International Review of Financial Analysis 78 (2021) 101920

Contents lists available at ScienceDirect

International Review of Financial Analysis

journal homepage: www.elsevier.com/locate/irfa

Accounting-based downside risk and expected stock returns: Evidence

from China
Yan Luo a, Xiaohuan Wang a, Chenyang Zhang a, c, Wei Huang b, *
a
School of Management, Fudan University, China
b
Nottingham University Business School China, University of Nottingham Ningbo, 199 Taikang East Road, Ningbo, China
c
Guotai Junan Securities, China

A R T I C L E I N F O A B S T R A C T

JEL classifications: We document that earnings downside risk contains information on firms’ future operating performance and is
G12 positively associated with expected stock returns in Chinese stock markets, and the return predictability of
G14 earning downside risk mainly comes from its accrual downside risk component. The pricing of earnings downside
G32
risk is especially evident among firms with more transparent information environment and stronger governance
Keywords: efficacy, such as large firms, non-high-tech firms, old firms, and firms with high analyst coverage. Lastly, we
Accounting-based downside risk
show that aggregated earnings downside risk and its components at the market level are all significantly and
Information environment
Stock returns
positively associated with subsequent stock market returns, which is consistent with the notion that the
Chinese stock markets accounting-based downside risk measures contain information about future macroeconomic conditions.

1. Introduction extent to which a company’s financial performance drops below its

expected level, and examine its pricing implications in the U.S. markets.
The literature has suggested that investors care more about downside The accounting-based downside risk differs from return downside risk,
losses than upside gain potentials and are therefore more sensitive to or stock price crash risk, in that the former is constructed based financial
losses than to gains (e.g., Gul, 1991; Kahneman & Tversky, 1979; Roy, information while the latter is constructed completely based on stock
1952). Accordingly, Koonce, McAnally, and Mercer (2005) show that return data. Therefore, the information content of these two types of
economic agents judge negative and positive expectations differently in downside risks could be quite different. Konchitchki et al. (2016)
risk management, placing more emphasis on potential loss outcomes. confirm this by showing that accounting-based downside risk is signif­
Ang, Chen, and Xing (2006) further show that stock market investors icantly positively associated with expected stock returns even when
place greater weight on downside risk and demand additional stock price crash risk measures are controlled.
compensation for holding stocks with greater sensitivities to downside In this study, we extend Konchitchki et al. (2016) by examining the
market movements. A bunch of studies thereafter have examined stock pricing of accounting-based downside risk in the Chinese A-share stock
price crash risk, which refers to an abrupt and large-scale drop in stock markets, which are becoming increasingly popular for investors around
prices, from various perspectives (e.g., Hutton, Marcus, & Tehranian, the world and have attracted much investments. Investigating potential
2009; Jin & Myers, 2006; Kim, Li, & Zhang, 2011a; Kim, Li, & Zhang, pricing factors in the Chinese A-share market thus has its own merit.
2011b; Kim & Zhang, 2016). Moreover, Konchitchki et al. (2016) contend that accounting-based
Although the downside risk in stock returns, especially stock price downside risk is priced in the U.S. markets as it contains information
crash risk, has been analyzed extensively in the literature, little research about firms’ future operating performance. In emerging markets, such as
has investigated the downside risk of accounting-based measures, which the Chinese stock markets, the information environment is much less
are important sources of information in capital markets. Konchitchki, transparent and firms’ governance efficiency is much weaker than that
Luo, Ma, and Wu (2016) are the first to construct measures of in the U.S. markets. Thus, whether accounting-based downside risk
accounting-based downside risk, which is defined as the probability and measures, which are constructed based on financial information

* Corresponding author.
E-mail addresses: [email protected] (Y. Luo), [email protected] (X. Wang), [email protected] (C. Zhang), wei.huang@
nottingham.edu.cn (W. Huang).

