Accounting Information and Left Tail Risk

Review of Quantitative Finance and Accounting (2022) 58:1709–1740
https://doi.org/10.1007/s11156-021-01036-6
ORIGINAL RESEARCH
Accounting information and left‑tail risk
Irfan Safdar1 · Michael Neel2 · Babatunde Odusami1
Accepted: 27 December 2021 / Published online: 9 February 2022

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2022
Abstract
Several recent studies attribute stock price crashes to firms withholding bad news from
financial disclosures before a stock price crash. Contrary to this notion, we find evidence
of a robust link between information in a firm’s financial disclosures and potential left-tail
risk. We document that the sophisticated equity options traders incorporate information
derived from financial statements about left-tail risk into prices of out-of-the-money put
options on a firm’s equity, implying that a firm’s financial disclosures contain significant
information relevant to pricing expected crash risk. However, we find that stock market
investors at large appear to overlook this link and fail to incorporate information in finan-
cial disclosures about left-tail risk into stock prices in a timely fashion, potentially con-
tributing to the severity of the eventual crash. These findings contradict the notion that
managers can fully conceal information pertinent to left-tail risks and highlight the role
of potential errors by investors in processing accounting information pertinent to left-tail
risks. Our study is amongst the first to link financial statement analysis to expected crash
risk.
Keywords Left-tail risk · Accounting information · Financial statement analysis · Market

efficiency · Crash risk
JEL Classification G11 · G12 · G14 · G40
1 Introduction
The prospect of sudden and large movements in equity prices and their implications for
asset pricing and portfolio management has recently risen to the forefront of financial
research. Of major interest to a growing number of researchers is the potential for large
* Irfan Safdar
musafdar@widener.edu
Michael Neel
michael.neel@unt.edu
Babatunde Odusami
boodusami@widener.edu
1
Department of Accounting, Economics, and Finance, Widener University, Chester, USA
2
Department of Accounting, University of North Texas, Denton, USA
13
Vol.:(0123456789)
1710 I. Safdar et al.
downside movements in firms’ stock prices, i.e. crashes, and what causes them. Given the
importance of left-tail risk to investment decisions, a swath of literature has examined the
role of various factors linked to price-crashes. Key findings from recent studies suggest that
the opacity of accounting information (Jin and Myers 2006; Hutton et al. 2009; Kim and
Zhang 2014), corporate tax avoidance (Kim et al 2011), degree of accounting conservatism
(Kim and Zhang 2016), degree of comparability in financial statements (Kim et al. 2016),
and transparency in public disclosures (Huang and Qiao 2017) are associated with the like-
lihood of a stock price crash.
The prevailing theory in the crash-risk literature espouses the notion that firms system-
atically manipulate financial disclosures to withhold large quantities of bad news (Habib
et al 2018). The theory posits that firms have incentives to introduce opacity, lack of trans-
parency, or other contaminants and omissions into their disclosures to conceal bad news
from investors while releasing good news on a timelier basis.1 After a significant amount of
bad news accumulates and can no longer be withheld, it is revealed to unsuspecting inves-
tors at once, causing a price crash.
We argue that to better understand what leads to stock price crashes, it is important to
understand not only how bad news is revealed to the market but also how investors process
this information. The aforementioned theory has underpinnings that attribute price crashes
to firms withholding negative information from financial disclosures prior to a crash. It dis-
counts the possibility that managers’ discretion to withhold large quantities of information
over an extended period is limited by GAAP or that incentives to withhold significant news
are likely to be small if the market has access to multiple sources of information. By exten-
sion, the theory does not allow room for an alternative scenario where delayed reaction
by investors to information pertinent to left-tail risk is a potential factor in price crashes.
As a consequence, despite significant accumulating evidence from capital markets research
that investors make information-processing errors, the role of investor errors in processing
information about left-tail risk has not been investigated in the crash-risk literature.2
In this study, we take a closer look at the link between a firm’s financial disclosures
and left-tail risk and develop evidence on whether investors evaluate this link correctly. If
firms are successful in concealing extreme bad news before a price crash, we should not
find a significant link between data in financial disclosures and expected crash-risk. Under
that scenario, we should not expect investor errors to play a role in price crashes. On the
other hand, if such a link is found, it undermines the notion that firms successfully conceal
significant bad news from investors. Furthermore, it raises the prospect of investor errors
in processing information in accounting data pertinent to left-tail risk. We believe that our
study is amongst the first to highlight the potential role of investor errors in studying stock
price crashes.
1
Support for this idea is drawn from studies that suggest that managers are incentivized to release news in
an asymmetric fashion. For example, Kothari et al. (2009) report evidence that managers delay the release
of bad news while readily leaking good news. Baginski et al. (2018) and Jiang et al. (2020) link this behav-
ior to managerial incentives to conceal bad news to prevent negative career outcomes. Furthermore, Hamm
et al. (2020) report release of bad news is frequently preceded by optimistic guidance by managers to dis-
guise the delay in revealing bad news.
2
Investors may react to information in a delayed fashion due to a host of cognitive biases documented in
earlier studies. Some examples include overconfidence (Daniel et al. 1998), conservatism bias (Barber et al.
1998), representativeness bias (Kahneman and Tversky 1982), and limited attention (Daniel et al 2002). To
our knowledge there is very limited work on whether such cognitive errors play a role in investors underes-
timating tail-risk.
13
Accounting information and left‑tail risk 1711
We take a two-pronged approach to examine evidence on the link between potentially

relevant information in financial disclosures and crash risk. We first test for and document a
robust link between accounting data from financial statements and expectations of sophis-
ticated equity options traders regarding left-tail probabilities embedded in option prices,
or ‘expected crash-risk’. Given that put options reflect insurance against downward price-
movements, out-of-the-money put options become relatively more valuable compared to at-
the-money call options when there is a greater risk of a large stock price drop.3 Operation-
ally, a greater expectation of a large price drop is inferred when the Black–Scholes implied
volatility derived from the price of an out-of-the-money put option becomes greater rela-
tive to the implied volatility derived from the price of an at-the-money call option on the
same stock. Based on this logic, Kim and Zhang (2014) label the difference in the volatility
implied from put option prices and volatility implied from call option prices on the same
stock (or volatility skewness) as a measure of ‘expected crash risk’; i.e., a higher difference
implies a greater expectation that the price of the underlying stock will crash.
We document a significant positive link (p-value < 0.01) between a comprehensive
measure of weakness in accounting fundamentals derived from financial statements (dis-
cussed below) and expected crash risk as described above. Given that options traders are
typically regarded as a class of sophisticated information-based traders, this evidence
refutes the notion that firms’ financial disclosures are devoid of information required to
assess left-tail risk. On the contrary, the activity of equity options traders implies that a
firm’s financial disclosures contain significant information pertinent to pricing expected
crash risk. Our findings are robust to controlling for a broad set of additional factors. We
believe that these findings motivate a re-examination of the mechanism driving stock price
crashes.
Given the link documented between accounting data and expected crash risk embedded
in option prices, we examine whether stock market investors at large recognize this link
and incorporate it correctly into stock prices. The logic underlying our tests follows the
standard argument in the efficient-markets literature. In a market where investors correctly
incorporate all available information into stock prices, past information should no longer
exhibit a relation to future returns (risk-adjusted). In other words, if investors correctly
price significant bad news in accounting data, this information should no longer be able to
predict extreme negative returns. On the other hand, if we find a significant link between
future crashes and past accounting data, it implies that stock market investors underweight
important information about left-tail risk found in firms’ financial disclosures.
Since our interest is specifically in left-tail risk, we use generalized logit models to
model the likelihood of an extreme negative return over the next 12-months, conditional on
a comprehensive measure of weakness in accounting fundamentals. We find a strong posi-
tive link (p-value < 0.0001) between weakness in accounting fundamentals and future price
crashes. This link suggests that stock market investors on average appear to underreact to
information in past accounting fundamentals pertinent to left-tail risk. These findings are
robust to alternative specifications of crashes, alternative models of probability estimation,
and several alternative explanations based on prior studies including opacity, comparabil-
ity, earnings management, and profitability. Our tests incorporate standard controls for risk,
3
Intuitively, all else equal, an out-of-the money put option is less valuable than an at-the-money call option
because the left tail of the stock price distribution is limited at $0 while no such limitation exists in the right
tail. As the probability of extreme downside risk increases, the out-of-the money put option becomes rela-
tively more valuable.
13
including market-beta, idiosyncratic volatility, leverage, size, and book-to-market. Our evi-
dence is consistent with the notion that delayed reaction by investors to bad news in finan-
cial disclosures is a potential factor in stock price crashes.4
Before moving to the next section, we briefly discuss our choice of accounting data
used in our tests. We focus on a measure that captures weakness in a broad set of account-
ing fundamentals. The reason is that if a firm experiences sufficient bad news to make a
price crash probable, then deterioration in its economic fundamentals should be reflected
by weakness in a broad and diverse set of accounting fundamentals. We focus on data in
financial statements because GAAP imposes limitations on managers’ discretion in prepar-
ing financial statements reviewed by professional auditors; therefore, while managers may
temporarily hide some bad news from a few salient signals (perhaps signals investors tend
to focus on), they are less likely to be successful in manipulating a comprehensive and
diverse set of accounting fundamentals simultaneously (absent outright fraud). Based on
this reasoning, we transform the well-documented F-score (Piotroski 2000) which captures
information from several categories of accounting fundamentals into a measure of weak-
ness in fundamentals to be used in our regressions.5
To summarize, our study documents that a comprehensive measure of accounting fun-
damentals captures significant information pertinent to left-tail risk embedded in option
prices. Furthermore, stock market investors on average appear to underweight this informa-
tion, potentially creating the impression that a firm’s financial disclosures failed to reveal
bad news to the market prior to a crash. This evidence is important because it suggests that
investor-underreaction to accounting information in assessing left-tail risk is a potential
factor in exacerbating stock price crashes. Collectively, this evidence is also inconsistent
with the notion that firms can systematically conceal large quantities of bad news before a
stock price crash. In essence, our study contributes to a better understanding of the role of
financial disclosures and investor errors in processing these disclosures prior to stock price
crashes.
In the next section, we discuss relevant current literature on crash risk in the context of
our study. Section 3 provides an overview of our sample and variables. Section 4 discusses
our empirical tests and results. Section 5 provides concluding remarks.
2 Related literature
2.1 Fundamentals and stock returns
If markets are informationally efficient, then information relating to a firm’s stock has been
incorporated into its price such that fundamental analysis should not result in superior risk-
adjusted returns. However, the literature is awash with evidence of return predictability
that are often attributed to cognitive errors in investor processing of relevant information.
Since the findings by Fama and French (1992) who show that the book-to-market (BM)
ratio predicts stock returns, there has been increasing interest in whether other firm-specific
fundamental signals can predict equity returns. Accordingly, Piotroski (2000) finds that
4
In our tests, we control for the previously documented relation between the fundamentals and stock
returns to ensure that our findings are incrementally informative about the left-tail of the return distribution.
5
Additional details are provided in Sect. 3.2 of the paper and in the appendix.
13
investors who purchase value (high B/M) stocks can significantly improve their portfolio
performance by buying firms with strong fundamentals.6 Fama and French (2006) provide
corroborating evidence for these findings. In their discussion, Fama and French (2006)
argue that financial fundamentals are able to predict future stock returns because they have
information about expected profitability and future asset growth.
Evidence supporting the relation between accounting fundamentals and the distribution
of future stock returns can also be found in recent literature. Piotroski and So (2012) find a
statistically significant increase in investment performance when value/glamour investment
strategies are intersected with fundamental analysis. Chen et al. (2016) and Ahmed and
Safdar (2018) show similar results for investment performance of momentum-based strate-
gies when overlaid with fundamental analysis. Recent studies also suggest that the ability
of fundamentals to predict future returns generalizes to international markets. For example,
Aspris et al. (2013) show that strategies based on favorable performance and credit signals
persist over time among Australian S&P/ASX 300 firms, while Quality Indices comprised
of fundamental accounting metrics predict abnormal returns in both the U.S. (Gallagher
et al 2014a) and Australia (Gallagher et al. 2014b). To our knowledge, the link between
fundamental analysis and left tail risk (incremental to the previously documented relations)
remains unexplored.
2.2 Accounting fundamentals, opaqueness, and crash risk: reconciling evidences
A growing interest in stock price crashes has generated a line of literature that examines
a broad set of variables linked to crash-risk, including earnings management (Xu et al.
2013), board monitoring (Cai et al. 2019), director liability protections (Choi and Jung
2020), vertical interlock among affiliated firms (Yang et al. 2020a,b), board social capital
(Jebran et al. 2021), disclosure timing (Li et al. 2020), and quality (Du et al. 2016), use of
promotion-based tournament incentives (Sun et al. 2019), quality of internal auditors (Park
and Park 2019), and patent generation (Ben-Nasr et al. 2019). Our study differs from the
above in that we are primarily interested in evaluating whether firms are able to conceal
bad news from their financial disclosures and whether investors make errors in processing
any such information.
In the extant literature, stock price crash risks are most notably linked to the opaque-
ness (overall opaqueness as in Jin and Meyer (2006) and earnings opaqueness, as in Hutton
et al. (2009)) of financial reporting. The general hypothesis in these papers is that manag-
ers have an incentive to hide bad news until all managerial discretion is exhausted, and
then all bad news is released, resulting in a significant drop in the stock price. We believe
these findings are consistent with the notion that fundamental signals can identify investor
errors. In essence, managers may be able to undertake certain actions to make financial
reports opaque and less transparent when bad news arrives but are ultimately unable to
conceal the news from a comprehensive set of accounting signals. Indeed, in the post-SOX
period, managers face significant legal costs for overtly aggressive accrual management
activities and are therefore unlikely to conceal significant material information through it.
Instead, managers can use opportunistic deviations in real operations to mask the under-
performance in real operation earnings. In such instances, a significant share of investors
6
Similarly, Banerjee and Deb (2017) show that the profitability of value strategies relies on the strong per-
formance of a few firms with strong fundamentals that outperform less robust value counterparts.
13
may be unable to distinguish such opportunistic actions from real performances even when
disclosed by managers.7 Thus, sophisticated traders may be able to evaluate the import of
these fundamentals for potential crash risk, while less sophisticated investors in the broader
market may be potentially misled by opaque financial statements into under-reacting to
relevant information.8 Motivated by prior studies, in our robustness tests we also explore
if fundamental analysis dominates opaqueness in accounting reports (Hutton et al. 2009),
accruals (Ohlson and Bilinski 2015), and accounting comparability (Kim et al. 2016) in
predicting stock price crashes.
The findings in the literature at large also suggest that, while accounting fundamentals
may be informative about the left tail of future equity returns, investors may not completely
incorporate this information into the current price of the stock. One potential answer to
why investors are unable to completely discount fundamental signals can be found in Dan-
iel et al. (2002).9 They argue that investors have limited attention and are subject to a sali-
ence effect. Due to limited attention investors are only able to conduct superficial assess-
ments of the firm’s financial condition (i.e. they may focus their attention only on earnings,
rather than its components).10 The salience effect theorized in Daniel et al (2002) suggests
that investors focus disproportionately on noticeable risks and only consider the firm’s dis-
closure as significant when consumed through basic means, such as the financial press.
Evidence in support of this theory can be found in Menon et al. (2010) who identify a
dichotomy in how different investor classes react to disclosure of the going concern audit
report. They find that while in general, stock price reaction to a going concern audit is
negative, the magnitude of the reaction varies with the level of institutional ownership.
They find no detectable reaction at low levels of institutional ownership and a more
negative reaction as the level of institutional-ownership increases. They attribute this to
sophisticated investors’ awareness of the firm’s capital constraints and debt covenants.
This may provide motivations for sophisticated investors to hedge their equity exposure
through option contracts and a linkage to the implied volatility smirk we observe in option
data. Further evidence in support of this result is shown in Atilgan et al. (2019) who find
that left-tail risk (crash risk) is disproportionately larger for stocks that are more likely
to be held by retail investors, that receive less investor attention, and that are costlier to
arbitrage.11
7
Francis et. al. (2016) finds that ability of discretionary accruals predict crash risk has essentially disap-
peared in the post-SOX period. Instead, they find that deviations from real operations to be positively cor-
related with future crash risk.
8
An extensive body of empirical literature that examines how financial markets react to a wide range of
news events has found that on the whole, markets appear to initially underreact to both positive and nega-
tive news. A few of these include the works of: Hong and Stein (1999), Hong et al. (2002). Ikenberry et al
(1995), and, Ikenberry and Ramnath (2002).
9
An extensive review of literature that has examined investor psychology in capital market can be found in
Daniel et al. (2002)
10
For example, Hodgson and Stevenson-Clarke (2000) examine interplay of multiple fundamental signals
and report that higher leverage is associated with earnings that are less informative for valuation. Thus, a
myopic focus on a single fundamental signal (i.e., high earnings) while disregarding an alternative funda-
mental signal (i.e. high leverage) can lead to valuation errors.
11
Some studies examine investor characteristics related to crash risk. For example, foreign ownership (Vo
2018), low liquidity (An et al. 2018), and greater margin trading volatility (Lv and Wu 2019) predict greater
crash risk. Additionally, Bai et al. (2020) report that superstition affects crash risk in the Chinese market
due to investor overreaction to negative news when a firm has an “unlucky” listing number. Moreover, stock
forum-induced panic predicts stock price crashes on the Chinese Growth Enterprise Market (Yang et al.
2020a, b).
13
3 Sample, descriptive statistics, and measurement of accounting

