Effects of managerial overconfidence on analyst recommendations

Abstract

This study investigates the relation between managerial overconfidence and analyst recommendations. The empirical finding shows that analysts are less likely to issue upgrade recommendations for firms managed by overconfident CEOs. Similarly, analysts spend a longer time to upgrade stocks associated with overconfident CEOs. The effect of CEO overconfidence on recommendation revisions is non-monotonic. Analysts are more reluctant to upgrade firms with highly-overconfident CEOs. More experienced analysts are less susceptible to managerial overconfidence. Moreover, investors exhibit stronger response to recommendations for firms with overconfident CEOs.

1 Introduction

Security analysts guide investor behavior by collecting, processing, interpreting and disseminating information about corporate finances, strategic decisions and industry trends, and by rendering summary judgments about the firms that they follow them (Hayward and Boeker 1998; Jenson 2004; Rao et al. 2001; Zuckerman 1999). Those judgments include recommendations about whether to buy, hold, or sell particular securities. There is extensive evidence that analysts’ stock recommendations have a material impact on trading behavior and stock market valuations (e.g., Ryan and Taffler 2006; Womack 1996). Changes in stock market valuations, in turn, affect a firm’s general reputation as well as its capacity to raise capital, which ultimately influence its corporate strategy (e.g., acquisition strategies), compensation policies (e.g., level of executive compensation); they also impact the reputation and career prospects of the firm’s top executives (Fombrun 1996; Hayward and Boeker 1998; Kuperman and Gibson 2003).

In this paper, we investigate how managers’ overconfidence affects analyst stock recommendations. Psychology literature suggests that overconfidence is related to the “better-than-average” effect (e.g., Larwood and Whittaker 1977; Svenson 1981; Alicke 1985). Overconfident managers tend to overestimate their wisdom or judgment compared to the average, and are convinced of their ability to create superior performance. Hribar and Yang (2016) find that overconfident CEOs are more likely to voluntarily issue optimistic earnings forecasts. In a related study, Mayew and Venkatachalam (2012) suggest that managers who exhibit positive affects release a lower proportion of bad news in the conference call. Therefore, we infer that overconfident CEOs tend to provide positive information of future prospects to analysts, and propose related questions to examine the effect of managerial overconfidence on analyst recommendations.

The questions this study will address are as follows: Does CEO overconfidence affect sell-side analysts’ recommendations? Do analysts take a longer time to revise recommendations for firms managed by overconfident CEOs? Do investors of stocks managed by overconfident managers rely relatively more on stock recommendation revisions from analysts than those by non-overconfident CEOs? This line of study is motivated by the following research. First, Jegadeesh et al. (2004) find that sell-side analysts are inclined to recommend glamour (e.g., positive momentum, high growth, and high volume) stocks. Hirshleifer et al. (2012) show that CEO overconfidence is associated with higher subsequent stock return volatility, which often appears among small, high growth, and high leveraged firms. Taken together, firm characteristics that sell-side analysts would like to recommend are similar to those operated by overconfident CEOs. Second, since overconfident CEOs are prone to overestimate future firm performance (Hribar and Yang 2016), they are thus more likely to provide positive information to analysts. Kramer and Liao (2012) propose that overconfidence affects the content of the information that CEOs provide to analysts, which in turn influences analysts’ forecasts. However, analysts may see through the drawbacks behind managerial overconfidence, and thus become more conservative in issuing recommendations. Third, because overconfident CEOs underestimate the likelihood of failure, they invest more aggressively than non-overconfident CEOs (Malmendier and Tate 2005). Furthermore, they have a higher tendency to invest more in innovation and riskier projects (Galasso and Simcoe 2011; Hirshleifer et al. 2012), engage in value-destroying mergers (Malmendier and Tate 2008), conduct unbeneficial long-term R&D investment (Chen et al. 2014), miss their voluntary forecast of earnings (Hribar and Yang 2016), thereby leading to greater earnings management (Hribar and Yang 2016; Schrand and Zechman 2012) and financial fraud (Schrand and Zechman 2012). Consequently, firms managed by overconfident CEOs would be embedded with riskier future prospects; thus, the magnitude of uncertainty is also greater. Given the corresponding higher risk investments and return volatility associated with overconfident CEOs (Galasso and Simcoe 2011; Hirshleifer et al. 2012), it becomes even more challenging to precisely evaluate the firm’s value (Godfrey et al. 2009).

Consistent with this view, we find that analysts are less likely to issue upgrade recommendations for stocks linked to overconfident CEOs. Moreover, analysts upgrade recommendations more slowly for companies managed by overconfident CEOs. Finally, investors exhibit a stronger response to stock recommendation revisions for firms with overconfident CEOs. Our results contribute to the finding of Loh and Stulz (2011) by showing that stock recommendations are more influential for firms with overconfident CEOs.

This study also performs a series of robustness checks to ensure that our results are solid. First, the empirical findings suggest that analysts are reluctant to provide upgrade recommendations to firms with overconfident CEOs, implying that the overestimation of future prospects and information release from overconfident CEOs have potential impact on analyst recommendations. Prior studies suggest that more experienced analysts have higher ability (Mikhail et al. 1997, 2003). This triggers the question as to whether the different ability of analysts identifies the influence from overconfident CEOs, or if all analysts are affected by overconfident CEOs. Therefore, we further classify analysts according to their experience, and find that less-experienced analysts are more susceptible to managerial overconfidence.

