Does Accounting Conservatism Make Good News Forecasts More Credible and Bad News Forecasts Less Alarming?

Abstract

I investigate whether conservatism in financial statements helps investors to interpret the information revealed through voluntary disclosure, specifically earnings guidance. Research shows that investor reaction to guidance is affected by two issues. First, managers’ incentives to please the market cause positive guidance to lack credibility. This makes investors underreact initially to good news forecasts. Second, the same incentives cause negative guidance to raise investor uncertainty amid mounting concerns that the bad news disclosure is incomplete and more bad news will soon follow. This makes investors overreact initially to bad news forecasts. I argue that conservatism can increase the credibility of good news forecasts by subjecting good news to stronger verification requirements before being recognized in future reports. Consistent with this argument, I find that, when conservatism is higher, positive forecast news is associated with (a) a more positive initial response and (b) a less positive subsequent drift. I also argue that conservatism, by reassuring investors that the recognition of bad news in periodic reports has been complete and timely, can mitigate the surge in uncertainty that follows negative guidance and contain investor overreaction. Consistent with this argument, I also find that, when conservatism is higher, negative forecast news is associated with (a) a less negative initial response, (b) a less positive subsequent correction, and (c) a smaller increase in investor uncertainty.

Keywords

accounting conservatism management forecasts earnings guidance market reaction investor uncertainty

Introduction

Conditional accounting conservatism (hereafter just conservatism) is commonly defined as the accountant’s tendency to require a higher degree of verification for the recognition of profits than for the recognition of losses (Basu, 1997). Conservatism results in the timely recognition of bad news and in the delayed recognition of difficult-to-verify good news, which is not recognized until stronger verification requirements have been met. This article investigates whether these properties of conservatism affect investor reaction to voluntary disclosure. In particular, I analyze whether the level of conservatism of a firm conditions the market’s response to management earnings forecast news as measured by the difference between the forecast estimate and preguidance analyst consensus.¹

The impact of conservatism on the equity market is a debated issue, made particularly relevant by studies showing that conservatism has increased in recent decades (Givoly & Hayn, 2000; Givoly, Hayn, & Katz, 2017). Standard setters have opposed conservative reporting, as it can provide biased information to investors (Watts, 2003). Some empirical studies support regulators’ concerns, reporting evidence of adverse valuation consequences of conservatism (Barth, Landsman, Raval, & Wang, 2017; Chen, Folsom, Paek, & Sami, 2014; Mensah, Song, & Ho, 2004; Penman & Zhang, 2002).

Other papers, however, find that conservatism is associated with beneficial outcomes for investors. For instance, LaFond and Watts (2008) show that conservatism arises as a response to mitigate information asymmetry in the stock market. This is consistent with empirical evidence suggesting that conservatism decreases the cost of capital and stock returns volatility and improves the firm’s information environment (D’Augusta, Bar-Yosef, & Prencipe, 2016; Garcìa Lara, Garcìa Osma, & Penalva, 2011, 2014). In addition, Y. Kim, Li, Pan, and Zuo (2013) report that conservatism mitigates information asymmetry around seasoned equity offerings. In a similar vein, B. H. Kim and Pevzner (2010) and J.-B. Kim and Zhang (2016) suggest that conservatism prevents the accumulation of bad news and thus reduces the likelihood of future price crashes. Moreover, Balakrishnan, Watts, and Zuo (2016) show that conservative firms enjoyed better stock performance during the financial crisis.

A recurring argument shared by many of these studies is the assumption that conservatism in financial statements benefits other “softer” information sources, such as voluntary disclosures (LaFond & Watts, 2008). Being subject to lower scrutiny than reported earnings, good news revealed through these disclosures can lack credibility in the eyes of investors, who can regard it as self-serving “cheap talk” by managers. In addition, managers’ reluctance to voluntarily disclose bad news can leave investors uncertain about whether they have been told the whole story. Both concerns can severely impair the usefulness of voluntary disclosure for investors. Conservative accounting numbers can mitigate this problem by serving as a “hard” benchmark against which investors can compare the information obtained from other disclosures (LaFond & Watts, 2008). Thus, by increasing the timeliness of bad news and the verification of good news that is revealed through recognition in mandatory financial statements, conservatism is assumed to help investors evaluate the information that managers convey through voluntary disclosure. However, this assumption has lacked empirical support so far. Is investor reaction to the release of voluntary disclosure influenced by conservatism? I address this issue by analyzing the stock market response to management earnings forecasts.

Research has established that forecasts are a value-relevant form of voluntary disclosure (Hirst, Koonce, & Venkataraman, 2008). Unlike reported earnings, forecasts are voluntary, forward-looking, unaudited, and often contain range estimates. Various studies show that these differences matter to investors when evaluating forecast news (Baginski, Conrad, & Hassell, 1993; Das, Kim, & Patro, 2012; Hutton, Miller, & Skinner, 2003; Jennings, 1987; Rogers, Skinner, & van Buskirk, 2009; Rogers & Stocken, 2005). Good news forecasts lack credibility: aware of managers’ incentives to please the market, investors react skeptically at first and wait for additional confirming signals to fully incorporate the information into prices. This translates to an initial underreaction followed by a positive drift. On the contrary, bad news forecasts are deemed credible and thus elicit a full immediate response, because of managers’ incentives to avoid bad news when they can. However, these same incentives can raise concerns that bad news has not been completely revealed and more is to come. The surge in investor uncertainty can cause an initial overreaction that is followed, again, by a positive drift as uncertainty is resolved over time and investors correct their valuation upward.

Therefore, the literature highlights that the lack of credibility of good news and the mounting uncertainty following bad news can impair the usefulness of earnings guidance. Some papers have investigated forecast attributes that can mitigate this problem (Hutton et al., 2003; Hutton & Stocken, 2009; Jennings, 1987; Ng, Tuna, & Verdi, 2013; Rogers et al., 2009; Yang, 2012). While mostly focusing on the characteristics of firms’ forecasting policies, this literature has dedicated less attention to whether financial statement characteristics can also improve the usefulness of guidance. In particular, to the best of my knowledge, the literature has not addressed whether conservatism can lend credibility to good news forecasts or mitigate uncertainty following bad news forecasts.

I fill this gap by analyzing the 3-day price response to earnings guidance and subsequent price drift. Prior studies show that investor response to new information can be conditioned by prior information that investors use to complement how they interpret the new information (Banker, Das, & Datar, 1993; Conrad, Cornell, & Landsman, 2002; Ho, Liu, & Ramanan, 1997; Liu, Ryan, & Wahlen, 1997). I argue that investors can combine their prior knowledge about a firm’s level of conservatism with information contained in the forecast to make investment decisions; therefore, conservatism can complement forecast news and condition the market’s response to it. First, I hypothesize that conservatism attenuates investor underreaction to good news forecasts. If managers know that good news is going to be subject to higher verification standards in periodic reports, they will be less likely to base positive guidance on less-verified information for fear of being constrained from meeting their own predictions. If better verified, good news will be less likely to be overstated or to incorporate uncertain, highly volatile components and, thus, will be more predictive of future profitability. This will enhance the credibility of good news forecasts as signals of underlying positive economic performance, thus eliciting a more positive immediate response and a smaller positive drift.

Second, I hypothesize that conservatism attenuates investor overreaction to bad news forecasts. When assessing the implications of unexpected negative guidance, investors can fear that more bad news, which has been withheld so far, will soon follow. These concerns can raise investor uncertainty (Rogers et al., 2009) and cause an initial overreaction to negative guidance followed by an upward correction (Das et al., 2012; Epstein & Schneider, 2008). By making the recognition of losses timelier in periodic reports, conservatism could prevent bad news from accumulating (B. H. Kim & Pevzner, 2010; J.-B. Kim & Zhang, 2016) and reassure investors that they possess a fuller picture of the negative situation when bad news is unexpectedly released. Conservatism will also make the benefits from releasing incomplete bad news forecasts short lived for managers, as bad news is likely to be quickly exposed in the next period’s financial statements. By relieving investors’ concerns and attenuating mounting uncertainty, conservatism will result in a less negative initial response and a smaller positive drift when uncertainty is resolved. Moreover, conservatism will be associated with a smaller increase in uncertainty after bad news forecasts.

I test my hypotheses on a sample of 9,529 point or closed range earnings forecasts released by industrial firms listed on the New York Stock Exchange (NYSE), American Stock Exchange (AMEX), and NASDAQ. Supporting the first hypothesis, I find that conservatism significantly augments the 3-day market response coefficient to good news forecasts. Also supporting the first hypothesis, I find that conservatism attenuates the positive drift following the forecast release. The results show that positive forecast news issued by firms in the bottom decile of conservatism is associated with positive abnormal returns in the postannouncement weeks and in the 3 days around the announcement of reported earnings (if such an announcement occurs within 45 days of the forecast’s release). The delayed response to forecast news becomes significantly weaker as conservatism increases. This suggests that conservatism makes good news more credible and thus helps investors to interpret positive guidance, leading to a stronger initial reaction and a smaller drift. Consistent with this argument, I also find that conservatism’s effect on post–good news drift tends to be stronger in subsamples characterized by higher levels of stock returns volatility, where credibility concerns are likely to be more acute.

Supporting the second hypothesis, I find a larger negative price response for bad news forecasts issued by firms in the bottom conservatism decile, followed by positive abnormal returns in the postforecast period, when investors correct the initial overreaction. Both the initial negative reaction and the following upward correction are progressively alleviated as conservatism increases. This suggests that conservatism can make negative guidance less alarming in the eyes of investors, thus mitigating their overreaction. Consistent with these results being driven by a spike in uncertainty that conservatism helps to mitigate, I find that negative forecast news is associated with an increase in several uncertainty proxies and that conservatism weakens this association. In addition, I find that conservatism’s effect on post–bad news drift tends to be stronger in subsamples characterized by higher levels of uncertainty.

This article contributes to the current literature in multiple ways. First, it provides evidence that conservatism in financial reports enhances the usefulness of other disclosure venues to investors. This is of interest to regulators and standard setters, as it adds to the current debate about whether conservatism’s consequences on the stock market are beneficial or detrimental.

This article also contributes to the literature on market reactions to earnings guidance. It is well established that guidance suffers from low credibility (when the news is good) and from mounting investor uncertainty (when the news is bad), two distinct problems that can significantly undermine the usefulness of this form of disclosure. However, the mechanisms that can address these problems are a less well-defined issue. By suggesting that conservatism can be one such mechanism, this article is of interest to both financial statement users and preparers. In particular, this study provides an incremental contribution over two papers by Das et al. (2012) and Ng et al. (2013). My findings agree with the results in these papers, both of which document drift patterns in the postforecast period similar to those that I find. However, neither Das et al. (2012) nor Ng et al. (2013) study the role of conservatism with respect to such drift patterns. By finding a more complete response and less of a price drift under conservatism, I expand beyond their insights.

In addition, this article contributes to the literature on the interaction between conservatism and management forecasts. This literature has focused on the association between conservatism and forecasts’ attributes and how it is mediated by firm fundamentals such as litigation risk, information asymmetry, and corporate governance (Hui, Matsunaga, & Morse, 2009; Jaggi & Xin, 2014). My article’s aim is not to examine how conservatism is related to forecast or firm characteristics, which are controlled for in all models. My goal, instead, is to address the different question of whether conservatism moderates the market response to forecast news of a given magnitude, all else being equal. Therefore, this study provides an incremental contribution over the work of Hui et al. (2009); while they focus on whether conservatism can substitute for guidance in addressing problems such as litigation risk, I study whether conservatism can also be a complement to guidance by conditioning the market’s response to it.

The remainder of this article is organized as follows. The following section reviews the literature and develops the hypotheses. Next, the research design is described. The main results and robustness tests are then presented, respectively, in the two subsequent sections. Concluding remarks are discussed in the final section.

Positioning and Hypothesis Development

Credibility of Good News Forecasts and Investor Underreaction

As their compensation and tenure are tied to economic performance, managers benefit from revealing positive outcomes and concealing negative ones. These asymmetric incentives could induce them to act strategically to meet expectations or to temporarily increase their firms’ stock prices (Ali & Gurun, 2009; Payne & Robb, 2000). For instance, managers could intentionally inflate good news, announce it before duly verifying it, or omit to disclose whether they expect it to have low persistence.

Due to these concerns, good news can lack credibility, which is defined as the extent to which investors believe the news (Jennings, 1987). Lack of credibility can impair the usefulness of a signal, as investors will react to it only to the extent that they believe it. This is especially true for management forecasts, which differ from reported earnings announcements in many respects (Ng et al., 2013). First, as noted by Hirst et al. (2008), the fact that forecasts are a “subjective and unaudited projection of future events” (p. 329) leads to concerns about the verifiability of their information, which caused early research to question whether guidance was relevant at all. Second, forecasts are often ranges of estimates rather than precise numbers. This can make them harder to interpret, as investors wonder about how much weight to attribute to the various possible realizations (e.g., upper bound, lower bound, midpoint) before updating their expectations.² The width of the forecast range also grants managers protection from litigation (and therefore stronger incentives to bias the estimates), as long as actual earnings are not materially below the lower bound. In addition, the forward-looking nature of guidance allows managers more leeway in justifying erroneous estimates to regulators. This can severely damage forecasts’ credibility if it provides firms with a “license to lie” (Rogers & Stocken, 2005, p. 1234). Moreover, managers’ forward-looking predictions can be biased upward by overconfidence (Hribar & Yang, 2016). All these considerations make investor reaction to management forecasts a powerful setting for testing which factors enhance the credibility of good news in the eyes of investors (Ng et al., 2013).

