1. Investor Beliefs and Investor Utility

Upon receiving the management guidance including the unexpected guidance bias $ b $ in $ {t}_1 $ , the investor fully incorporates this information into her beliefs. More specifically, she believes in a payoff distribution that is shifted by $ b $ compared to the rational benchmark such that her updated beliefs imply a uniform distribution of $ {P}_2 $ in the interval $ \left[{E}_1\left({P}_2\right)+b\pm 0.5\sigma \right] $ .Footnote ⁴ Hence, she does not detect the bias such that she expects a final stock payoff of

(1) $$ {E}_1^{Inv}\left({P}_2\right)={E}_1\left({P}_2\right)+b, $$

where $ {E}_1^{Inv}\left(\cdot \right) $ denotes investor expectations and $ {E}_1\left(\cdot \right) $ objective expectations, both conditional on the information in $ {t}_1 $ .

The investor’s utility function incorporates two central features: reference point dependence and loss aversion. These two features have been frequently documented and applied in both the asset pricing and corporate finance literature (see, e.g., Shefrin and Statman (Reference Shefrin and Statman1985), Burgstahler and Dichev (Reference Burgstahler and Dichev1997), Degeorge, Patel, and Zeckhauser (Reference Degeorge, Patel and Zeckhauser1999), Graham et al. (Reference Graham, Harvey and Rajgopal2005), Ben-David and Hirshleifer (Reference Ben-David and Hirshleifer2012), An et al. (Reference An, Wang, Wang and Yu2020), and Riley, Summers, and Duxbury (Reference Riley, Summers and Duxbury2020)). More specifically, when evaluating her investment in $ {t}_1 $ and $ {t}_2 $ , the representative investor uses the stock prices from the previous period (i.e., $ {P}_0 $ and $ {P}_1 $ , respectively) as reference points. Following Baker et al. (Reference Baker, Mendel and Wurgler2016), we posit that she perceives negative deviations from these reference points as particularly bad. Hence, we scale up the evaluation of losses using a loss aversion parameter of $ \left(1+\lambda \right) $ with $ \lambda >0 $ . Such an asymmetric evaluation of gains versus losses is well-founded in the psychological literature on loss aversion and a core feature of prospect theory (Kahneman and Tversky (Reference Kahneman and Tversky1979), Tversky and Kahneman (Reference Tversky and Kahneman1992)). For simplicity, we assume linear utility in the gain and loss domain as well as the same utility function in $ {t}_1 $ and $ {t}_2 $ . Consequently, the representative investor’s realized utility from holding one stock is given as

(2) $$ {u}_1=\left({P}_1-{P}_0\right)\left(1+\lambda {\mathbf{1}}_{P_1<{P}_0}\right)\hskip2em \mathrm{and}\hskip2em {u}_2=\left({P}_2-{P}_1\right)\left(1+\lambda {\mathbf{1}}_{P_2<{P}_1}\right) $$

in $ {t}_1 $ and $ {t}_2 $ , respectively. This combination of reference point dependence and loss aversion will turn out to be a key determinant for the manager’s choice of guidance bias. Moreover, based on Equation (2), in $ {t}_1 $ , the representative investor will allocate her investments between stock and risk-free asset such that she maximizes her expected level of utility $ {u}_2 $ , that is, $ {E}_1^{Inv}\left({u}_2\right) $ . When trading the stock, the investor is assumed to be a price taker.

2. Manager Utility

We assume that the manager follows an investor perspective such that her utility depends on both $ {u}_1 $ and $ {u}_2 $ (the investor’s utility from holding the stock from $ {t}_0 $ to $ {t}_1 $ and from $ {t}_1 $ to $ {t}_2 $ , respectively). This focus on both intermediate and long-term shareholder value is in line with Leland and Pyle (Reference Leland and Pyle1977), Miller and Rock (Reference Miller and Rock1985), Bebchuk and Stole (Reference Bebchuk and Stole1993), and Baker et al. (Reference Baker, Mendel and Wurgler2016). Moreover, it is motivated by the notion that a poor intermediate investor mood can already have detrimental effects for managers such as reduced compensation and, potentially, CEO turnover. In line with this logic, we assume that the manager cares about the incumbent investor, who can determine managerial compensation and dismissal through their voting rights, instead of the marginal potential investor. Beyond this investor perspective, we assume that the manager suffers personal costs if her forecast from $ {t}_1 $ turns out to be imprecise in $ {t}_2 $ . More specifically, we assume that deviations of the final payoff $ {P}_2 $ from the provided guidance $ {E}_1\left({P}_2\right)+b $ induce costs $ c $ .Footnote ⁵ Hence, the manager’s intertemporal utility is given as

