Earnings Guidance, Bias, and Stock Price Crash Risk Sophia J.W. Hamm Fisher College of Business The Ohio State University Columbus, OH 614 292 2529 [email protected]Edward Xuejun Li Zicklin School of Business Baruch College New York, NY 646 312 3235 [email protected]Jeffrey Ng School of Accounting and Finance The Hong Kong Polytechnic University Kowloon, Hong Kong +852 2766 7099 [email protected]August 5, 2018 ____________________ We appreciate helpful comments from an anonymous reviewer, Anne Beatty, Jeremy Bertomeu, Ilan Guttman, Darren Roulstone, Lakshmanan Shivakumar (Editor), Nemit Shroff, Andy Van Buskirk, Rodrigo Verdi, Ro Verrecchia, Jerry Zimmerman, seminar participants at Kent State University, and conference participants at the University of Minnesota Empirical Accounting Conference 2012, FARS 2013, and the Baruch–SWUFE Accounting Conference 2013.
69
Embed
Earnings Guidance, Bias, and Stock Price Crash Risk
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Earnings Guidance, Bias, and Stock Price Crash Risk
Data Availability: The data used in this study are available from the public sources identified in
the paper.
1
1. Introduction
In response to a series of major corporate scandals (e.g., Enron, AIG, etc.) and the recent
financial crisis, investigating the cause of extreme price declines has become a source of
considerable interest to regulators, practitioners, and researchers. In her testimony before the
Financial Crisis Inquiry Commission, then-SEC Chairman Mary Schapiro contended that “[a]
central question [...] is whether investors received timely and accurate disclosure concerning
deteriorating business conditions” (Schapiro, 2010). While many recent studies in finance and
accounting investigate how characteristics of the earnings generation process (e.g., accrual quality,
conservatism, and comparability) are related to future stock price crashes (e.g., Hutton et al., 2009;
Kim and Zhang, 2014, 2016; DeFond et al., 2015; Kim et al., 2016), there is limited evidence on
the impact of voluntary disclosure on stock price crash risk.
A careful analysis of a prominent voluntary disclosure mechanism, earnings guidance, and
the bias therein could enhance our understanding of the relation between earnings information and
crash risk in two ways. First, prior studies highlight guidance as a primary outlet of timely earnings
news. For example, Ball and Shivakumar (2008, p.1009) estimate that guidance, if issued, explains
20%-25% of the total quarterly stock return variance as compared to only 3.5%-4.5% from
earnings releases. Similarly, Beyer et al. (2010) show that earnings releases and SEC filings
account for less than 12% of the total stock return variance explained by financial disclosures,
compared to over 55% from guidance. Given its impact on stock returns, guidance could have
important implications for crash risk. Second, while prior studies document significant
associations between various earnings characteristics and crash risk, there is limited evidence on
how disclosure bias gets impounded into price and leads to future crashes. One difficulty is that
the metrics of disclosure bias often have significant measurement errors and interpretive ambiguity
2
(Dechow et al., 2010). For example, discretionary accrual models have a low explanatory power
and can attribute changes in business fundamentals to misreporting (Owens et al., 2017). Earnings
guidance, however, provides a nice setting for capturing forecast bias and hence allows for a
cleaner and more direct test. The purpose of our study, therefore, is to investigate how guidance
and its bias are related to stock price crash risk.
Our empirical investigation is important because, ex ante, there are no clear predictions on
the relation between guidance and crash risk. Intuitively, one might expect that issuing guidance
lowers crash risk because a vast literature views guidance as an opportunity for managers to reveal
private information and adjust market expectations towards their beliefs (Ajinkya and Gift, 1984;
Kasznik and Lev, 1995; Matsumoto, 2002). Prior studies also suggest that guidance allows for
better monitoring, which curbs managers’ value-destroying behaviors (e.g., Bushman and Smith,
2001; Healy and Palepu, 2001; Nagar et al., 2003). Further, litigation risk prompts managers to
forewarn investors about bad news (Skinner, 1994). Collectively, this typical view predicts that
managers issue guidance to reduce the risk of a future stock price crash. We refer to this prediction
as the crash preemption hypothesis.
However, considering that guidance is voluntary and not audited before issuance, there are
growing concerns about guidance contributing to inflated market expectations (Core, 2001; Healy
and Palepu, 2001), which could, in turn, raise crash risk. One possibility is managers’ intentional
misuse of guidance.1 Specifically, if declining business conditions prompt a manager’s career
concerns, it could incentivize her to deviate from truthful reporting (Jin and Myers, 2006; Bleck
and Liu, 2007; Benmelech et al., 2010). Instead, she could sugarcoat poor results with optimistic
1 For prior studies, see, for example, Aboody and Kasznik (2000), Amel-Zadeh and Meeks (2018), Bergman and
Roychowdhury (2008), Cheng and Lo (2006), Cotter et al. (2006), Feng et al. (2015), Noe (1999), Rees et al. (2014),
and Shroff et al. (2013).
3
guidance that creates an illusion of stability while betting that a future revival will hide the
discrepancy (Graham et al., 2005; Kothari et al., 2009). Given asymmetric information, it is often
difficult for investors to immediately detect changes in guidance incentives (Fischer and
Verrecchia, 2000; Hutton et al., 2003; Rogers and Stocken, 2005).2 Another possibility is that
managers issue unintentionally biased guidance. For example, in periods of high sentiment,
managers can overinvest and justify their decision with optimistic forecasts. Investors influenced
by the same high sentiment can keep the consequent bubble alive for some time. In either case, to
the extent that a firm cannot produce earnings to meet inflated expectations and the future
revelation of bad news triggers an abrupt decline in stock price, more guidance could engender a
higher crash risk. We refer to this predication as the inflated expectation hypothesis.
Collectively, the above hypotheses highlight the ex-ante tension in the research question
about the net impact of guidance on crash risk. To address this question, we analyze a sample of
71,909 firm years from 1997 to 2015. We follow Hutton et al. (2009) to measure crash risk after
controlling for both market and industry returns. Such a firm-specific measure helps alleviate the
concern that any result is purely driven by market-wide shocks. To test the link between guidance
and future crashes, we focus on long-horizon guidance (i.e., annual earnings guidance with its
realized value reported in the following year) and use its incidence and frequency within a fiscal
year as our initial measures of guidance.3
Our first analysis shows a positive relation between a firm’s guidance frequency and its
crash risk, after controlling for information characteristics such as accruals quality (Hutton et al.,
2 The financial reporting fraud at Qwest Communications (United States v. Nacchio (No. 07-1311, March 17, 2008)
offers a good illustration. In September 2000, Qwest’s CEO Joseph Nacchio issued an annual revenue forecast of
$21.3 to $21.7 billion for fiscal year 2001, despite an internal memo hinting at a disappointing $20.4 billion or worse.