Second, since Goel and Thakor (2008) suggest that the influence of managerial overconfidence on corporate decision would be a non-linear effect, we additionally examine the relation between analyst recommendations and the degrees of overconfidence. The findings suggest that different degrees of managerial overconfidence have non-monotonic impacts on analyst recommendations. Analysts are more reluctant to upgrade firms with highly-overconfident CEOs, and this phenomenon is less likely for moderately-overconfident CEOs. The implication is qualitatively consistent with Goel and Thakor (2008), who show that moderately-overconfident CEOs increase firm value by mitigating underinvestment associated with rational CEOs, while high overconfidence generates overinvestment and decreases firm value.

Our study contributes to the recent and growing literature in several ways. First, prior literature on market reaction to stock recommendations usually focuses on whether it is affected by variations in firm-specific characteristics or recommendation types (e.g., upgrade, reiteration, or downgrade). This study extends prior studies by examining the impacts of managerial overconfidence on analyst recommendations. Second, prior literature suggests that analyst experience matters in the context of earnings forecasting accuracy (Mikhail et al. 1997; Clement 1999; Clement et al. 2007). We complement the finding that analysts experience is related to the difference in their recommendation revisions between overconfident and non-overconfident CEOs. Third, McNichols and O’Brien (1997) find that analysts take a longer time to downgrade than to upgrade stocks. We show a distinct but related finding that analysts spend a longer time to revise stocks with overconfident CEOs than with non-overconfident ones. Fourth, past research shows that the market places much more weight on experienced analyst forecasts than on novice analyst forecasts (Mikhail et al. 1997). Differently, by examining the market reaction to recommendation revisions, our finding suggests that the market reacts more strongly to revisions of stocks managed by overconfident CEOs than by non-overconfident ones.

The rest of the paper is organized as follows. In the next section, we describe the data and methodology employed in our empirical tests. Section 3 presents the empirical results. Section 4 discusses the robust tests carried out herein, and Sect. 5 presents our conclusions.

2 Data and methodology

2.1 Data

The data on the stock recommendations examined in this paper are taken from IBES. Recommendations in the IBES database are coded as follows: 1 = Strong Buy, 2 = Buy, 3 = Hold, 4 = Sell, 5 = Strong Sell. The recommendation changes include upgrades, downgrades, or reiterations. To interpret the quantitative results more intuitively, we invert the recommendations so that more favorable recommendations receive a higher score (e.g., 5 = strong buy, 3 = hold, 1 = strong sell). This study focuses on recommendation revisions since prior research shows that recommendation changes are more informative than mere levels (e.g., Boni and Womack 2006; Jegadeesh and Kim 2009; Papakroni 2018; Aggarwal et al. 2018). The recommendation change, Revision, is calculated as the current rating minus the prior rating by the same analyst. By construction, Revision ranges between − 4 and + 4.

We retrieved annual accounting and financial data from the Compustat database. We also obtained transaction-related information, i.e. price and trading volume, and firm size, i.e. market capitalization from the Center for Research in Security Prices (CRSP) database. Firms with stock price lower than $5 and with negative shareholders’ equity are excluded from the sample. CRSP Daily Stock File is used to calculate abnormal returns around recommendation days. We obtain quarterly data on institutional holdings from Thompson 13F Institutional Holdings database. The ExecuComp database provides manager-related information like age and tenure, with coverage from 1992 to 2014. Because IEBS recommendation data are available starting from 10/29/1993, our sample starts from 1994 and is limited to firms for which we have measures of CEO overconfidence, as described by Hirshleifer et al. (2012).

2.2 Measures of CEO overconfidence

Our empirical proxy for overconfidence is proposed by Malmendier and Tate (2005, 2008), and based on the stock-option holding decisions of CEOs. This option-based measure is widely adopted in literature.¹ This set of measures exploits the fact that CEOs are often under-diversified. If a particular CEO holds his stock options until the year of expiration, even though the options are relatively high monetarily, this behavior can be interpreted as overestimation of future firm performance; thus, the CEO is classified as overconfident.

However, the CEO overconfidence sample in Malmendier and Tate (2005, 2008) requires full information of CEOs’ option grant packages. Therefore, Campbell et al. (2011) and Hirshleifer et al. (2012) use a modified methodology derived from Core and Guay (2002) to measure the average exercise price of option grants. In this study, we adopt the construction criterion in Hirshleifer et al. (2012). Following Malmendier and Tate, we identify a CEO as overconfident once the CEO does not exercise his/her vested options that are at least 67% in-the-money. We assign a binary variable to a CEO who takes value as one when he/she is defined as overconfident, and zero otherwise. It is worth noting that once a CEO is identified as overconfident under this options-based measure, he/she remains the same type throughout the rest of the tenure. This feature is consistent with the concept that overconfidence is a persistent trait (Hirshleifer et al. 2012).

This study follows Hirshleifer et al. (2012) to compute the average moneyness of the CEO’s option portfolio for each year. First, we compute the average realizable value per option, which is computed from the total realizable value of the options over the number of total exercisable options held by the CEO for each CEO-year. Next, we use fiscal year end price minus the average realizable value to obtain the estimated average exercise price. Finally, we use the fiscal year end price over the estimated average exercise price minus 1 to obtain the average moneyness of the options. Since we judge whether or not a CEO is overconfident depending on his/her option exercising behavior, we only consider the vested option held by the CEO.