fundamentals
In this section, we discuss our sample and associated descriptive statistics followed
by measurement of accounting fundamentals. For context, we also briefly discuss the
empirical frequency of negative stock returns in the US stock market.
3.1 Sample and descriptive statistics
Our empirical tests require data from the US markets for stock options and stock returns,
as well as data from financial statements to calculate our measure of accounting funda-
mentals. We obtain option price data and associated variables from OptionMetrics. We
draw our overall sample of firms based on the intersection of CRSP and S&P’s Capital
IQ databases. We identify all firms over the period 1991 to 2018 that have stock return
data and accounting variables available to calculate each company’s F-score described
in the appendix. To ensure that our results are not driven by extremely small firms, we
drop firms with stock prices below $5/share as of the fiscal year-end. This produces a
sample of 74,395 firm years over the 1991 to 2018 period. Beyond this, the size of the
sample varies depending upon our tests and control variables. To prepare stock return
data for analysis, we compute annual stock returns using 12 monthly returns starting
four months after each fiscal year-end for each firm. If a firm delists, we use its delist-
ing return as this last monthly return and calculate the annual return using the avail-
able monthly returns. This provides a series of annual stock returns for each firm. The
three-month delay after fiscal year-end is to ensure that financial statement information
is available to the market and available for use by investors.
Panel A of Table 1 provides descriptive statistics for our overall sample. The average
firm in the sample has a market cap of $5.56 billion with a median of $784 million. The
mean book-to-market value of equity ratio is 0.49. The average annual stock return in
our sample is 14.56%. Panel A also provides statistics on each of the variables used to
calculate our measure of fundamentals, the F-score (calculation described in the appen-
dix). The F-score of a firm can range from a minimum of 0 to a maximum of 9, with
a higher F-score indicating stronger fundamentals; the average F-score in our overall
sample is 5.06.
We report descriptive statistics for the sub-sample of 24,797 firm years used in the
tests of expected crash risk in panel B of Table 1. The median F-score of 5 indicates
a sample of firms with an average level of fundamentals, comparable to our overall
sample. The other statistics for the sample are generally comparable to those docu-
mented in prior studies. In panel C of Table 1, we report the mean and median values of
both measures of expected crash risk, conditioned on categories of firm fundamentals
(defined as Weak (F-scores 0–3), Neutral (F-scores 4–6), and Strong (F-scores 7–9)).
Expected crash risk over both the three-month (IV_Skew3) and 12-month (IV_Skew12)
time horizons increases monotonically as fundamentals become weaker. For example,
mean IV_Skew3 increases by about 30% (p-value < 0.01) when moving from the Strong
fundamentals (0.070) to Weak fundamentals category (0.090). We find a similar effect
for mean IV_Skew12, as well as for median values of both measures. This provides an
initial descriptive indication of the relation we test for and observe in our formal tests.
13
Table 1 Descriptive statistics
Variable Mean Median Std. Dev
Panel A: Fundamental Variables – Full Sample

MVE ($mil.) $4,814.7 $607.4 $18,506.6
B/M 0.50 0.43 0.57
ΔMargin 0.51% 0.09% 23.3%
ΔROA 0.03% 0.06% 10.5%
ΔTurn − 0.012 0.002 0.215
ΔLiquid − 0.09 -0.01 1.52
ΔLever 0.42% 0.00% 8.2%
ROA 2.57% 4.43% 12.9%
CFO 7.91% 8.78% 11.6%
Accruals − 2.29% − 3.79% 14.6%
Issuance ($mil.) $44.8 $2.36 $256.7
F-score 5.14 5.00 1.64
Log(MV) 6.52 6.41 1.90
Log(Btom) − 0.90 − 0.81 0.76
Log(Lev) 0.16 0.14 0.15
Log(Vol) − 2.16 − 2.17 0.49
Opacity 0.26 0.17 0.27
Ann. Rett+1 12.66% 6.23% 63%
Adj. Ann. Rett+1 0.00% − 5.93% 59%
Total Available Firm Years 74,012
Panel B: Descriptive Statistics - Expected Crash Risk Sample
IV_Skew3 0.078 0.051 0.096
IV_Skew12 0.081 0.056 0.083
F-Score 5.212 5.000 1.533
ATM_IV 0.435 0.402 0.169
Ln(MktVal) 7.732 7.549 1.548
Ln(Leverage) 0.155 0.144 0.137
Ln(M/B) 1.059 0.996 0.746
Cashflow_Vol 0.074 0.048 0.085
Earnings_Vol 0.078 0.042 0.111
Sales_Vol 0.267 0.164 0.309
Stock_Turn 0.239 0.188 0.180
Beta 1.339 1.212 0.812
Idosy_Vol 0.053 0.047 0.027
Total_Vol 0.060 0.053 0.028
Neg_Skew 0.014 0.006 0.727
Ret 0.127 0.083 0.488
HHI 0.162 0.104 0.157
Strategy 0.184 0.180 0.034
Total Firm Years 24,797
13
Table 1 (continued)
IV_Skew3 IV_Skew12
Panel C: Univariate Tests of Expected Crash Risk Conditional on Fundamentals

Means
Weak 0.090 0.094
Neutral 0.078 0.081
Strong 0.070 0.072
Strong—Weak − 0.020*** − 0.022***
Medians
Weak 0.059 0.066
Neutral 0.051 0.056
Strong 0.046 0.051
Strong—Weak − 0.013*** − 0.015***
Panel A provides descriptive statistics for the overall sample of firms used in this study. We first draw
our sample of firms from the intersection of CRSP and S&P’s Capital IQ databases. We identify all firms
over the period 1991 to 2018 that have stock return data and accounting variables available to calculate
each company’s fundamentals as described in the appendix. To ensure that our results are not driven by
extremely small firms, we drop firms with stock prices below $5/share as of the fiscal year-end. Panel
B reports descriptive statistics for all variables used in our primary tests of expected crash risk. Panel C
reports the mean and median values for our two measures of expected crash risk (IV_Skew) partitioned
on fundamentals. IV_Skew3 (IV_Skew12) is measured over the three months (12 months) starting in the
fourth month after the fiscal year-end. Following past studies (Piotroski and So 2012), we group firms with
fiscal year-end F-scores of 0–3 as Weak, F-scores of 4–6 as Neutral, and F-scores of 7–9 as Strong. Vari-
able definitions are provided in the appendix.
3.2 Measurement of accounting fundamentals
Our measure of accounting fundamentals is based on the F-score proposed by Piotroski