He reaffirmed the guidance several times later in spite of deteriorating prospects. When Qwest finally revealed the
bad news in June 2001, its stock price plunged about 21%. 3 Untabulated analysis finds no significant relation between crash risk and short-horizon guidance (i.e., earnings
guidance, annual or quarterly, with its realized value reported in the same year).
et al., 2016), real earnings management (Francis et al., 2016), and 10-K readability (Ertugrul et al.,
2017). Further analysis indicates that this relation is economically significant. This finding is
surprising given that prior literature often relates more disclosure to greater transparency.
To provide direct evidence on the mechanism of this relation, we classify each guidance as
optimistic or pessimistic by comparing guidance with its realized value. We then examine how a
firm’s crash risk is related to its issuance of optimistic and pessimistic guidance. We find a
significant positive relation between the issuance of optimistic guidance and crash risk but no
significant relation between pessimistic guidance and crash risk.4 These results suggest that on
average, the inflated expectation hypothesis is descriptive of the link between guidance and crash
risk. Specifically, optimistic guidance is related to inflated expectations, which lead to a higher
crash risk. However, there is no general evidence to support the crash preemption hypothesis.
We then attempt to address various robustness and endogeneity issues that could arise in
the analysis of the relation between optimistic guidance and crash risk.5 First, we show that our
finding is generalizable across most years and not specific to years with high sentiment or market
crashes. Second, we use a Conditional Logit model and Chamberlain’s Random Effects (CRE)
probit model (Wooldridge, 2002) to control for firm fixed effects and find robust results. Third,
we exploit a natural experiment under Regulation SHO (Reg SHO) for the relation between
guidance optimism and crash risk. Chen et al. (2014) and Li and Zhang (2015) show that while
Reg SHO has no significant impact on the pilot firms’ guidance frequency and bias, it increases
4 Neutral guidance is left out of this analysis because it comprises only a small portion (6%) of the long-horizon
guidance sample. Including neutral guidance does not change the tenor of results. In particular, untabulated analysis
shows that neutral guidance exhibits no significant association with crash risk. 5 All our results related to guidance optimism are robust to including the corresponding guidance pessimism
measures as control variables.
5
short selling pressure and price sensitivity to bad news for such firms. Such increased sensitivity
to disappointing realized earnings predicts a stronger relation between optimistic guidance and
crash risk for the pilot firms during the Reg SHO treatment period. However, such a differential
relation should not be present in either the pre- or post-Reg SHO periods. We find results consistent
with our predictions. Given the randomness in selecting the pilot firms, this analysis provides
strong evidence that the relation between guidance optimism and crash risk is not purely driven by
omitted correlated variables. Lastly, we consider Regulation Fair Disclosure (Reg FD) as an
exogenous shock for guidance since firms issue more guidance to replace the selective disclosure
that Reg FD prohibits (Bailey et al., 2003; Heflin et al., 2003; Heflin et al., 2016).6 Because the
idea behind Reg FD is “leveling the playing field” and boosting confidence in the capital market,
it is unlikely to directly raise the probability of future crashes.7 Our two stage method yields a
similar positive relation between optimistic guidance and crash risk. A caveat is that many market-
wide events surrounding Reg FD could contaminate this result.
While guidance optimism could arise either intentionally or unintentionally, one might
expect investor perception of the latter origin to engender a lower crash risk. To proxy for
conditions under which investors would perceive bias as more unintentional, we use high litigation
risk, which is expected to curb intentional bias. We find that when bias is more likely to be
unintentional, the positive relation between guidance optimism and crash risk becomes more
attenuated. To further examine the issue of unintentional and intentional bias, we decompose
guidance bias into a predictable portion that arises from serial correlation and an innovation in bias,
6 To give a sense of the shock, we show that the percentage of guidance firms jumped from 27.2% in 2000 to 37.6%
in 2001 and the percentage of firms that issue optimistic guidance rose from 13.7% to 24.3%. 7 On the contrary, Kothari et al. (2009) provide evidence that firms reduced the extent of bad news withholding
relative to good news after Reg FD, which implies that the regulation may have indirectly decreased crash risk. This
speaks to the strength of Reg FD as our instrumental variable (Larcker and Rusticus, 2010).
6
as Gong et al. (2011) argue that bias due to serial correlation is associated with managers’
unintentional information processing bias rather than opportunistic forecasting behaviors. We find
that crash risk is positively related to both types of guidance bias.
We also conduct several more analyses to triangulate our results. First, we test market
reactions to guidance issued prior to crashes and find no evidence that investors undo the guidance
bias. This outcome offers direct evidence on how disclosure bias is impounded into price before
crashes. Second, we find a negative relation between guidance optimism and future stock returns.
We note that there is an important distinction between this type of negative returns and price
crashes, as theories suggest that investors have a strong aversion to large occasional crashes and
demand extra compensation for bearing such a high level of crash risk (Bates, 1991; Pan, 2002).
Nevertheless, the similarity in the results with stock price crash risk and future stock returns
suggest that the negative impact of stock price crashes are not transitory.
While we find no significant relation between pessimistic guidance and crash risk in
general, it is difficult to fully dismiss the preemptive role of guidance. In our final analysis, we
follow Kim and Park (2012) in identifying a subset of pessimistic guidance issued for downward
expectation management and show that such guidance is negatively related to future crashes,
lending some support to the crash preemption hypothesis.
Taken as a whole, our study extends prior research on corporate disclosure and crash risk.
Our results suggest that earnings guidance plays an important role in crash risk, one that is
incremental to the many earnings characteristics prior literature examines. We also provide the
first direct evidence on the mechanism by which the forecast bias gets impounded into price and
leads to future crashes. Finally, our investigation of a long-horizon capital market outcome of
guidance also adds to the management forecast literature that typically focuses on short term
7
market reactions (e.g., Patell, 1976; Penman, 1980; Rogers et al., 2009). On a broader note, our
study highlights an important link between voluntary disclosure and future stock price downside
risk. We caution that our evidence does not suggest that managers have a general tendency to
inflate their forecasts or that guidance optimism is prevalent, because a stock price crash, by
construction, is an extraordinary event. A more appropriate, albeit narrower, conclusion is that if
a firm provides guidance that later turns out to be optimistic, there is a higher stock price crash
risk and predictable variations exist in this relation.
