could potentially serve as another proxy, these forecasts are voluntary and come in different forms: point, range, and qualitative estimates. Hence, the availability of usable quantitative management forecasts is limited. Overall, there are 4,606 firm-year observations with management forecasts made between month t− 11 and month t− 8 from 1993 to 2005 (we include range estimates by taking the midpoint of the range), compared with 17,034 observations used in our main analyses.
We conduct some robustness checks using management earnings forecast errors (MFEt− 1) as a proxy for CEO optimism. 14 Specifically, we measure MFEt− 1 as management forecasts minus actual earnings scaled by lagged stock price, so a higher value of MFEt−1 corresponds to a higher level of CEO optimism. Given the limited number of observations and the uneven distribution of MFEt− 1 across years, we use ordinary least squares (OLS) regressions with firm-fixed effects. Untabulated analysis shows that the results are qualitatively similar (most results are even stronger), with the notable exception that the coefficient of the interaction term (MFEt− 1×REV1) becomes insignificant in the REV2 regression (Model 2 in Table 6). However, we hesitate to draw strong conclusions from this analysis due to the scarcity and self-selection nature of management earnings forecasts.
Conclusion
This article examines the implications of CEO optimism for the bias in analyst forecasts. CEOs exert influence on analysts’ forecasts via their communication with these analysts. Analysts go along with optimistic CEOs on the basis of their cost/benefit analysis. Using insider-trading behavior to capture CEO optimism, we find that 8-month-ahead consensus forecasts are biased upward, and that the bias (actual minus forecast) is negatively related to the level of CEO optimism. This negative relation is stronger for firms with a poor or uncertain information environment, as these CEOs exert more influence on analysts. We also show that, when analysts revise down their forecasts for the current year’s earnings, their revision of the next year’s earnings forecasts is smaller for firms with more optimistic CEOs, a result consistent with the notion that CEOs remain optimistic about their companies’ futures and influence analysts’ forecast revisions accordingly. Finally, we find that investors understand the effect of CEO optimism on the bias in analysts’ forecasts and show a smaller reaction to the downward revisions in analysts’ forecasts for firms with optimistic CEOs.
Footnotes
Acknowledgements
The authors would like to thank Zahi Ben-David, Sharon Hudson, Lisa Koonce, Ulrike Malmendier, Patricia O’Brien, Jake Thomas, an anonymous reviewer, and workshop participants at Baruch College, Emory University, HKUST Summer Symposium, London Business School, McMaster University, University of British Columbia, University of Texas at Austin, and University of Waterloo for their helpful comments and suggestions. They gratefully acknowledge the contribution of I/B/E/S International Inc. for providing earnings forecast data, available through the International Brokers Estimate System.
Authors’ Note
All errors remain the authors’ responsibility. This project was completed while Wong was on faculty at INSEAD and on unpaid leave of absence from the University of Toronto.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: We acknowledge the financial support of INSEAD, the Rotman School of Management, the Social Sciences and Humanities Research Council of Canada (Grant 484989), and Yale University.