(2000). While the literature on fundamentals has broadened over the past two decades,
the F-score has appeared in several prominent studies on financial statement analysis,
it is straightforward to calculate, and does not require a large sample of firms to estab-
lish relative rankings. It is a comprehensive summary measure calculated based on
nine binary signals derived from a firm’s financial statements. Broadly, these signals
are drawn from data in several categories of fundamentals such as operating efficiency,
performance, liquidity, and leverage. Each fiscal year, for each signal, Piotroski (2000)
assigns a value of 1 to a firm if the signal indicates strong fundamentals and 0 other-
wise. The sum of the nine binary signals forms a firm’s F-score, ranging from 0 (weak-
est fundamentals) to a maximum of 9 (strongest fundamentals). Details for calculating
the F-score are provided in the appendix. To adapt the F-score into a variable that can
be deployed in a regression with an intuitive interpretation, we transform it into a vari-
able labeled
( )
WeakFundi,t = 9 − F − scorei,t ∕9 (1)
WeakFundi,t ranges from a minimum of 0 to a maximum of 1, with larger values

implying greater weakness in accounting fundamentals. We calculate this measure for
each fiscal year of a firm in which data is available to calculate the F-score.
13
Fig. 1 Distribution of negative one-year stock returns
Table 2 Frequency of Negative One-Year Stock Returns (Pooled Sample: 1991–2018)

Ret < -90% < -80% < -70% < -60% < -50% < -25% < 0% All Firm Years
Panel A – Frequency of One-Year Negative Returns

Number of Firm Years
444 1234 2503 4300 6690 16,075 31,950 74,395
Frequency of Firm Years
0.6% 1.7% 3.4% 5.8% 9.0% 21.6% 42.9% 100.0%
Panel B – Conditional Distribution of Negative Returns Based on Accounting Fundamentals
Fundamentals Ret < -90% < -80% < -70% < -60% < -50% < -25% < 0%

Weak 1.6% 3.9% 7.7% 12.1% 17.4% 33.3% 52.2%
Neutral 0.5% 1.5% 2.9% 5.3% 8.3% 20.7% 42.3%
Strong 0.2% 0.6% 1.3% 2.6% 4.7% 15.9% 38.5%
This table provides the frequency distribution of negative one-year stock returns computed from the pooled
sample of firm-year observations. Panel A provides empirical frequencies without conditioning on funda-
mentals. Panel B provides empirical frequencies within categories based on grouping firms with similar
fundamentals. Following past studies (Piotroski and So 2012), we group firms with fiscal year-end F-scores
of 0–3 as Weak, F-scores of 4–6 as Neutral, and F-scores of 7–9 as Strong
3.3 A look at the empirical frequency of negative stock returns in the US market
As a matter of general interest, we present a look at the frequency distribution of stock

return for the pooled sample of firm-years over the 1991–2018 period in a non-parametric
framework. These frequencies are plotted graphically in Fig. 1 and documented in Table 2.
Panel A of Table 2 identifies the empirical frequency of one-year negative stock returns in
the pooled sample. For example, 26,094 firm years exhibit a negative one-year return out
of the 74,395 total firm years, for an empirical frequency of 43%; i.e., in 43% of firm-years,
the stock return was negative. Panel A also provides the empirical frequency for increas-
ingly negative cutoffs; in 21% of the firm-years, the one-year stock return was worse than
13
Fig. 2 Conditional distribution of negative one-year stock returns conditional on fundamentals (Sample:

1991–2018)
-25%, while in 9% of the firm-years, the one-year stock return was worse than -50% (i.e.,
a greater than 50% drop in the market value of equity). A notable observation from this
figure is that despite large volatilities observed in the prices of individual stocks in the US
market, extreme negative returns are infrequent. For example, less than 4% of firms experi-
ence a negative return worse than -70%, and less than 1% experience a return worse than
-90%. Overall, if we consider a drop of 50% or worse in the stock price over 12 months as a
crash, then about 10% of firms experience a stock price crash on average.
3.3.1 The frequency of negative stock returns conditional on fundamentals
In panel B of Table 2, we provide a simple depiction of the conditional empirical frequency

of negative returns. To simplify this analysis, we group firms with similar fundamentals
into categories of Weak, Medium, and Strong fundamentals using the F-score. Following
past studies (Piotroski and So 2012), we group firms with fiscal year-end F-scores of 0–3
as Weak, F-scores of 4–6 as Neutral, and F-scores of 7–9 as Strong. In our sample, approx-
imately 17% of firms fall into the Weak category, 62% fall into the Neutral category, and
20% fall into the Strong category which is similar to earlier studies. We measure the stock
returns of firms over the 12 months subsequent to the three months that follow the end of
the fiscal year from which the F-score is calculated (as shown in the diagram in Sect. 4.2).
Panel B of Table 2 evaluates the empirical frequencies of the one-year negative stock
returns within fundamentals-based categories. For ease of exposition, we plot these fre-
quencies in Fig. 2 which shows that 33% of firms in the Weak-fundamentals category expe-
rience a market value drop of −25% or worse while only 16% of firms in the Strong-funda-
mentals category experience such a drop. The difference becomes starker when considering
a market value drop of 50% or greater; Weak firms (17.4%) are almost 4 times more likely
than Strong firms (4.6%) to experience such a large drop in market value. The contrast
becomes even greater for very large negative returns; Weak firms experience a one-year
drop of 70% or greater with a frequency of 7.7% compared to only 1.3% for Strong firms.
The analysis of empirical frequencies in Table 2, graphically depicted in Figs. 1
and 2, leads to a basic conclusion – the likelihood of a firm’s stock price crashing (i.e.,
13
experiencing a very large negative return) appears to be much larger for firms with Weak
fundamentals as opposed to Strong fundamentals. Notably, this frequency differential
becomes larger for more extreme negative return levels.
4 Empirical tests and results
In this section, we describe each of our main empirical tests followed by a discussion of
our findings.
4.1 Tests of accounting fundamentals and expected crash risk
Santa-Clara and Yan (2010) report that sophisticated investors demand a 70% higher risk
premium for expected crash risk than for realized historical crash risk. Our tests described
below focus on the link between accounting information and a well-documented measure
of ‘expected’ crash risk based on the option-implied volatility smirk. This measure reflects
ex-ante crash risk derived from prices of stock options, capturing the market’s expecta-
tions regarding an anticipated stock price drop. The literature has expanded to include
notable studies that rely on the notion of expected crash risk (Kim and Zhang 2014, Kim
et al. 2016, and Kim et al. 2019). Based on these studies and the theoretical work of Bates
(2000) and Pan (2002), we describe below the measurement of expected crash risk, fol-
lowed by a description of our empirical tests using this metric. In particular, we are inter-
ested in learning whether there is a link between our measure of accounting fundamentals
and expectations of crash risk embedded in options prices.
4.1.1 Measurement of expected crash risk
We follow Kim and Zhang (2014) who measure ex-ante stock price crash risk using the vola-
tility smirk that is implicit in the prices of stock options. The volatility smirk arises when
the implied volatility of low strike-price options exceeds the implied volatility of high strike
price options. In the context of crash risk, the volatility smirk will become steeper as out-of-
the-money (OTM) put options become more expensive relative to at-the-money (ATM) call
options. If options traders believe that weak accounting fundamentals increase the probabil-
ity of a large future drop in the stock price, one solution is to purchase OTM put options as
insurance. As a result, OTM put options will become more expensive relative to ATM call
options, leading to a steeper volatility smirk. Under such a scenario, we should find a positive
link between weakness in accounting fundamentals and our measure of expected crash risk.
Using data from OptionMetrics, we measure the implied volatility smirk (IV_Skewit)
of stock i’s option as the difference between the implied volatility of an out of the money
(OTM) put on day t (IVitOTMP) and the implied volatility of an at the money (ATM) call on
the same day t (IVitATMC):
IV_Skewit = IVitOTMP − IVitATMC (2)
Following prior studies, we define option moneyness using the option delta value, with
OTM puts defined as put options with a delta value between -0.375 and -0.125 and ATM calls
defined as call options with a delta value between 0.375 and 0.625. Consistent with Kim and
Zhang (2014) we additionally apply the following restrictions to the options: (1) the implied
13
volatility of the option is not missing and is between 0.03 and 2.00; (2) the open interest of the
option is not missing and is greater than zero; (3) the total volume of options contracts is not
missing; and (4) the best offer price is equal to or greater than the best bid price and the best bid
price is not zero. When there are multiple put or call option contracts for stock i on the same day
t, we calculate IV_Skewit as the weighted average of the implied volatilities for the put or call
options using the option open interest. Following Kim and Zhang (2014), we average the daily
IV_Skew over longer windows. These include either the three-month period or 12-month period
beginning on the fourth month after the fiscal year-end. By beginning the measurement window
on the fourth month following the fiscal year-end we are able to assess investors’ perception of
future crash risk after the annual financial statements become available.
4.1.2 Tests using expected crash risk
To test the association between weak accounting fundamentals and the implied volatil-
ity smirk, we follow Kim and Zhang (2014) and estimate the following pooled regression
model (maintaining their primary specification and control variables for comparison):
IV_Skewi,t = 𝛼̂ + 𝛽̂1 WeakFundi,t + 𝛾̂1 ATM_IVi,t + 𝛾̂2 Ln(Mktval)i,t

( )
M
+ 𝛾̂3 Ln(Leverage)i,t + 𝛾̂4 Ln + 𝛾̂5 Cashflow_Voli,t
B i,t
+ 𝛾̂6 Earnings_Voli,t + 𝛾̂7 Sales_Voli,t + 𝛾̂8 Stock_Turni,t (3)
+ 𝛾̂9 MktBetai,t + 𝛾̂10 Idosy_Voli,t + 𝛾̂11 Total_Voli,t + 𝛾̂12 Neg_Skewi,t
+ 𝛾̂13 Reti,t + 𝛾̂14 HHIi,t + 𝛾̂15 Strategyi,t + 𝜀̂ i,t
We also include year and industry fixed effects as in prior studies. The variable
WeakFundi,t is as defined in Sect. 3.2. We follow Kim and Zhang (2014) and include sev-
eral control variables, including volatility in several financial variables, various measures of
risk, the contemporaneous stock return, and a measure of industry concentration. All vari-
ables are defined in the appendix. In the model above, the dependent variable IV_Skewi,t is a
proxy for expected crash risk. Intuitively, it reflects the degree to which an out-of-the-money
put option is more valuable relative to an at-the-money call. A statistically significant posi-
tive value of the coefficient 𝛽̂1 would indicate that our measure of accounting fundamentals
contains information pertinent to expected crash risk as evaluated by options traders.
4.1.3 Results from tests of expected crash risk – discussion of Tables 3
Our first test investigates the association between firm fundamentals and ex-ante crash
risk, which captures the market’s expectations about potential future stock price crashes
embedded in option prices. We report the results from estimating Eq. (3) in Table 3. Our
first dependent variable is IV_Skew3 measured over three months (t + 4 to t + 6) after fis-
cal year-end t. A significant positive association to weakness in fundamentals would sug-
gest that the pricing of expected crash risk in options reflects this information. We assess
significance based on standard errors two-way clustered by firm and year. The estimated
coefficient on WeakFund is positive (0.012) and significant (p-value = 0.045), indicat-
ing that during the three months immediately following the availability of financial data,
options traders perceive the risk of a future sudden drop in stock price to be higher among
firms with weaker accounting fundamentals. Next, we examine a longer time horizon and
13
Table 3 Accounting fundamentals and expected crash risk