The rest of the paper is organized as follows. Section 2 summarizes the related research
and our predictions. In Section 3, we discuss our data and the basic research design. Section 4
presents empirical results on the replication of prior studies and the relation between guidance and
crash risk. In Section 5, we conduct a direct test of forecast bias and crash risk. Section 6 provides
several supplemental analyses. We conclude in Section 7.
2. Prior research and predictions
2.1 Prior research on corporate disclosure and crash risk
Beginning with Jin and Myers (2006), researchers have been concerned about whether the
information asymmetry between managers and shareholders, coupled with managers’ self-interest,
could be related to stock price crash risk. As Taleb (2007) indicates, a good understanding of these
extreme outcomes can offer valuable insight into their true nature. The recent literature on crash
risk argues that a stock price crash occurs when investors realize that stock prices have been
(severely) inflated and that a crash’s occurrence could be an indicator of prior agency problems.8
While several analytical works use different models and settings, the underlying themes are largely
8 While recent research has focused on the agency problems that lead to extreme price declines, the early literature
examines a few equity market-based explanations for price crashes (e.g., Chen et al., 2001; French et al., 1987; Hong
and Stein, 2003; Romer, 1993).
8
similar: managers’ career concerns give them an incentive to conceal bad news (i.e., job security
or compensation) and opacity allows managers to hoard bad news, which subsequently leads to a
price crash (Jin and Myers, 2006; Bleck and Liu, 2007; Benmelech et al., 2010).9
Seeking to explore the precise nature of the agency problems, recent empirical studies have
investigated how crashes arise from managers’ bad news hoarding, which could be due to tax
avoidance (Kim et al., 2011a), equity-based compensation (Kim et al., 2011b), and opaque
reporting practices (Jin and Myers, 2006; Hutton et al., 2009; Kim and Zhang, 2014; Kim and
Zhang, 2016; Kim et al., 2016; Kim et al., 2018). More specifically, based on information from
earnings reports, Hutton et al. (2009) demonstrate that poor accruals quality in reported annual
earnings allows managers to conceal bad news, which leads to future price crashes. Kim and Zhang
(2014) confirm Hutton et al.’s (2009) inference using expected crash risk. Kim and Zhang (2016)
also show that the level of accounting conservatism inferred from reported earnings is negatively
associated with crash risk. Further, based on information from 10-K filings, Kim et al. (2018) and
Ertugrul et al. (2017) find that less readable 10-Ks predict a higher future price crash risk. Finally,
Kim et al. (2016) find that financial statement comparability can discipline managers from
concealing bad news and shows a negative relation between comparability and expected crash risk.
Despite these efforts, the bad news hoarding mechanisms examined to date have largely
been confined to characteristics of the earnings generation process or disclosures within the
mandatory reporting system. This is limited evidence on the role played by voluntary disclosure.
However, prior studies have shown that voluntary disclosure has a significant price impact (Ball
9 For example, the model in Jin and Myers (2006) predicts the following link between opacity and crash risk. Self-
interested managers have incentives to hide bad news about cash flow innovations because their informational
advantage allows them to exploit shareholders. Opacity about firm operations helps managers conceal information.
When the news is negative, managers would personally absorb losses and conceal the bad news to keep their jobs.
However, when the accumulated losses become excessive, they exercise the abandonment option and reveal the
accumulated bad news all at once, leading to extreme price declines.
9
and Shivakumar, 2008; Beyer et al., 2010), which is particularly relevant when managers’ motives
in disclosing are tied to stock price outcomes. While prior studies document associations between
earnings characteristics and crash risk, there is little evidence on how disclosure bias gets
incorporated into stock price, subsequently leading to crashes. A major hurdle faced by prior
studies may be the lack of clean measures on disclosure bias. Prior research shows that
discretionary accrual models often misattribute changes in business fundamentals to opportunistic
reporting (Owens et al., 2017). Dechow et al. (2010) also contend that many estimation models
(e.g., accruals, conservatism, etc.) are plagued by low explanatory power and interpretative
ambiguity.
To address these issues, we extend prior studies by investigating a prominent type of
voluntary disclosure, namely earnings guidance. Two unique features make this an important
disclosure setting to examine. First, as discussed above, guidance has a considerable impact on
stock price. Second, it is relative easy and straightforward to measure disclosure bias by comparing
a forecast with the subsequently realized value. The objective of our paper is to provide a more
complete picture on the relation between corporate disclosure and crash risk.
2.2 Earnings guidance and crash risk
In addition to the aforesaid reasons, our empirical investigation into the relation between
management earning guidance and stock price crash risk is important because, ex ante, there are
no clear predictions on how guidance and its forecast bias affect future stock price crashes.
Therefore, we adopt a more balanced approach to carefully evaluate the different effects that
guidance and its bias could have on crash risk.
Specifically, in line with traditional disclosure theory (Verrecchia, 2001), the expectation
adjustment hypothesis by Ajinkya and Gift (1984) posits that managers issue guidance to narrow
10
the gap between managers’ and investors’ expectations about future earnings. The idea here is that
information asymmetry exists between the firm and market participants. Managers, seeking to
reduce this information asymmetry, issue earnings guidance to synchronize investors’ earnings
expectations with managers’ beliefs. A series of studies present evidence in support of the
expectations adjustment hypothesis (e.g., Hassell and Jennings, 1986; Kasznik and Lev, 1995;
Matsumoto, 2002). Consistent with guidance reducing information asymmetry, Coller and Yohn
(1997) find that bid-ask spreads decrease after guidance is issued. Frankel et al. (1995) offer further
evidence that managers issue more guidance before accessing capital markets to lower the costs of
raising capital. If managers provide guidance to forewarn investors when their firms face a
downturn in business, we would expect guidance to reduce the risk of a future stock price crash.
In addition, the information contained within earnings guidance also allows for better monitoring
and reduces managers’ incentives to shirk or engage in value-destroying behaviors that are likely
to trigger price crashes (Bushman and Smith, 2001; Healy and Palepu, 2001; Nagar et al., 2003).
Furthermore, Skinner (1994) suggests that managers have strong incentives to avoid litigation risk
by issuing bad news guidance that preempts large negative earnings surprises, which would trigger
significant price declines and considerable litigation costs. Collectively, these arguments suggest
that guidance could have a crash preemption role, which would thus create a negative relation
between guidance and stock price crash risk.
However, recent literature has considered the possibility that earnings guidance could lead
to a misalignment between market expectations and firm fundamentals because of possible bias in
earnings guidance. The bias could be either intentional or unintentional. First, prior studies provide
extensive discussion and examination of the opportunistic use of voluntary disclosure. As Healy
and Palepu (2001, p. 425) caution, “the extent to which voluntary disclosure mitigates resource
11
misallocation in the capital market depends on the degree of credibility of information [...].