2.3 Methodology

2.3.1 Recommendation revisions and CEO overconfidence

This section focuses on the revisions of stock recommendations because they contain more information than stock recommendation levels (Jegadeesh et al. 2004; Green 2006; Boni and Womack 2006). We use Heckman’s (1979) two-stage model to examine whether managerial overconfidence affects analysts’ revisions of stock recommendations. This model avoids the self-selection problem where analysts might be strategic in choosing which companies to cover (McNichols and O’Brien 1997). The first step uses the following probit model to yield an inverse Mills ratio (IMR):

$$ P(Rec_{it} = 1|\varvec{Z}_{it - 1} ) = \varPhi (\alpha_{0} +\varvec{\beta}^{{\prime }} \varvec{Z}_{it - 1} + \varepsilon_{it} ), $$

(1)

where the dependent variable is one if analysts provide research recommendations for any given company i during one month, and zero otherwise (Ljungqvist 2007). The Φ is the cumulative distribution function of the standard normal distribution. ${\mathbf{Z}}$ is a vector of the variables which can possibly affect the analyst’s recommendations, including industry and year fixed effects; $\varepsilon$ is the random error. The control variables include: $Ret_{t - 5,t - 1}$,$Ret_{t - 21,t - 6}$,$Ret_{m - 6,m - 2}$, $Ret_{m - 12,m - 7}$, natural log of market value (ln(ME)), market-to-book ratio (MB), firm age (FirmAge), turnover ratio (Turnover), idiosyncratic volatility (Ivol), and institutional ownership (Inst) (Loh and Stulz 2011).

$Ret_{t - 5,t - 1}$ is the cumulative return over the last five trading days in month t − 1. Similarly, $Ret_{t - 21,t - 6}$,$Ret_{m - 6,m - 2}$ and $Ret_{m - 12,m - 7}$ are the cumulative returns from day t − 21 to t − 6 in month t − 1, month − 6 to − 2 and month − 12 to − 7, respectively. These four variables are used to control for the momentum effect on revisions (Jegadeesh et al. 2004). We also control for firm characteristics that may affect analyst coverage shown by existing literature. For example, we include firm size and growth (Bhushan 1989; O’Brien and Bhushan 1990; Brennan and Hughes 1991; Lang and Lundholm 1996; Yu 2008). The firm size is measured by the logarithm of market value of equity to capture the nonlinear relation between analyst coverage and firm size. In addition, we include MB (ratio of market value to book value) which accounts for analysts’ attention to the growth attributes (Frankel and Lee 1998). Firm age (FirmAge) is the year that firms appeared in CRSP. Analysts prefer to issue recommendations for high liquid firms because trading on these firms generates more trading commissions (Barth et al. 2001). Therefore, we also include turnover ratio (Turnover). Extant literature presents evidence that idiosyncratic volatility is positively related to corporate growth options (Andrés‐Alonso et al. 2006; Cao et al. 2008) and technological revolutions (Jovanovic and MacDonald 1994; Schwert 2002; Pastor and Veronesi 2009). As a result, we include idiosyncratic volatility (Ivol), which is calculated using Carhart (1997) four-factor model (Kumar 2009). Institutional investor is highly correlated with analyst coverage (O’Brien and Bhushan 1990); to account for the effect of institutional investor, we control for institutional ownership (Inst) (Yu 2008). Inst is the fraction of the firm’s shares held by institutions. We control for industry and year fixed effects since the effects of analysts’ recommendations can vary across industry and with time. We use three-digit SIC codes as proxies for industry.

In the second step, we run the following Eq. (2) with the resulting IMR as one additional explanatory variable

$$ Revision_{it} = \alpha_{0} + \alpha_{1} OC_{it - 1} + {\mathbf{\beta^{\prime}Z}}_{it - 1} + \alpha_{1} IMR_{it} + \varepsilon_{it} , $$

(2)

where $Revision$ is the change in stock recommendations; $OC$ is a dummy variable, which is 1 if the firm’s CEO is overconfident, and 0 otherwise. In addition to the controlling variables in Eq. (1), we add the firm’s characteristics like the analysts’ coverage (Coverage) (Loh and Stulz 2011) and standardized unexpected earnings (SUE).

Westphal and Clement (2008) document that a negative recommendation may prompt top executives’ (CEOs) negative reciprocity toward an analyst through their social influence tactics. This retaliation may deter other analysts’ negative recommendations. Therefore, we also include CEO characteristics like gender (Gender), age (CEOAge) and tenure as a CEO (CEOTenure) in the regression.

We also add analyst characteristics, including analyst experience and the dummy variable of lead analysts to proxy for analysts’ ability, in our analysis. We consider two types of analyst experience: general and firm-specific (Yu 2008). We measure analyst general experience in terms of the number of years between the current year and the year the analysts first appear on the recommendation database of IBES (GeneralExp). Following Clement (1999), firm-specific experience is calculated using the number of years spent in a particular firm (FirmExp). We identify lead analysts based on Cooper et al. (2001) to represent their reputation. $Lead$ is a dummy that equals one if the analyst is the lead analyst, and zero otherwise. The standard errors are adjusted for clustering at the firm level (Peterson 2009). The definitions of all variables are shown in “Appendix”; positive (negative) $\alpha_{1}$ indicates that analysts issue more favorable (unfavorable) revisions in stock recommendations for firms managed by overconfident CEOs.