Empirical evidence is consistent with good news forecasts suffering from low credibility. Jennings (1987) shows that the market underreacts to positive guidance that analysts view as less credible. Rogers and Stocken (2005) document that managers’ incentives cause good news forecasts to be biased upward and that investors, aware of the risk of bias, underreact to positive guidance. Extending the analysis to the postforecast period, Das et al. (2012) show that the initial underreaction to positive guidance gives rise to a positive drift in the following weeks, as investors wait for confirming signals to corroborate the good news. Exploring what mechanisms can lend credibility to positive guidance so that it elicits a stronger reaction, researchers have focused on forecast-related attributes, such as a smaller forecast range (Baginski et al., 1993), higher prior forecasting accuracy (Hutton & Stocken, 2009; Ng et al., 2013; Yang, 2012), and whether the forecast is accompanied by ex post verifiable claims as opposed to just talk (Hutton et al., 2003). Different from these studies and contributing to this literature, my article focuses on the disciplining role of mandatory financial statements’ numbers as an additional mechanism to enhance good news forecasts’ credibility. In particular, to the extent of my knowledge, no previous study has examined whether a firm’s level of conservatism boosts investor reaction to positive guidance.

Uncertainty About Bad News Forecasts and Investor Overreaction

Because disclosing bad news goes against managers’ incentives, negative guidance is considered to be “inherently more credible” (Hutton et al., 2003, p. 871): investors fully react to it without wondering whether they should believe it or not (Rogers & Stocken, 2005). Consistently, empirical studies show that price reactions to bad news forecasts are stronger (Ng et al., 2013), whether the forecasts are accompanied by verifiable statements or not (Hutton et al., 2003).³

While they make negative guidance credible, managers’ incentives to avoid bad news can also raise investors’ concerns that what they have been provided is not the whole picture. Therefore, the release of bad news can cause a spike in investor uncertainty as investors weigh the probability that additional bad news of an unknown magnitude will be eventually revealed.⁴ This concern is robustly supported by empirical research. Kothari, Shu, and Wysocki (2009) show that managers tend to release bad news only if they believe it to be irreversible; otherwise, they tend to withhold bad news and let it accumulate, gambling that “subsequent corporate events will allow them to bury the bad news” (Kothari et al., 2009, p. 242). Kasznik and Lev (1995) report that warnings tend to be issued for permanent earnings disappointments, rather than transitory disappointments, and Tucker (2007) suggests that managers’ decisions to warn investors are associated with the existence of other bad news that is not incorporated in the earnings warning. Moreover, Rogers et al. (2009) show that only bad news forecasts increase investor uncertainty, while good news forecasts tend to temporarily reduce it.

The surge in uncertainty that follows negative guidance is increased by the same considerations that cause positive guidance to lack credibility. Different from losses recognized in financial statements, bad news forecasts are unaudited, forward-looking ranges of estimates that are more exposed to managerial subjectivity and error and, therefore, represent a more ambiguous signal. In addition, negative guidance is voluntary. Investors will react not only to its content but also to the manager’s decision to disclose it, which the investors need to interpret (Das et al., 2012). If they see such a decision as a potential sign that bad news has reached the level where it can no longer be withheld, their uncertainty will grow. Das et al. (2012) show that these considerations make investors initially overreact to negative guidance. In the following weeks, as uncertainty is gradually resolved, the overreaction is corrected and the stock price experiences a positive drift. However, Das et al. do not explore what factors could relieve investor uncertainty and mitigate the overreaction. In particular, to the extent of my knowledge, no previous papers have directly addressed whether any property of the numbers recognized in periodic reports, such as the level of conservatism, can play a role in preventing the spike in uncertainty and investor overreaction.

Complementariness Between Conservatism and Forecast News

Unlike forecast news, the level of conservatism of the forecasting firm is likely to be a known piece of information to investors on the release day. Prior literature suggests that the level of conservatism of a firm’s financial statements tends to be stable over time (e.g., Garcìa Lara, Garcìa Osma, & Penalva, 2016; Khan & Watts, 2009). Moreover, the literature that examines the effect of conservatism on the decisions of capital providers (e.g., Balakrishnan et al., 2016; Garcìa Lara et al., 2011; Y. Kim et al., 2013) is consistent with the assumption that financial statement users, such as investors, are able to discern the firm type (i.e., high conservatism vs. low conservatism). Various studies suggest that such an assumption is plausible. For instance, managers are reportedly willing to reveal their type to maximize their utility and build a reputation for the timely disclosure of losses (Ball & Shivakumar, 2005; Guay & Verrecchia, 2007). This is in line with evidence that analysts are able to assess the degree of a firm’s conservative reporting with sufficient precision when forming their earnings forecasts (Garcìa Lara et al., 2014; Sohn, 2012). Similarly, evidence suggests that financial statement users are able to assess the level of a firm’s conservatism as early as the initial public offering (Ball & Shivakumar, 2008). Thus, a firm’s level of conservatism is likely to be known to investors prior to receiving the forecast news.

Previous research shows that investor response to new information (e.g., forecast news) can be conditioned by prior information (e.g., the firm’s level of conservatism), as investors use prior information to complement how they interpret the new information. This notion of complementariness between prior and new information is supported by various empirical studies. For instance, Banker et al. (1993) show that the response to dividend announcements is moderated by investors’ knowledge of prior accounting information. Similarly, Conrad et al. (2002) suggest that investor response to reported earnings can be conditioned by investors’ assessment of prior market information. Additional evidence of complementarity between prior and new information has been documented in the context of open-market repurchase announcements (Ho et al., 1997) and reported loan loss provisions (Liu et al., 1997).

Based on the considerations outlined above, I argue that investors’ knowledge of a firm’s level of conservatism will condition the investor response to the forecast news if conservatism lends credibility to (mitigates the uncertainty about) the voluntary disclosure of good (bad) news. In the next section, I discuss how conservatism can have such an effect.

Hypothesis Development

Under conservatism, the recognition of good news in reported earnings is delayed until it meets higher verification standards, whereas the recognition of bad news is more timely and complete (Basu, 1997). The higher verification requirements for the recognition of good news in periodic financial statements provide a benchmark for the voluntary disclosure that anticipates such good news (LaFond & Watts, 2008). Consequently, although conservatism may not directly constrain the inclusion of unverified profits in a forecast, it will prevent such unverified good news from being recognized in periodic reports. This will reduce managers’ incentives to base their forecasts on the unverified good news in the first place, as failing to meet good news forecasts can increase litigation and audit costs (Krishnan, Pevzner, & Sengupta, 2012; Rogers & van Buskirk, 2009). Being better verified, the good news will be less likely to be overstated (Garcìa Lara, Garcìa Osma, & Penalva, 2017) or based on unreliable or volatile components, which would decrease the good news’ persistence (Richardson, Sloan, Soliman, & Tuna, 2005). Both effects make good news forecasts more credible signals of positive economic performance.⁵

Because good news forecasts generally lack credibility, investors initially underreact to their release, weakening the association between forecast news and stock returns around the forecast release. Investors will then wait for additional confirming signals before fully incorporating the forecast news into prices. This will cause a positive association between forecast news and postforecast returns (i.e., a positive price drift). If investors perceive conservatism to be a mechanism that increases the credibility of a good news forecast, the initial price response to the signal will be more positive (i.e., lower underreaction) and the delayed price response weaker (i.e., lower positive drift). Therefore, I formulate the following hypotheses.

Hypothesis 1a (H1a): The higher the conservatism, the more positive the 3-day market response to the magnitude of positive forecast news.

Hypothesis 1b (H1b): The higher the conservatism, the less positive the delayed market response to the magnitude of positive forecast news (i.e., the less positive the postforecast drift).

As bad news disclosure goes against managers’ incentives, it is inherently more credible and arguably does not require conservatism to elicit a full response. However, the disclosure of bad news increases market uncertainty, as investors fret that more bad news will follow in the near future. The spike in uncertainty further depresses the stock price, as a larger discount factor is applied to future expected cash flows and investors focus on the worst-case scenario (Epstein & Schneider, 2008). As uncertainty is resolved over time and reverts to preguidance levels, investors’ initial overreaction is corrected and a positive drift arises.

Conservatism can help to contain the spike in uncertainty and the associated overreaction. By inducing a more timely and complete recognition of negative outcomes in periodic reports, conservatism prevents bad news from accumulating for long periods (B. H. Kim & Pevzner, 2010; J.-B. Kim & Zhang, 2016). Investors can therefore use the information that was previously recognized in financial statements to interpret the implications of the bad news that has been voluntarily disclosed. In addition, conservatism can encourage the complete disclosure of bad news by making the benefits from hiding the bad news short lived, as it will be soon exposed through income recognition. If investors are reassured that they have been provided with a complete picture of the negative situation, their overreaction to negative guidance (and the following upward drift) will be contained. This leads to the following hypotheses.

Hypothesis 2a (H2a): The higher the conservatism, the less negative the 3-day market response to the magnitude of negative forecast news.

Hypothesis 2b (H2b): The higher the conservatism, the less positive the delayed price correction that follows the release of negative forecast news (i.e., the less positive the postforecast drift).

Moreover, if investor overreaction to bad news is caused by a spike in uncertainty that conservatism helps to mitigate, then I expect (a) the magnitude of negative⁶ forecast news to be associated with an increase in observable proxies of investor uncertainty and (b) such an association to become weaker as conservatism increases. Therefore, I formulate the following hypothesis.

Hypothesis 2c (H2c): The higher the conservatism, the less positive the association between the magnitude of negative forecast news and the postforecast increase in investor uncertainty.

Research Design

Sample Construction

The sample is composed of forecast announcements by industrial firms listed on the NYSE, AMEX, or NASDAQ with sufficient required data. Consistent with prior studies (Das et al., 2012; Ng et al., 2013), the sample is made of point or closed range forecasts. The data are obtained from the Wharton Research Data Services website. Financial statement data are retrieved from the Center for Research in Security Prices (CRSP)/Compustat Merged Fundamentals Annual. Market data are obtained from the CRSP daily stock database, analysts’ forecast data from the Institutional Brokers’ Estimate System Detail History files, and management forecast data from First Call. The initial data set comprises 46,547 management forecasts issued by industrial companies related to annual⁷ earnings in the fiscal years from 1991 to 2011.⁸ Of these forecasts, 31,283 were released in the 3-day window around an earnings announcement (i.e., bundled forecasts). I cannot test my hypotheses on these observations, as the reaction to the forecast is confounded by the reaction to the earnings announcement, so I remove them from the sample.⁹ I also delete 2,328 non-U.S. dollar observations. To avoid bias from very small or thinly traded stock, I exclude observations with market capitalization and stock prices lower than US$20 million and US$2, respectively, and also require observations to have no missing values for the variables needed for the models. The final sample consists of 9,529 observations. All continuous variables are winsorized at the first and 99th percentiles.

Sample Selection Issues

The sample excludes firm years with unavailable forecast data. To ensure that this exclusion does not impair the validity of the results, I follow previous research and employ the Heckman (1979) two-step procedure to control for potential sample selection bias. In the first step, I estimate the following probit regression:

\Pr {(FORECAST)}_{it} = Φ (\begin{matrix} λ_{0} + λ_{1} EARNVO L_{it} + λ_{2} LogMV E_{it} + λ_{3} M B_{it} + λ_{4} F D_{it} + λ_{5} RE G_{it} + λ_{6} SDRE T_{t} sd \\ + λ_{7} BIG 4_{it} + λ_{8} AG E_{it} \end{matrix})

where FORECAST is a dummy variable equal to one if a firm issued at least one forecast relative to a given fiscal year’s earnings, which makes the firm-year observation available from First Call. I include in the model variables that have been shown to be associated with both conservatism and the likelihood of releasing earnings guidance¹⁰ (Khan & Watts, 2009; Lennox & Park, 2006; Penman & Zhang, 2002). EARNVOL captures the volatility of reported performance, measured as the standard deviation of earnings changes during the 16 quarters ending in the year before the forecast announcement; LogMVE is the logarithm of market value of equity; megabyte (MB) is the market to book ratio; FD and REG are dummy variables identifying firms that belong to high a regulated industry or fiscal years after the introduction of Regulation Fair Disclosure in the year 2000; AGE represents the age of the company; and BIG4 identifies firms whose auditor belong to one of the big four audit companies.¹¹ In the second step, I calculate the inverse Mills ratio (MILLS) as

MILL S_{it} = \frac{φ (- λ^{T} Z_{it})}{Φ (- λ^{T} Z_{it})}

where ϕ() and Φ() represent the standard normal probability density function and the cumulative distribution function, respectively; Z is the vector of covariates; and λ the vector of parameter estimates. Then, MILLS is added to the ordinary least squares regressions as a control variable.