Because managers have incentives to make self-serving voluntary disclosures, it is unclear whether
management disclosures are credible.” Core (2001) adds to this agency point of view by noting
that in addition to the informational role of disclosure, it is important to jointly consider managers’
incentives and corporate governance structure to understand firms’ optimal disclosure policies and
their enforcement. Fischer and Verrecchia (2000) further demonstrate that managers have
incentives to bias reports if there is sufficient uncertainty about their reporting objectives.
Hermalin and Weisbach (2012) also predict that career concerns (i.e., job security or
compensation) could induce managers to opportunistically distort disclosure if they are to be
evaluated against it. As Kothari et al. (2009) argue, managers face asymmetric payoffs in
disclosure because good news increases compensation and extends tenure, whereas bad news leads
to adverse outcomes such as reduced compensation, termination of employment, and a tarnished
reputation in labor markets. Consequently, managers have incentives to issue optimistic guidance
that camouflages bad news in the hope that their firm’s business conditions will improve insofar
as to nullify the need to ever report such news (Graham et al., 2005). Even if the guidance is bound
to be verified later against realized values, it cannot completely discourage managers’
opportunistic guidance.10
One might also expect unintentional upward bias in guidance to contribute to inflated
expectations. Hurwitz (2017) finds that guidance optimism increases with investor sentiment and
10 A general concern about the opportunistic disclosure conjecture is that managers should ex-ante engage in truthful
disclosure out of a rational belief that the truth will be revealed in the future. In fact, an extensive literature provides
evidence that managers are, on average, not manipulative. Our study, however, narrows down to the context in which
managers are not always perfectly truthful in their disclosure. Likely rationales include career concerns, as well as the
notion that the optimal disclosure choice is not always the perfect transparency because eliminating all manipulations
can be too costly to stockholders (Watts and Zimmerman, 1986; Lambert et al., 1991). Another possibility is that
(some) managers are not completely rational and/or they believe that their misrepresentation will not be detected
(Dichev et al., 2013).
12
that such sentiment-driven optimism is most likely unintentional. In periods of high sentiment,
managers holding optimistic views would overinvest and issue optimistic guidance to justify their
decision. Hribar and Yang (2016) find consistent evidence that managerial overconfidence
increases the amount of optimism in management forecasts. When the sentiment-driven bubble
subsequently bursts, investor and managers revise their beliefs, causing a stock price crash. Kim,
et al. (2016) present a similar argument that when managers overestimate their investment returns,
more crashes ensue. While they do not test guidance, we conjecture that unintentional guidance
optimism would play an important role. In other words, to the extent that the market does not
immediately unravel unintentional optimism in guidance, there could be inflated market
expectations and a higher likelihood of a future stock price crash.11
Taken together, regardless of whether the inflated expectations are related to intentional or
unintentional optimistic guidance, there will be a positive relation between guidance and crash
risk. We refer it as the inflated expectation hypothesis. In sum, the average effect of guidance and
its bias on stock price crash risk is an open empirical question. We investigate this question first
providing an initial analysis of the relation between guidance and stock price crash risk. We then
provide in-depth analyses on how guidance optimism, the guidance characteristic most likely to
contribute to inflated expectations, could explain the relation between guidance and stock price
crash risk.
3. Data and basic research design
11 In the real world, it is ex-ante difficult for investors to price-protect against bias in information, intentional or
unintentional, especially in the presence of information asymmetry between the firm and its investors. The nature of
voluntary disclosure is likely to make price-protection even more difficult. The assumption that investors can easily
undo the intentional optimistic bias at the time of the disclosure begs the question of why managers would make the
effort to create bias, thereby exposing themselves to litigation and reputation risk. If the disclosure contains
optimistic bias because both the managers and market participants are overly optimistic about the firm’s prospects,
investors are even more likely to fail to infer and price the bias.
13
3.1 Data
We draw our primary sample from the intersection of CRSP and Compustat. Our sample
begins in 1997, the first year that the First Call CIG provides a comprehensive coverage of earnings
guidance, and ends in 2015. Because First Call CIG was discontinued in 2011, we supplement it
with the new I/B/E/S guidance data, which cover guidance from 2002 and onward12. As we
examine the relation between guidance and future stock price crashes, we focus on 62,817 long-
horizon annual EPS guidance with points, range, and open-ended estimates from the union of these
two databases. Long-horizon guidance refers to management forecasts whose realized values are
to be reported in the next year. Consistent with prior literature, we exclude pre-announcements.
Similar to Hutton et al. (2009), we exclude low-priced stocks and firms in the financial and utilities
industries and calculate crash risk using CRSP data. We further use I/B/E/S estimates, Thomson
Reuters Insiders, and Institutional Holdings (13f) databases for control variables. Our main sample
consists of 71,909 firm-year observations.
3.2 Basic research design
To test our hypothesis on how guidance is related to stock price crash risk, we follow
Hutton et al. (2009) in adopting the following basic research design for our regression analyses:
where Guide is a dummy variable indicating whether the firm issues any guidance during fiscal
year t. A positive (negative) coefficient on Guide indicates that guidance firms have a higher (lower)
probability of a future stock price crash than do non-guidance firms. GuideFreq is the total number
of guidance defined as above. A positive (negative) coefficient on GuideFreq indicates that firms
that issue more guidance have a higher (lower) crash risk. These two variables capture the guidance
incidence and frequency. 13 To control for previously documented influential information
characteristics on crash risk, we include five measures: accrual quality (Opaq), financial statement
comparability (Comparability), accounting conservatism (C_Score), real earnings management
(DRO), and 10-K readability (10KFilesize). Opaq is the three-year moving sum of the absolute
annual discretionary accruals estimated from the modified Jones model (Dechow et al., 1995) from
Hutton et al. (2009). C_Score is the firm-year accounting conservatism measure estimated from
the model by Khan and Watts (2009). Comparability is the average comparability scores calculated
with all other firms in the same industry, downloaded from Rodrigo Verdi’s website. Following
13 In untabulated sensitivity analyses, we find qualitatively similar results using variations of the guidance frequency
variables: i) the natural logarithm of the number of earnings guidance, and ii) the number of days with at least one
forecast over the course of the year.
17
Kim et al. (2016), we convert raw comparability scores into decile ranks standardized between
zero and one. DRO is the real earnings management measure calculated as a moving three-year
sum of the absolute values of abnormal discretionary expenses and production following Francis
et al. (2016) and 10KFilesize is the size of 10-Ks, downloaded from Bill McDonald's website
(Ertugrul et al. 2017).14 Prior studies predict a positive coefficient on Opaq, DRO, and 10KFilesize,
but negative coefficients on C_Score and Comparability.