2.3.2 Speed in incorporating news into their stock recommendations

In this section, we take a further step to test whether managerial overconfidence affects the decision by analysts to upgrade or downgrade the ratings, by running the Cox proportional hazard model. The Cox proportional hazard regression model, introduced by Cox (1972), makes no assumption regarding the nature of the hazard function h itself, and the coefficients can be estimated without having to specify the baseline hazard function h₀. The Cox proportional hazard model is represented as:

$$ h_{i} = h_{0,i} { \exp }\left[ {{\text{X}}_{i}^{ '} \beta } \right] , $$

(3)

where $h_{i}$ is the hazard function for the ith subject; $h_{0,i}$ is referred to as the baseline hazard function, which corresponds to the case where ${\text{X}}_{i}$ are covariate values corresponding to the ith subject; and $\beta$ are the parameters to be estimated.

The Cox proportional hazard model does not assume the probability of changing the recommendation levels, and employs a non-parametric estimate of the baseline hazard. The model is performed with time-varying covariates, which include some variables associated with firm-specific, analyst and CEO characteristics, which are used in Eq. (2). Specifically, to know if the time analysts take to revise recommendations is influenced by CEO overconfidence, we estimate the following Cox proportional hazard model:

$$ h_{it} = h_{0,i} \left[ {\alpha_{0} + \alpha_{1} Revision_{it} + \alpha_{2} OC_{it - 1} + \alpha_{3} Revision_{it} \times OC_{it - 1} + \beta^{\prime}{\mathbf{Z}}_{it - 1} } \right] . $$

(4)

All variables are the same as defined in Eq. (2).

2.3.3 Returns following recommendation revisions and CEO overconfidence

To measure the market reaction after the revision of stock recommendations, we compute the buy-and-hold abnormal returns (BHARs). We use the reference portfolio approach proposed by Daniel et al. (1997) to compute the benchmark portfolio returns. For each event firm i, the buy-and-hold abnormal return is calculated as the difference between the firm’s buy-and-hold return ($R_{i,t}$) and the buy-and-hold return of the corresponding reference portfolio p ($R_{p,t}$):

$$ BHAR\left( {0,T} \right) = \mathop \prod \limits_{t = 0}^{T} \left( {1 + R_{i,t} } \right) - \mathop \prod \limits_{t = 0}^{T} \left( {1 + R_{p,t} } \right) , $$

(5)

where T is the holding period in months. We compare $BHAR$ between overconfident and non-overconfident CEOs to examine whether investors’ response to revisions of stock recommendations varies with CEO overconfidence.

3 Empirical results

3.1 Univariate analysis

We focus exclusively on firms that have executive compensation data on Compustat Executive Compensation Database.² We also exclude data without observations for the determinants of the analyst stock recommendations. After the filtering, the sample covers only recommendations by brokerage houses which have CEO compensation data and complete financial data; we are left with 58,776 revisions in stock recommendations, which include 37,505 overconfident observations and 21,271 non-overconfident observations.

Table 1 presents the summary statistics for all variables used in the following regression analysis. The results show that firms managed by overconfident CEOs tend to be larger in size, growth, and young stocks, with a higher turnover ratio, idiosyncratic risk, institutional ownership, and analyst coverage. Firms that are managed by overconfident CEOs have lower standardized unexpected earnings compared to their non-overconfident counterparts. In regard to managerial characteristics, overconfident CEOs are older and have longer tenure as CEOs. Analysts who are more likely to recommend firms managed by overconfident CEOs are more experienced in terms of their years appearing in the IBES. However, analysts with less firm-specific experience tend to recommend firms with overconfident CEOs. Lead analysts are also more inclined to cover firms managed by overconfident CEOs.

Table 1 Summary statistics of related variables

Table 2 presents the Spearman rank correlation of independent variables used in the following regressions. Turnover is highly correlated with idiosyncratic risk. Analysts’ general experience and firm-specific experience are also highly correlated. To avoid the multicollinearity problem, following the approach of Hong et al. (2000), we run Ivol over Turnover to obtain residual Ivol (Res_Ivol), FirmExp over GeneralExp to obtain residual FirmExp (Res_Firmexp), and then run the subsequent equations with Ivol and FirmExp being replaced by Res_Ivol and Res_FirmExp, respectively.

Table 2 Pearson correlation coefficients among related variables

3.2 Analysts’ recommendation revisions and managerial overconfidence

The left column of Table 3 has the regression results from the first stage of the Heckman model. The results confirm that analysts are more likely to cover large, growth, old, and liquid stocks. Moreover, the larger the institutional ownership is, the more the analysts cover the stock. Coverage is also more likely for stocks whose prices run up within a month but run down within one year.