Variable Measurement

I test the hypotheses employing multiple proxies for conservatism. The first is the conservatism score, CSCORE, proposed by Khan and Watts (2009). The measure CSCORE is estimated by interacting several known determinants of conservatism¹² with Basu’s (1997) model to obtain a predicted value of conservatism for each firm-year observation. I estimate CSCORE in the fiscal year prior to the year in which the management forecast is issued. The second measure, SKEW, developed by Givoly and Hayn (2000), is the difference between the skewness in operating cash flows and in earnings over the 20 quarters ending in the fiscal year prior to the year in which the management forecast is issued.

While there is some controversy over measures of conservatism (see Callen & Segal, 2013), I use CSCORE and SKEW because they are appropriate proxies to test whether the market reaction to guidance is moderated by investors’ appreciation of a firm’s propensity to verify profits and recognize losses in a timely fashion—that is, by the firm’s reputation to be conditionally conservative. Both CSCORE and SKEW capture aspects of conservatism that are stable over time, as evidenced by their strong serial correlation, and are thus likely correlated with the firm’s reputation for conservatism. Similarly, both proxies are constructed based on factors (i.e., past earnings, past cash flows, and market forces demanding conservatism) that are known to investors. Following previous studies (e.g., Aier, Chen, & Pevzner, 2014; Y. Kim et al., 2013), I construct the main conservatism proxy employed in this study, labeled CONS, as the average decile rank of CSCORE and SKEW. The variable CONS is then adjusted to range from zero to nine. For the sake of brevity, I mainly report and discuss the results of the composite measure CONS and present the results of the other proxies when discussing robustness tests.

To test H1a and H1b, the initial investor reaction to the forecast release is measured by the cumulative market-adjusted abnormal returns (RET) over the 3-day window centered on the trading day of the forecast release.¹³ To test H2a and H2b, I measure the investor-delayed reaction with the postforecast return drift (DRIFT), obtained by cumulating market-adjusted returns over the 25 trading days starting 2 days after the forecast announcement. To test H2c, I need to measure the increase in investor uncertainty caused by the release of negative guidance. To do so, I use the change in stock return volatility relative to the preannouncement levels. I therefore calculate abnormal volatility (AVOL) as the increase in the standard deviation of daily stock returns from the preforecast window (–55, –5) to the forecast window (–1, +25).

For further robustness, I repeat the test using two additional proxies of investor uncertainty. The first is based on the bid–ask spread. If the release of negative guidance increases uncertainty, market makers will react to the higher trading risk by raising the spread. I therefore calculate ΔSPREAD as the difference between the average daily bid–ask spread in the 3-day window around the forecast release, and the average spread in the (–55, –5) preguidance raises uncertainty, investor disagreement about the interpretation of the forecast will cause abnormal trading volumes in the weeks after the release. I cumulate the standardized unexplained trade volume (SUV), calculated according to Garfinkel and Sokobin (2006), from the day before the announcement up to 25 days later.

All models include the variables specified as follows. I measure the news contained in the earnings forecast as the difference between the value of the forecast (or its midpoint, in the case of a range forecast) and the mean analysts’ consensus prior to the announcement, scaled by the stock price. To capture the sign of the news, I create the variable NEG, which is equal to one if the forecast news is negative and zero otherwise. To capture the magnitude of the forecast news, I include the variable FNEWS, equal to the absolute value of the forecast news. I use the logarithm of the total sales volume to control for the size and visibility of a firm’s business (SIZE). I also include control variables related to the information environment, including analysts’ forecast dispersion (DISP) on the day before the forecast release, the annual standard deviation of daily returns (SDRET), and the number of analysts following the firm (FOLLOW). To control for aspects related to the firm’s reputation as a forecaster, I also include the management forecast frequency (FREQ), measured as the total number of forecasts issued relative to a given fiscal year earnings, and the average error of a firm’s earnings guidance (ERROR). Moreover, I add control variables related to the forecast being released: RANGE measures the imprecision of the forecast, while HORIZON measures how close the forecast release is to the announcement of the actual earnings figure. Finally, I add the inverse Mills ratio (MILLS), calculated as explained in the previous section. Detailed definitions of all variables are provided in the appendix.

Table 1 presents the descriptive statistics. The mean, standard deviation, and quartile values of CSCORE and SKEW are in line with those reported by other studies (Khan & Watts, 2009; Y. Kim et al., 2013; B. H. Kim & Pevzner, 2010). Table 2 (Panel A) presents the correlations among the variables. The composite measure of conservatism (CONS) is equally correlated with both of its components (the correlation being approximately 70%), suggesting that both CSCORE and SKEW contribute equally to the composite proxy. Consistent with the findings of previous literature, CONS is negatively correlated with SIZE and positively correlated with SDRET and DISP, conforming to the notion that conservatism arises as a response to higher information uncertainty (LaFond & Watts, 2008). The variable CONS is also negatively associated with the frequency and specificity of the guidance.¹⁴ The low correlation between CSCORE and SKEW is congruent with the results of several previous studies (Y. Kim et al., 2013; Sohn, 2012), suggesting the potentially noisy nature of conservatism variables (B. H. Kim & Pevzner, 2010). Panel B of Table 2 reports the serial correlation of the conservatism proxies, up to five lags. Consistent with conservatism being stable over time, all three proxies show strong serial correlation for all five lags, ranging between 0.858 (SKEW, first lag) and 0.367 (CSCORE, fifth lag).¹⁵

Table 1.

Descriptive Statistics.

	n observations	M	SD	p25	p50	p75
AVOL	9,529	0.117	1.134	−0.474	−0.020	0.526
CONS	9,529	4.507	2.138	3.000	4.500	6.000
CSCORE	8,808	0.056	0.113	−0.011	0.051	0.116
DISP	9,529	0.281	0.395	0.076	0.152	0.309
DRIFT	9,529	0.013	0.095	−0.039	0.011	0.063
ERROR	9,529	1.882	4.993	−0.734	−0.012	2.561
FNEWS	9,529	2.647	5.759	0.063	0.248	2.412
FOLLOW	9,529	11.549	6.892	6.000	10.000	16.000
FREQ	9,529	6.616	3.701	4.000	6.000	8.000
HORIZON	9,529	−34.365	22.360	−49.143	−33.857	−16.000
MILLS	9,529	1.307	0.295	1.081	1.284	1.495
NEG	9,529	0.474	0.499	0.000	0.000	1.000
RANGE	9,529	0.392	0.480	0.076	0.251	0.517
RET	9,529	−0.681	7.769	−2.889	0.087	2.947
SDRET	9,529	2.491	1.108	1.699	2.287	3.056
SIZE	9,529	21.556	1.625	20.425	21.552	22.704
SKEW	8,911	0.607	1.523	−0.301	0.460	1.479
SUV	9,529	4.745	18.162	−7.808	3.947	16.162
ΔSPREAD	9,369	0.003	0.228	−0.052	−0.010	0.030

Note. AVOL is the increase in the abnormal volatility of daily returns after the forecast release. CONS is the average decile ranking of the conservatism measures developed by Khan and Watts (2009; CSCORE) and by Givoly and Hayn (2000; SKEW). DISP is the analyst forecast dispersion on the day before the forecast announcement, scaled by price. DRIFT is the market-adjusted abnormal returns cumulated over the (+2, +25) window after the forecast release. ERROR is the difference between a firm’s average error of past earnings guidance in the last 4 years and its industry median, scaled by stock price. FNEWS is the absolute value of the difference between the value of the forecast and the analysts’ consensus on the day before the announcement, scaled by stock price. FOLLOW is the number of analysts following on the day before the forecast announcement. FREQ is the number of forecasts issued by a firm relative to the annual earnings per share of a given fiscal year. HORIZON is the number of weeks to the earnings announcement (the higher the value of the variable, the closer to the announcement is the forecast). MILLS is the inverse Mills ratio. NEG is the dummy variable equal to 1 if the signed value of FNEWS is negative, and 0 otherwise. RANGE is the absolute value of the difference between the upper and lower bound of a forecast, scaled by price. RET is the market-adjusted abnormal returns cumulated over the 3-day window centered on the forecast announcement date. SDRET is the standard deviation of daily stock returns. SIZE is the natural logarithm of sales. SUV is the standardized unexplained volume after the forecast release, calculated as per Garfinkel and Sokobin (2006). ΔSPREAD is the change in bid/ask spread after the forecast release. A detailed description of all variables in provided in the appendix. AVOL, RET, FNEWS, RANGE, ΔSPREAD, ERROR, DISP, and SDRET are expressed in % points (i.e., multiplied by 100) to facilitate the interpretation of coefficients.

Table 2.

Panel A: Pearson (Lower Diagonal) and Spearman (Upper Diagonal) Correlations.
		1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19
1	AVOL	1	.023*	.027*	−.051*	−.015	.004	.070*	−.092*	−.097*	.023*	.053*	.045*	−.027*	−.135*	−.063*	−.060*	−.001	.357*	.070*
2	CONS	.037*	1	.709*	.155*	.041*	.012	.075*	−.227*	−.051*	.013	.298*	0	.133*	.003	.240*	−.209*	.712*	.018	−.011
3	CSCORE	.030*	.680*	1	.196*	.047*	−.048*	.076*	−.366*	−.032*	.042*	.426*	.023*	.188*	0	.311*	−.313*	−.028*	.001	−.014
4	DISP	−.048*	.150*	.167*	1	.038*	.005	.263*	−.140*	−.094*	−.330*	.238*	−.002	.389*	.006	.240*	−.062*	.038*	0	−.02
5	DRIFT	−.015	.044*	.036*	.030*	1	.053*	.084*	−.043*	−.029*	−.004	.060*	−.045*	.021*	.012	.085*	−.046*	.01	.011	.003
6	ERROR	−.018	.038*	−.005	.136*	.056*	1	.427*	.007	−.050*	−.016	.057*	−.243*	.103*	.029*	.055*	−.004	.054*	.005	.007
7	FNEWS	−.007	.040*	.01	.124*	.060*	.750*	1	−.199*	−.157*	−.083*	.243*	−.288*	.269*	−.040*	.178*	−.136*	.036*	.080*	−.021*
8	FOLLOW	−.094*	−.191*	−.288*	−.124*	−.038*	−.018	−.080*	1	.261*	.013	−.631*	.027*	−.168*	.014	−.246*	.628*	.016	−.101*	.064*
9	FREQ	−.089*	−.065*	.009	−.111*	−.035*	−.101*	−.103*	.211*	1	−.054*	−.384*	0	.129*	.091*	−.208*	.284*	−.032*	−.097*	.021*
10	HORIZON	.017	.018	.055*	−.191*	.004	−.016	−.036*	.038*	−.097*	1	−.001	.030*	−.257*	−.049*	−.001	.021*	−.021	−.022*	−.004
11	MILLS	.093*	.290*	.336*	.257*	.057*	.122*	.131*	−.579*	−.365*	.001	1	−.046*	.080*	−.014	.606*	−.801*	.029*	.053*	−.058*
12	NEG	.075*	.006	.061*	.015	−.050*	−.293*	−.349*	.016	.013	.037*	−.039*	1	−.056*	−.202*	−.024*	.032*	−.02	.01	.036*
13	RANGE	−.033*	.126*	.167*	.364*	.045*	.265*	.310*	−.173*	.045*	−.219*	.133*	−.109*	1	−.006	.089*	−.006	.005	.009	−.016
14	RET	−.305*	−.002	.001	−.002	−.003	.029*	.056*	.036*	.107*	−.044*	−.058*	−.245*	.014	1	.019	−.006	.018	−.105*	−.039*
15	SDRET	−.033*	.235*	.270*	.299*	.095*	.087*	.063*	−.207*	−.175*	−.002	.602*	−.007	.161*	−.030*	1	−.444*	.061*	−.008	−.046*
16	SIZE	−.079*	−.199*	−.220*	−.087*	−.045*	−.013	−.052*	.593*	.262*	.021*	−.785*	.030*	−.037*	.027*	−.429*	1	−.006	−.057*	.058*
17	SKEW	.001	.675*	−.034*	.044*	.011	.047*	.044*	.032*	−.075*	−.013	.050*	−.041*	.012	.012	.068*	−.015	1	.019	.013
18	SUV	.341*	.019	−.009	−.007	−.004	.008	.028*	−.097*	−.091*	−.029*	.048*	.018	0	−.160*	−.019	−.057*	.025*	1	−.004
19	ΔSPREAD	.084*	−.014	−.011	.024*	.026*	.007	−.012	.01	−.027*	−.019	0	.035*	−.008	−.091*	−.035*	.018	.006	.005	1

Panel B: Autocorrelation of Conservatism Proxies.
	LAG1	LAG2	LAG3	LAG4	LAG5
CONS	.797*	.660*	.548*	.463*	.400*
CSCORE	.774*	.625*	.514*	.429*	.367*
SKEW	.858*	.757*	.675*	.606*	.551*

p values lower than 10%.