Table 3 presents the results of the above regression. In Columns (1) and (2), we document
that stock price crash risk is positively associated with the respective issuance and frequency of
guidance after controlling for various economic and institutional characteristics, industry fixed
effects, and year fixed effects. In the remaining columns, we show that these results are robust to
controlling for different information characteristics. In Columns (3) and (4), we introduce the
financial reporting opacity proxy, Opaq, used in Hutton et al. (2009). We show that our results are
robust to controlling for financial reporting opacity. Consistent with Hutton et al. (2009), we find
a positive coefficient on financial reporting opacity. More importantly, we find that the guidance
variables have significantly positive coefficients. In Columns (5) and (6), we add additional
information characteristics as control variables: Comparability, C_Score, DRO, and 10KFilesize.
We note that the number of observations tend to be much smaller in these columns due to the data
requirements. We continue to find a significant positive association between earnings guidance
and stock price crash risk. In Column (7), we report consistent results from estimating the
regression on a reduced sample with at least one guidance issued during the fiscal year.15
14 DRO is equivalent to DRO_1 in Francis et al. (2016). The results are largely similar if we use the real earnings
management measures from Francis et al. (2016) or those from Khurana et al. (2018). Our results are also
unchanged if we use the 10-K readability measures from Loughran and McDonald (2011). 15 Our results with regard to the information-related variables are not directly comparable to those in prior studies,
for two reasons. First, there are differences in our sample due to the addition of other variables, each with its own
data requirements. Our sample becomes significantly smaller once these variables are added. Second, there is
evidence that the results can vary over time, e.g., before and after SOX (Hutton et al., 2009; Francis et al., 2016).
18
The above results suggest that on average, guidance is related to inflated expectations and
the unravelling of these expectations in the future leads to a stock price crash. This finding is
interesting and possibly counter-intuitive for two reasons. First, prior literature often regards more
disclosure as a reflection of greater transparency, which is expected to preempt a future stock price
crash. Second, the descriptive evidence above on guidance bias indicates that optimistic guidance
is a non-pervasive phenomenon and less frequent than pessimistic guidance. If pessimistic and
optimistic guidance have comparable effects on future crashes, we should observe an average
negative relation between guidance frequency and crash risk. The surprising average positive
effect prompts us to dig deeper into what guidance properties lead to a higher crash risk.
In the next section, we provide sharper analyses to examine the effect of guidance bias on
stock price crash risk. Intuitively, among different types of forecast characteristics, the one that is
most likely to lead to inflated expectations is guidance optimism. Such a focus on optimistic bias
in guidance is also consistent with the notion commonly expressed in the stock price crash risk
literature that when investors are unaware that the firm’s true state is worse than projected, there
will be a higher likelihood of a stock price crash. As noted earlier, regardless of whether
optimistically biased forecasts are the result of managers’ intention to mislead investors or of their
unintentional optimism, the end result is inflated expectations on the part of market participants.
5. Analysis on forecast bias and crash risk
5.1 Guidance optimism and crash risk
In the last section, we show that guidance issuance is positively related to crash risk, but
we are still unclear about the mechanism that underlies this relation. In this section, we closely
examine the relation by investigating the role played by management forecast bias. More
specifically, we examine the inflated expectation hypothesis, which predicts that firms issue
19
guidance that inflates investors’ expectations as a way of camouflaging bad news; a stock price
crash occurs in future periods when investors realize their expectations are inflated.
We first present univariate analyses in Panel A of Table 4 where we divide 71,909 sample
firm-years into three groups based on the frequency of optimistic guidance: no optimistic guidance
(GuideOpt=0), low frequency (1<=GuideOptFreq<4), and high frequency (GuideOptFreq>=4).
For each group, we calculate the percentage of firm-years with stock price crashes in the following
year. We observe that the percentage increases monotonically from the no optimistic guidance
group (20.64%) to the high frequency group (28.47%) and that the differences are highly
significant. We document a similar pattern for subsamples divided by the sign of the bias of the
last guidance of the year. This univariate evidence is in line with the inflated expectation
where ri,w is the current weekly return for firm i; rmkt,w (rmkt,w-1, rmkt,w+1) is the weekly
market return in the current (prior, next) week; and rind,w (rind,w-1, rind,w+1) is the
weekly industry return in the current (prior, next) week.21 We then compute Wi,w,
the natural logarithm of one plus the residual return, εi,w, from the regression;
Crash is a dummy variable equaling one if, in year t+1, there is at least one
extremely low Wi,w, which is defined as a Wi,w smaller than [Mean (Wi,w) - 3.09 ×
Std Dev (Wi,w)] and zero otherwise.
Disclosure variables (t)
Guide Indicator variable equaling one if a firm issued at least one annual earnings
forecast in fiscal year t, for earnings announced after year t (long-horizon
guidance).
GuideFreq The number of annual earnings forecasts issued over the course of fiscal year t, for
earnings announced after year t (long-horizon guidance). All the guidance
variables defined below are based on the annual long-horizon earnings guidance.
GuideOpt Indicator variable equaling one if a firm issued at least one optimistic forecast in
fiscal year t. A forecast is defined as optimistic if the actual earnings fall short of
the forecast value or the minimum of the forecast range.
GuideOptFreq The number of optimistic earnings forecasts issued in fiscal year t.
GuideOptLast Indicator variable equaling one if the last guidance issued in fiscal year t is
optimistic.
GuidePes Indicator variable equaling one if a firm issued at least one pessimistic forecast in
fiscal year t. A forecast is defined as pessimistic if the actual earnings exceed the
forecast value or the maximum of the forecast range.
GuidePesFreq The number of pessimistic earnings forecasts issued in fiscal year t.
GuidePesLast Indicator variable equaling one if the last guidance issued in fiscal year t is
pessimistic.
JointOpt Indicator variable equaling one if a firm issued at least one optimistic forecast in
fiscal year t that is followed by optimistic analyst forecasts.
JointOptFreq The number of optimistic earnings forecasts issued in fiscal year t, followed by
optimistic analyst forecasts.
JointOptLast Indicator variable equaling one if the last guidance issued in fiscal year t is
optimistic and followed by optimistic analyst forecasts.
RGuideOpt Residual value from a 10-year rolling autoregressive model of GuideOpt.
RGuideOptFreq Residual value from a 10-year rolling autoregressive model of GuideOptFreq.
RGuideOptLast Residual value from a 10-year rolling autoregressive model of GuideOptLast.
PGuideOpt Predicted value from a 10-year rolling autoregressive model of GuideOpt.
PGuideOptFreq Predicted value from a 10-year rolling autoregressive model of GuideOptFreq.