Table 3 Recommendation change and managerial overconfidence

The right three columns of Table 3 report the regression results from the second stage of Heckman model. Model (1) is the result controlling for firm characteristics, Model (2) controls both firm and CEO characteristics, and Model (3) controls firm, CEO and analyst characteristics. The negative and significant coefficients of OC in Models (1) to (3) are primary evidence that analysts are less likely to upgrade a stock with an overconfident CEO. Hribar and Yang (2016) and Schrand and Zechman (2012) show that overconfident CEOs are too optimistic about future prospects. Moreover, overconfident CEOs may engage in higher risky investment projects, which leads to greater return volatility (Galasso and Simcoe 2011; Hirshleifer et al. 2012). Taking these two lines of literature together, our results collaborate with previous findings and reveal that analysts tend to be conservative when revising recommendations for firms managed by overconfident CEOs.³

3.3 Time to revise recommendations and managerial overconfidence

Table 4 contains regression results from estimating whether the time that analysts spend to revise prior recommendations varies with managerial overconfidence, after controlling for other potential effects. The significantly positive coefficient of OC indicates that analysts spend more time when revising recommendations on stocks with overconfident CEOs. The coefficient on the interaction term of revision and overconfidence, Revision × OC, is also significantly positive. This implies that the tendency for analysts to take a longer time to upgrade recommendations on a firm increases with overconfident CEOs. To the extent that overconfident CEOs are more likely to miss their voluntary forecast of earnings (Hribar and Yang 2016), we find that analysts tend to be more hesitant to upgrade stocks with CEO overconfidence as a factor.

Table 4 Time in revising stock recommendations and managerial overconfidence

3.4 Returns following recommendation revisions

Prior studies show that analyst recommendations affect stock prices at the time of their release (Elton et al. 1986; Stickel 1995; Womack 1996; Barber et al. 2001). To examine whether there are differences in market reaction to analyst recommendations with respect to overconfident and non-overconfident CEOs, Table 5 summarizes 126-day BHARs subsequent to the revision of stock recommendations.

Table 5 Difference in performance of overconfident and non-overconfident samples

In Table 5, over the period of six months following recommendation revisions, the average abnormal returns following upgrades are positive, while the returns following downgrades are negative. Consistent with prior studies (e.g., Elton et al. 1986; Womack 1996; Barber et al. 2001), analyst recommendations have predictive power regarding subsequent returns. In comparison, there are significantly higher abnormal returns for stocks with CEO overconfidence within two months subsequent to the upgrade recommendations. Given a downgrade recommendation, stocks of overconfident CEOs have significantly lower negative abnormal returns. This phenomenon holds for as long as six months after a downgrade, indicating that recommendation revisions on stocks with overconfident CEOs are more influential than on non-overconfident CEOs.

The findings in Table 5 show that there are significantly higher (lower) abnormal returns for stocks with overconfident CEOs within two (six) months after the upgrade (downgrade) recommendations. Thus, we can create long/short portfolios to profit from the apparent persistence in the overconfidence of CEOs. A long position for upgrades of stocks with overconfident CEOs combined with a short position for upgrades of stocks with non-overconfident CEOs yields an estimated abnormal return of 0.671% over the two-month trading period after the stock recommendations. A short position for downgrades of stocks with overconfident CEOs combined with a long position for downgrades of stocks with non-overconfident CEOs yields an estimated abnormal return of 1.018% (1.201%) over the one(six)-month trading days after the stock recommendations. However, one condition for a riskless arbitrage is that we can buy an asset at a lower price and immediately sell a similar asset for a higher price. To the extent that analysts do not issue recommendations for all stocks at the same time, the arbitrage portfolio is not riskless. Furthermore, the higher abnormal returns for the stocks of firms with overconfident CEOs from the upgrade recommendations only persist for two months. The rebalancing costs of this portfolio might be too large and therefore offset the potential profits.

Prior research shows that overconfident CEOs tend to invest more aggressively (Malmendier and Tate 2005) and make greater investment in innovation and riskier projects (Galasso and Simcoe 2011; Hirshleifer et al. 2012). Thus, the firm managed by overconfident CEOs would have higher uncertainty. These uncertain features associated with overconfident CEOs might explain why the sensitivity of market response to analysts’ revisions in stock recommendations gets stronger with CEO overconfidence. This is also in line with prior literature concluding that firms with more uncertainty rely more on analyst forecasts to transmit information (Lin et al. 2014).

4 Robustness checks

This paper’s findings suggest that: (1) analysts are more likely to issue downward revisions in recommendations for overconfident CEOs, (2) analysts spend more time in upgrading stocks with CEO overconfidence, and (3) the revisions in stock recommendations of overconfident-managed firms have higher predictive power related to future firm performance. This section explores the robustness of the main results. In particular, do the results remain unchanged using a logit model? Are the results correlated with analyst experience? Are these findings monotonic in CEO overconfidence?

4.1 Upgrades and downgrades

Since changes in recommendation take ordered values from − 4 to + 4, we take a further step to test whether analysts have a less (more) likelihood of issuing upgrade (downgrade) stock recommendations for firms with overconfident CEOs, using the following logit model:

$$ P(Upgrade_{it} = 1|OC_{it - 1} , \varvec{Z}_{it - 1} ) = L\left( {\alpha_{0} + \alpha_{1} OC_{it - 1} + \varvec{\beta^{\prime}Z}_{it - 1} + \alpha_{2} IMR_{it} + \varepsilon_{it} } \right) , $$

(6)

where L is a logistic function; Upgrade is a dummy variable which takes a value of 1 if the recommendation change is positive and 0 otherwise. Other variables are as defined in Eq. (2). The same regression is run with the dependent variable being replaced by Downgrade; the value of the dependent variable equals 1 for downgrades, and 0 otherwise. Table 6 shows the regression results. The coefficient of OC is significantly negative for upgrades at the 10% confidence level and significantly positive for downgrades at the 5% level. This confirms the above finding that analysts are more likely to downgrade and reluctant to upgrade for stocks whose CEO exhibits overconfidence. This result means that our findings, in which analysts are less likely to upgrade stocks with overconfident CEOs, are less confounded by using an alternative model.