(3) $$ U={u}_1+\beta {u}_2-\beta c\mid {P}_2-\left({E}_1\left({P}_2\right)+b\right)\mid, $$

where $ \beta \in \left(0,1\right] $ reflects the manager’s intertemporal utility trade-off between $ {t}_1 $ and $ {t}_2 $ with lower values of $ \beta $ indicating stronger discounting and a stronger present preference. We assume a rational manager who chooses $ b $ such that it maximizes her expected level of intertemporal utility $ {E}_1(U) $ . However, the manager’s choice is subject to the plausibility constraint $ b\in \left[-0.5\sigma, +0.5\sigma \right] $ , that is, we assume that the costs $ c $ are sufficiently high such that the manager cannot communicate a payoff forecast for $ {P}_2 $ that lies outside the range of possible $ {P}_2 $ realizations.

1. Implications for the Stock Price in $ {t}_1 $

In $ {t}_1 $ , the investor allocates her funds between stock and risk-free asset such that her expected utility from $ {t}_2 $ payoffs is maximized. Hence, in equilibrium, the investor’s marginal expected utility from buying more stocks equals the marginal utility from investing more at the risk-free rate. Given the investor’s loss aversion (Equation (2)), the stock price in $ {t}_1 $ equalsFootnote ⁶

(4) $$ {P}_1={E}_1\left({P}_2\right)+b-\frac{\sigma }{2}+\frac{\sigma }{\lambda}\left(\sqrt{1+\lambda }-1\right). $$

Hence, the stock price depends on rational payoff expectation $ {E}_1\left({P}_2\right) $ and guidance bias $ b $ , where the latter reflects the stock’s temporary mispricing. Such mispricing implications are qualitatively in line with prior catering models that also relax the assumption of market efficiency (Baker and Wurgler (Reference Baker and Wurgler2004), Baker et al. (Reference Baker, Greenwood and Wurgler2009)). The remaining parts of Equation (4) imply $ {P}_1<{E}_1\left({P}_2\right)+b $ such that the investor expects a positive stock return between $ {t}_1 $ and $ {t}_2 $ . This subjective risk premium compensates the loss-averse investor for payoff risk and increases in $ \lambda $ , that is, for a given guidance bias $ b $ , a higher $ \lambda $ implies a lower stock price. Vice versa, for the limiting case $ \lambda \to 0 $ , $ {P}_1={E}_1\left({P}_2\right)+b $ , such that the investor expects a return of zero.

2. Implications for Management Guidance

Based on Equations (1) to (4), the manager’s expected level of intertemporal utility is given as

(5) $$ {\displaystyle \begin{array}{c}{E}_1(U)=\left({P}_1^{unbiased}+b-{P}_0\right)\left(1+\lambda {\mathbf{1}}_{P_1^{unbiased}+b<{P}_0}\right)+\beta \left({E}_1\left({P}_2\right)-{P}_1^{unbiased}-b\right)\\ {}-\beta \left(\frac{\lambda }{2\sigma }{\left({P}_1^{unbiased}-{E}_1\left({P}_2\right)+b+\frac{\sigma }{2}\right)}^2+\frac{{c b}^2}{\sigma }+\frac{c\sigma}{4}\right),\end{array}} $$