Results

Effect of Conservatism on Market Reactions to the Forecast Release (Test of H1a and H2a)

To examine the initial investor response to management forecasts, I regress RET on FNEWS. The coefficient of FNEWS, therefore, represents the market response to the magnitude of the forecast news. To gauge whether such a response is affected by the level of conservatism of the firm, I interact FNEWS with CONS, as in the following model:

RE T_{it} = β_{0} + β_{1} CON S_{it} + β_{2} FNEW S_{it} + β_{3} CONS * FNEW S_{it} + β_{k} CONTROL S_{kit} + ε_{i}

The use of an interaction specification is motivated by previous research (e.g., Banker et al., 1993; Conrad et al., 2002; Ho et al., 1997; Liu et al., 1997) that shows that investors, when reacting to new signals, can complement new information with previously acquired knowledge about preguidance the coefficient β₃ will be significantly different from zero.

To separately test H1a and H2a, I need to modify Equation 2 and interact NEG (i.e., the negative forecast news identifier) with all the other variables,¹⁶ to allow their coefficients to vary depending on the news sign. The transformed regression equation becomes

\begin{matrix} RE T_{it} = β_{0} + β_{1} CON S_{it} + β_{2} FNEW S_{it} + β_{3} CONS * FNEW S_{it} \\ + β_{4} CON S_{it} \times NE G_{it} + β_{5} FNEW S_{it} \times NE G_{it} + β_{6} CONS \times FNEW S_{it} \times NE G_{it} \\ + β_{7} NEG + β_{k} CONTROL S_{kit} + β_{k} CONTROL S_{kit} \times NE G_{it} + ε_{i} \end{matrix}

where β₂ represents the market response coefficient to positive forecast news for firms that belong in the bottom decile of conservatism. The estimate of β₃ represents the change in the market response coefficient in response to positive news as CONS increases by one decile and H1a predicts it to be positive (i.e., the higher the conservatism, the more positive the investor response to good news).

To gauge how conservatism affects the reaction to negative forecast news, one needs to look at the total coefficients, which result from the sum of the coefficient of a variable and the coefficient of the same variable when interacted with the bad news identifier. For instance, the sum β₂+β₅ (i.e., the total coefficient for FNEWS+FNEWS×NEG) represents the market response coefficient to negative forecast news for firms in the bottom decile of conservatism, whereas the sum β₃+β₆ (i.e., the total coefficient for CONS×FNEWS+CONS×FNEWS×NEG) represents the change in response to bad news as CONS increases by one decile. The sum β₃+β₆ is predicted to be positive by H2a (i.e., the higher the conservatism, the less negative the investor response to bad news). Figure 1 provides a graphic example of these tests.

Figure 1.

Test of H1a and H2a.

The results of the regression analysis are reported in Table 3. The left side of the table reports the partial coefficients used to analyze the market reaction to positive guidance and how it is affected by conservatism (H1a). The coefficient β₂ (FNEWS) is positive but insignificant (coeff. = .020, p value = .34), suggesting that good news forecasts released by low-conservatism firms are not associated with a significant market response. This result is consistent with these forecasts lacking credibility in the eyes of investors, who therefore do not react to the release but, rather, wait for additional confirming signals. However, the coefficient β₃, which represents the effect of conservatism on the market response coefficient to good news forecasts, is positive (coeff. = .019) and significant at the 1% level. This result supports the hypothesis that conservatism improves the credibility of positive guidance, resulting in a smaller underreaction to good news forecasts and thus a more positive association between forecast news and abnormal returns around the news release.

Table 3.

Test of H1a (H2a): Does Conservatism Make the 3-Day Response to Good (Bad) News Forecasts More Positive (Less Negative)?RET = β₀+β₁CONS+β₂FNEWS+β₃CONS×FNEWS+β₄CONS ×NEG+β₅FNEWS×NEG+β₆CONS×FNEWS×NEG+β₇NEG+β_kCONTROLS +β_kCONTROLS ×NEG+ϵIn the equation above, β₂ (the sum β₂+β₅), bold faced in the table, represents the market response coefficient to positive (negative) forecast news for firms in the bottom decile of conservatism. β₃ (the sum β₃+β₆), bold faced and shaded in the table, represents the change in the market response coefficient as CONS increases by one decile.

Test of coefficient estimates—Good news forecasts					F test of significance of total coefficients (main effect + interaction)—Bad news forecasts
Variables	Test	Expected sign:	Estimate	p values	Test	Expected sign:	Estimate	p values
CONS	β₁ ≠ 0	?	0.008	(.905)	β₁ + β₄ ≠ 0	?	−0.019	(.816)
FNEWS	β₂ > 0	+	0.020	(.338)	β₂+β₅ < 0	−	–3.687 ***	(.002)
			Test of H1a				Test of H2a
CONS×FNEWS	β₃ > 0	+	0.019 ***	(.004)	β₃+β₆ > 0	+	0.380 ***	(.001)
CONS×NEG	β₄ ≠ 0	?	−0.027	(.800)
FNEWS×NEG	β₅ < 0	−	−3.706***	(.003)
CONS×FNEWS×NEG	β₆ ≠ 0	?	0.361***	(.004)
NEG	β₇ ≠ 0	?	1.867	(.793)
MILLS	β₈ ≠ 0	?	0.086	(.939)	β₈ + β_κ ≠ 0	?	−0.755	(.603)
ERROR	β₉ ≠ 0	?	−0.109***	(.005)	β₉ + β_κ ≠ 0	?	−0.134*	(.065)
FREQ	β₁₀ ≠ 0	?	0.024	(.444)	β₁₀ + β_κ ≠ 0	?	0.193***	(.000)
HORIZON	β₁₁ ≠ 0	?	−0.011**	(.015)	β₁₁ + β_κ ≠ 0	?	−0.014**	(.029)
RANGE	β₁₂ ≠ 0	?	−0.608**	(.017)	β₁₂ + β_κ ≠ 0	?	−0.868	(.107)
SIZE	β₁₃ ≠ 0	?	0.027	(.860)	β₁₃ + β_κ ≠ 0	?	−0.092	(.627)
DISP	β₁₄ ≠ 0	?	0.475	(.201)	β₁₄ + β_κ ≠ 0	?	3.306***	(.000)
SDRET	β₁₅ ≠ 0	?	0.857***	(.000)	β₁₅ + βκ ≠ 0	?	−0.846***	(.001)
FOLLOW	β₁₆ ≠ 0	?	−0.045**	(.032)	β₁₆ + β_κ ≠ 0	?	0.070**	(.012)
Constant	β₀ ≠ 0	?	−0.010	(.998)	β₀ + β₇ ≠ 0	?	1.858	(.740)
CONTROLS ×NEG	β_κ ≠ 0	?	YES	(omitted)
Year/Ind. FE			YES	(omitted)
Observations			9,529
Adjusted R²			.149

Note. CONS is the average decile ranking of the conservatism measures developed by Khan and Watts (2009; CSCORE) and by Givoly and Hayn (2000; SKEW). ERROR is the difference between a firm’s average error of past earnings guidance in the last 4 years and its industry median, scaled by stock price. FNEWS is the absolute value of the difference between the value of the forecast and the analysts’ consensus on the day before the announcement, scaled by stock price. FOLLOW is the number of analysts following on the day before the forecast announcement. FREQ is the number of forecasts issued by a firm relative to the annual earnings per share of a given fiscal year. HORIZON is the number of weeks to the earnings announcement (the higher the value of the variable, the closer to the announcement is the forecast). MILLS is the inverse Mills ratio. NEG is the dummy variable equal to 1 if the signed value of FNEWS is negative, and 0 otherwise. RANGE is the absolute value of the difference between the upper and lower bound of a forecast, scaled by price. RET is the market-adjusted abnormal returns cumulated over the 3-day window centered on the forecast announcement date. SDRET is the standard deviation of daily stock returns. SIZE is the natural logarithm of sales. A detailed description of all variables in provided in the appendix. The p values are based on one-tail test when a directional prediction is made, and on a two-tail test otherwise. RET, FNEWS, RANGE, ERROR, DISP, and SDRET are expressed in % points (i.e., multiplied by 100) to facilitate the interpretation of coefficients. Standard errors are heteroskedasticity robust (White, 1980) and clustered at the firm level. FE = fixed effects.

p values lower than 10%. **p values lower than 5%. ***p values lower than 1%.

The right side of Table 3 reports the total coefficients (i.e., the sum of a variable’s coefficient and the coefficient of its interaction with NEG), which allows the analysis of the market reaction to negative guidance and how it is affected by conservatism (H2a). The sum (β₂+β₅) is statistically significant and equal to −3.687. This result suggests that, when a firm belonging in the bottom conservatism decile releases negative forecast news equal to 1% of its stock price, cumulative abnormal returns are, on average, –3.69%. Consistent with H2a, such a negative reaction is significantly attenuated as we move up in the conservatism decile rankings: The sum of β₃ and β₆, which estimates the effect of a one-decile increase in conservatism on the market response coefficient to bad news, is positive and significant at the 1% level (coeff. = .380, p value < .01). As the CONS deciles range from zero to nine, the stock returns are estimated to be higher by 3.42% in the top decile relative to the bottom one. These results support the hypothesis that conservatism, by reassuring investors about a forecast’s implications for future profitability, decreases the uncertainty associated with bad news releases and therefore mitigates the negative market reaction to the bad news.

Postforecast Drift Patterns

Effect of conservatism on the market drift after the forecast release (test of H1b and H2b)

The results discussed in the previous section support both H1a and H2a. In this section, I test H1b and H2b by analyzing stock returns patterns in the postforecast weeks. If conservatism enhances the credibility of good news forecasts, leading to a fuller initial response, more conservative firms will experience a smaller delayed response (i.e., a less positive drift) after the forecast release. Similarly, if conservatism mitigates the post–bad news surge in investor uncertainty so that the initial overreaction is smaller, more conservative firms will experience a smaller price correction (i.e., a less positive drift) after the forecast release.

To study the existence of a delayed response (or partial reversal) after positive (negative) forecast news, I modify Equation 3, replacing RET with DRIFT, which represents the price drift after the forecast release and is measured by the market-adjusted returns cumulated over the (+2, +25) window. Therefore, I test the following model:

\begin{matrix} DRIF T_{it} = β_{0} + β_{1} CON S_{it} + β_{2} FNEW S_{it} + β_{3} CONS \times FNEW S_{it} \\ + β_{4} CON S_{it} \times NE G_{it} + β_{5} FNEW S_{it} \times NE G_{it} + β_{6} CONS \times FNEW S_{it} \times NE G_{it} \\ + β_{7} NEG + β_{k} CONTROL S_{kit} + β_{k} CONTROL S_{kit} \times NE G_{it} + ε_{i} \end{matrix}

where the estimate of β₂ represents the drift following positive forecast news for firms in the bottom decile of conservatism. Based on Das et al. (2012), the coefficient β₂ is expected to be positive, suggesting a delayed reaction after an initial underreaction to positive forecast news. The estimate of β₃ represents the change in drift as CONS increases by one decile and H1b predicts it to be negative, suggesting that conservatism leads to a fuller initial reaction to good news forecasts.

Similarly, the sum β₂+β₅ represents the drift following negative forecast news for firms in the bottom decile of conservatism. If these firms experience a delayed correction after an initial overreaction to negative forecast news, the sum β₂+β₅ is expected to be positive. The sum β₃+β₆ represents the change in drift as CONS increases by one decile and H2b predicts it to be negative, suggesting that conservatism leads to a smaller price correction after the release of bad news forecasts.

The results are reported in Table 4. As in Table 3, the left side reports the partial coefficients used to analyze the drift following positive guidance and how it is affected by conservatism (H1b). The right side reports the total coefficients relative to the drift that follows bad news forecasts (H2b).

Table 4.

Test of H1b (H2b): Does Conservatism Make the Delayed Response to (the Price Correction After) Good (Bad) News Forecasts Less Positive?DRIFT = β₀+β₁CONS+β₂FNEWS+β₃CONS×FNEWS+β₄CONS×NEG+β₅FNEWS×NEG+β₆CONS×FNEWS×NEG+β₇NEG+β_kCONTROLS +β_kCONTROLS ×NEG+ϵIn the equation above, β₂ (the sum β₂+β₅), bold faced in the table, represents the price drift after good news (bad news) forecasts for firms in the bottom decile of conservatism. β₃ (the sum β₃+β₆), bold faced and shaded in the table, represents the change in the drift as CONS increases by one decile.