Table 6 Upgrades, downgrades and managerial overconfidence

To know if the time analysts take to upgrade and downgrade differs, and if the difference is influenced by CEO overconfidence, we estimate the following Cox proportional hazard model:

$$ h_{it} = h_{0,i} \left[ {\alpha_{0} + \alpha_{{1{\text{a}}}} Upgrade_{it} + \alpha_{{1{\text{b}}}} Downgrade_{it} + \alpha_{2} OC_{it - 1} + \alpha_{{3{\text{a}}}} Upgrade_{it} \times OC_{it - 1} + \alpha_{{3{\text{b}}}} Downgrade_{it} \times OC_{it - 1} + \varvec{\beta^{\prime}Z}_{it - 1} } \right]. $$

(7)

All the variables are the same as defined in Eq. (2). Table 7 reports the results. The coefficient of Upgrade × OC is significantly positive for upgrades and insignificantly negative for downgrades at the 5% confidence level. This confirms the above findings that analysts take a longer time to upgrade stocks with managerial overconfidence.

Table 7 Managerial overconfidence and time in upgrading and downgrading recommendations

4.2 Analyst experience

Do analysts differ in their ability to identify CEOs’ tendency to overestimate their abilities and probabilities of success. Analysts may become better at analyzing financial statements and recognizing economic trends as they gain experience. Thus, analysts with greater experience are more capable of identifying which CEOs are just overestimating their managerial performance. By comparing more-experienced with less-experienced analysts, we are able to observe a relationship between analyst experience and revisions in stock recommendations. We consider two types of analyst experience: general and firm-specific (Yu 2008). We sort the samples for each year, and referred the observations as reflecting more experienced analysts if the value was larger than the median, and less experienced analysts for values less than the median.

The left part of Table 8 shows the results for general experience, and the right part presents the results for firm-specific experience. When analyst experience is measured in terms of years appearing in IBES, less-experienced analysts are less likely to upgrade stocks for overconfident CEOs. In addition to obtaining general skills, analysts also obtain firm-specific skills over time. For example, experienced analysts might better understand the idiosyncratic parts of a particular firm’s reporting practices or establish better relationships with insiders and thereby gain better access to managers’ private information. Thus, analysts with more firm-related experience might be less likely to be affected by CEO overconfidence. As expected, firm-specific inexperienced analysts have a lower tendency to issue upgrade recommendations for overconfident CEOs at the 10% confidence level.

Table 8 Recommendation change and managerial overconfidence: analysts’ experience

To examine whether analyst experience has an effect on the time it takes to revise recommendations, the results are reported in Table 9. As shown, inexperienced analysts take longer time to upgrade the ratings of a stock with overconfident CEOs. This holds regardless of analysts’ general and firm-specific experience. Our findings indicate that less-experienced analysts may think that they lack the ability to fully see through the benefits and drawbacks behind CEO overconfidence. This makes less-experienced analysts more hesitant to upgrade stocks linked to overconfident CEOs. Conversely, more experienced analysts are more capable of analyzing the advantages and disadvantages associated with CEO overconfidence. Thus, the time that analysts spend to revise recommendations decreases as they gain more experience.

Table 9 Time to revise recommendation and analyst experience

4.3 Multiple level of managerial overconfidence

Goel and Thakor (2008) suggest that moderate CEO overconfidence mitigates underinvestment and increases firm value, while high CEO overconfidence induces overinvestment and reduces firm value. This implies a non-linear relationship between CEO overconfidence and firm value. Accordingly, we also examine whether there is a nonlinear relation between favorable recommendation revisions and the degree of overconfidence, by creating four groups of CEOs based on their option holding behavior.

Following the classification in Hirshleifer et al. (2012), group 1 classifies managers who never hold options that are at least 67% in the money as non-overconfident CEOs. Group 2 classifies managers as low overconfident CEOs after the time that they have held options that the moneyness level is at least greater than 67% and less than 130%. Group 3 classifies managers as moderately overconfident CEOs after the time that the moneyness level is at least greater than 130% and less than 250%. Group 4 classifies managers as highly overconfident CEOs that they have held options that the moneyness level is at least greater than 250%.

We use three dummy variables regarding the three levels of overconfidence to substitute for the indicator variable for overconfidence in Eq. (1). The base group is non-overconfident CEOs. $OC\_Low$ is a dummy variable being one if managers are classified as low overconfident CEOs, and 0 otherwise; $OC\_Mod$ is one if managers are classified as moderately overconfident CEOs, and 0 otherwise; $OC\_High$ is a dummy variable being one if managers are classified as highly overconfident CEOs, and 0 otherwise. Other variables are as defined in Eq. (1). We then run the regression as follows:

$$ Revision_{it} = \alpha_{0} + \alpha_{1a} OC\_Low_{it - 1} + \alpha_{1b} OC\_Mod_{it - 1} + \alpha_{1c} OC\_High_{it - 1} + \varvec{\beta^{\prime}Z}_{it - 1} + \alpha_{2} IMR_{it} + \varepsilon_{i,t} . $$

(8)

Table 10 shows the results for different levels of overconfidence on recommendation revisions. In regard to all samples, the coefficient of the overconfident measure is only significantly negative at the 10% confidence level for OC_High. This indicates that analysts are less likely to upgrade stocks whose managers are highly overconfident. The coefficients on the three indicator variables display an inverse U-shaped of pattern. In particular, it increases from low overconfidence to moderate overconfidence, and then decreases for high overconfidence. This result, though insignificant, shows that the less likelihood for analysts to upgrade stocks would be weaker for firms managed by moderately overconfident CEOs.