where $ {P}_1^{unbiased} $ is the hypothetical price for the case of unbiased guidance ( $ b=0 $ ), that is, $ {P}_1^{unbiased}={P}_1-b $ . As evident from Equation (5), the manager’s choice of guidance bias $ b $ results from the intertemporal utility trade-off between $ {t}_1 $ and $ {t}_2 $ . On one hand, providing excessively optimistic forecasts in $ {t}_1 $ increases immediate utility. This positive impact of $ b $ is particularly strong if the stock trades in the investor’s loss domain (i.e., if $ {P}_1={P}_1^{unbiased}+b<{P}_0 $ ). On the other hand, raising expectations and thus reference prices in $ {t}_1 $ entails three negative effects in $ {t}_2 $ . First, the positive shift in expectations is reversed, implying a utility loss. Second, with respect to the investor’s loss aversion, both the likelihood to incur a loss from $ {t}_1 $ to $ {t}_2 $ and the potential loss magnitude linearly increase in $ b $ . The combination of these effects results in the first term that is quadratic in $ b $ . Third, similar arguments imply that expected personal costs are also quadratic in $ b $ as a higher level of $ b $ increases both the likelihood of large forecast errors and also the potential magnitude of the corresponding costs (see also the quadratic cost function in Beyer, Guttman, and Marinovic (Reference Beyer, Guttman and Marinovic2019)).

A formal investigation of Equation (5) implies that the following guidance bias $ {b}^{\ast } $ maximizes the manager’s expected level of intertemporal utility $ {E}_1(U) $ :

(6) $$ {b}^{\ast }=\left\{\begin{array}{llllll}\frac{1-\beta \sqrt{1+\lambda }}{\beta /\sigma (\lambda +2c)}& \mathrm{f}\mathrm{o}\mathrm{r}& & & {P}_0-{P}_1^{unbiased}& <\frac{1-\beta \sqrt{1+\lambda }}{\beta /\sigma (\lambda +2c)}\\ {}{P}_0-{P}_1^{unbiased}& \mathrm{f}\mathrm{o}\mathrm{r}& \frac{1-\beta \sqrt{1+\lambda }}{\beta /\sigma (\lambda +2c)}& \le & {P}_0-{P}_1^{unbiased}& \le \frac{1+\lambda -\beta \sqrt{1+\lambda }}{\beta /\sigma (\lambda +2c)}\\ {}\frac{1+\lambda -\beta \sqrt{1+\lambda }}{\beta /\sigma (\lambda +2c)}& \mathrm{f}\mathrm{o}\mathrm{r}& \frac{1+\lambda -\beta \sqrt{1+\lambda }}{\beta /\sigma (\lambda +2c)}& <& {P}_0-{P}_1^{unbiased}.& \end{array}\right.\operatorname{} $$

The manager is incentivized to issue a guidance bias $ {P}_0-{P}_1^{unbiased} $ in order to move $ {P}_1 $ toward $ {P}_0 $ , that is, the manager tries to smooth the stock’s price path. More specifically, due to the representative investor’s loss aversion, the manager will try to avoid negative stock returns from $ {t}_0 $ to $ {t}_1 $ . Vice versa, she might also try to avoid and postpone positive returns in order to have some good news in reserve such that the probability for a loss from $ {t}_1 $ to $ {t}_2 $ is reduced. However, the manager’s inclination to provide biased forecasts and cater to the investor’s preferences is bound by the costs associated with biased guidance. More specifically, these costs imply that the optimal guidance bias $ {b}^{\ast } $ always lies between the first optimal solution (lower boundary) and the last optimal solution (upper boundary) shown in Equation (6).

According to Equation (6), if both $ \lambda =0 $ (no loss aversion) and $ \beta =1 $ (no discounting), the manager would not deceive the market ( $ {b}^{\ast }=0 $ ). Keeping $ \lambda =0 $ , a manager’s discounting ( $ \beta <1 $ ) implies a positive forecast bias $ {b}^{\ast } $ as the manager prefers immediate over later utility. In particular, without loss aversion, $ {b}^{\ast } $ does not depend on the stock prices $ {P}_0 $ and $ {P}_1^{unbiased} $ . On the contrary, with $ \lambda >0 $ , the optimal bias $ {b}^{\ast } $ is a function of $ {P}_0-{P}_1^{unbiased} $ . This key model prediction also remains qualitatively the same if the manager does not apply any discounting ( $ \beta =1 $ ).