IV_Skew3 (t + 4, t + 6) IV_Skew12 (t + 4, t + 15)
Coefficient T-stat P-value Coefficient T-stat P-value
Intercept 0.139 (8.799) 0.000 0.134 (7.713) 0.000

WeakFund 0.012 (2.133) 0.045 0.013 (2.875) 0.009
ATM_IV 0.104 (2.905) 0.008 0.094 (2.937) 0.008
Ln(MktVal) − 0.011 (− 6.047) 0.000 − 0.010 (− 5.592) 0.000
Ln(Leverage) 0.008 (1.477) 0.155 0.013 (2.349) 0.029
Ln(M/B) − 0.005 (− 2.898) 0.009 − 0.006 (− 2.900) 0.009
Cashflow_Vol 0.022 (1.891) 0.073 0.014 (1.445) 0.163
Earnings_Vol 0.005 (0.508) 0.617 0.001 (0.067) 0.947
Sales_vol − 0.006 (− 2.126) 0.046 − 0.002 (− 0.810) 0.427
Stock_Turn − 0.055 (− 6.073) 0.000 − 0.041 (− 4.550) 0.000
Beta − 0.001 (− 0.337) 0.739 − 0.001 (− 0.535) 0.598
Idosy_Vol 0.300 (2.123) 0.046 0.455 (5.410) 0.000
Total_Vol −0.319 (− 1.888) 0.073 − 0.452 (− 3.122) 0.005
Neg_Skew −0.000 (− 0.175) 0.863 0.000 (0.339) 0.738
Ret 0.002 (0.745) 0.464 − 0.010 (− 4.795) 0.000
HHI −0.005 (− 1.256) 0.223 − 0.001 (− 0.397) 0.696
Strategy −0.025 (− 1.201) 0.243 − 0.003 (− 0.155) 0.879
F 14.50 14.93
R-squared 19.23% 27.83%
No. of Obs 24,797 24,797
Industry & Year Yes Yes
Fixed Effects
This table provides the results from modeling a proxy of expected crash risk on fundamentals
( )
̂ + 𝛽̂1 WeakFundi,t + ̂
IVSKEW i,t = 𝛼 𝛾2 ATMIV i,t + ̂
𝛾3 Ln(Mktval)i,t + ̂ 𝛾5 Ln MB
𝛾 4 Ln(Leverage)i,t + ̂ +
i,t
𝛾6 CashflowVoli,t
̂
+̂ 𝛾7 EarningsVoli,t + ̂
𝛾8 SalesVoli,t + ̂
𝛾9 StockTurni,t +̂
𝛾 10 Betai,t + ̂𝛾11 IdiosVoli,t + ̂
𝛾12 TotalVoli,t
+̂ 𝛾13 NEGSKEW i,t + ̂𝛾14 Reti,t + ̂
𝛾15 HHI i,t + ̂
𝛾16 Strategyi,t + 𝜀̂i,t
The dependent variable IV_SKEWi,t is a proxy for expected crash risk measured as the average daily
implied volatility skew over either the three months ( IV_SKEW3i,t ) or 12 months ( IV_SKEW12i,t ) begin-
ning in the fourth month following the fiscal year-end
We measure fundamentals based on the F-score devised by Piotroski (2000). The F-score is a summary
measure calculated based on nine binary signals derived from financial statement data. Calculation of
the F-score is described in the appendix. We transform the F-score into a variable labeled WeakFundi,t:
WeakFundi,t = (9—F-scorei,t)/9.
WeakFundi,t ranges from a minimum of 0 to a maximum of 1, with larger values implying weaker fun-
damentals. All other variables are defined in the appendix. Continuous independent variables (except
HHI) are winsorized at the top and bottom 1 percentile. Year and industry (two-digit SIC) fixed effects are
included but not tabulated. We compute standard errors two-way clustered by firm and year. Coefficient
estimates along with associated t-statistics and p-values are provided
13
Table 4 Accounting Fundamentals and Expected Crash Risk – Robustness Tests

IV_Skew (t + 4, t + 6) IV_Skew (t + 4, t + 15)
Coefficient T-stat P-value Coefficient T-stat P-value
Panel A: Opacity
WeakFund 0.011 (2.098) 0.048 0.013 (2.899) 0.009
Opacity − 0.006 (− 2.267) 0.034 − 0.005 (− 1.543) 0.138
Controls Yes Yes
R-squared 19.7% 27.8%
No. of Obs 23,208 23,208
Panel B: Accruals
WeakFund 0.012 (2.136) 0.045 0.013 (2.931) 0.008
Accruals − 0.010 (− 2.179) 0.041 − 0.003 (− 0.643) 0.527
Controls Yes Yes
R-squared 19.8% 27.9%
No. of Obs 24,540 24,540
Panel C: CompAcctInd
WeakFund 0.009 (1.463) 0.158 0.013 (2.610) 0.016
CompAcctInd −0.001 (−1.779) 0.090 −0.001 (−1.017) 0.321
Controls Yes Yes
R-squared 20.3% 28.2%
No. of Obs 16,994 16,994
Panel D: Bid-Ask Spread
WeakFund 0.012 (2.100) 0.048 0.013 (2.807) 0.011
Bid-Ask Spread -0.135 (-0.268) 0.791 -0.137 (-0.319) 0.753
Controls Yes Yes
R-squared 20.1% 28.1%
No. of Obs 24,641 24,641
Panel E: Amihud (2002) Illiquidity
WeakFund 0.012 (2.178) 0.041 0.014 (2.897) 0.009
Illiquidity 0.235 (0.789) 0.439 0.232 (0.794) 0.436
Controls Yes Yes
R-squared 20.1% 28.2%
No. of Obs 24,641 24,641
This table provides results from modeling a proxy of expected crash risk on fundamentals, with the inclu-
sion of additional control variables. See Table 3 for a full model description. In panel A, we control for
the opacity of financial reports as computed in Hutton et al. (2009). In panel B, we control for accruals
as in Ohlson and Bilinski (2015). In panel C we control for financial statement comparability (Kim and
Zhang 2014). In panel D we control for liquidity using the mean daily relative bid-ask spread over the year.
In panel E we control for liquidity using the median daily price impact over the year, where price impact
equals the daily absolute price change in percent divided by US$ trading volume measured in thousands
(Amihud 2002). All variables are defined in the appendix. Year and industry (two-digit SIC) fixed effects,
controls, and intercepts are included but not tabulated. We compute standard errors two-way clustered by
firm and year. Coefficient estimates along with associated t-statistics and p-values are provided.
measure the dependent variable, IV_Skew, over twelve months (t + 4, t + 15). Again, the
coefficient on WeakFund is positive (0.013) and significant (p-value = 0.009), suggest-
ing that the initial perceived crash risk of weak fundamental firms persists, and becomes
stronger, over the 12-month horizon.
13
The evidence from our tests of ex-ante crash risk is consistent with a class of sophisti-
cated investors (option traders) employing information contained in accounting fundamen-
tals to evaluate the potential for stock price crashes before they occur and embedding this
information into option prices. This finding is inconsistent with the notion that firms suc-
cessfully obfuscate information pertinent to left-tail risk found in accounting statements
before a crash. We examine robustness tests next.
4.1.4 Additional sensitivity and robustness tests – discussion of Table 4
In this section, we discuss additional sensitivity tests to ensure that our findings regarding
the relation of accounting fundamentals to expected crash risk are robust to controls for
findings in earlier studies. In particular, we control for transparency in financial reporting
or opacity (Hutton et al. 2009), accruals, financial statement comparability, and liquidity.
4.1.4.1 Controlling for financial reporting opacity Hutton et al. (2009) is a well-cited
study that examined the relation between opacity in financial reporting and realized ex-
post crashes. They found that opacity in financial reports measured as the lagged three-year
moving average of the absolute value of discretionary accruals is related to the likelihood of
stock price crashes. We follow the definition of opacity in Hutton et al. (2009) and introduce
it as an additional control in Eq. (3) to assess whether our results are robust to this control
in our main tests.
We measure opacity for firm i as of fiscal year-end t as in Hutton et al (2009):
( ) ( ) ( )
Opacityi,t = abs DiscAcci,t + abs DiscAcci,t−1 + abs DiscAcci,t−2 (4)
As in Hutton et al. (2009), discretionary accruals ( DiscAcci,t ) are measured using the
modified Jones model (Dechow et al 1995).12
We add the Hutton et al. (2009) measure and estimate Eq. (3). Panel A of Table 4 shows
the results from this estimation. All previous controls are retained but we do not report
estimates for control variables in Table 4 for brevity. Opacity in financial reporting exhibits
a weakly negative association with expected crash risk after controlling for fundamentals.
While this is not the focus of our study, we believe this is an interesting result that is wor-
thy of investigation in a separate study given the findings in Hutton et al. (2009). Turning
to our main variable, the coefficient on WeakFund continues to be significantly positive
for crash risk estimated over the 3-month horizon (p-value 0.048) and 12-month horizon
(p-value = 0.009). Controlling for opacity does not impact our findings.
12
As in Hutton et al. (2009), discretionary accruals ($$D{iscAcc}_{i,t}$$) are measured using the modi-
fied Jones model (Dechow et al, 1995):
TA ΔSales PPE
Assetsi,t = 𝛼0 Assets
1
+ 𝛼1 Assets i,t + 𝛼1 Assets i,t + 𝜀i,t
i,t−1 i,t−1 i,t−1 i,t−1
where TAi,t are the total accruals, ΔSalesi,t is the change in sales, and PPEi,t is the property, plant, and
equipment value for firm i during year t. We measure total accruals as in Sloan (1996): Total Accru-
als = [Change in Current Assets (ACT) – Change in Cash (CH) – Change in Current Liabilities
(LCT) + Change in Debt in Current Liabilities (DLC) – Depreciation (DP)]/Assetsi,t-1. Compustat variables
are in parentheses.
13
4.1.4.2 Controlling for accruals For our next robustness test, we control for accounting
accruals because Ohlson and Bilinski (2015) find a relation between lagged accruals and
extreme future annual stock returns. While the focus of their study was not on forecasting
crash risk, they document that accruals are a strong predictor of extreme negative returns.
Therefore, we add total accruals (calculated as in Sloan, 1996) as a control variable to
Eq. (3). Again, the coefficient on WeakFund continues to be significantly positive over the
3-month horizon (p-value 0.045) and 12-month horizon (p-value = 0.008). Controlling for
accruals does not impact our findings.
4.1.4.3 Controlling for accounting comparability We next control for financial statement