https://doi.org/10.1016/j.irfa.2021.101920
Received 18 January 2021; Received in revised form 14 September 2021; Accepted 11 October 2021
Available online 22 October 2021
1057-5219/© 2021 Elsevier Inc. All rights reserved.
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

disclosed by firms, still convey useful information about firms’ future operating performance as it only recognizes the inflows and outflows of
operating performance and whether such information is properly cash flows but does not match the expenses with revenues when they are
incorporated into stock prices in the Chinese A-share markets warrants realized. In contrast, accruals match expenses with revenues by altering
further investigation. the timing of cash flow recognition in the earnings, thus providing more
We differ from Konchitchki et al. (2016) mainly in three ways. First, relevant information for predicting firms’ future cash flow and expected
we examine the pricing of accounting-based downside risk in large stock returns. OCF downside risk, therefore, might only be a noisy
developing stock markets, the Chinese A-share markets, which differ predictor of future cash flow and contains less information about vari­
from the U.S. markets investigated by Konchitchki et al. (2016) in that ations in expected returns due to a mismatching problem.2
its information environment is less transparent and the governance Moreover, we examine the pricing of earnings downside risk cross-
systems are less effective. Such investigation can help investors to better sectionally. If earnings downside risk affects expected stock returns
understand the role played by accounting-based downside risk in less due to the fact that it contains information about firms’ future operating
developed markets sharing similar characteristics as the Chinese mar­ performance, we expect the positive relation between earnings down­
kets. In the investigation of the pricing of accounting-based downside side risk and expected stock returns to be stronger among stocks with
risk, we focus on earnings-downside risk as well as its two components, greater information transparency and governance efficacy. We partition
the accrual downside risk and operating cash flow downside risk (OCF our sample firms into subsamples conditional on firm size, industry, age,
downside risk). Second, we examine the cross-sectional variation in the and analyst coverage. We expect firms with larger size, belong to non-
pricing of accounting-based downside risk. Specifically, we examine high-tech industries, with a longer listing history, and with more ana­
whether and how information environment and governance efficacy lysts following to have more transparent information environment and
affects the relation between earnings downside risk and stock returns at greater governance efficacy than their counterparts, and thus expect the
the individual stock level in the Chinese A-share markets. Such inves­ positive relation between earnings downside risk and expected stock
tigation not only help to further differentiate our study from Konchitchki returns to be stronger in this group of stocks. Our empirical results are
et al. (2016), but also helps to deepen investors’ understanding of factors consistent with our expectation. Such pieces of evidence echo our pre­
that would affect the pricing of accounting-based downside risk. Third, vious findings that although earnings downside risk processes robust
while Konchitchki et al. (2016) examines the pricing of accounting- return predictability in the Chinese A-share markets overall, the
downside risk at the individual stock level only, we extend their study magnitude of its influence on stock returns is relatively lower than that
by investigating the pricing of such a risk at the market level as well. We in the U.S. markets, where firms’ information environment and gover­
aggregate accounting-based downside risk at the firm level into an nance efficiency is of higher standards in general.
aggregated measure at the market level and examine the extent to which Lastly, we examine the pricing of earnings downside risk at the
it predicts future stock market returns. If accounting-based downside market level. Past studies have shown that cross-sectional predictors of
risk captures information on firms’ future operating performance and stock returns, such as accruals and cash flows, can be aggregated to the
contains a systematic component, we expect it to be able to predict market level to predict market returns (e.g., Hirshleifer, Hou, & Teoh,
subsequent market returns. 2009; Kang, Liu, & Qi, 2010). By analogy, we value-weight individual
We utilize a large sample of Chinese listed firms to perform the stocks’ earnings downside risk to come up with measures at the aggre­
investigation. We first show that earnings downside risk is associated gated market level. We contend that high earnings downside risk at the
with firms’ future operating performance, and significantly positively aggregated level implies that a large proportion of firms are likely to
predicts expected stock returns. The return predictability of earnings have deteriorating operating performance, which is a signal of wors­
downside risk is robust and not subsumed by various earnings attributes, ening macroeconomic conditions. In such periods, investors would
such as accrual quality, earnings persistence and predictability, value require higher expected market returns to participate in the stock mar­
relevance, smoothness, timeliness, conservatism, and other conven­ kets. We therefore expect aggregated earnings downside risk to be
tional return explanatory variables. Its return predictability is also positively associated with expected stock market returns, which is sup­
robust to the control of stock price crash risk measures. The evidence ported by our empirical results.
suggests that earnings downside risk contains unique information. The The rest of the paper is organized as follows. Section 2 develops
magnitude of the influence of earnings-downside risk on stock returns, hypotheses. Section 3 introduces data and measures. Sections 4 reports
however, is relatively smaller in the Chinese A-share markets than in the results of the empirical investigation. Section 5 concludes.
U.S. markets. A one standard deviation increase in earnings downside
risk is followed by a 0.32% increase in monthly stock returns in the U.S. 2. Hypothesis development
markets, and a 0.24% increase in monthly stock returns in the Chinese
A-share markets. Such a difference might be attributable to the lower As Fishburn (1977) states, decision makers in investment contexts
information transparency and weaker governance efficacy in general in frequently associate risk with failure to attain a target return. Given that
the Chinese A-share markets than in the U.S. markets, which impairs the risk is more likely to manifest through downside states than upside
effect of earnings-downside risk on stock returns to some degree. states, downside risk should be valued differently from the upside as a
Further, we decompose earnings downside risk into accrual down­ risk measurement. Konchitchki et al. (2016) define earnings downside
side risk and OCF downside risk. We find that the return predictability of risk as the probability and magnitude that a firms’ earnings fall short of
earnings downside risk in China mainly comes from its accrual downside investors’ expectation, and confirm that it is priced in the U.S. markets.
risk component, and the relation between OCF downside risk and ex­ We follow Konchitchki et al. (2016) in measuring earnings downside
pected stock returns is ambiguous.1 We contend that such findings could risk. Specifically, we calculate earnings downside risk based on residuals
results from the different information contents of accruals and OCF. from an earnings expectation model. This measure captures the proba­
Although some argue that OCF is a more reliable indicator for firms’ bility and extent to which a firms’ future earnings will fall below
future earnings than accruals as it cannot be easily manipulated by
managers, OCF does have its disadvantages in predicting firms’ future
2
Another possible explanation for the weak and ambiguous relation between
OCF downside risk and expected stock returns is that OCF is associated with
1
Konchitchki et al. (2016) examines the pricing of earnings downside risk firms’ upside opportunities. Firms with lower OCF than expected may be those
only. However, in an early version of the paper, the authors also examine the that have made great investments. Shareholders may require less compensation
pricing of accrual and OCF downside risk, and they also find that the pricing of for such firms due to their upside potentials, resulting in weak relation between
earnings downside risk mainly from its accrual downside risk component. OCF downside risk and expected stock returns.