The three-part optimal guidance bias $ {b}^{\ast } $ from Equation (6) is also depicted in Figure 2. As evident from this figure, our stylized model yields several testable predictions with respect to the optimal guidance bias $ {b}^{\ast } $ . Most importantly, $ {b}^{\ast } $ strongly depends on the stock’s previous performance. If the fair value $ {P}_1^{unbiased} $ implies a loss relative to the purchase price $ {P}_0 $ , managers will provide upward-biased forecasts to reduce the magnitude of incurred losses. For stocks in the gain domain, managers might still communicate a positive bias $ b $ if their present preference is very strong, but might also issue a negative bias to have some good news in reserve. Notwithstanding, $ b $ will be lower compared to the loss domain because the management does not face strong pressure from discontent investors. These considerations imply our first hypothesis.

Hypothesis 1. Management guidance biases decrease in the investors’ return since purchase.

FIGURE 2 Managerial Choice of Optimal Guidance Bias

Figure 2 depicts the relationship between the stock’s previous performance (i.e., the hypothetical stock return $ {P}_1^{unbiased}/{P}_0-1 $ ) and the guidance bias $ {b}^{\ast } $ chosen by the manager to maximize her expected level of intertemporal utility. The baseline scenario uses $ {P}_0=100 $ as the investor’s initial stock purchase price, $ \lambda =1.25 $ to reflect loss aversion, a discount factor of $ \beta =0.75 $ , personal costs for issuing biased forecasts of $ c=2 $ , and final payoff uncertainty $ \sigma =40 $ . Beyond this baseline scenario, the “lower $ \beta $ -scenario” applies $ \beta =0.65 $ , the “higher $ \sigma $ -scenario” applies $ \sigma =50 $ , and the “lower $ c $ -scenario” applies $ c=1 $ .

In addition to this baseline effect, our model predicts in which situations the relationship between guidance bias and previous stock returns should be particularly strong. For example, a stronger managerial preference for immediate compared to future utility (stronger discounting via lower $ \beta $ ) carries two implications (see Equation (6) and Figure 2). First, managers will issue more optimistic forecasts to transfer later utility to today. Second, the bias will increase in magnitude as loss aversion and personal cost considerations from $ {t}_2 $ play a smaller role.Footnote ⁷

Further, if the uncertainty with respect to the final payoff is higher (higher $ \sigma $ ), the magnitude of guidance biases increases. To illustrate this, consider a given positive level of guidance bias $ b $ . Then, a higher level of $ \sigma $ reduces the probability that this positive bias leads to negative stock returns between $ {t}_1 $ and $ {t}_2 $ and a corresponding utility loss. Moreover, if $ \sigma $ is high and guidance is imprecise anyway, issuing biased guidance is comparably less costly.

Finally, managers will issue more biased forecasts if the costs $ c $ for providing inaccurate estimates are low.

We empirically examine these four hypotheses in the following.

1. Dependent Variables

To measure biases in managerial earnings guidance, we define the $ Guidance\ Bias $ for fiscal year $ t $ as the difference between forecasted and realized earnings per share $ {EPS}_t $ , deflated by the stock price $ {P}_{t-1} $ at the beginning of the associated fiscal year (Ajinkya et al. (Reference Ajinkya, Bhojraj and Sengupta2005), Rogers and Stocken (Reference Rogers and Stocken2005), Baik et al. (Reference Baik, Farber and Lee2011)):

(7) $$ Guidance\hskip0.42em {Bias}_t=\frac{EPS_{forecasted,t}-{EPS}_{realized,t}}{P_{t-1}}\times 100. $$

Hence, the $ Guidance $ $ Bias $ measures which proportion of the firm’s market capitalization is earned more or less than predicted. Positive biases indicate excessively optimistic forecasts, while negative values indicate managerial pessimism relative to the ex post realization.

To measure the market’s reaction toward both guidance (announcement of $ {EPS}_{forecasted} $ ) and actual earnings (announcement of $ {EPS}_{realized} $ ), we calculate cumulative abnormal announcement returns in symmetric 3-day windows around these event dates to determine $ Guidance $ $ CAR $ and $ Earnings $ $ Announcement $ $ CAR $ , respectively. Importantly, we focus on the subsequent annual earnings announcement where the earnings, which the guidance refers to, are announced. Abnormal returns are calculated relative to the Fama and French (Reference Fama and French1993) 3-factor model. Following Savor (Reference Savor2012), the underlying factor loadings are estimated using the 255 trading days prior to the event date with a 31-day gap. We require at least 30 valid return observations to estimate factor loadings and event returns.