Test of coefficient estimates—Good news forecasts					F test of significance of total coefficients (main effect + interaction)—Bad news forecasts
Variables	Test	Expected sign:	Estimate	p values	Test	Expected sign:	Estimate	p values
CONS	β₁ ≠ 0	?	0.052	(.564)	β₁ + β₄ ≠ 0	?	0.223**	(.014)
FNEWS	β₂ > 0	+	0.128**	(.019)	β₂+β₅ > 0	+	1.330 ***	(.003)
			Test of H1b				Test of H2b
CONS×FNEWS	β₃ < 0	−	–0.016*	(.060)	β₃+β₆ < 0	−	–0.225 ***	(.001)

CONS×NEG	β₄ ≠ 0	?	0.171	(.173)	β₂ + β₃ + β₆ ≠ 0	?	−0.097	(.318)
FNEWS×NEG	β₅ ≠ 0	?	1.202**	(.014)
CONS×FNEWS×NEG	β₆ ≠ 0	?	−0.209***	(.005)
Constant	β₀ ≠ 0	?	3.935	(.591)
CONTROLS, NEG			YES	(omitted)
CONTROLS×NEG			YES	(omitted)
Year / Ind. FE			YES	(omitted)
Observations			9,529
Adjusted R²			.034

Note. CONS is the average decile ranking of the conservatism measures developed by Khan and Watts (2009; CSCORE) and by Givoly and Hayn (2000; SKEW). FNEWS is the absolute value of the difference between the value of the forecast and the analysts’ consensus on the day before the announcement, scaled by stock price. NEG is the dummy variable equal to 1 if the signed value of FNEWS is negative, and 0 otherwise. DRIFT is the market-adjusted abnormal returns cumulated over the (+2, +25) window after the forecast release. CONTROLS is the vector of control variables included in all equations but omitted here for brevity (i.e., MILLS, ERROR, FREQ, HORIZON, RANGE, SIZE, DISP, SDRET, FOLLOW). A detailed description of all variables in provided in the appendix. The p values are based on one-tail test when a directional prediction is made, and on a two-tail test otherwise. FNEWS and DRIFT are expressed in % points (i.e., multiplied by 100) to facilitate the interpretation of coefficients. Standard errors are heteroskedasticity robust (White, 1980) and clustered at the firm level. FE = fixed effects.

p values lower than 10%. **p values lower than 5%. ***p values lower than 1%.

Consistent with expectations, the coefficient of FNEWS is significantly positive (coeff. = .128, p value = .02). This result suggests that the lower returns experienced by less conservative firms around the release of good news forecasts are explained by an initial market underreaction to news that is viewed as less credible. The positive drift arises as the market waits to verify the credibility of good news forecasts before fully reacting to them.

A significantly positive drift is also detected for bad news forecasts. The sum of the coefficients of FNEWS and FNEWS×NEG is equal to 1.330, significant at the 1% level. In other words, firms in the lowest conservatism decile experience a positive cumulated return equal to around 1.33% in the weeks that follow the release of a forecast that was 1% below analysts’ expectations. This result suggests that investors initially overreact to negative guidance that is released by less conservative firms, amid concerns that more bad news could soon follow.

Supporting H2a and H2b, the magnitude of the drift is progressively attenuated as conservatism increases. In particular, the effect of a one-decile increase in conservatism on the bad news drift (i.e., the sum of CONS×FNEWS and CONS×FNEWS×NEG) is significantly negative (coeff. = –.225, p value < .01). This result is consistent with conservatism reducing investor uncertainty following the release of negative forecasts and therefore attenuating the market overreaction and subsequent reversion (H2b). A similar but weaker pattern emerges following good news forecasts (H2a): The coefficient of CONS×FNEWS is negative and significant at the 10% level. This result suggests that the market underreaction is particularly concentrated among less conservative firms, whose good news is initially considered less credible. Overall, the results reported in this section show a positive price drift after both good and bad news forecasts. This result is consistent with the findings of Das et al. (2012) and Ng et al. (2013), both of whom document similar drift patterns in the postforecast period. However, the results of Table 4 provide additional insights by showing that conservatism can attenuate the postforecast drift, which is an issue that neither Das et al. (2012) nor Ng et al. (2013) consider.

Additional insights into the results of H1b and H2b: Do drift attenuation patterns vary across different levels of market uncertainty? By showing that conservatism attenuates postguidance drift, the results reported in Table 4 suggest an additional question¹⁷: Is the drift attenuation effect stronger when investor reaction to the forecast occurs in the presence of higher levels of uncertainty? On one hand, when uncertainty is low and the information playing field is level, investors could be more likely to possess adequate information to interpret the forecast news. Under such circumstances, negative guidance could be less alarming and positive guidance could pose fewer credibility concerns. Consequently, the forecast news could take less time to be correctly priced, which will reduce the postforecast drift and conservatism’s effect on it. On the other hand, when uncertainty is high, investors could be less capable of fully assessing the credibility of good news and the full extent of bad news, making management guidance a more ambiguous signal. As investors take longer to correctly price the forecast news, the drift (and conservatism’s effect on it) will be stronger.

To investigate this additional question, I partition the sample based on the level of uncertainty affecting the market after the forecast release (i.e., when investors are pricing the new information). Because high daily volatility is a symptom of market uncertainty, I compute the tercile ranks of the standard deviation of daily stock returns over the window (–1, +25) around the forecast release. I then repeat the analysis of Table 4 within each tercile, testing for differences in the estimates of β₃ and β₃+β₆. The results are reported in Panel A of Table 5.

Table 5.

Additional Insights Into the Results of Table 4: Do Drift Attenuation Patterns Vary Across Different Levels of Market Uncertainty?DRIFT = β₀+β₁CONS+β₂FNEWS+β₃CONS×FNEWS+β₄CONS×NEG+β₅FNEWS×NEG+β₆CONS×FNEWS×NEG+β₇NEG+β_kCONTROLS +β_kCONTROLS ×NEG+ϵIn the equation above, β₂ (the sum β₂+β₅) represents the price drift after good news (bad news) forecasts for firms in the bottom decile of conservatism. β₃ (the sum β₃+β₆) represents the change in the drift as CONS increases by one decile.

Panel A: Test of H1b and H2b Across Different Levels of Stock Volatility.
Test	Expected sign:	Low tercile		Middle tercile		High tercile		Difference: High – low
		Estimate	p values	Estimate	p values	Estimate	p values	Test	Expected sign:	Estimate	p values
β₃ < 0Test of H1b	–	−0.001	(.495)	−0.002	(.449)	−0.032*	(.051)	Diff. < 0	–	−0.031*	(.086)
β₃+β₆ < 0Test of H2b	–	0.008	(.466)	−0.231**	(.029)	−0.279***	(.003)	Diff. < 0	–	−0.287**	(.015)
Panel B: Test of H1b and H2b Across Different Levels of Bid/Ask Spread.
Test	Expected sign:	Low tercile		Middle tercile		High tercile		Difference: Top – bottom
		Estimate	p values	Estimate	p values	Estimate	p values	Test	Expected sign:	Estimate	p values
β₃ < 0Test of H1b	–	0.006	(.390)	−0.037**	(.030)	−0.015	(.139)	Diff. < 0	–	−0.021	(.182)
β₃+β₆ < 0Test of H2b	–	0.091	(.368)	−0.381**	(.027)	−0.242**	(.014)	Diff. < 0	–	−0.333	(.125)
Panel C: Test of H1b and H2b Across Different Levels of Unexpected Trade Volume.
Test	Expected sign:	Low tercile		Middle tercile		High tercile		Difference: Top – bottom
		Estimate	p values	Estimate	p values	Estimate	p values	Test	Expected sign:	Estimate	p values
β₃ < 0Test of H1b	–	−0.018	(.121)	−0.012	(.268)	−0.028*	(.081)	Diff. < 0	–	−0.010	(.361)
β₃+β₆ < 0Test of H2b	–	−0.369**	(.015)	−0.221**	(.042)	−0.258*	(.092)	Diff. < 0	–	0.111	(.328)

Note. This table reports the estimates obtained for the coefficients of interest when testing H1b and H2b across different levels of returns volatility, bid/ask spread, and unexpected trade volume. Volatility is measured by the standard deviations of daily stock returns over the window (–1, +25) around the announcement. Bid/ask spread is measured by the average daily spread, scaled by its midpoint, over the window (-1, +1) around the announcement. Standardized unexpected trade volume is measured as per Garfinkel and Sokobin (2006) over the window (-1, +25) around the announcement. For brevity, the coefficients of the other variables are not tabulated. CONS is the average decile ranking of the conservatism measures developed by Khan and Watts (2009; CSCORE) and by Givoly and Hayn (2000; SKEW). FNEWS is the absolute value of the difference between the value of the forecast and the analysts’ consensus on the day before the announcement, scaled by stock price. NEG is the dummy variable equal to 1 if the signed value of FNEWS is negative, and 0 otherwise. DRIFT is the market-adjusted abnormal returns cumulated over the (+2, +25) window after the forecast release. CONTROLS is the vector of control variables included in all equations but omitted here for brevity (i.e., MILLS, ERROR, FREQ, HORIZON, RANGE, SIZE, DISP, SDRET, FOLLOW). A detailed description of all variables in provided in the appendix. The p values are based on one-tail test when a directional prediction is made, and on a two-tail test otherwise. Standard errors are heteroskedasticity robust (White, 1980) and clustered at the firm level.

p values lower than 10%. **p values lower than 5%. ***p values lower than 1%.

Consistent with expectations, as market uncertainty grows, the strength of the drift attenuation effects after positive (β₃) and negative (β₃+β₆) guidance increases monotonically in both magnitude and significance. As shown in the last column of Table 5, the difference between the top and bottom terciles is significant for both positive and negative guidance (at the 10% and 1% levels, respectively).

For robustness, I repeat the test using two alternative proxies of uncertainty levels: the bid–ask spread and unexplained trade volumes. The results are reported in Panels B and C of Table 5, respectively. Using the bid–ask spread yields results that are similar to those of Panel A, though the variation is not monotonic and the significance of the differences is somewhat lower. When using unexplained volume, however, the difference across terciles becomes too weak to be significant. Overall, these results suggest that higher market uncertainty, especially when measured by daily volatility, contributes to postforecast drift and to conservatism’s effect on it.

Effect of Conservatism on Investor Uncertainty Following Bad News Forecasts (Test of H2c)

The results discussed in the previous section provide evidence of the link between investor uncertainty and postguidance drift patterns: When investors face high levels of uncertainty, management guidance becomes harder to interpret, giving rise to a price drift. Previous research suggests that this link is stronger for negative guidance. Different from good news forecasts, bad news forecasts increase uncertainty relative to preforecast levels (Rogers et al., 2009), as investors grow concerned that more bad news could soon follow. I argue that conservatism, by reassuring investors that the disclosure of economic losses has been timely and complete, alleviates such an increase in uncertainty (H2c), thus containing the overreaction (H2a) and the subsequent correction (H2b). In this section, I directly test H2c by examining whether conservatism mitigates the increase in investor uncertainty caused by negative forecast news. To do so, I substitute DRIFT with AVOL, which measures the postforecast increase in daily return volatility, as the dependent variable in Equation 4. The regression equation is as follows:

\begin{matrix} AVO L_{it} = β_{0} + β_{1} CON S_{it} + β_{2} FNEW S_{it} + β_{3} CONS \times FNEW S_{it} \\ + β_{4} CON S_{it} \times NE G_{it} + β_{5} FNEW S_{it} \times NE G_{it} + β_{6} CONS \times FNEW S_{it} \times NE G_{it} \\ + β_{7} NEG + β_{k} CONTROL S_{kit} + β_{k} CONTROL S_{kit} \times NE G_{it} + ε_{i} \end{matrix}

I expect the sum of β₂ and β₅ to be positive, indicating that negative forecast news that is released by less conservative firms produces a surge in investor uncertainty. I also expect the sum of β₃ and β₆ to be negative, indicating that conservatism reduces the uncertainty associated with the unexpected revelation of bad news. I do not expect β₂ or β₃ to be significantly different from zero, as they estimate the coefficients relative to the release of positive forecast news, where an increase in uncertainty is not predicted.

Panel A of Table 6 reports the regression results. As expected, the coefficients of FNEWS and CONS×FNEWS are not significant, indicating that good news announcements do not significantly raise market uncertainty. Instead, the sum of FNEWS and FNEWS×NEG is significantly positive (coeff. = .370, p value < .01). This result suggests that, when firms belonging to the bottom decile of conservatism release a bad news forecast, a 1% increase in the magnitude of the news raises uncertainty by an amount equal to 36% of the standard deviation of AVOL. This effect is significantly attenuated by conservatism: a one-decile increase in CONS is associated with a reduction in the aforementioned coefficient that is equal to –.044 (p value < .01), that is, approximately 11% of β₂+β₅. For robustness, I repeat the analysis using two additional proxies of uncertainty: the increase in the bid–ask spread (ΔSPREAD) and the standardized unexpected trade volume (SUV). The results (reported in Panel B) are qualitatively similar.

Table 6.

Test of H2c: Does Conservatism Mitigate the Post–Bad News Forecast Increase in Investor Uncertainty?AVOL, ΔSPREAD, or SUV = β₀+β₁CONS+β₂FNEWS+β₃CONS ×FNEWS+β₄CONS ×NEG+β₅FNEWS×NEG+β₆CONS ×FNEWS×NEG+β₇NEG+β_kCONTROLS +β_kCONTROLS ×NEG+ϵIn the equation above, the sum β₂+β₅, bold faced in the table, represents the effect of negative forecast news on alternative uncertainty proxies for firms in the bottom decile of conservatism. The sum β₃+β₆, bold faced and shaded in the table, represents the change in this effect as CONS increases by one decile.