Table 10 Recommendation change and multiple-level managerial overconfidence

We also rerun Eq. (2) with the overconfident dummy replaced by three levels of overconfident dummies for more- and less-experienced analysts. The results on the right part of Table 10 reveal that less-experienced, instead of more-experienced, analysts exhibit the tendency to downgrade stocks whose managers are highly overconfident. In the context of less-experienced analysts, there is also an inverse U-shaped coefficient with respect to CEO overconfidence. This reveals that the effect of overconfidence on changes in recommendations by less-experienced analysts is also non-monotonic, and the moderately overconfident CEOs are less likely to receive downgrade recommendations.

Table 11 examines whether the effects of CEO overconfidence on the time analysts take to revise recommendations is also non-monotonic. The equation is as follows:

Table 11 Time to revise recommendation and degrees of overconfidence

$$ h_{it} = h_{0,i} \left[ {\alpha_{0} + \alpha_{1} Revision_{it} + \alpha_{{2{\text{a}}}} OC\_Low_{it - 1} + \alpha_{{2{\text{b}}}} OC\_Mod_{it - 1} + \alpha_{{2{\text{c}}}} OC\_High_{it - 1} + \alpha_{{3{\text{a}}}} Revision_{it} \times OC\_Low_{it - 1} + \alpha_{{3{\text{b}}}} Revision_{it} \times OC\_Mod_{it - 1} + \alpha_{{3{\text{c}}}} Revision_{it} \times OC\_High_{it - 1} + {\mathbf{\beta^{\prime}Z}}_{{{\text{it}} - 1}} } \right] , $$

(9)

As shown, for all samples, among the three overconfident levels, only the coefficient of Revision × OC_High is positive and significant. This confirms the view that analysts take a longer time to upgrade stocks whose managers are highly overconfident.

In regard to analyst experience, inexperienced analysts take a longer time to upgrade firms managed by overconfident CEOs. This result holds for both moderately and highly overconfident CEOs based on analysts’ general experience. In terms of analysts’ firm-specific experience, the result remains significant for all CEO overconfidence levels. The coefficients on the three indicator variables display a U-shape pattern for firm-specific inexperienced analysts. This indicates that the effect of overconfidence on the time to revise recommendations is non-monotonic, and that the moderately overconfident CEOs are associated with the shortest duration to upgrade recommendations. In addition, analysts with more firm-specific experience also take a longer time to upgrade a stock linked to highly overconfident CEOs. However, this phenomenon does not hold when measured based on general experience, or for analysts with low and moderate experience. In sum, inexperienced analysts tend to be more affected by CEO overconfidence, and specifically by high levels of CEO overconfidence.

5 Conclusions

An overconfident CEO tends to be more optimistic about his/her firm’s future performance and overestimates the precision of his/her private information. This would affect analysts’ stock recommendations. We focus on revisions in recommendation levels and CEO overconfidence because recommendation revisions have significant information contents (Womack 1996; Jegadeesh et al. 2004).

Overconfident managers tend to overestimate their wisdom or judgment compared to the average, and are more likely to voluntarily issue optimistic earnings forecasts (Libby and Rennekamp 2012), and have a greater likelihood of missing their voluntary forecast of earnings (Hribar and Yang 2016). That is, overconfidence affects the information content that CEOs provide to analysts, which in turn influences analysts’ recommendations. Consistent with this view, we find that analysts are less likely to upgrade stocks linked with overconfident CEOs and take a longer time to upgrade firms managed by overconfident CEOs. Our results remain unchanged using an alternative model for testing.

In sum, overconfident CEOs tend to overestimate returns and underestimate risk that leads to analysts being reluctant to upgrade recommendations, especially for highly overconfident CEOs. Given these findings, if the board of directors can introduce diverse viewpoints to curb the actions of overconfident CEOs, this restraint would not only benefit shareholders but also reduce analysts’ reluctance to issue upgrade recommendations for firms managed by overconfident CEOs.

The effects of overconfident CEOs on recommendation revisions are non-monotonic; analysts are less inclined to upgrade stocks for CEOs with high overconfidence. Similarly, analysts spend a longer time to upgrade a stock managed by highly overconfident CEOs.

We compared the difference in incremental value of changes in analyst recommendations between overconfident and non-overconfident CEOs. The results show that investors react more strongly to recommendation revisions on firms managed by overconfident CEOs than by non-overconfident CEOs. This can be attributed to the greater uncertain prospects associated with overconfident CEOs; thus, investors rely more on analysts’ recommendations. Our results contribute to the finding of Loh and Stulz (2011) by showing that stock recommendations are more influential for firms with overconfident CEOs.