comparability (CompAcctInd), a measure that has been shown to have a strong association
with expected crash risk (Kim et al. 2016).13 Consistent with prior research we find a nega-
tive coefficient on CompAcctInd in panel C of Table 4, indicating less expected crash risk
among firms with greater accounting comparability. However, this effect is only significant
at conventional levels over the shorter three-month horizon. Again, WeakFund continues
to bear a positive association with expected crash risk, however, this effect is only signifi-
cant over the 12-month horizon. (p-value = 0.016). The lack of significance over the shorter
3-month window, in the presence of comparability, might indicate that comparability domi-
nates as a signal immediately following the release of annual financial information, but that
over the long-term accounting fundamentals play a dominant role in sophisticated trader
beliefs. Alternatively, the inclusion of the comparability measure reduces the sample size
by over 30% which might have an impact on the results. Nevertheless, weak fundamentals
continue to be positively associated with expected crash risk based on the typical 12-month
measurement window used in prior studies.
4.1.4.4 Controlling for liquidity We also control for liquidity based on two measures widely
used in the literature. These are the average relative bid-ask spread over the firm’s fiscal year
and the Amihud (2002) measure of illiquidity. Both variables are defined in the Appendix.
We control for the bid-ask spread in panel D and Amihud (2002) illiquidity measure in panel
E. Under both specifications, WeakFund continues to be positively associated with expected
crash risk over both the 3-month and 12-month horizons (p-value < 0.05). Thus, weak fun-
damentals are not simply reflecting low liquidity.
In summary, our primary finding in Table 5 essentially remains unchanged despite these
additional controls.
4.2 Tests of investor’ processing of accounting fundamentals and left‑tail risk
The test described above provides evidence regarding whether financial disclosures of
firms contain valuable information about left-tail risk. However, a limitation of this test
is that it is restricted to firms for which options are traded on their stock. Furthermore,
it is useful to note that options traders are often likely to be sophisticated who may be
more apt to engage in informed trading. The link tested above does not inform us about
how the average stock investor in the market evaluates the same signals in financial disclo-
sures regarding left-tail risk. Therefore, to extend our tests of expected crash risk described
13
We refer the reader to Kim et al. 2016 for a description of how this variable is calculated. We follow
their approach in calculating this variable.
13
Table 5 Generalized Logit Regressions using Maximum-Likelihood Estimation
1726
Panel A: Future One-Year Return < -50% Panel B: Future One-Year Return < -70%
Model 1 Model 2 Model 1 Model 2
13
Prob. Of R
ett+1 < -50% Prob. Of A
djRett+1 < -50% Prob. Of R
djRett+1 < -70%
Estimate z-score P-value Estimate z-score P-value Estimate z-score P-value Estimate z-score P-value
Intercept − 0.20 − 1.01 0.3106 0.14 0.78 0.4354 − 2.14 − 6.42 < .0001 − 0.90 − 3.43 0.0006
Weak Fund 1.879 17.54 < .0001 0.783 7.95 < .0001 2.52 15.35 < .0001 1.41 9.95 < .0001
Ln(Volatility) 1.143 20.16 < .0001 0.516 10.07 < .0001 1.03 12.3 < .0001 0.50 6.9 < .0001
Beta − 0.01 − 0.43 0.668 − 0.01 − 0.62 0.539 0.00 0.05 0.9568 − 0.05 − 1.90 0.057
Ln(MktVal) 0.00 0.01 0.991 0.02 1.45 0.1464 0.03 1.41 0.1601 0.01 0.49 0.626
Ln(B/M) − 0.15 − 5.89 < .0001 − 0.17 − 7.32 < .0001 − 0.17 − 4.73 < .0001 − 0.22 − 6.71 < .0001
Ln(Leverage) 0.98 7.37 < .0001 0.47 3.77 0.0002 1.69 8.49 < .0001 0.53 2.98 0.0029
Log(Ret) 0.295 8.14 < .0001 0.404 11.74 < .0001 0.409 8.17 < .0001 0.562 12.19 < .0001
Bid-Ask Spread − 2.675 − 1.97 0.0494 − 0.273 − 0.23 0.819 − 3.664 − 1.73 0.0839 − 1.593 − 0.92 0.3569
Model Fit Statistics
Wald χ2 10,280.1 7580.9 7025.2 5763.3
p(χ2) < .0001 < .0001 < .0001 < .0001
Pseudo R2 24.80% 18.30% 22.39% 19.38%
No. of Obs 63,338 63,338 63,338 63,338
Fixed Year Effects Yes Yes Yes Yes
Odds Ratio Esti- Estimate 95% Wald Conf. Limits Estimate 95% Wald Conf. Limits Estimate 95% Wald Conf. Limits Estimate 95% Wald Conf. Limits
mates
Low High Low High Low High Low High
Weak Fund 6.54 5.31 8.07 2.19 1.80 2.65 12.46 9.03 17.20 4.09 3.10 5.39
I. Safdar et al.
Panel C: Future One-Year Return < -50% Panel D: Future One-Year Return < -70%
Prob. Of R
djRett+1 < -50% Prob. Of R
djRett+1 < -70%
Intercept − 0.18 − 0.97 0.3303 0.18 1.07 0.285 − 2.13 − 6.60 < .0001 − 0.88 − 3.51 0.0004
Weak Fund 1.861 17.37 < .0001 0.772 7.83 < .0001 2.51 15.25 < .0001 1.40 9.86 < .0001
Ln(Volatility) 1.117 19.57 < .0001 0.504 9.76 < .0001 1.00 11.9 < .0001 0.48 6.6 < .0001
Beta − 0.01 − 0.58 0.562 − 0.02 − 0.76 0.447 0.00 − 0.04 0.9659 − 0.06 − 2.01 0.0444
Ln(MktVal) − 0.01 − 0.52 0.6011 0.01 0.96 0.3379 0.03 1.22 0.2229 0.00 0.21 0.8368
Ln(B/M) − 0.15 − 5.90 < .0001 − 0.17 − 7.29 < .0001 − 0.17 − 4.76 < .0001 − 0.22 − 6.72 < .0001
Ln(Leverage) 0.98 7.35 < .0001 0.48 3.81 0.0001 1.68 8.49 < .0001 0.53 2.99 0.0028
Log(Ret) 0.291 8.14 < .0001 0.406 11.90 < .0001 0.403 8.16 < .0001 0.561 12.27 < .0001
Illiquidity − 0.225 − 4.62 < .0001 − 0.077 − 2.35 0.0187 − 0.304 − 3.27 0.0011 − 0.139 − 2.51 0.0122
Wald χ2 10,276.0 7585.4 7017.6 5764.4
p(χ2) < .0001 < .0001 < .0001 < .0001
Pseudo R2 24.84% 18.32% 22.42% 19.40%
No. of Obs 63,338 63,338 63,338 63,338
Fixed Year Yes Yes Yes Yes
Effects
Odds Ratio Estimate 95% Wald Conf. Limits Estimate 95% Wald Conf. Limits Estimate 95% Wald Conf. Limits Estimate 95% Wald Conf. Limits
Estimates
Weak Fund 6.43 5.21 7.93 2.16 1.78 2.62 12.27 8.89 16.93 4.04 3.06 5.33
1727
13
1728
Panel E – Additional Robustness Tests
Model 1: Prob. Of AdjRet < -50% Model 2: Prob. Of AdjRet < -70%
13
Estimate z-score P-value Estimate z-score P-value
Intercept − 0.07 − 0.53 0.5932 − 1.63 − 7.41 < .0001

WeakFund 0.976 12.24 < .0001 1.602 13.15 < .0001
Opacity 0.307 6.45 < .0001 0.352 5.28 < .0001
Ln(Volatility) 1.361 31.29 < .0001 1.424 21.94 < .0001
Beta - 0.06 − 3.33 0.001 − 0.07 − 2.59 0.010
Ln(MktVal) - 0.07 − 6.65 < .0001 − 0.10 − 6.55 < .0001
Ln(B/M) - 0.12 − 6.11 < .0001 − 0.15 − 5.22 < .0001
Ln(Leverage) 0.96 9.64 < .0001 1.23 8.20 < .0001
Log(Ret) 0.231 8.36 < .0001 0.3262 8.45 < .0001
Model fit statistics
Wald χ2 5009.0 3389.4
p(χ2) < .0001 < .0001
Pseudo R2 18.38% 23.15%
No. of Obs 60,488 60,488
Fixed Year Effects Yes Yes
Odds Ratio Estimates 95% Wald 95% Wald
Conf. Limits Conf. Limits

Estimate Low High Estimate Low High
Weak Fund 2.65 2.27 3.10 4.96 3.91 6.30

I. Safdar et al.
This table shows results from modeling stock price crash risk probabilities using GENERALIZED logit models. To identify crashes, we measure the stock return of each firm
over the 12-month period that follows the release of financial statements, each fiscal year. We allow a three-month delay between the fiscal year-end of a firm and the begin-
ning of the stock return measurement period to ensure that financial statement information is available to the market. If this return for firm i during year t,t + 12 qualifies as a
crash (< −50% in panel A and < −70% in panel B), we code the variable Crashi,t,t+12 as 1 and -1 if the firm’s return is > + 50% (Panel A) or > + 70% (Panel B), and zero other-
wise. Panels C and D introduce an additional measure of liquidity as a control while Panel E adds a control for opacity in financial reporting. In model 1, we use the raw one-
year return to code Crashi,t,t+12 while in model 2 we use an adjusted return, calculated by subtracting from each firm’s raw one-year return the return of a portfolio of firms
with similar fundamentals (Sect. 3.5 provides details). Our measure of fundamentals is based on the F-score devised by Piotroski (2000). The F-score is a summary measure
calculated based on nine binary signals statement data. Calculation of the F-score is described in the appendix. We transform the F-score into a variable
labeled WeakFundi,t WeakFundi,t = 9 − F − scorei,t ∕9
( derived from financial
)
WeakFundi,t ranges from a minimum of 0 to a maximum

Prob. Crashi,t,t+12 = 1 B
( of 1, with )larger values implying weaker fundamentals. To model the probability of a stock’s price crashing, the sim-
ln + 𝛾̂3 Ln(Vol)i,t
( )
ple logit model takes the following form: Prob. Crashi,t,t+12 = −1 M i,t
( ) = 𝛼̂ + 𝛽̂1 WeakFundi,t + 𝛾̂1 Log(Mktval)i,t + 𝛾̂2 Ln
+ 𝛾̂4 MktBetai,t + 𝛾̂5 Ln(Leverage)i,t + 𝛾̂6 Ln(Ret)i,t + 𝜀̂ i,t

All variables are defined in the appendix. We estimate the equation above using maximum likelihood estimation. Coefficient estimates along with associated p-values and
̂
overall model statistics are provided. The odds ratio estimates are calculated as e𝛽 , provided along with 95% confidence intervals
1729
13
above, we also test the link between the likelihood of a future crash and our measure of
weakness in accounting fundamentals. If the average investor evaluates information about
left-tail probabilities contained in financial disclosures accurately, then there should not be
a further positive link between accounting fundamentals and subsequent crashes. In other
words, any information regarding an expected crash should already be reflected in stock
prices and we should not be able to predict a left-tail event using this information.
Below, we first describe how we define a price crash, followed by our probability mod-
els that test the link described above.
4.2.1 Defining price crashes
We follow Jang and Kang (2019) and define a stock-price crash as a large negative return
over the next one-year period. As in their study, we define crashes as a capital loss of at
least 50% over a one-year period (stock return < −50%). However, for robustness, we also
repeat our tests based on a capital loss of at least 70%.
To identify a crash, we examine the stock return of each firm over the 12-month period
following the release of financial statements for each fiscal year. We allow a three-month
delay after the fiscal year-end of a firm before stock return measurement to ensure that
financial statements are available to the market. We use this return for firm i during year
t,t + 12 to identify a crash as described above. The diagram below illustrates the identifica-
tion of a crash in the stock returns following a firm’s fiscal year-end.
Financial statements Measure stock returns over

released. next 12-month period.
Fiscal year-end month t month t+12

(month t-3)
4.2.2 Estimation of generalized logit models to evaluate likelihood of a stock price