2. Main Explanatory Variable

All explanatory variables are based on firm-level information available at the last turn of month before the management forecast. Annual accounting information is updated each year at the end of June based on the fiscal period that ends in the previous calendar year; this time lag of at least 6 months ensures that the accounting information is publicly available when using it as predictive variable (Fama and French (Reference Fama and French1993)). Since we are interested in the long-run valuation effects of biased forecasts on earnings announcements, we link the later $ Earnings $ $ Announcement $ $ CAR $ to the same explanatory variable values as the corresponding $ Guidance $ $ CAR $ .

Our model predicts that the $ Guidance $ $ Bias $ depends on the investors’ return since purchase. Therefore, we follow Grinblatt and Han (Reference Grinblatt and Han2005) in estimating the aggregate investor’s purchase price to calculate the capital gains overhang measure $ CGO $ , which measures the average investor’s stock return since share purchase. Following Grinblatt and Han (Reference Grinblatt and Han2005), the estimation is based on weekly price and trading volume data over the previous 5 years such that the market’s average purchase price in week $ t $ is given as

(8) $$ {P}_{pur,t}=\frac{1}{k_t}\sum \limits_{n=1}^{260}\left({Vol}_{t-n}\prod \limits_{\tau =1}^{n-1}\left[1-{Vol}_{t-n+\tau}\right]\right){P}_{t-n}, $$

where $ {Vol}_t $ and $ {P}_t $ represent the stock’s turnover ratio and split-adjusted stock price in week $ t $ , respectively. We truncate the turnover ratios at a maximum level of 1 such that the term in parentheses reflects the proportion of outstanding shares that was purchased for $ {P}_{t-n} $ in week $ t-n $ . We scale the overall sum by the sum of these proportions $ {k}_t $ such that the 260 weekly weights add up to 1. In order to estimate $ {P}_{pur,t} $ , we require at least 100 weekly observations. Based on this aggregate cost basis, we calculate $ CGO $ as

(9) $$ {CGO}_t=\frac{P_{t-1}-{P}_{pur,t}}{P_{pur,t}}. $$

Our event study uses the last weekly $ CGO $ estimate in the month prior to the first guidance announcement. High values of $ CGO $ indicate that managers face an investor base that is comparably satisfied with the respective stock investment prior to the guidance announcement. The prior empirical literature has emphasized the importance of the purchase price as reference price for investors, the strong influence of $ CGO $ on the evaluation of their investments, and thus the impact of $ CGO $ on trading behavior (Shefrin and Statman (Reference Shefrin and Statman1985), Ben-David and Hirshleifer (Reference Ben-David and Hirshleifer2012), An et al. (Reference An, Wang, Wang and Yu2020), and Riley et al. (Reference Riley, Summers and Duxbury2020)).Footnote ⁹ Thus, even if managers do not tally the purchase price of each investor to compute their respective capital gains, there is strong evidence that $ CGO $ captures investors’ perception of past performance (which managers are likely to perceive and react to).

3. Control Explanatory Variables

Our multivariate analyses control for various stock characteristics that have been shown to predict guidance accuracy and guidance biases. For brevity, we shortly introduce these variables in the following, but provide detailed definitions in the Supplementary Material only.

We include the stocks’ market $ Beta $ because prior studies show that increased market risk reduces forecast accuracy (Feng et al. (Reference Feng, Li and McVay2009), Baik et al. (Reference Baik, Farber and Lee2011)). Moreover, we control for $ Size $ as larger firms have been shown to display a smaller $ Guidance $ $ Bias $ and a higher forecast accuracy (Ajinkya et al. (Reference Ajinkya, Bhojraj and Sengupta2005)). Similarly, we employ the $ Book $ - $ to $ - $ Market $ ratio, which arguably proxies for a firm’s growth opportunities and has been related to forecast properties by the prior literature (Bamber and Cheon (Reference Bamber and Cheon1998), Rogers and Stocken (Reference Rogers and Stocken2005)).