Test of coefficient estimates—Good news forecasts					F test of significance of total coefficients (main effect + interaction)—Bad news forecasts
Variables	Test	Expected sign:	Estimate	p values	Test	Expected sign:	Estimate	p values
Panel A: Test of H2c using abnormal volatility (AVOL) as a proxy for market uncertainty
CONS	β₁ ≠ 0	?	0.005	(.589)	β₁ + β₄ ≠ 0	?	0.008	(.410)
FNEWS	β₂ ≠ 0	?	−0.001	(.898)	β₂+β₅ > 0	+	0.370 ***	(.000)
							Test of H2c
CONS×FNEWS	β₃ ≠ 0	?	−0.001	(.523)	β₃+β₆ < 0	−	−0.044 ***	(.000)
CONS×NEG	β₄ ≠ 0	?	0.003	(.835)
FNEWS×NEG	β₅ ≠ 0	?	0.370***	(.000)
CONS×FNEWS×NEG	β₆ ≠ 0	?	−0.043***	(.000)
Constant	β₀ ≠ 0	?	−0.229	(.780)
CONTROLS			YES	(omitted)
CONTROLS ×NEG	βκ ≠ 0	?	YES	(omitted)
Year/Ind. FE			YES	(omitted)
Observations			9,529
Adjusted R²			.085
Panel B: Robustness tests of H2c using alternative proxies of market uncertainty (ΔSPREAD and SUV)—only the main coefficients of interest are reported for brevity.
DEPVAR = ΔSPREAD
CONS	β₁ ≠ 0	?	−0.005	(.628)	β₁ + β₄ ≠ 0	?	−0.002	(.866)
FNEWS	β₂ ≠ 0	?	−0.003	(.742)	β₂ + β₅ > 0	+	0.116*	(.052)
							Test of H2c
CONS×FNEWS	β₃ ≠ 0	?	−0.001	(.567)	β₃+β₆ < 0	−	−0.019**	(.046)
DEPVAR = SUV
CONS	β₁ ≠ 0	?	0.005	(.618)	β₁ + β₄ ≠ 0	?	0.021**	(.031)
FNEWS	β₂ ≠ 0	?	0.007	(.255)	β₂+β₅ > 0	+	0.273 ***	(.000)
							Test of H2c
CONS×FNEWS	β₃ ≠ 0	?	−0.000	(.927)	β₃+β₆ < 0	−	−0.036 ***	(.000)

Note. AVOL is the increase in the abnormal volatility of daily returns after the forecast release. SUV is the standardized unexplained volume after the forecast release, calculated as per Garfinkel and Sokobin (2006), ΔSPREAD is the change in bid/ask spread after the forecast release. CONTROLS is the vector of control variables included in all equations but omitted here for brevity (i.e., MILLS, ERROR, FREQ, HORIZON, RANGE, SIZE, DISP, SDRET, FOLLOW). A detailed description of all variables in provided in the appendix. The p values are based on one-tail test when a directional prediction is made, and on a two-tail test otherwise. FNEWS is expressed in % points (i.e., multiplied by 100), and AVOL and SUV are divided by their sample standard deviation, to facilitate the interpretation of coefficients. Standard errors are heteroskedasticity robust (White, 1980) and clustered at the firm level. FE = fixed effects.

p values lower than 10%. **p values lower than 5%. ***p values lower than 1%.

Robustness Tests

Alternative Proxies for Conservatism

I repeat the analyses alternatively using the decile rank of CSCORE or SKEW instead of the composite measure (CONS) in the regression equations. Table 7 reports the estimate of the coefficients of interest for all hypotheses (the other variables are omitted for brevity). The coefficients are always directionally consistent with those reported in the main analysis. Moreover, the coefficient magnitudes are generally similar: neither CSCORE nor SKEW appears to be the sole driver of the main results, which, instead, are equally driven by both proxies. In addition, the results are always significant, with the only exception of the test of H1b, where significance is slightly above the 10% threshold.

Table 7.

Sensitivity Analysis: Estimates and Significance of the Coefficients of Interest, Using Alternative Proxies to Measure Accounting Conservatism When Testing H1a, H1b, H2a, H2b, H2c.DEPVAR = β₀+β₁CONS_PROXY+β₂FNEWS+β₃CONS_PROXY×FNEWS+β₄CONS_PROXY×NEG+β₅FNEWS×NEG+β₆CONS_PROXY×FNEWS×NEG+β₇NEG+β_kCONTROLS +β_kCONTROLS ×NEG+ϵIn the equation above, β₂ (the sum β₂+β₅) represents the association of the dependent variable (measured by DEPVAR) with positive (negative) forecast news for firms in the bottom decile of conservatism (measured by CONS_PROXY). β₃ (the sum β₃+β₆) represents the change in such association as CONS_PROXY increases by one decile.

CONS_PROXY	DEPVAR	Test	Expected sign:	Estimate	p values	CONS_PROXY	DEPVAR	Test	Expected sign:	Estimate	p values
Test of H1a: three-day market response to good news						Test of H2a: three-day market response to bad news
CONS	RET	β₃ > 0	+	0.019***	(.004)	CONS	RET	β₃+β₆ > 0	+	0.380***	(.001)
CSCORE	RET	β₃ > 0	+	0.017***	(.002)	CSCORE	RET	β₃+β₆ > 0	+	0.267**	(.023)
SKEW	RET	β₃ > 0	+	0.007*	(.092)	SKEW	RET	β₃+β₆ > 0	+	0.239***	(.001)
Test of H1b: delayed market response to good news						Test of H2b: delayed market response to bad news
CONS	DRIFT	β₃ < 0	–	−0.016*	(.060)	CONS	DRIFT	β₃+β₆ <λ0	–	−0.225***	(.001)
CSCORE	DRIFT	β₃ < 0	–	−0.008	(.162)	CSCORE	DRIFT	β₃+β₆ < 0	–	−0.103*	(.082)
SKEW	DRIFT	β₃ < 0	–	−0.009	(.118)	SKEW	DRIFT	β₃+β₆ < 0	–	−0.145**	(.014)
CONS_PROXY	DEPVAR	Test	Expected sign:	DEPVAR: AVOL		DEPVAR: SUV		DEPVAR: ΔSPREAD
				Estimate	p values	Estimate	p values	Estimate	p values
Test of H2c: increase in market uncertainty following the release of bad news
CONS		β₃+β₆ < 0	–	−0.044***	(.000)	−0.036***	(.000)	−0.019**	(.046)
CSCORE		β₃+β₆ < 0	–	−0.029***	(.009)	−0.038***	(.000)	−0.017**	(.027)
SKEW		β₃+β₆ < 0	–	−0.023***	(.002)	−0.009*	(.082)	−0.002	(.44)

Note. CONS is the average decile ranking of the conservatism measures developed by Khan and Watts (2009; CSCORE) and by Givoly and Hayn (2000; SKEW). FNEWS is the absolute value of the difference between the value of the forecast and the analysts’ consensus on the day before the announcement, scaled by stock price. NEG is the dummy variable equal to 1 if the signed value of FNEWS is negative, and 0 otherwise. RET is the market-adjusted abnormal returns cumulated over the three-day window centered on the forecast announcement date. DRIFT is the market-adjusted abnormal returns cumulated over the (+2, +25) window after the forecast release. AVOL is the increase in the abnormal volatility of daily returns after the forecast release. SUV is the standardized unexplained volume after the forecast release, calculated as per Garfinkel and Sokobin (2006). ΔSPREAD is the change in bid/ask spread after the forecast release. CONTROLS is the vector of control variables included in all equations but omitted here for brevity (i.e., MILLS, ERROR, FREQ, HORIZON, RANGE, SIZE, DISP, SDRET, FOLLOW). A detailed description of all variables in provided in the appendix. The p values are based on one-tail test when a directional prediction is made, and on a two-tail test otherwise. FNEWS and DRIFT are expressed in % points (i.e., multiplied by 100), and AVOL and SUV are divided by their sample standard deviation, to facilitate the interpretation of coefficients. Standard errors are heteroskedasticity robust (White, 1980) and clustered at the firm level. FE = fixed effects.

p values lower than 10%. **p values lower than 5%. ***p values lower than 1%.

Alternative Model to Calculate the Inverse Mills Ratio

I recalculate the inverse Mills ratio after including several variables associated with firm corporate governance characteristics, which could influence both voluntary disclosure policies and investor reactions to such disclosures. In particular, I add to Equation 1 the number of board members, the number of independent auditors, the percentage of executive directors on the board, an indicator equal to one if the chief executive officer is also the chair of the board, and the percentage of shares held by institutional investors. Because of the scarce availability of these data, obtained from the Thomson Reuters and RiskMetrics databases, the sample size is reduced by more than 60%, which greatly decreases the power of the tests. Overall, the results are robust. The coefficient estimates are always consistent with all hypotheses and similar (or even larger) in magnitude. Despite the large loss of observations, the p values are generally significant and the highest value is 14% (H2c when using AVOL).

Bundled Announcements

The results described so far are based on a sample that excludes forecasts issued on the same day of an earnings announcement (i.e., bundled forecasts). Several significant concerns motivate this choice. First, reactions to the announcement would confound reactions to the forecast. This would add noise and possibly bias to the tests, which aim to capture the effect of conservatism on investor reaction to voluntary disclosure. Second, bundled forecasts differ from standalone forecasts in several aspects, such as their consequences (Billings, Jennings, & Lev, 2015) and managers’ incentives to release them. In addition, managers’ practice of routinely releasing guidance on the earnings announcement day mitigates and possibly confounds the effect of negative guidance on uncertainty (Rogers et al., 2009), making bundled forecasts a poor setting to test H2a through H2c. Because of these concerns, it is common to exclude bundled forecasts from the analysis of the market reaction to management guidance (Das et al., 2012; Ng et al., 2013; Rogers et al., 2009).

After bundled forecasts are included in the sample, the coefficients of interest are always consistent with the hypotheses, although sometimes smaller in magnitude. Most of the p values are significant, ranging between .01 and .05; the only exceptions are related to the test of H2a (p value = .11) and the test of H1b (p value = .31). Overall, these results suggest that including bundled forecasts in the sample does not change the coefficient signs and only partially affects their magnitude but adds noise that reduces the power of the tests.

Potential Bias Originating From Early Years and Observations With a Small Analyst Following

The sample necessarily excludes observations for which forecast data are not available (i.e., not present in the First Call data set). To mitigate potential bias due to this sample selection requirement, all models include the inverse Mills ratio, which controls for the likelihood that a firm is covered by First Call. For further robustness, I perform two additional tests following Chuk, Matsumoto, and Miller (2013), who suggest that early years (e.g., before 1997) and firms with a small analyst following are more likely to be affected by potential biases.

First, I repeat the tests using only observations whose analyst following is larger than five, which, as Chuk et al. (2013) show, are less affected by potential biases. If the coefficient estimates reported in the main analysis were the result of a sampling bias concentrated among firms with a small analyst following, removing these firms would have a significant impact on the coefficients’ magnitudes and would make them inconsistent with the hypotheses. Contrary to this concern, all coefficients of interest are practically identical in magnitude and sign. Significance levels are also qualitatively similar, despite the reduction in sample size, with the only exception being the test for H2a (p value 12%).

Second, I remove from the sample all observations before 1997. Again, all the coefficients of interest remain practically the same, which rejects the notion that the main results were driven by a sampling bias related to early years.

Cross-Sectional Variation in the Credibility-Enhancing Effect of Conservatism

By showing that the market response to good news forecasts is stronger when conservatism is higher, the results of Table 3 suggest that conservatism lends more credibility to positive guidance. An additional question is whether this credibility-enhancing effect of conservatism affects some forecasting firms more than others.¹⁸

On one hand, firms characterized by higher visibility (e.g., larger firms or those with a greater analyst following) could have established a better reputation for credible disclosure. This could attenuate investor concerns regarding the credibility of positive guidance. Similarly, investors could be less skeptical of managers’ ability to predict good news for firms that operate in less turbulent and volatile markets. This would cause conservatism’s effect on the market response to positive guidance (i.e., coefficient β₃ in Equation 3) to be smaller for more visible and less volatile firms, whose good news would be more credible, regardless of conservatism. On the other hand, Zhang (2006) shows that behavioral biases and incomplete information can cause investor response to be slower when small, small-analyst-following, high-volatility firms issue a new signal, whether credible or not. This would cause conservatism’s effect to be smaller for less visible and more volatile firms, which would experience a weak reaction, regardless of conservatism. Finally, the effects described above could offset each other, resulting in no significant differences across subsamples.

To shed light on this additional question, I partition the sample based on proxies of firm visibility (i.e., SIZE and FOLLOW) and market volatility (i.e., SDRET and whether the stock is listed on NASDAQ). The results are reported in Table 8.

Table 8.

Cross-Sectional Analysis: Does the Credibility-Enhancing Effect of Conservatism on Positive Guidance Affect Some Firms More Than Others?RET = β₀+β₁CONS+β₂FNEWS+β₃CONS×FNEWS+β₄CONS×NEG+β₅FNEWS×NEG+β₆CONS×FNEWS×NEG+β₇NEG+β_kCONTROLS +β_kCONTROLS ×NEG+ϵIn the equation above, β₂ represents the market response coefficient to positive forecast news for firms in the bottom decile of conservatism. β₃ represents the change in the market response coefficient as CONS increases by one decile.