crash
We implement generalized logit models to test for a relationship between accounting fun-
damentals and left-tail risks. A logit model allows us to evaluate how variations in account-
ing fundamentals derived from recently released financial statements impact the likelihood
of a future price crash while controlling for multiple variables. A simple logit model takes
the form:
( )
Prob. Crashi,t,t+12 = 1
ln ( ) = 𝛼̂ + Xi,t + Controls + 𝜀̂ i,t (5)
where Xi,t is the conditioning variable (e.g. fundamentals) and 𝛽̂1 is the coefficient of inter-
est, and the variable Crashi,t,t+12 takes on a value of one if the stock return qualifies as a
13
crash and 0 otherwise. While past studies often rely on the simple logit model, a potential
shortcoming is that it divides the sample into only two groups: firms that experience a crash
and firms that do not. Inferences drawn from this model do not take into account whether
the results are driven by firms with greater future volatility. Therefore, in our study, we
focus on a generalized logit model to provide a better control for ex-post volatility.14
A generalized logit model allows us to control for ex-post volatility by defining three
categories of firms: (1) firms that experience a crash (e.g. one-year stock return < −50%,),
(2) firms that experience extreme positive returns of similar magnitude (e.g. one-year stock
return > + 50%), and (3) firms that experience a stock return in between. In this model, the
variable Crashi,t,t+12 takes on a value of −1 to define the category of firms that experience
extreme positive returns of similar magnitude as extreme negative returns. Equation (5) is
generalized to estimate the following logit regression:
( )
ln ( ) = 𝛼̂ + 𝛽̂1 WeakFundi,t + 𝛾̂1 Ln(Mktval)i,t
Prob. Crashi,t,t+12 = −1
( ) (6)
B
+ 𝛾̂2 Ln + 𝛾̂3 Ln(Vol)i,t + 𝛾̂4 MktBetai,t
M i,t
+ 𝛾̂5 Ln(Leverage)i,t + 𝛾̂6 Ln(Ret)i,t + 𝜀̂ i,t
Equation (6) above includes standard risk controls potentially related to crash risks such
as size, volatility, market-risk, and leverage. We add additional controls in the robustness
tests which are presented in subsequent sections of the paper. All variables are defined in
the appendix. Parameters in Eq. (6) are estimated using maximum likelihood estimation.
As is typical of logit regressions, Eq. (6) models the log odds ratio, i.e. the ratio of odds of
a crash to the odds of an extreme right-tail return, as a function of independent variables.
Once the estimates are obtained, the coefficient 𝛽̂1 can be interpreted as the change in log
odds ratio due to weak fundamentals, holding other variables constant. After some basic
algebra and holding control variables constant, we can state that:
( )
Prob. Crashi,t,t+12 = 1|WeakFundi,t
Δln ( ) = 𝛽̂1 ΔWeakFundi,t (7)
Prob. Crashi,t,t+12 = −1|WeakFundi,t
Inserting 1 for weak fundamentals ( Δ WeakFundi,t = 1) and exponentiating both sides to
remove the logarithm, we obtain:
( ) ( )
Prob. Crashi,t,t+12 = 1|WeakFundi,t = 1 𝛽̂
Prob. Crashi,t,t+12 = 1|WeakFundi,t = 0
( ) =e × ( )
Prob. Crashi,t,t+12 = −1|WeakFundi,t = 1 Prob. Crashi,t,t+12 = −1|WeakFundi,t = 0
or more simply,
̂
Odds ratio of Crash |Weak Fund. = e𝛽 × Odds ratio of Crash |Strong Fund. (8)
To test whether accounting information is linked to the likelihood of future stock price
crashes, our hypothesis would be supported if we obtain a statistically significant positive
̂
coefficient on WeakFundi,t; i.e., 𝛽̂ > 0 or equivalently, e𝛽 > 1. Intuitively, this would imply
̂
that the odds of firms with weak fundamentals crashing are e𝛽 times the odds of a firm with
strong fundamentals crashing, holding all other variables constant.
14
We do run tests using simple logit models and find stronger results but do not tabulate them in the paper.
13
4.2.3 Discussion of Table 5 – results from generalized logit models
Equation (6) in the previous section relates the likelihood of a crash to a set of independent
variables, including our primary variable WeakFundi,t. We control for a host of variables
that prior studies have identified as relevant to stock price crashes. They include market
risk, size, and leverage. Recent studies also demonstrate that stock price liquidity is an
important variable that is associated with price crash risk. In particular, Chang et al. (2017)
show that liquidity increases stock price crash risk. They attribute this to the trading of
transient investors who are quick to exit on bad news and can magnify the ensuing price
shock. We control for the liquidity effect using two different variables: the relative bid-ask
spread and the Amihud (2002) measure of illiquidity. In all cases, our primary results are
robust to controls for stock price liquidity.
Results from the estimation of Eq. (6) are reported in panels A-E of Table 5. In panels A and
B, we control for liquidity using the bid-ask spread. In panel A, we define a crash as a negative
return of −50% or worse. We use the pooled sample to estimate two versions of the model: (1)
using the raw future annual return (model 1) to define a crash and (2) using each firm’s future
annual return adjusted by subtracting the return of a portfolio of similar fundamentals (model
2) to define a crash.15 The parameter estimates from the logit model are presented along with
z-scores and p-values.16 The coefficient estimate 𝛽̂1 along with Eq. (8) can be used to calcu-
late the relative odds of a stock price crash in firms with weak fundamentals compared to firms
with strong fundamentals. For example, in model 1 of panel A of Table 5, the coefficient of
1.879 (p-value < 0.0001) on WeakFundi,t implies that holding other variables constant, the odds
of a stock price crash in a firm with weak fundamentals are predicted to be 6.54 times greater
than a firm with strong fundamentals.17 The overall model is statistically significant with a Wald
model-test p-value < 0.0001 and a pseudo-Rsq of 24.80%. As a robustness check, in model 2
of panel A, when we use the firm’s fundamentals-adjusted annual stock return, we find that the
odds of a firm with weak fundamentals crashing are greater by a factor of 2.19; the model is
again statistically significant with a Wald-test p-value < 0.0001.
In panel B of Table 5, we use a cutoff of -70% to define stock price crashes and find
even stronger results. Using raw returns (model 1), the estimated coefficient of 2.52
(p-value < 0.0001) on WeakFundi,t implies that the odds of a stock price crash in a firm with
weak fundamentals are 12.46 times greater than a firm with strong fundamentals. When we
use the firm’s fundamentals-adjusted annual stock return (model 2), we find that the odds of
a firm with weak fundamentals crashing are 4.09 times greater; both models are again statis-
tically significant with a Wald-test p-value < 0.0001. These results provide strong statistical
15
Accounting fundamentals, including the F-score, have previously been shown to predict cross-sectional
differences in future stock returns; a portfolio of firms with weak fundamentals achieves lower future
returns while a portfolio of firms with strong fundamentals achieves higher future returns. While our tests
focus on extreme negative returns, we would like to ensure that we are not simply picking up previously
documented effects. To address this, we follow Piotroski and So (2012) to form portfolios of firms based on
the F-score and adjust the annual return of each firm by subtracting the return of a portfolio of firms with
similar fundamentals. This procedure is similar to numerous prior studies that subtract returns of size-based
portfolios to control for the size effect.
16
Also shown are coefficient estimates for several control variables, defined in the appendix, along with
overall model fit statistics.
17 ̂
Calculated as e𝛽 per Eq. (4) in Sect. 3.2. This statistic is shown at the bottom of the panel along with a
95% Wald confidence interval.
13
Table 6 Logit regression tests based on liquidity-based sub-samples
Panel A: HIGH Liquidity Stocks Panel B: LOW Liquidity Stocks
Prob. Of A
djRett+1 < − 50% Prob. Of A
djRett+1 < − 70%
Intercept 0.33 0.89 0.3722 − 1.65 − 2.67 0.0077 − 0.17 − 0.63 0.5291 − 0.94 − 2.43 0.0153
Weak Fund 0.554 2.93 0.0034 1.354 4.90 < .0001 0.97 5.97 < .0001 1.80 7.68 < .0001
Ln(Volatility) 0.290 2.69 0.0071 0.205 1.30 0.1932 0.50 6.2 < .0001 0.48 4.2 < .0001
Beta 0.01 0.16 0.873 − 0.02 − 0.38 0.707 − 0.04 − 1.20 0.2313 − 0.09 − 1.85 0.0643
Ln(MktVal) − 0.04 − 1.20 0.2309 − 0.02 − 0.40 0.6855 0.01 0.39 0.6938 − 0.01 − 0.26 0.7967
Ln(B/M) − 0.11 − 2.55 0.0107 − 0.17 − 2.74 0.0062 − 0.26 − 6.37 < .0001 − 0.31 − 5.59 < .0001
Ln(Leverage) 0.71 3.03 0.0025 0.64 1.90 0.0578 0.24 1.11 0.265 0.40 1.31 0.1918
Log(Ret) 0.468 7.37 < .0001 0.541 6.30 < .0001 0.383 6.16 < .0001 0.594 7.13 < .0001
Wald χ2 2898.3 2074.8 2357.6 1853.2
p(χ2) < .0001 < .0001 < .0001 < .0001
Pseudo R2 21.76% 23.02% 16.33% 17.89%
No. of Obs 22,091 22,091 20,755 20,755
Fixed Year Yes Yes Yes Yes
Effects
Odds Ratio Estimate 95% Wald Conf. Limits Estimate 95% Wald Conf. Estimate 95% Wald Conf. Limits Estimate 95% Wald Conf.
Estimates Limits Limits
Weak Fund 1.74 1.20 2.52 3.87 2.25 6.66 2.64 1.92 3.64 6.03 3.81 9.53
1733
13
Table 6 continued
1734
This table shows results from modeling stock price crash risk probabilities using GENERALIZED logit models for subsamples based on stock price liquidity. Using the Ami-
hud (2002) measure of illiquidity (defined in the appendix), we split the sample into three groups. The top 1/3 of stocks are classified as Low liquidity stocks while the bottom
13
1/3 of stocks are classified as high liquidity stocks. For each of these groups, we estimate a logit regression. To identify crashes, we measure the stock return of each firm over
the 12-month period that follows the release of financial statements, each fiscal year. We allow a three-month delay between the fiscal year-end of a firm and the beginning
of the stock return measurement period to ensure that financial statement information is available to the market. If this return for firm i during year t,t + 12 qualifies as a crash
(< -50% in model 1 and < -70% in model 2), we code the variable Crashi,t,t+12 as 1 and -1 if the firm’s return is > + 50% (model 1) or > + 70% (model 2), and zero otherwise.
Panel A shows results for the High liquidity stocks and Panel B shows results for the Low liquidity stocks. Our measure of fundamentals is based on the F-score devised by
Piotroski (2000). The F-score is a summary measure calculated based on nine binary signals statement data. Calculation of the F-score is described in
the appendix. We transform the F-score into a variable labeled WeakFundi,t: WeakFundi,t = 9 − F − scorei,t ∕9
( derived from financial
)
WeakFundi,t ranges from a minimum of 0 to a maximum of 1, with larger values implying weaker fundamentals. To model the probability of a stock’s price crashing, the sim-
Prob. Crash =1 B
ple logit model takes the following form: ln Prob. (Crash i,t,t+12=−1) = 𝛼̂ + 𝛽̂1 WeakFundi,t + 𝛾̂1 Log(Mktval)i,t + 𝛾̂2 Ln M
+ 𝛾̂3 Ln(Vol)i,t + 𝛾̂4 MktBetai,t + 𝛾̂5 Ln(Leverage)i,t + 𝛾̂6 Ln(Ret)i,t + 𝜀̂ i,t
( )
( i,t,t+12 ) i,t
All variables are defined in the appendix. We estimate the equation above using maximum likelihood estimation. Coefficient estimates along with associated p-values and
̂
overall model statistics are provided. The odds ratio estimates are calculated as e𝛽 , provided along with 95% confidence intervals
I. Safdar et al.
evidence that accounting fundamentals are informative about the relative likelihood of a firm’s
stock price crashing over the next 12 months.
The statistically significant coefficients for our measure of weakness in accounting funda-
mentals shown in results from our tests and discussed above indicate that investors on average
do not appear to fully incorporate the information embedded in deteriorating accounting fun-
damentals into their expectations of left-tail risk in a timely manner.
We also perform a number of sensitivity and robustness tests on the results obtained from
our logit model. In panels C and D, we re-estimate Eq. (6) by using another measure of stock
price liquidity (Amihud, 2002) and find very similar results. In panel E, we add the Hutton
et al. (2009) measure of opacity (described in Sect. 4.1.4) into Eq. (6) as a control. For brev-
ity, we use the fundamentals-adjusted annual return AdjReti,t,t+12 (described in the previous
section) to define stock-price crashes; results using raw returns are stronger (but not tabu-
lated). In both models 1 and 2 of Panel E of Table 5 (the models differ in the magnitude of
returns used to define crashes), the coefficient on opacity is positive and statistically signifi-
cant (p-value < 0.01). This corroborates the findings of Hutton et al. (2009). More notably, the
coefficient on WeakFundi,t becomes larger, providing odds ratios greater than those found in
our previous results (panels A-D).
To further examine our results and the role of liquidity, we split the sample into terciles each
year based on the Amihud (2002) measure of illiquidity. Tercile 1 stocks are high liquidity stocks
while Tercile 3 stocks are the least liquid based on this grouping. We re-estimate our models for
the Tercile 1 and Tercile 3 groups to see if the results we document differ for these two groups.
These results are documented in Panels A (high liquidity stocks) and B (low liquidity stocks) of
Table 6. For brevity, we focus on adjusted returns (results using raw returns are stronger). Model
2 of Panel A shows that the coefficient on WeakFund is 1.35 (p-value < 0.0001) for high liquid-
ity stocks when we use a threshold of −70% to define price crashes. The coefficient is greater in
magnitude (1.80 with p-value < 0.0001) for low liquidity stocks as shown in model 2 of panel B.
Thus, our results remain strong regardless of liquidity but the probability of weak stocks crashing
is greater for low liquidity stocks than for high liquidity stocks.
We also use accruals, discretionary accruals as a measure of earnings management, and
accounting comparability as additional controls in our regressions and find that our results
remain unaffected. We do not tabulate these results.
5 Conclusions
Several prior crash-risk studies rely on the notion that firms withhold significant quan-
tities of bad news from investors. In this study, we undertake the task of determining
whether information derived from financial statements is linked to expectations of future
crash risk. We find a robust and strong relationship between our measure of weakness
in accounting fundamentals and expected crash risk as priced by options traders. This
evidence suggests that sophisticated traders understand this link and incorporate it into
the prices of equity options traded on a stock. Furthermore, we document a strong posi-
tive link between weakness in accounting fundamentals and the likelihood of an ex-post
crash. Collectively, these findings contradict the notion that firms can entirely conceal
bad news from their financial disclosures and help clarify the role of investor interpreta-
tion of accounting information in driving stock price crashes. Our results also show that
fundamental analysis is useful for identifying stocks that are more exposed to left tail
13
risk. Thus, our study also contributes to the literature regarding the utility of financial
statement analysis.
Appendix
Variable Definitions
ATM_IVi,t – is the daily open interest-weighted implied volatility of at the money