Since forecasts made earlier within a given year have been shown to be less accurate and more biased (Ajinkya et al. (Reference Ajinkya, Bhojraj and Sengupta2005), Feng et al. (Reference Feng, Li and McVay2009), Hribar and Yang (Reference Hribar and Yang2016)), we control for the forecast’s time until the period end, $ Horizon $ . Further, we include a $ Loss\ Indicator $ based on the firm’s income before extraordinary items as loss firms’ forecasts have been found to be less accurate (Baik et al. (Reference Baik, Farber and Lee2011), Hilary and Hsu (Reference Hilary and Hsu2011)). As Rogers and Stocken (Reference Rogers and Stocken2005) relate industry-specific litigation risk to management guidance, we include a $ Process\ Risk $ dummy based on the firm’s SIC code (Cheng and Warfield (Reference Cheng and Warfield2005)). In addition, we control for the $ Prior\ Error $ , which is the one-period lag of $ Guidance\ Bias $ to capture autocorrelation and persistence in guidance biases (set to 0 if missing).

Furthermore, the broader accounting performance has also been related to guidance properties, prompting us to include Operating Profitability as additional control. Since analyst attention has been shown to constrain management’s willingness to issue inaccurate forecasts (Rogers and Stocken (Reference Rogers and Stocken2005)), we control for the number of analysts ( $ Analyst\ Coverage $ ) and their disagreement ( $ Analyst\ Dispersion $ ) calculated according to Johnson (Reference Johnson2004). We also control for the effect of a stock’s return volatility $ VOLA $ (see, e.g., Houston, Lev, and Tucker (Reference Houston, Lev and Tucker2010)). Similarly, firms accessing capital markets might be incentivized to issue positively biased guidance (Hribar and Yang (Reference Hribar and Yang2016)), leading us to control for a $ Net\; Equity\ Issuer $ dummy variable. Moreover, Bergman and Roychowdhury (Reference Bergman and Roychowdhury2008) illustrate how managers adjust their disclosure strategy to the prevailing investor sentiment, while Seybert and Yang (Reference Seybert and Yang2012) show how sentiment affects $ Guidance\; CAR $ . Following these articles, we use the $ Investor\ Sentiment $ index from the Michigan Consumer Confidence Index as control.

In additional analyses, we use $ Share\ Turnover $ and $ Relative $ $ Short $ $ Interest $ , both calculated relative to shares outstanding, as measures of the present preference of guiding managers in our interaction tests. Further, we control for stock returns over different windows prior to the guidance date—Return 1M, Return 3M, Return 1Y, Return 3Y, and Return 5Y—to differentiate between the investors’ average return since purchase ( $ CGO $ ) and the recent stock performance in general. Moreover, we estimate EW CGO, a capital gains overhang measure where we assume identical turnover weights in the cross-section of stocks. More explicitly, for each week $ t $ , we estimate the median weekly stock turnover over the previous 5 years over all common ordinary U.S. stocks. We use this median value as $ Vol $ in Equation (9). Hence, EW CGO reflects a stock’s past returns with exponentially decaying weights.

Finally, we control for industry-fixed effects via 2-digit SIC codes in our regressions. All continuous variables are winsorized at the 1% and 99% levels.Footnote ¹⁰ Our sample includes all firm-year observations with available data for the three main dependent variables (bias in management guidance as well as cumulative abnormal announcement returns around guidance and earnings announcement dates) and the main independent variable $ CGO $ .

1. CGO Versus Past Returns

The prior tests consistently establish that $ CGO $ predicts $ Guidance\ Bias $ as implied by our model. However, one might argue that our regressions just capture correlations between $ CGO $ and $ Guidance\ Bias $ and thus fail to identify a causal relation between investors’ experienced returns and managerial behavior. It is important to note that we do not claim that managers sit down with a calculator and compute each investor’s rate of return as basis for their guidance strategy. Rather, we theoretically identify $ CGO $ as a central determinant of investors’ utility such that it should affect managerial catering because managers should be broadly aware of their investors’ experienced returns. However, one might argue that $ CGO $ also reflects other economic mechanisms since it is closely related to the stock return over the previous 5 years by construction. If this raw return or related measures of firm performance had a direct impact on management guidance, these effects might be captured by $ CGO $ . We therefore add the firm’s stock returns over several event windows prior to the guidance date—1 month, 3 months, 1 year, 3 years, and 5 years—as control variables. While these return variables should capture general arguments related to firm performance, the remaining effect associated with $ CGO $ exhibits a clear connection to investors’ experienced returns since $ CGO $ reflects the turnover-weighted returns of the previous 5 years.