PGuideOptLast Predicted value from a 10-year rolling autoregressive model of GuideOptLast.
21 The weekly stock (market) returns are computed using the daily stock (value-weighted market) returns from CRSP. The weekly value-weighted
Fama-French industry returns are computed using the daily industry returns available from Kenneth French’s website: http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.
37
GuideUEM
Indicator variable equaling one if a firm issued at least one guidance that is more
optimistic but less accurate than the analyst consensus in fiscal year t (Kim and
Park, 2012). GuideUEMFreq and GuideUEMLast are the frequency and last of the
fiscal year dummy for the variable, respectively.
GuideComm
Indicator variable equaling one if a firm issued at least one guidance that is more
accurate in fiscal year t (Kim and Park, 2012). GuideCommFreq and
GuideCommLast are the frequency and the last of the fiscal year dummy for the
variable, respectively.
GuideDEM
Indicator variable equaling one if a firm issued at least one guidance that is more
pessimistic but less accurate than the analyst consensus in fiscal year t (Kim and
Park, 2012).GuideCommFreq and GuideCommLast are the frequency and the last
of the fiscal year dummy for the variable, respectively.
Control variables (t)
ROE Ratio of net income to the book value of equity at the fiscal year end (Compustat
IB/CEQ).
SIZE Natural logarithm of the market value of equity at the fiscal year end (log of CRSP
Abs(PRC)/SHROUT ).
MB Ratio of the market value of equity (CRSP Abs(PRC)/SHROUT) to the book value
of equity at the fiscal year end (Compustat CEQ * 1000).
Leverage Ratio of total debt to total assets (Compustat (DLTT+DLC)/AT).
MeanRet Average of weekly returns.
StdRet Standard deviation of weekly returns.
Insiderown Shares owned by insiders as a percentage of the total shares outstanding.
Nanalyst Number of analysts issuing one-year-ahead EPS forecasts at the end of fiscal year
t.
Instown % of Institutional ownership.
Nseg Number of business segments in which the firm operates.
Other variables (t)
Sentiment Fiscal year average of monthly investor sentiment (Baker and Wurgler, 2006,
2007), downloaded from Jeffery Wurgler's webpage
(http://people.stern.nyu.edu/jwurgler/).
Overconfidence Indicator variable equaling one if a CEO has served for at least two years and the
average moneyness of the vested options held by the CEO is over 67%, at least
twice in the sample (Campbell et al., 2011), zero otherwise. The moneyness of
options is the ratio of the end of the fiscal year stock price to the average exercise
price, minus one.
Overinvestment Abnormal investment calculated as the residual from the model of total investment
on lagged sales growth, adjusted with the abnormal industry average investment
(Richardson, 2006; Biddle et al., 2009).
Pilot Indicator variable equaling one for the short sale restriction pilot test firms and
zero for the other firms in the Russell 3000 index during the pilot test period.
RegFD Indicator variable equaling one if fiscal year t is after the effective date of
Regulation Fair Disclosure (i.e., October 2000), zero otherwise.
LitRisk_High Indicator variable equaling one for a higher-than-fiscal-year median of the
predicted litigation risk estimated from a probit model of the likelihood of being
sued (Kim and Skinner, 2012), zero otherwise.
Opaq Moving three-year average of the absolute values of abnormal discretionary
accruals measured from the modified Jones model (Hutton et al., 2009).
38
Comparability The average comparability scores with the four firms that are most comparable
(m4_acctcomp) with firm i, downloaded from Rodrigo Verdi’s webpage
(http://www.mit.edu/~rverdi/). Following Kim et al. (2016), we convert raw scores
into decile ranks standardized between zero and one.
C_Score The firm-year conservatism measure estimated from the model by Khan and Watts
(2009).
DRO Deviation in real operations calculated as a moving three-year sum of the absolute
values of abnormal discretionary expenses and production following Francis et al.
(2016).
10Kfilesize Log of the gross 10-K file size in megabytes (Loughran and McDonald, 2011),
downloaded from Bill McDonald's webpage (https://sraf.nd.edu/data/).
Event-study variables (quarterly)
CAR Cumulative abnormal return over a five day window surrounding a management
forecast.
Last4Q Indicator variable equaling one if the quarter to which a guidance issuance date
belongs is one of the last four fiscal quarters prior to the fiscal year with stock
price crashes, and zero otherwise.
GuideBias Bias in management forecasts, measured as management forecast minus actual
earnings, scaled by the pre-window stock price on trading day -3.
TrueNews True news in management forecasts, measured as actual earnings minus the pre-
guidance consensus analyst estimates, scaled by the pre-window stock price on
trading day -3.
ConcEANews Quarterly earnings news announced on the same day as a bundled earnings
guidance (Rogers and Van Buskirk, 2013), measured as actual quarterly earnings
minus the most recent analyst consensus (or quarterly guidance if there is no
analyst consensus, or (q-4) earnings if there is neither analyst consensus nor
guidance), scaled by the pre-window stock price on trading day -3.
39
References
Aboody, D., Kasznik, R., 2000. CEO stock option awards and the timing of corporate voluntary
disclosures. Journal of Accounting and Economics 29, 73-100.
Wooldridge, J., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press,
Cambridge, MA.
46
Figure 1 Management guidance bias distribution This figure plots frequencies of annual long-horizon guidance in each bias interval with width of 0.0005 between the
10% and 90% percentile bias range (-0.0113, 0.0277). Bias is defined as (forecast EPS-Actual EPS) deflated by the
fiscal year end stock price. Zero on the horizontal axis is marked by a dotted line.
0
500
1000
1500
2000
2500
3000
3500
4000
4500
-0.0105 -0.005 0 0.0055 0.011 0.0165 0.022
Fre
qu
en
cy
Guidance Bias
47
Figure 2 The relation between stock price risk (t+1) and guidance optimism (t) by year This graph depicts the coefficient estimates and the associated 95% confidence intervals, from a probit regression of
crash risk on guidance optimism estimated by year.
Panel A: Coefficient estimates from Table 4, Panel B, Column (1) and their confidence intervals.
Panel B: Coefficient estimates from Table 4, Panel B, Column (2) and their confidence intervals.
Panel C: Coefficient estimates from Table 4, Panel B, Column (3) and their confidence intervals.
This table presents the distribution of firms across the years for the sample period from 1997 to 2012.