Panel A: Conservatism Effect on the Market Response to Positive Guidance Across Different Levels of Firm Visibility.
Test	Exp.	Estimate	p values	Estimate	p values	Estimate	p values	Test	Exp	Estimate	p values
Full sample				Small SIZE		Large SIZE		Diff.: Large – small
β₃ > 0	+	0.019***	(.004)	0.014*	(.073)	0.017*	(.057)	Diff. ≠ 0	?	0.003	.838
Full sample				Low FOLLOW		High FOLLOW		Diff.: High – low
β₃ > 0	+	0.019***	(.004)	0.014*	(.086)	0.029***	(.004)	Diff. ≠ 0	?	0.015	.330
Panel B: Conservatism Effect on the Market Response to Positive Guidance Across Different Levels of Market Volatility.
Test	Exp.	Estimate	p values	Estimate	p values	Estimate	p values	Test	Exp.	Estimate	p values
Full sample				Low SDRET		High SDRET		Diff.: High – low
β₃ > 0	+	0.019***	(.004)	0.012	(.154)	0.019**	(.025)	Diff. ≠ 0	?	0.006	.657
Full sample				NYSE/AMEX		NASDAQ		Diff. NASD.—N/A
β₃ > 0	+	0.019***	(.004)	0.017**	(.012)	0.024*	(.063)	Diff. ≠ 0	?	0.007	.697

Note. This table reports the estimates obtained for the coefficients of interest when testing H1a across different levels of firm visibility (measured by the firm size and analyst following) and market volatility (measured by the standard deviation of daily stock returns and by an indicator variable to indentify firms that are listed on NASDAQ). To divide the sample based on the level of SIZE, FOLLOW, and SDRET, the median is used as a cutoff point. For brevity, the coefficients of the other variables are not tabulated. CONS is the average decile ranking of the conservatism measures developed by Khan and Watts (2009; CSCORE) and by Givoly and Hayn (2000; SKEW). FNEWS is the absolute value of the difference between the value of the forecast and the analysts’ consensus on the day before the announcement, scaled by stock price. NEG is the dummy variable equal to 1 if the signed value of FNEWS is negative, and 0 otherwise. RET is the market-adjusted abnormal returns cumulated over the 3-day window centered on the forecast announcement date. SDRET is the standard deviation of daily stock returns. SIZE is the natural logarithm of sales. FOLLOW is the number of analysts following on the day before the forecast announcement. CONTROLS is the vector of control variables included in all equations but omitted here for brevity (i.e., MILLS, ERROR, FREQ, HORIZON, RANGE, SIZE, DISP, SDRET, FOLLOW). A detailed description of all variables in provided in the appendix. The p values are based on one-tail test when a directional prediction is made, and on a two-tail test otherwise. Standard errors are heteroskedasticity robust (White, 1980) and clustered at the firm level.

p values lower than 10%. **p values lower than 5%. ***p values lower than 1%.

In all subsamples, the coefficient estimates are in line with those of obtained in the full sample (the significance is somewhat lower due to the loss of half of the sample). This result suggests that the main results are robust to cross-sectional partitioning. Moreover, the coefficient estimates tend to be larger for firms that are more visible and volatile: however, this difference is never statistically significant. Overall, these results suggest that cross-sectional differences in conservatism’s effect are present but too weak to significantly affect the results, which appear robust to sample partitioning.¹⁹

Delayed Response on the Earnings Announcement Day

Similar to the postforecast drift test described in the section on postforecast drift patterns, I also test whether the news contained in the forecast explains future stock returns on the day when the company announces the earnings that they had forecast (EANN_RET). The rationale for this analysis follows the lines of H1b and H2b. If investors underreact to good news forecasts that suffer from low credibility, they are likely to wait for additional confirmatory signals before fully incorporating the forecast news into the stock price. The announcement of the actual earnings could be one of these signals, as it is an audited, backward-looking, nonestimated number. Consequently, part of investors’ delayed reaction could happen upon the earnings announcement. This, however, is likely to hold for announcements occurring soon after the forecast release; otherwise, their confirmatory value would probably be preempted by other information from analysts or industry peers. Similarly, bad news forecasts will raise uncertainty if investors are afraid that more bad news will soon be uncovered. When the actual earnings are announced, investors learn with better precision the extent of the bad news and uncertainty is (at least partially) resolved.

Therefore, whether guidance was positive or negative, the magnitude of the news contained in the forecast will be positively associated with market returns during the earnings announcement days. I test these predictions by replacing DRIFT with EANN_RET, the cumulative abnormal return in the 3-day window centered on the earnings announcement. This yields the following equation:

\begin{matrix} EANN_RE T_{it} = β_{0} + β_{1} CON S_{it} + β_{2} FNEW S_{it} + β_{3} CONS \times FNEW S_{it} \\ + β_{4} CON S_{it} \times NE G_{it} β_{5} FNEW S_{it} \times NE G_{it} + β_{6} CONS \times FNEW S_{it} \times NE G_{it} \\ + β_{7} NEG + β_{k} CONTROL S_{kit} + β_{k} CONTROL S_{kit} \times NE G_{it} + ε_{i} \end{matrix}

Table 9 (Panel A) reports the results when analyzing forecasts released no more than 45 days before the earnings announcement. I focus on these short-horizon forecasts because the confirmatory value of the earnings announcement is less likely to have been preempted by other information sources. On the downside, this restriction greatly diminishes the sample size (only 935 observations left), which impairs the power of the test. Despite this problem, the coefficient of FNEWS is significantly positive (0.153, p value = .07), suggesting that positive forecast news is indeed associated with a delayed positive reaction when actual earnings are announced. Similarly, the fact that the sum of FNEWS and FNEWS×NEG is also positive (1.804, p value = .06) suggests that earnings announcements occurring within 45 days of bad news forecasts are associated with partial reversal of the initial market overreaction. This result suggests that earnings announcements provide confirmatory information that partly mitigates investor uncertainty about how much trouble the firm is really experiencing, consistent with Rogers et al. (2009).

Table 9.

Additional Analysis: Is There a Delayed Response (a Partial Reversal) on the Day When a Firm Announces the Earnings That Were Predicted by a Good News (Bad News) Forecast? Does Conservatism Reduce the Delayed Response (the Partial Reversal)?EANN_RET = β₀+β₁CONS+β₂FNEWS+β₃CONS×FNEWS+β₄CONS×NEG+β₅FNEWS×NEG+β₆CONS×FNEWS×NEG+β₇NEG+β_kCONTROLS +β_kCONTROLS ×NEG+ϵIn the equation above, β₂ (the sum β₂+β₅), bold faced in the table, represents the association between positive (negative) forecast news and stock returns during the subsequent earnings announcement for firms in the bottom decile of conservatism. β₃ (the sum β₃+β₆), bold faced and shaded in the table, represents the change in the association as CONS increases by one decile.

Panel A: Earnings Announcement Within 45 Days of the Forecast Release.
Test of coefficient estimates—Good news forecasts					F test of significance of total coefficients (main effect + interaction)—Bad news forecasts
Variables	Test	Expected sign:	Estimate	p values	Test	Expected sign:	Estimate	p values
CONS	β₁ ≠ 0	?	0.153	(.455)	β₁ + β₄ ≠ 0	?	−0.128	(.497)
FNEWS	β₂ > 0	+	0.153*	(.070)	β₂+β₅ > 0	+	1.804*	(.058)
CONS×FNEWS	β₃ < 0	−	–0.028**	(.036)	β₃+β₆ < 0	−	−0.301**	(.035)
CONS × NEG	β₄ ≠ 0	?	−0.281	(.305)
FNEWS × NEG	β₅ ≠ 0	?	1.651	(.149)
CONS × FNEWS × NEG	β₆ ≠ 0	?	−0.273*	(.098)
Constant	β₀ ≠ 0	?	3.935	(.591)
CONTROLS, NEG			YES	(omitted)
CONTROLS × NEG			YES	(omitted)
Year/Ind. FE			YES	(omitted)
Observations			935
Adjusted R²			.026
Panel B: Earnings Announcement Between 45 and 180 Days of the Forecast Release.
Test of coefficient estimates—Good news forecasts					F test of significance of total coefficients (main effect + interaction)—Bad news forecasts
Variables	Test	Expected sign:	Estimate	p values	Test	Expected sign:	Estimate	p values
CONS	β₁ ≠ 0	?	−0.052	(.675)	β₁ + β₄ ≠ 0	?	−0.124	(.225)
FNEWS	β₂ > 0	+	0.121*	(.079)	β₂+β₅ > 0	+	1.534 ***	(.005)
CONS×FNEWS	β₃ < 0	–	−0.009	(.262)	β₃+β₆ < 0	−	–0.204 ***	(.005)
All other variables and year/Ind. FE			YES	(omitted)
Observations			2,904
Adjusted R²			.018

Note. CONS is the average decile ranking of the conservatism measures developed by Khan and Watts (2009; CSCORE) and by Givoly and Hayn (2000; SKEW). FNEWS is the absolute value of the difference between the value of the forecast and the analysts’ consensus on the day before the announcement, scaled by stock price. NEG is the dummy variable equal to 1 if the signed value of FNEWS is negative, and 0 otherwise. EANN_RET is the market-adjusted abnormal returns cumulated over the (–1, +1) window center on the announcement date of the actual earnings that have been predicted by the management forecast. CONTROLS is the vector of control variables included in all equations but omitted here for brevity (i.e., MILLS, ERROR, FREQ, HORIZON, RANGE, SIZE, DISP, SDRET, FOLLOW). A detailed description of all variables in provided in the appendix. The p values are based on one-tail test when a directional prediction is made, and on a two-tail test otherwise. FNEWS and EANN_RET are expressed in % points (i.e., multiplied by 100) to facilitate the interpretation of coefficients. Standard errors are heteroskedasticity robust (White, 1980) and clustered at the firm level. FE = fixed effects.

p values lower than 10%. **p values lower than 5%. ***p values lower than 1%.

Of particular relevance is the fact that the estimates of β₃ (–.028, p value = .04) and of the sum β₃+β₆ (–.301, p value = .03) are both negative and significant. This adds robustness to H1b and H2b, indicating that investors’ delayed reaction to good news forecasts (and partial correction after bad news forecasts) is attenuated by conservatism.

Panel B of Table 9 reports the results for forecasts with longer horizons (up to 180 days). In this case, other information sources are likely to have already preempted the confirmatory value of the earnings announcement. Consistent with this intuition, the association between forecast news and announcement returns is weaker in magnitude (though still significant). Conservatism’s effect in attenuating this association is also weaker and only significant for bad news forecasts. When extending the forecast horizon beyond 180 days, untabulated results show that all these coefficients become practically insignificant. Overall, the results reported in this section corroborate the findings discussed in the section on postforecast drift patterns.

Concluding Remarks

In this article, I analyze how investor response to good and bad news forecasts is affected by the firm’s level of conservatism. The results suggest that conservatism makes positive guidance more credible, thus reducing investor underreaction to it, and negative guidance less alarming, thus attenuating investor uncertainty and overreaction. By doing so, the article adds to the debate on the information effects of conservatism on the stock market. Some studies have suggested that such effects could be detrimental (e.g., Barth et al., 2017; Chen et al., 2014; Mensah et al., 2004; Penman & Zhang, 2002; Sen, 2005), while others have highlighted beneficial outcomes (e.g., Ahmed, Morton, & Schaefer, 2000; Garcìa Lara et al., 2011; B. H. Kim & Pevzner, 2010; J.-B. Kim & Zhang, 2016; LaFond & Watts, 2008; Mason, 2004; Ryan, 2000). In light of these different arguments and results, whether conservatism would help investors make use of voluntary disclosure has remained an empirical question.

In addition, the article expands current knowledge on the relation between conservatism and management guidance. A previous paper by Hui et al. (2009) suggests that conservatism can be a substitute for frequent and specific guidance to mitigate litigation risk and information asymmetry about losses. A recent working paper by Jaggi and Xin (2014) corroborates the findings of Hui et al. (2009). However, neither Hui et al. nor Jaggi and Xin consider the different question of whether, keeping constant all cross-sectional differences in a firm’s forecasting history and other characteristics, conservatism can complement guidance by facilitating investors’ pricing of forecast news. This latter, so far unexplored aspect of the interaction between conservatism and guidance, is the focus of my article. Therefore, my results are not in contrast with those of Hui et al. In fact, the descriptive statistics discussed earlier support them, strengthening the two papers’ coexistence in the literature. However, my article provides an incremental contribution over that of Hui et al. by suggesting that conservatism, though a substitute for guidance in reducing litigation risk and information asymmetry, can also be a complement to guidance by conditioning market reactions to it.

Moreover, this article contributes to the literature on market reactions to guidance. Research has shown that forecast attributes can improve the credibility of positive guidance and mitigate investor underreactions to it (Hutton & Stocken, 2009; Ng et al., 2013; Yang, 2012). Little attention has been paid to whether the attributes of financial statements, such as conservatism, can play such a role, too. Research has also documented that negative guidance generates a surge in investor uncertainty (Rogers et al., 2009) and a market overreaction (Das et al., 2012). My findings suggest that conservatism can help investors interpret and correctly price the unexpected voluntary disclosure of bad news.