We also examined whether analyst experience affects the results. We find that less-experienced analysts are less likely to upgrade their recommendations for overconfident CEOs than for non-overconfident CEOs, and take a longer time to upgrade stocks linked to CEO overconfidence. In sum, inexperienced analysts are more affected by CEO overconfidence. This means that an analyst’s ability to be immune from behavioral bias increases with experience. Lees (1981) posits that as analysts have more access to management information, the learning-by-doing effect improves. Han et al. (2018) take a further step and show that inexperienced analysts can mitigate their behavioral biases and gain an informational advantage by visiting listed companies and having more private interactions with their management. Taken together, inexperienced analysts can mitigate the influence of an overconfident CEO through engaging in more interactions with the firm’s managers.

Appendix: Variable definitions

This appendix shows the definitions of the variables used in this study. Accounting data are from Compustat, stock return data from CRSP, analyst recommendation data from IBES, compensation data and CEO characteristics from Execucomp, and institutional ownership data from Thomson 13f

Variables	Definition
Recommendation variables
Recommendation	Analyst recommendations. 5 is strong buy; 4 is buy; 3 is hold; 2 is sell; 1 is strong sell
Revision	Change in recommendation
Upgrade	Binary variable that equals one if the analyst upgrade from previous recommendation, and zero otherwise
Downgrade	Binary variable that equals one if the analyst downgrade from previous recommendation, and zero otherwise
Overconfidence variables
OC	Binary variable that equals one if the firm’s CEO is overconfident, and zero otherwise. A CEO is classified as overconfident after the time that he foregoes exercising 67% in the money options. The moneyness of options is measured as the fiscal year end price divided by the estimated exercise price minus 1. The estimated exercise price is computed from the fiscal year end price minus the average realizable value which is calculated by dividing the total realizable value of the options by the number of options held by the CEO for each CEO-year
OC_Low	Binary variable that equals one if the firm’s CEO is classified as low overconfident, and zero otherwise. A CEO is classified as low overconfident after the time that he have held options that the moneyness level is at least greater than 67% and less than 130%
OC_Mod	Binary variable that equals one if the firm’s CEO is classified as moderately overconfident, and zero otherwise. A CEO is classified as moderately overconfident after the time that he have held options that the moneyness level is at least greater than 130% and less than 250%
OC_High	Binary variable that equals one if the firm’s CEO is classified as highly overconfident, and zero otherwise. A CEO is classified as highly overconfident after the time that he have held options that the moneyness level is at least greater than 250%
Control variables
Ret t− 5, t− 1	Cumulative return from day t − 1 to t − 5 before recommendation revision date t
Ret t− 21, t− 6	Cumulative return from day t − 21 to t − 6 before recommendation revision date t
Ret m− 6, m− 2	Cumulative return from month −6 to − 2 before recommendation revision date t
Ret m− 12, m− 7	Cumulative return from month − 12 to −7 before recommendation revision date t
ME	Natural logarithm of market equity value. Market equity value equals outstanding shares multiplied by the stock price
MB	The ratio of market value of equity over book value of equity; MB for July of year t to June of year t + 1 is ME for June of year t divided by the book equity for the last fiscal year end in year t − 1
FirmAge	The year that firms appeared in CRSP
Turnover	Average monthly turnover over the previous year
Ivol	Idiosyncratic volatility: the standard deviation of the residuals of the Carhart 4-factor model estimated using the daily stock returns over the past 1 year
Inst	Institutional ownership. A firm’s shares held by institutions scaled by shares outstanding from CRSP
Analyst level variables
Coverage	The number of analyst stock recommendations during 1 year
SUE	Standardized unexpected earnings: unexpected earnings (actual quarterly earnings per share minus the analyst consensus of quarterly earnings per share) divided by the stock price over the end of prior month
GeneralExp	The number of years between the current year and the year that analysts first appeared on the recommendation database of IBES
FirmExp	The number of years that analyst follows the same firm
Lead	We identify lead analysts based on Cooper, Day and Lewis (2001) to represent their reputation. Lead is a dummy that equals one if the analyst is the lead analyst, and zero otherwise
CEO characteristics
CEOAge	Age of CEO
Gender	Gender of CEO
CEOTenure	CEO tenure in years
Performance variable
BHAR	Buy-and-hold abnormal returns: $BHAR\left( {0,T} \right) = \prod\nolimits_{t = 0}^{T} {\left( {1 + R_{it} } \right) - \prod\nolimits_{t = 0}^{T} {\left( {1 + R_{pt} } \right)} }$. $R_{it}$ refers to the raw return of sample firms. $R_{pt}$ refers to the returns of reference portfolio. T refers to the holding period. The reference portfolio is constructed from size, market-to-book and momentum portfolios. At the end of each June, we first sort stocks into quintile based on market value on the last day of June. The breakpoints use NYSE stocks only. Within each size quintile, we then form quintile book-to-market breakpoints. The book-to-market ratio is the book-value at the end of the fiscal year t − 1 to the market value at the end of December of the year t − 1. We subtract the long-term average book-to-market ratio from an individual firm’s ratio to “industry adjust” the book-to-market ratio. Each of the 25 size and book-to-market portfolios is then split into price momentum quintiles. Price momentum is computed as the preceding twelve-month return one month before the formation date (Carhart 1997). Thus, a total of 125 reference portfolios are formed. For each benchmark portfolio, its equal-weighted portfolio return is calculated