Year
Number of
firms
Number of
firms with at
least one
guidance
Percentage of
firms with at
least one
guidance
Number of firms
with at least one
optimistic
guidance
Percentage of firms
with at least one
optimistic guidance
1997 5,716 365 6.39% 171 2.99%
1998 5,375 635 11.81% 308 5.73%
1999 5,174 715 13.82% 327 6.32%
2000 4,946 775 15.67% 422 8.53%
2001 4,293 1,142 26.60% 767 17.87%
2002 3,873 1,145 29.56% 656 16.94%
2003 3,721 1,176 31.60% 679 18.25%
2004 3,766 1,288 34.20% 737 19.57%
2005 3,674 1,117 30.40% 678 18.45%
2006 3,637 1,133 31.15% 628 17.27%
2007 3,588 1,062 29.60% 591 16.47%
2008 3,319 949 28.59% 651 19.61%
2009 3,057 750 24.53% 346 11.32%
2010 3,050 771 25.28% 288 9.44%
2011 2,962 790 26.67% 359 12.12%
2012 2,855 813 28.48% 429 15.03%
2013 2,908 802 27.58% 438 15.06%
2014 3,036 807 26.58% 408 13.44%
2015 2,959 740 25.01% 356 12.03%
Total 71,909 16,975 9,239
49
Table 2 Descriptive statistics Panel A: Descriptive statistics This table presents the descriptive statistics of the variables in the main analysis. Variable definitions are in Appendix
1.
Main variables N Mean Std Dev Minimum 25th Pctl Median 75th Pctl Maximum
Table 4 (continued) Panel C: Probit regression by year This table presents the relation between a stock price crash and guidance optimism from the probit model in Panel A
estimated each year. Variable definitions are in Appendix 1. Industry dummies and control variables are included in
all the regressions but their coefficients are not tabulated. The z-statistic for each coefficient is provided in parentheses
below. Significance levels are based on two-tailed tests. ***, **, and * denote significance at the 1%, 5%, and 10%
levels, respectively.
(1) (2) (3)
GuideOpt GuideOptFreq GuideOptLast
1997 0.073 0.073 0.049
(0.65) (0.65) (0.41)
1998 0.087 0.093 0.114
(0.97) (1.39) (1.19)
1999 0.166** 0.060 0.240***
(2.03) (1.26) (2.69)
2000 -0.029 -0.002 -0.084
(-0.37) (-0.05) (-0.99)
2001 0.144** 0.055** 0.197***
(2.46) (2.27) (3.14)
2002 0.135** 0.027 0.196***
(2.07) (1.14) (2.76)
2003 0.164*** 0.075*** 0.144**
(2.65) (3.33) (2.02)
2004 0.214*** 0.056*** 0.262***
(3.63) (3.04) (4.04)
2005 0.114* 0.050*** 0.112
(1.83) (2.69) (1.57)
2006 0.127** 0.049** 0.187**
(1.97) (2.53) (2.53)
2007 0.021 0.004 -0.043
(0.33) (0.21) (-0.59)
2008 0.109 0.025 0.211***
(1.57) (1.41) (2.74)
2009 0.047 0.006 0.028
(0.54) (0.21) (0.24)
2010 0.176** 0.064** 0.392***
(1.99) (2.32) (3.70)
2011 0.203*** 0.044** 0.225**
(2.62) (1.96) (2.39)
2012 0.247*** 0.059*** 0.321***
(3.29) (2.86) (3.52)
2013 0.188** 0.069*** 0.266***
(2.47) (3.31) (2.99)
2014 0.024 0.007 0.008
(0.31) (0.32) (0.09)
2015 0.107 0.030 0.133
(1.31) (1.38) (1.28)
55
Table 4 (continued) Panel D: Tests using alternative regression models
This table presents the regression results of a stock price crash on guidance optimism from the Fama-MacBeth model in Columns (1)-(3), a conditional logit model
in Columns (4)-(6), and the main model run on the reduced sample with at least one guidance in Columns (7)-(9). Variable definitions are in Appendix 1. The t-
statistic for each coefficient is provided in parentheses below. Significance levels are based on two-tailed tests. ***, **, and * denote significance at the 1%, 5%,
and 10% levels, respectively.
Fama-MacBeth regression Conditional logit regression Firm-years with at least one guidance
Table 5 Endogeneity tests Panel A: Controlling for other variables
This table presents the results of estimating a probit model of a stock price crash on guidance optimism after controlling for other possible predictors of a stock
price crash risk. Variable definitions are in Appendix 1. Year and industry dummies are included in all the regressions but their coefficients are not tabulated.
Standard errors are clustered by firm and by year. The z-statistic for each coefficient is provided in parentheses below. Significance levels are based on two-tailed
tests. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
Table 5 (continued) Panel B: Chamberlin Random Effect (CRE) model regression
This table presents coefficients on guidance variables from a CRE probit model. Variable definitions are in Appendix
1. Year dummies are included in all the regressions but their coefficients are not tabulated. Standard errors are
clustered by firm and by year. The z-statistic for each coefficient is provided in parentheses below. Significance levels
are based on two-tailed tests. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
(1) (2) (3)
Intercept -0.315** -0.316** -0.337***
(-2.46) (-2.46) (-2.62)
GuideOpt 0.044**
(2.24) GuideOptFreq 0.014**
(2.16) GuideOptLast 0.086***
(2.91)
ROE 0.018** 0.018** 0.018**
(2.41) (2.41) (2.40)
Size 0.116*** 0.115*** 0.115***
(6.00) (5.99) (6.06)
MB 0.001 0.001 0.001
(1.00) (0.99) (1.00)
Leverage 0.162*** 0.161*** 0.161***
(2.75) (2.76) (2.75)
MeanRet 8.308*** 8.297*** 8.355***
(6.97) (6.95) (7.02)
StdRet -1.193*** -1.192*** -1.216***
(-4.14) (-4.15) (-4.21)
Insiderown 0.225 0.227 0.220
(1.14) (1.15) (1.11)
Nanalyst 0.105*** 0.106*** 0.105***
(6.16) (6.16) (6.10)
Instown 0.063 0.063 0.061
(1.47) (1.48) (1.40)
NSeg -0.001 -0.001 -0.001
(-0.25) (-0.25) (-0.26)
Observations 71,909 71,909 71,909
Pseudo R-squared 0.0262 0.0262 0.0264
58
Table 5 (continued) Panel C: Difference-in-difference test using Regulation SHO
This table presents the relation between a stock price crash and guidance optimism using a probit model for the Russell 3000 firms in our sample during the Rule
202T—Pilot Program period (i.e., July 2004 through Aug 2007). The sample is composed of the first fiscal years affected by Reg SHO, with 2,169 firm-years in
fiscal year 2004 and 132 in fiscal year 2005. Out of these firm-years, 736 are pilot stocks whereas the others are controls. The pre- (post-) period sample is composed
the same firms’ fiscal year 2003 (2007) observations. Variable definitions are in Appendix 1. Industry dummies are included in all the regressions but their
coefficients are not tabulated. Standard errors are clustered by firm and by year. The z-statistic for each coefficient is provided in parentheses below. Significance
levels are based on two-tailed tests. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
Wald test of exogeneity 3.032 3.032 3.429 3.429 2.920 2.920
60
Table 6 Refinement of guidance optimism measures Panel A: Joint optimism of managers and analysts
This table presents the relation between a stock price crash and guidance in conjunction with optimism in subsequent
analyst forecasts. Variable definitions are in Appendix 1. Year and industry dummies are included in all the regressions
but their coefficients are not tabulated. Standard errors are clustered by firm and by year. The z-statistic for each
coefficient is provided in parentheses below. Significance levels are based on two-tailed tests. ***, **, and * denote
significance at the 1%, 5%, and 10% levels, respectively.