As the focus of this article is on market reactions to voluntary disclosure, I explore whether conservatism mitigates the low credibility and rising uncertainty problems that affect positive and negative guidance, respectively. Therefore, I do not investigate or hypothesize about the effect of conservatism on reactions to other, nonvoluntary information releases, such as earnings announcements. Unlike forecast net income, reported net income is an audited, backward-looking number that is never presented as a range estimate. For these reasons, the announcement of positive earnings is unlikely to suffer much from lack of credibility, which makes it a poor setting to test credibility-related hypotheses, as noted by Ng et al. (2013). In addition, the announcement of a loss, being part of a predictable mandatory routine that managers cannot avoid, will not be as alarming as the unexpected release of negative guidance (Rogers et al., 2009), especially considering, again, its audited and backward-looking nature. In the absence of significant problems related to the low credibility of good news or to rising uncertainty following bad news, the arguments developed in my article do not represent reasons for predicting an effect of conservatism on market reactions to earnings announcements. Such effects may still exist, though, for different reasons. For instance, the post–earnings-announcement drift (PEAD) was shown to be caused by investors’ poor understanding of future earnings persistence (Bernard & Thomas, 1990)²⁰; this could potentially mediate an association between conservatism and postannouncement returns. As this type of research question falls outside of the scope of this article, I leave the issue to other studies.²¹

Footnotes

Appendix

Acknowledgements

I am grateful to the Editor, Bharat Sarath, the Associate Editor, Dan Givoly, and two anonymous reviewers for their insightful comments and suggestions. Special thanks to Sasson Bar-Yosef, Matt DeAngelis, Robbie Moon, Annalisa Prencipe, Steve Ryan, James Wilhelm, and Ari Yezegel. I also wish to thank workshop participants at Bentley University, Georgia State University, and the 2015 annual meeting of the Canadian Academic Accounting Association. I thank Middle Tennessee State University and Georgia State University for support. An earlier draft of this article was written when I was at Georgia State University.

Authors’ Note

Data are available from public sources identified in the text.

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) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Carlo D’Augusta

Notes

References

Ahmed

A. S.

Morton

R. M.

Schaefer

T. F.

(2000). Accounting conservatism and the valuation of accounting numbers: Evidence on the Feltham-Ohlson (1996) model. Journal of Accounting, Auditing & Finance, 15, 271-292.

Aier

J. K.

Chen

Pevzner

(2014). Debtholders’ demand for conservatism: Evidence from changes in directors’ fiduciary duties. Journal of Accounting Research, 52, 993-1027.

Ali

Gurun

(2009). Investor sentiment, accruals anomaly, and accruals management. Journal of Accounting, Auditing & Finance, 24, 415-431.

Baginski

S. P.

Conrad

E. J.

Hassell

J. M.

(1993). The effects of management forecast precision on equity pricing and on the assessment of earnings uncertainty. The Accounting Review, 88, 913-927.

Balakrishnan

Watts

R. L.

Zuo

(2016). The effect of accounting conservatism on corporate investment during the global financial crisis. Journal of Business Finance & Accounting, 43, 513-542.

Ball

Shivakumar

(2005). Earnings quality in UK private firms: Comparative loss recognition timeliness. Journal of Accounting and Economics, 39, 83-128.

Ball

Shivakumar

(2008). Earnings quality at initial public offerings. Journal of Accounting & Economics, 45, 324-349.

Banker

R. D.

Das

Datar

S. M.

(1993). Complementarity of prior accounting information: The case of stock dividend announcements. The Accounting Review, 68, 28-47.

Barth

M. E.

Landsman

W. R.

Raval

Wang

(2017). Asymmetric timeliness and the resolution of investor disagreement and uncertainty at earnings announcements (SSRN 2348905). Retrieved from https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2348905

10.

Basu

(1997). The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting & Economics, 24, 3-37.

11.

Bernard

V. L.

Thomas

J. K.

(1990). Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting & Economics, 13, 305-340.

12.

Billings

M. B.

Jennings

Lev

(2015). On guidance and volatility. Journal of Accounting & Economics, 60, 161-180.

13.

Callen

J. L.

Segal

(2013). An analytical and empirical measure of the degree of conditional conservatism. Journal of Accounting, Auditing & Finance, 28, 215-242.

14.

Chen

L. H.

Folsom

Paek

Sami

(2014). Accounting conservatism, earnings persistence and pricing multiples on earnings. Accounting Horizons, 28, 233-260.

15.

Cho

Kwon

S. Y.

(2014). Trading volume and investor disagreement around management forecast disclosures. Journal of Accounting, Auditing & Finance, 29, 3-30.

16.

Chuk

Matsumoto

Miller

G. S.

(2013). Assessing methods of identifying management forecasts: CIG vs. researcher collected. Journal of Accounting & Economics, 55, 23-42.

17.

Ciconte

Kirk

Tucker

J. W.

(2014). Does the midpoint of range earnings forecasts represent managers’ expectations?Review of Accounting Studies, 19, 628-660.

18.

Conrad

Cornell

Landsman

W. R.

(2002). When is bad news really bad news?Journal of Finance, 57, 2507-2533.

19.

Das

Kim

Patro

(2012). On the anomalous stock price response to management earnings forecasts. Journal of Business Finance & Accounting, 39, 905-935.

20.

D’Augusta

Bar-Yosef

Prencipe

(2016). The effects of conservative reporting on investor disagreement. European Accounting Review, 25, 451-485.

21.

Epstein

L. G.

Schneider

(2008). Ambiguity, information quality, and asset pricing. Journal of Finance, 63, 197-229.

22.

Francis

Lafond

Olsson

Schipper

(2007). Information uncertainty and post-earnings-announcement-drift. Journal of Business Finance & Accounting, 34, 403-433.

23.

Garcìa Lara

J. M.

Garcìa Osma

Penalva

. (2011). Conditional conservatism and cost of capital. Review of Accounting Studies, 16, 247-271.

24.

Garcìa Lara

J. M.

Garcìa Osma

Penalva

. (2014). Information consequences of accounting conservatism. European Accounting Review, 23, 173-198.

25.

Garcìa Lara

J. M.

Garcìa Osma

Penalva

. (2016). Accounting conservatism and firm investment efficiency. Journal of Accounting & Economics, 61, 221-238.

26.

Garcìa Lara

J. M.

Garcìa Osma

Penalva

. (2017). Conditional conservatism and the limits to earnings management (SSRN 2165694). Retrieved from https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2165694

27.

Garfinkel

J. A.

Sokobin

(2006). Volume, opinion divergence, and returns: A study of post–earnings announcement drift. Journal of Accounting Research, 44, 85-112.

28.

Givoly

Hayn

(2000). The changing time-series properties of earnings, cash flows and accruals: Has financial reporting become more conservative?Journal of Accounting & Economics, 29, 287-320.

29.

Givoly

Hayn

Katz

(2017). The changing relevance of accounting information to debt holders over time. Review of Accounting Studies, 22, 64-108.

30.

Guay

W. R.

Verrecchia

R. E.

(2007). Conservative disclosure (SSRN 1383642). Retrieved from https://papers.ssrn.com/sol3/papers.cfm?abstract_id=817108

31.

Heckman

(1979). Sample selection bias as a specification error. Econometrica, 47, 153-161.

32.

Hirst

D. E.

Koonce

Venkataraman

(2008). Management earnings forecasts: A review and framework. Accounting Horizons, 22, 315-338.

33.

L. C. J.

Liu

C.-S.

Ramanan

(1997). Open-market stock repurchase announcements and revaluation of prior accounting information. The Accounting Review, 72, 475-487.

34.

Hribar

Yang

(2016). CEO overconfidence and management forecasting. Contemporary Accounting Research, 33, 204-227.

35.

Hui

K. W.

Matsunaga

S. R.

Morse

(2009). The impact of conservatism on management earnings forecasts. Journal of Accounting & Economics, 47, 192-207.

36.

Hutton

A. P.

Miller

Skinner

D. J.

(2003). The role of supplementary statements with management earnings forecasts. Journal of Accounting Research, 41, 867-891.

37.

Hutton

A. P.

Stocken

P. C.

(2009). Prior forecasting accuracy and investor reaction to management earnings forecasts (SSRN 817108). Retrieved from https://papers.ssrn.com/sol3/papers.cfm?abstract_id=817108

38.

Jaggi

Xin

(2014). Accounting conservatism and management earnings forecasts (SSRN 2246168). Retrieved from https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2246168

39.

Jennings

(1987). Unsystematic security price movements, management earnings forecasts, and revisions in consensus analyst earnings forecasts. Journal of Accounting Research, 25, 90-110.

40.

Kasznik

Lev

(1995). To warn or not to warn: Management disclosures in the face of an earnings surprise. The Accounting Review, 70, 113-134.

41.

Khan

Watts

R. L.

(2009). Estimation and empirical properties of a firm-year measure of accounting conservatism. Journal of Accounting and Economics, 48, 132-150.

42.

Kim

B. H.

Pevzner

(2010). Conditional accounting conservatism and future negative surprises: An empirical investigation. Journal of Accounting and Public Policy, 29, 311-329.

43.

Kim

J.-B.

Zhang

(2016). Accounting conservatism and stock price crash risk: Firm-level evidence. Contemporary Accounting Research, 33, 412-441.

44.

Kim

Pan

Zuo

(2013). The role of accounting conservatism in the equity market: Evidence from seasoned equity offerings. The Accounting Review, 88, 1327-1356.

45.

Kothari

S. P.

Shu

S. Z.

Wysocki

P. D.

(2009). Do managers withhold bad news?Journal of Accounting Research, 47, 241-276.

46.

Krishnan

G. V.

Pevzner

Sengupta

(2012). How do auditors view managers’ voluntary disclosure strategy? The effect of earnings guidance on audit fees. Journal of Accounting and Public Policy, 31, 492-515.

47.

LaFond

Watts

R. L.

(2008). The information role of conservatism. The Accounting Review, 83, 447-478.

48.

Lennox

C. S.

Park

C. W.

(2006). The informativeness of earnings and management’s issuance of earnings forecasts. Journal of Accounting & Economics, 42, 439-458.

49.

Liang

(2003). Post-earnings announcement drift and market participants’ information processing biases. Review of Accounting Studies, 8, 321-345.

50.

Liu

C. C.

Ryan

S. G.

Wahlen

J. M.

(1997). Differential valuation implications of loan loss provisions across banks and fiscal quarters. The Accounting Review, 72, 133-146.

51.

Mashruwala

(2014). Is there a “torpedo effect” in earnings announcement returns? The role of short-sales constraints and investor disagreement. Journal of Accounting, Auditing & Finance, 29, 519-546.

52.

Mason

(2004). The impact of accounting conservatism on the magnitude of the differential information content of cash flows and accruals. Journal of Accounting, Auditing & Finance, 19, 249-282.

53.

Mensah

Y. M.

Song

S. S. M.

(2004). The effect of conservatism on analysts’ annual earnings forecast accuracy and dispersion. Journal of Accounting, Auditing & Finance, 19, 159-183.

54.

Miller

(1977). Risk, uncertainty, and divergence of opinion. The Journal of Finance, 32, 1151-1168.

55.

Tuna

Verdi

R. S.

(2013). Management forecast credibility and underreaction to news. Review of Accounting Studies, 18, 956-986.

56.

Payne

Robb

(2000). Earnings management: The effect of ex ante earnings expectations. Journal of Accounting, Auditing & Finance, 15, 371-392.

57.

Penman

S. H.

Zhang

X.-J.

(2002). Accounting conservatism, the quality of earnings, and stock returns. The Accounting Review, 77, 237-264.

58.

Richardson

S. A.

Sloan

R. G.

Soliman

M. T.

Tuna

(2005). Accrual reliability, earnings persistence and stock prices. Journal of Accounting & Economics, 39, 437-485.

59.

Rogers

J. L.

Skinner

D. J.

van Buskirk

(2009). Earnings guidance and market uncertainty. Journal of Accounting & Economics, 48, 90-109.

60.

Rogers

J. L.

Stocken

P. C.

(2005). Credibility of management forecasts. The Accounting Review, 80, 1233-1260.

61.

Rogers

J. L.

van Buskirk

(2009). Shareholder litigation and changes in disclosure behavior. Journal of Accounting & Economics, 47, 136-156.

62.

Ryan

S. G.

(2000). Discussion of “Accounting conservatism and the valuation of accounting numbers: Evidence on the Feltham-Ohlson (1996) model.”Journal of Accounting, Auditing & Finance, 15, 293-299.

63.

Sen

P. K.

(2005). Reported earnings quality under conservative accounting and auditing. Journal of Accounting, Auditing & Finance, 20, 229-256.

64.

Sohn

B. C.

(2012). Analyst forecast, accounting conservatism and the related valuation implications. Accounting & Finance, 52(Suppl.), 311-341.

65.

Tang

Zarowin

Zhang

(2015). How do analysts interpret management range forecasts?Accounting, Organizations and Society, 42, 48-66.

66.

Tucker

J. W.

(2007). Is openness penalized? Stock returns around earnings warnings. The Accounting Review, 82, 1055-1087.

67.

Watts

R. L.

(2003). Conservatism in accounting part I: Explanations and implications. Accounting Horizons, 17, 207-221.

68.

White

(1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48, 817-838.

69.

Yang

H. I.

(2012). Capital market consequences of managers’ voluntary disclosure styles. Journal of Accounting and Economics, 53, 167-184.

70.

Zhang

F. X.

(2006). Information uncertainty and stock returns. Journal of Finance, 61, 105-136.