Table 4 reports regression results of Guidance Bias on $ CGO $ and the five different return variables separately in columns 1 to 5 and jointly in column 6. Notably, CGO remains statistically significant at the 1% level and economically large (coefficients between −1.11 and −1.89) across all specifications. One of the five return variables (Return 1Y) is also significantly related to Guidance Bias, yet does not lead to a loss of significance in the CGO–Guidance Bias relation. These results emphasize that managers are clearly more responsive to the returns experienced by investors than the overall return over the past years, which provides further evidence for the proposed catering channel.Footnote ¹⁴

TABLE 4 Guidance Bias and Capital Gains Overhang: Past Return

Finally, we insert EW CGO alongside CGO in column 7 of Table 4. This specification constitutes a particularly demanding test of the catering theory of earnings guidance: EW CGO is also computed based on weekly stock returns over the preceding 260 weeks, yet these returns are weighted with market-wide median (instead of firm-specific) turnover. Consequently, EW CGO captures all arguments related to weighted past returns. Hence, the remaining effect of CGO essentially stems from firms with the same past returns but different investor capital gains that we infer from different purchase dates based on the different turnover dynamics. Controlling for EW CGO—which exhibits a correlation with CGO of 80.50%—reduces the CGO coefficient to −0.76. Nonetheless, the effect of CGO remains both statistically significant at the 1% level and economically large.

2. Catering Versus Biased Beliefs

As alternative underlying mechanism, the documented effect of $ CGO $ on $ Guidance\ Bias $ might also reflect a systematic managerial bias akin to the investor bias associated with $ CGO $ (Grinblatt and Han (Reference Grinblatt and Han2005)). In addition, the managers’ own beliefs might be excessively optimistic if $ CGO $ is low such that our key findings might reflect managers’ judgment biases rather than catering. However, this alternative line of argument is difficult to reconcile with the interaction tests in Table 3 that support the more nuanced catering implications of our model. To further bolster the evidence for intentional managerial catering, we explicitly address three channels through which managerial biases could be tied to $ CGO $ :

i) Biases in beliefs are frequently associated with high levels of sentiment (Seybert and Yang (Reference Seybert and Yang2012), Stambaugh, Yu, and Yuan (Reference Stambaugh, Yu and Yuan2012)). If biased managerial beliefs caused the effect of $ CGO $ on Guidance Bias, we would expect it to intensify substantially in times of high investor sentiment. Interaction tests in our Supplementary Material show that the CGO effect does not depend strongly on sentiment and remains economically large in times of low sentiment.

ii) Overconfidence is a frequently studied distortion of managerial beliefs (see, e.g., Malmendier and Tate (Reference Malmendier and Tate2005), (Reference Malmendier and Tate2008)). Such overconfidence might manifest in excessively optimistic earnings forecasts after poor stock returns because managers disregard the cause of the stock price decline or overestimate their ability to address it. Then, the effect of $ CGO $ on Guidance Bias should be particularly strong for overconfident managers. In fact, we show in our Supplementary Material that the effect of $ CGO $ on earnings guidance is—if anything—smaller for overconfident managers. This result indicates that biased managerial beliefs constrain $ CGO $ -dependent catering (instead of causing it).

iii) The effect of $ CGO $ on earnings guidance could emerge because $ CGO $ might also reflect the manager’s capital gains, which in turn could relate to her beliefs—for example, due to motivated reasoning (Zimmermann (Reference Zimmermann2020)). We document in our Supplementary Material that the effect of $ CGO $ is unaffected by the inclusion of the CEO’s equity share as interaction term.Footnote ¹⁵ This finding is at odds with the assumption that managerial capital gains drive the $ CGO $ effect as even managers without equity share engage in catering. Overall, our results support the notion that intentional managerial catering is the source of the $ CGO $ effect in earnings guidance.