(1) (2) (3)
Intercept -0.496*** -0.497*** -0.505***
(-4.34) (-4.34) (-4.46)
JointOpt 0.136***
(7.22) JointOptFreq 0.046***
(5.39) JointOptLast 0.309***
(3.11)
ROE 0.016** 0.016** 0.017**
(2.12) (2.13) (2.18)
Size -0.012 -0.012 -0.011
(-1.30) (-1.30) (-1.20)
MB 0.003*** 0.003*** 0.003***
(2.69) (2.66) (2.63)
Leverage 0.006 0.006 0.010
(0.19) (0.20) (0.31)
MeanRet 3.807*** 3.786*** 3.590***
(4.86) (4.81) (4.64)
StdRet -0.961*** -0.969*** -0.960***
(-3.43) (-3.49) (-3.50)
Insiderown 0.315** 0.319** 0.308**
(2.34) (2.37) (2.31)
Nanalyst 0.093*** 0.094*** 0.097***
(7.35) (7.52) (7.51)
Instown 0.056** 0.059*** 0.065***
(2.55) (2.70) (2.96)
NSeg -0.006*** -0.006*** -0.005***
(-3.65) (-3.63) (-3.43)
Observations 71,909 71,909 71,909
Pseudo R-squared 0.0208 0.0206 0.0204
61
Table 6 (continued) Panel B: Tests using different type of guidance to construct guidance optimism measures
This table presents the result from estimating a probit model of a stock price crash on guidance optimism constructed with different types of guidance. Each
coefficient is from a separate model of crash risk on individual guidance variable and controls. A forecast is defined as bundled if the announcement day falls
within the [-2, +2] window surrounding an earnings announcement date (Rogers and Van Buskirk 2013). A forecast is defined as reaffirming (updating) if the
forecast is not the first of the fiscal year, and the forecasted earnings is within (out of) the +/− 10% range of the previous guidance. A firm is a routine guider for a
fiscal year if it, on any guidance date in the fiscal year, has issued guidance in three out of the four calendar quarters prior to that date (Rogers et al. 2009); it is a
sporadic guider otherwise. Variable definitions are in Appendix 1. Year and industry dummies are included in all the regressions but their coefficients are not
tabulated. Standard errors are clustered by firm and by year. The z-statistic for each coefficient is provided in parentheses below. Significance levels are based on
two-tailed tests. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
Table 7 Unintentional Bias Panel A: Cross-sectional variation in proxies for unintentional bias This table examines the relation between crash risk and guidance optimism, when bias is likely unintentional because
of high litigation risk. Variable definitions are in Appendix 1. Year and industry dummies are included in all the
regressions but their coefficients are not tabulated. Standard errors are clustered by firm and by year. The z-statistic
for each coefficient is provided in parentheses below. Significance levels are based on two-tailed tests. ***, **, and *
denote significance at the 1%, 5%, and 10% levels, respectively.
Table 7 (continued) Panel B: Analyses of predicted and residual guidance optimism
This table examines how predicted and residual optimism in guidance, measured as residuals from a rolling AR(1)
model, are associated with stock price crash risk. Variable definitions are in Appendix 1. Year and industry dummies
are included in all the regressions but their coefficients are not tabulated. Standard errors are clustered by firm and by
year. The z-statistic for each coefficient is provided in parentheses below. Significance levels are based on two-tailed
tests. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
(1) (2) (3)
Intercept -0.400** -0.388** -0.409**
(-2.38) (-2.32) (-2.42)
RGuideOpt 0.118***
(3.94) RGuideOptFreq 0.041***
(4.30) RGuideOptLast 0.191***
(4.50)
PGuideOpt 0.172***
(5.22) PGuideOptFreq 0.036***
(5.64) PGuideOptLast 0.211***
(4.45)
ROE 0.006 0.006 0.006
(0.43) (0.39) (0.43)
Size -0.026** -0.027** -0.025**
(-2.15) (-2.22) (-2.06)
MB -0.000 -0.000 -0.000
(-0.14) (-0.18) (-0.16)
Leverage 0.031 0.032 0.037
(0.61) (0.63) (0.73)
MeanRet 3.909*** 3.935*** 3.920***
(3.11) (3.07) (3.07)
StdRet -0.453 -0.461 -0.485
(-1.28) (-1.31) (-1.40)
Insiderown 0.317* 0.329* 0.309*
(1.80) (1.89) (1.79)
Nanalyst 0.097*** 0.101*** 0.099***
(5.13) (5.48) (5.18)
Instown 0.013 0.021 0.016
(0.39) (0.63) (0.49)
NSeg -0.005*** -0.005** -0.004**
(-2.60) (-2.58) (-2.43)
Observations 24,168 24,168 24,168
Pseudo R-squared 0.0160 0.0159 0.0164
64
Table 8 Supplemental tests Panel A: Market reaction to forecast bias in guidance
This table examines how markets react to the news and bias in long-horizon guidance using the last earnings guidance issued in a fiscal quarter. The first column
presents the results from an OLS regression estimated for the full firm-quarter panel, and the second column presents the results from an OLS regression estimated
for guidance that is not bundled with an earnings announcement. CAR is measured for the five day window surrounding the guidance date. Last4Q is the dummy
variable assigned to the four fiscal quarters prior to a crash year. The concurrent quarterly earnings news (ConcEAnews) is calculated as an earnings surprise for
guidance bundled with announcements of quarter t-1 earnings and is set to zero for non-bundled guidance. More detailed variable definitions are in Appendix 1.
Year and industry dummies are included in all the regressions but their coefficients are not tabulated. Standard errors are clustered by firm and by quarter. The t-
statistic for each coefficient is provided in parentheses below. Significance levels are based on two-tailed tests. ***, **, and * denote significance at the 1%, 5%,