(ATM) call options, computed as the average over the fiscal year.
Betai,t – is the market beta estimated from the CAPM using daily stock and market
returns over the fiscal year (used in Table 6).
BidAskSpreadi,t – is the mean daily relative bid-ask spread over the year, where rela-
tive bid-ask spread as the absolute difference in daily ending bid and ask divided by
the average of daily ending bid and ask.
Cashflow_Voli,t – is the standard deviation of operating cash flow (OANCF) (scaled
by lagged total assets) over the past five years.
CRASHi,t,t+12 – an indicator variable that takes on a value of 1 if firm i experiences
a negative one-year stock return of -50% or worse (in other models we use -70% or
worse) during year t,t + 12, and 0 otherwise. Year t,t + 12 is defined as the 12 months
starting with four months after the most recent fiscal year-end. For generalized
logit models, a value of -1 is assigned to firms with a one-year positive stock return
of + 50% or greater (in other models + 70% or greater).
Earnings_Voli,t – is the standard deviation of income before extraordinary items (IB)
(scaled by lagged total assets) over the prior five years.
HHIi,t – is the Herfindahl–Hirschman Index measured within each three-digit SIC
industry-year.
Idosy_Voli,t – is the standard deviation of weekly firm-specific returns over the fiscal
year.
Illiquidityi,t − is the median daily price impact over the year where price impact equals
the daily absolute price change in percent divided by US$ trading volume (measured
in thousands) from Amihud (2002). We omit zero return days to avoid misclassifying
low trading activity days as highly liquid and multiply Illiquidity by 1,000. Smaller
values indicate greater liquidity.
IV_SKEWi,t – a proxy for expected crash risk calculated as the average daily implied vol-
atility skew over either the three months ( IV_SKEW3i,t ) or 12 months ( IV_SKEW12i,t )
beginning in the fourth month following the fiscal year-end.
Ln(Mktval)i,t – the natural log of the market value of firm i’s equity calculated as the
shares
( )outstanding (CSHO) times the stock price (PRCC_F) as of the fiscal year-end t.
Ln MB – the natural log of the ratio of firm i’s book-value of equity (CEQ) to its mar-
i,t
ket value as of fiscal year-end t.
Ln(Leverage)i,t – the natural log of the ratio of firm i’s long-term total debt to its balance
sheet value of assets as of fiscal year-end t.
Ln(Ret)i,t – the natural log of one plus the one-year stock return during fiscal year t.
13
Ln(Vol)i,t – the natural log of the ratio of firm i’s stock return volatility is measured as
the standard deviation of its monthly stock returns ending over a 36-month period end-
ing at fiscal year-end t.
MktBetai,t – the beta of a firm estimated using its past 36 monthly returns regressed
( ) the value-weighted market index as of the end of fiscal year-end t.
against
M
B
– is the ratio of the market value of equity to the book value of equity at the end
i,t
of the year.
NEG_SKEWi,t – is the negative skewness of weekly stock returns over the fiscal year.
Sales_Voli,t – is the standard deviation of sales revenue (scaled by lagged total assets)
over the prior five years.
Stock_Turni,t – is the average monthly share turnover over the fiscal year.
Strategyi,t – is the business strategy composite measure of Bentley et al. (2013), scaled
by 100.
Total_Voli,t – is the standard deviation of weekly stock returns over the fiscal year.
WeakFundi,t – defined as (9 – F-score)/9; ranges from 0 to 1, with larger values implying
weaker fundamentals. (construction described in Sect. 3.2).
Calculation of the F‑score
Following Piotroski and So (2012), the F-score is calculated for each firm using nine sig-
nals derived from its annual financial statements:
F − Score = I_ROA + I_CFO + I_ACC + I_DROA + I_DLEV
(A.1)
+ I_LIQ + I_SSTK + I_DM + I_DTURN.
Variable definitions follow; COMPUSTAT variables are in parentheses:
ROAi,t – income before extraordinary items (IB) divided by the beginning of the year
total assets (ATi,t-1). The indicator variable I_ROAi,t equals 1 if R OAi,t > 0 and 0 other-
wise.
CFOi,t – measured as the cash flow from operations (OANCFi,t, measured as funds from
operations when OANCF is not available) scaled by the beginning of year total assets
(ATi,t-1). The indicator variable I_CFOi,t equals 1 if O ANCFi,t > 0 and 0 otherwise.
ACCRUALi,t – measured as the difference between income before extraordinary items
(IBi,t) scaled by the beginning of year total assets (ATi,t-1) and cash flow from operations
as described above. The indicator variable I_ACC equals 1 if ACCRUALi,t < 0 and 0
otherwise.
DROAi,t – measured as the difference between the current year’s R OAi,t, and the previ-
ous year’s R OAi,t-1. The indicator variable I_DROA equals 1 if D ROAi,t > 0 and 0 oth-
erwise.
DLEVERi,t – measured as the difference between the current year’s debt-to-assets ratio
and the previous year’s debt-to-assets ratio. The debt-to-assets ratio is measured as
long-term debt ( DLTTi,t) divided by total assets (ATi,t). The indicator variable I_DLEV
equals 1 if D LEVERi,t < 0 and 0 otherwise.
DLIQUIDi,t – measured as the difference between the current year’s current ratio and
the previous year’s ratio. The current ratio is measured as current assets (ACTi,t) divided
13
by current liabilities (CLTi,t). The indicator variable I_DLIQ equals 1 if DLIQUIDi,t > 0