3. Further Robustness Tests

We provide a number of further robustness tests in the Supplementary Material. In particular, we document that the effect of $ CGO $ on $ Guidance\ Bias $ is economically large and of comparable magnitude for point and range forecasts based on subsample tests. Additionally, we document that the CGO–Guidance Bias relation is large irrespective of the share of institutional ownership. Moreover, our findings could be related to Baginski et al. (Reference Baginski, Campbell, Ryu and Warren2023) who document that a positive (negative) $ Guidance\ Bias $ is associated with negative (positive) earnings surprises revealed at concurrent earnings announcements. We do not control for earnings surprises in our prior analyses because our sample also includes guidance announcements without concurrent earnings announcements. All results presented in this article, however, remain qualitatively unchanged if we limit our sample to guidance announcements with concurrent earnings announcements and control for earnings surprises. We report two central pieces of evidence in our Supplementary Material: i) We show that the effect of $ CGO $ on $ Guidance\ Bias $ remains significant in the subsample of guidance announcements without concurrent earnings announcements. ii) Controlling for earnings surprises does not change the effect of $ CGO $ on $ Guidance\ Bias $ .

Figure 2 implies that the effect of the investor’s past returns is large for moderate values of past returns and flattens out for extreme positive and negative values. We provide two pieces of empirical evidence, which are consistent with such a reduced sensitivity toward extreme returns. i) Escalating the levels of CGO winsorization increases the negative effect of CGO on Guidance Bias. ii) Interacting CGO with a dummy variable, which is equal to 1 if the absolute value of CGO is smaller than 10%, and 0 otherwise, confirms that the effect is strongest for moderate CGO values.

Finally, our catering theory rests on the assumption that discontented investors exert pressure on the manager such that the manager’s intertemporal utility depends on both short-term and long-term firm performance. We argue that CEOs might be particularly inclined to engage in $ CGO $ -dependent catering if the stock performance reflected by $ CGO $ can be attributed to them. In this case, the link between investor and manager utility should be stronger because, for example, investors’ utility losses might result in investor rebellion and CEO dismissal. Analyses based on recently employed CEOs confirm that the effect is driven by established managers and insignificant for newly appointed managers.

4. Ex Post Consequences of Catering

CEO-Specific Consequences of Catering. Our theoretical model predicts that Guidance Bias depends on CGO because managers respond to investors’ reference point-dependent loss aversion. The prior empirical evidence strongly supports this proposition, while also pointing toward substantial heterogeneity in this propensity to cater. In addition to the stock market consequences, which we examine in Section V, this also raises the question, whether catering via earnings guidance has any implications for CEOs’ careers. In line with the intuition laid out in our theoretical model, we conjecture that CEOs’ responsiveness to the investor mood is advantageous for them.

We test this hypothesis by identifying CEOs whose biases in forecasts are consistent with catering, that is, they tend to issue excessively optimistic (pessimistic) guidance when investors are in their loss (gain) domain. We provide two pieces of evidence in our Supplementary Material that suggest that these CEOs are indeed better off. First, comparing average career outcomes, we find that catering CEOs are associated with higher total compensation—in particular driven by incentive-based pay—and longer tenures. Second, we examine the annual likelihood of CEO turnover, CEO dismissal, and total compensation in a firm-year panel. The results confirm that catering CEOs are less likely to be replaced and receive higher compensation.Footnote ¹⁶ These results suggest that catering via earnings guidance pays off for CEOs.

Firm-Specific Consequences of Catering. Earnings guidance is a particularly well-suited laboratory to test theories of managerial catering, given the comparably low personal costs for managers if forecasts do not materialize exactly as predicted. Whether such catering via earnings guidance coincides with other forms of misstatement is an empirical question, which we test using the Dechow et al. (Reference Dechow, Ge, Larson and Sloan2011) dataset on material accounting misstatements. We conjecture that managers of low-CGO firms might not only publish too optimistic earnings guidance, but might also have the strongest incentives to publish incorrect accounting figures. In line with this conjecture, our Supplementary Material shows that catering goes along with a negative relationship between CGO and the likelihood of a concurrent misstatement of financial accounting data. We interpret this as initial evidence that the incentives that cause firms to cater via earnings guidance might also affect other, more costly forms of biased disclosure.Footnote ¹⁷ Consequently, this result points toward the wider implications of investors’ reference point-dependent loss aversion for corporate policies.