and 0 otherwise.
ISSUANCEi,t – measured as the amount of stock issued by a firm in a given year
(SSTKi,t). The indicator variable I_SSTK equals 1 if S STKi,t < = 0 and 0 otherwise.
DMARGINi,t – measured as the difference between the current year’s gross margin ratio
and the previous year’s ratio. The gross margin ratio is measured as one minus the ratio
of cost of goods sold (COGSi,t) and net sales (SALEi,t). The indicator variable I_DM
equals 1 if D MARGINi,t > 0 and 0 otherwise.
DTURNi,t – measured as the difference between the current year’s asset turnover ratio
and the previous year’s turnover ratio. The asset turnover ratio is measured as net sales
(SALEi,t) divided by total assets (ATi,t). The indicator variable I_DTURN equals 1 if
DTURNi,t > 0 and 0 otherwise.
References
Ahmed AS, Safdar I (2018) Dissecting stock price momentum using financial statement analysis. Account
Financ 8:3–43
Aman H (2013) An analysis of the impact of media coverage on stock price crashes and jumps: evidence
from Japan. Pac Basin Financ J 24:22–38
Amihud Y (2002) Illiquidity and stock returns: cross-section and time-series effects. J Financ Markets
5:31–56
An Z, Gao W, Li D, Zhu F (2018) The Impact of firm-level illiquidity on crash risk and the role of media
independence: international evidence. Int Rev Financ 18:547–593
Aspris A, Finch N, Foley S, Meyer Z (2013) Fundamental-based market strategies. Aust Account Rev
6:380–392
Atilgan Y, Bali TG, Demirtas KO, Gunaydin AD (2019) Left-tail momentum: underreaction to bad news,
costly arbitrage and equity returns. J Financ Econ 135:725
Baginski SP, Campbell JL, Hinson LA, Koo DS (2018) Do career concerns affect the delay of bad news
disclosure. Account Rev 93:61–95
Bai M, Xu L, Yu CF, Zurbruegg R (2020) Superstition and stock price crash risk. Pacific-Basin Financ J
60:101287
Banerjee P, Deb S (2017) Abnormal returns using accounting information within a value portfolio. Account
Res J 30:73–88
Bates S (2000) Post-’87 crash fears in the S&P 500 futures option market. J Econ 94(1/2):181–238
Ben-Nasr H, Bouslimi L, Zhong R (2019) Do patented innovations reduce stock price crash risk. Int Rev
Financ 21:3
Cai G, Xu Y, Yu D, Zhang J, Zheng G (2019) Strengthened board monitoring from parent company and
stock price crash risk of subsidiary firm. Pac Basin Financ J 56:352–368
Chang X, Chen Y, Zolotoy L (2017) Stock liquidity and stock price crash risk. J Financ Quant Anal
52(4):1605–1637
Chen HY, Lee CF, Shih W (2016) Technical, fundamental, and combined information for separating win-
ners from losers. Pac Basin Financ J 39:224–242
Choi S, Jung H (2020) Director liability reduction and stock price crash risk: evidence from Korea. Int Rev
Financ 21:1492
Daniel K, Hirshleifer D, Teoh SH (2002) Investor psychology in capital markets: evidence and policy impli-
cations. J Monet Econ 49:139–209
Dechow P, Sloan R, Sweeny A (1995) Detecting earnings management. Account Rev 70(2):193–225
Du C, Song L, Wu J (2016) Bank accounting disclosure, information content in stock prices, and stock crash
risk: global evidence. Pac Account Rev 28:260–278
Fama E, French KR (1992) The cross-section of expected returns. J Financ 47(2):427–465
Fama E, French K (2006) Profitability, investment, and average returns. J Financ Econ 82:491–518
Francis B, Hasan I, Li L (2016) Abnormal real operations, real earnings management, and subsequent
crashes in stock prices. Rev Quant Financ Account 46:217–260
13
Gallagher D, Gardner P, Schmidt C, Walter T (2014a) Portfolio quality and mutual fund performance. Int
Rev Financ 14:485–521
Gallagher D, Gardner P, Schmidt C, Walter T (2014b) Quality investing in an Australian context. Aust J
Manag 39:615–643
Habib A, Hasan M, Jiang H (2018) Stock price crash risk: review of the empirical literature. Account Financ
58:211–251
Hamm S, Li EX, Ng J (2018) Management forecasts and bad news hoarding: evidence from stock price
crashes. Available at SSRN 2055008
Hodgson A, Stevenson-Clarke P (2000) Accounting variables and stock returns: the impact of leverage. Pac
Account Rev 12:37–64
Hong H, Lim T, Stein JC (2000) Bad news travels slowly: size, analyst coverage, and the profitability of
momentum strategies. J Financ 55(1):265–295
Hong H, Stein JC (1999) A unified theory of underreaction, momentum trading, and overreaction in asset
markets. J Financ 54(6):2143–2184
Huang S, Qiao Z (2017) Shadows in the sun: crash risk behind earnings transparency. J Bank Financ
83:1–18
Hutton AP, Alan JM, Tehranian H (2009) Opaque financial reports, R2, and crash risk. J Financ Econ
94(1):67–86
Ikenberry D, Lakonishok J, Vermaelen T (1995) Market underreaction to open market share repurchases. J
Financ Econ 39(181–208):45
Ikenberry DL, Ramnath S (2002) Underreaction to self-selected news events: the case of stock splits. Rev
Financ Stud 15(2):489–526
Jebran K, Chen S, Zhang R (2021) Board social capital and stock price crash risk. Rev Quant Financ
Account. https://doi.org/10.1007/s11156-021-01001-3
Jang J, Kang J (2019) Probability of price crashes, rational speculative bubbles, and the cross section of
stock returns. J Financ Econ 132(1):222–247
Jiang F, Cia X, Nofsinger JR, Zheng X (2021) Can reputation concern restrain bad news hoarding in family
firms? J Bank Financ 114(105808):0378–4266
Jin Li, Myers CS (2006) R2 around the world: new theory and new tests. J Financ Econ 79(2):257–292
Kahneman D, Slovic P, Tversky A (1982) Judgment under uncertainty: heuristics and biases. Cambridge
University Press, Cambridge
Kim JB, Li Y, Zhang L (2011) Corporate tax avoidance and stock price crash risk: firm-level analysis. J
Financ Econ 100(3):639–662
Kim J-B, Zhang L (2014) Financial reporting opacity and expected crash risk: evidence from implied vola-
tility smirks. Contemp Account Res 31:851–875
Kim J-B, Zhang L (2016) Accounting conservatism and stock price crash risk: firm-level evidence. Con-
temp Account Res 33:412–441
Kim Y, Li H, Li S (2014) Corporate social responsibility and stock price crash risk. J Bank Finance 43:1–13
Kim J-B, Li L, Lu L, Yu Y (2016) Financial statement comparability and expected crash risk. J Account
Econ 61:294–312
Kim J-B, Lu L, Yu Y (2019) Analyst coverage and expected crash risk: evidence from exogenous changes in
analyst coverage. Account Rev 94(4):345–364
Kothari SP, Shu S, Wysocki PD (2009) Do managers withhold bad news? J Account Res 47(1):241–276
Li EX, Ramesh K (2009) Market reaction surrounding the filing of periodic SEC reports. Account Rev
84(4):1171–1208
Li T, Xiang C, Liu Z, Cai W (2020) Annual report disclosure timing and stock price crash risk. Pacific-
Basin Finance J 62:101392
Lv D, Wu W (2019) Margin-trading volatility and stock price crash risk. Pac Basin Financ J 56:179–196
Menon KL, Williams DD (2010) Investor reaction to going concern audit reports. Account Rev
85(6):2075–2105
Ng C, Shen J (2016) Screen winners from losers using simple fundamental analysis in the Pacific-Basin
stock markets. Pac Basin Financ J 39:159–177
Ohlson JA, Bilinski P (2015) Risk versus anomaly: a new methodology applied to accruals. Account Rev
90(5):2057–2077
Pan J (2002) The jump-risk premia implicit in options: evidence from an integrated time-series study. J
Financ Econ 63(2):3–50
Park SY, Park HY (2019) Do statutory internal auditors influence stock price crash risk? Firm-level evi-
dence from Korea. Manag Audit J 35:207–237
Piotroski JD (2000) Value investing: the use of historical financial statement information to separate winners
from losers. J Account Res 38:1–41
13
Piotroski JD, So EC (2012) Identifying expectation errors in value/glamour strategies: a fundamental analy-
sis approach. Rev Financ Stud 25(9):2841–2875
Santa-Clara P, Yan S (2010) Crashes, volatility, and the equity premium: lessons from S&P 500 options.
Rev Econ Stat 92(2):435–451
Shengmin H, Zhen Q (2017) Shadows in the sun: crash risk behind earnings transparency. J Bank Finance
83:1–18
Sloan RG (1996) Do stock prices fully reflect information in accruals and cash flows about future earn-
ings? Accounting Review : 289-315
Stice EK (1991) The market reaction to 10-K and 10-Q filings and to subsequent the wall street journal
earnings announcements. Account Rev 66(1):42–55
Sun S, Habib A, Huang H (2019) Tournament incentives and stock price crash risk: evidence from China.
Pac Basin Financ J 54:93–117
Vo X (2018) Foreign investors and stock price crash risk: evidence from Vietnam. In Press, International
Review of Finance
Xu N, Jiang X, Chan KC, Yi Z (2013) Analyst coverage, optimism, and stock price crash risk: evidence
from China. Pac Basin Financ J 25:217–239
Yang C, Chen X, Chen X (2020) Vertical interlock and stock price crash risk. Pacific-Basin Finance J
68:101387
Yang X, Zhu Y, Cheng TY (2020) How the individual investors took on big data: The effect of panic from
the internet stock message boards on stock price crash. Pacific-Basin Finance J 59:101245
Zhou W, Wu L, Wang H (2016) The consequences of increasing the scope of managerial judgement in
accounting standards. Abacus 52:404–440
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
13

Accounting Information and Left Tail Risk

Uploaded by

Copyright:

Available Formats

Accounting Information and Left Tail Risk

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Accounting Information and Left Tail Risk

Uploaded by

Copyright:

Available Formats

Review of Quantitative Finance and Accounting (2022) 58:1709–1740

Accounting information and left‑tail risk

Irfan Safdar1 · Michael Neel2 · Babatunde Odusami1

Accepted: 27 December 2021 / Published online: 9 February 2022

Keywords Left-tail risk · Accounting information · Financial statement analysis · Market

JEL Classification G11 · G12 · G14 · G40

We take a two-pronged approach to examine evidence on the link between potentially

2.1 Fundamentals and stock returns

2.2 Accounting fundamentals, opaqueness, and crash risk: reconciling evidences

3 Sample, descriptive statistics, and measurement of accounting

3.1 Sample and descriptive statistics

Panel A: Fundamental Variables – Full Sample

Panel C: Univariate Tests of Expected Crash Risk Conditional on Fundamentals

3.2 Measurement of accounting fundamentals

Our measure of accounting fundamentals is based on the F-score proposed by Piotroski

WeakFundi,t ranges from a minimum of 0 to a maximum of 1, with larger values

Fig. 1 Distribution of negative one-year stock returns

Table 2 Frequency of Negative One-Year Stock Returns (Pooled Sample: 1991–2018)

Panel A – Frequency of One-Year Negative Returns

Fundamentals Ret < -90% < -80% < -70% < -60% < -50% < -25% < 0%

3.3 A look at the empirical frequency of negative stock returns in the US market

As a matter of general interest, we present a look at the frequency distribution of stock

Fig. 2 Conditional distribution of negative one-year stock returns conditional on fundamentals (Sample:

3.3.1 The frequency of negative stock returns conditional on fundamentals

In panel B of Table 2, we provide a simple depiction of the conditional empirical frequency

4 Empirical tests and results

4.1 Tests of accounting fundamentals and expected crash risk

4.1.1 Measurement of expected crash risk

IV_Skewit = IVitOTMP − IVitATMC (2)

4.1.2 Tests using expected crash risk

IV_Skewi,t = 𝛼̂ + 𝛽̂1 WeakFundi,t + 𝛾̂1 ATM_IVi,t + 𝛾̂2 Ln(Mktval)i,t

4.1.3 Results from tests of expected crash risk – discussion of Tables 3

Table 3 Accounting fundamentals and expected crash risk

Intercept 0.139 (8.799) 0.000 0.134 (7.713) 0.000

Table 4 Accounting Fundamentals and Expected Crash Risk – Robustness Tests

4.1.4 Additional sensitivity and robustness tests – discussion of Table 4

4.1.4.3 Controlling for accounting comparability We next control for financial statement

4.2 Tests of investor’ processing of accounting fundamentals and left‑tail risk

Panel A: Future One-Year Return < -50% Panel B: Future One-Year Return < -70%

Model 1 Model 2 Model 1 Model 2

Model 1 Model 2 Model 1 Model 2

Panel E – Additional Robustness Tests

Model 1: Prob. Of AdjRet < -50% Model 2: Prob. Of AdjRet < -70%

Intercept − 0.07 − 0.53 0.5932 − 1.63 − 7.41 < .0001

Conf. Limits Conf. Limits

Weak Fund 2.65 2.27 3.10 4.96 3.91 6.30

WeakFundi,t ranges from a minimum of 0 to a maximum

+ 𝛾̂4 MktBetai,t + 𝛾̂5 Ln(Leverage)i,t + 𝛾̂6 Ln(Ret)i,t + 𝜀̂ i,t

4.2.1 Defining price crashes

Financial statements Measure stock returns over

Fiscal year-end month t month t+12

4.2.2 Estimation of generalized logit models to evaluate likelihood of a stock price

4.2.3 Discussion of Table 5 – results from generalized logit models

Model 1 Model 2 Model 1 Model 2

ATM_IVi,t – is the daily open interest-weighted implied volatility of at the money

Variable definitions follow; COMPUSTAT variables are in parentheses:

by current liabilities ­(CLTi,t). The indicator variable I_DLIQ equals 1 if ­DLIQUIDi,t > 0

You might also like

2.1 Fundamentals and stock returns

2.2 Accounting fundamentals, opaqueness, and crash risk: reconciling evidences

3 Sample, descriptive statistics, and measurement of accounting

3.1 Sample and descriptive statistics

3.2 Measurement of accounting fundamentals

Fig. 1 Distribution of negative one-year stock returns

Table 2 Frequency of Negative One-Year Stock Returns (Pooled Sample: 1991–2018)

3.3 A look at the empirical frequency of negative stock returns in the US market

Fig. 2 Conditional distribution of negative one-year stock returns conditional on fundamentals (Sample:

3.3.1 The frequency of negative stock returns conditional on fundamentals

4 Empirical tests and results

4.1 Tests of accounting fundamentals and expected crash risk

4.1.1 Measurement of expected crash risk

4.1.2 Tests using expected crash risk

4.1.3 Results from tests of expected crash risk – discussion of Tables 3

Table 3 Accounting fundamentals and expected crash risk

Table 4 Accounting Fundamentals and Expected Crash Risk – Robustness Tests

4.1.4 Additional sensitivity and robustness tests – discussion of Table 4

4.2 Tests of investor’ processing of accounting fundamentals and left‑tail risk

4.2.1 Defining price crashes

4.2.2 Estimation of generalized logit models to evaluate likelihood of a stock price

4.2.3 Discussion of Table 5 – results from generalized logit models

by current liabilities (CLTi,t). The indicator variable I_DLIQ equals 1 if DLIQUIDi,t > 0