2
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

expectations. Earnings expectation models directly reflect expectations expected stock returns.
of firm-specific risk in operating performance, indicating potential If earnings downside risk affects stock returns as it contains infor­
future operational losses and downward earnings patterns. The validity mation on firms’ future operating performance, we expect such effect to
of the earnings downside risk measure suggests a negative association be affected by firms’ information environment and governance efficacy.
between earnings downside risk and firms’ future operating perfor­ Among firms where information is more transparent and governance
mance, which leads to our first hypothesis: efficacy is stronger, the earnings downside risk measure, which is con­
structed based financial information that has been released by firms, is
H1. Earnings downside risk is negatively related to firms’ future
more likely to precisely capture information about firms’ future per­
operating performance.
formance, and vice versa. We thus expect the positive relation between
Extant literature suggests that information risk is non-diversifiable earnings downside risk and expected stock returns to be stronger among
and earnings uncertainties increase firm cost of capital (Eliwa, Has­ firms with more transparent information environment and stronger
lam, & Abraham, 2016; Lugo, 2019). More importantly, investors’ re­ governance efficacy. We therefore propose our third hypothesis as
actions to losses and gains are asymmetric. Roy (1952) suggests that follows:
individuals care more about downside losses than upside gain potentials.
H3. The positive relation between earnings downside risk and ex­
In proposing their prospect theory, Kahneman and Tversky (1979)
pected stock returns is stronger among firms with more transparent in­
contend that investors are more sensitive to losses than to gains and thus
formation environment and stronger governance efficacy.
exhibit a loss-aversion tendency. These studies suggest that individuals
assess negative and positive outcomes differently in making investment Hirshleifer et al. (2009) and Kang et al. (2010) have shown that re­
decisions, with more emphasis put on potential losses than on gains. turn predictors at the firm level, such as accruals and cash flows, can be
Earnings downside risk possesses the unique property of depicting aggregated to predict market returns. Similarly, we extend our investi­
negative shocks to firms’ future operating performance, and should thus gation of the pricing effect of accounting-based downside risk from the
affect investors’ expectations about firms’ future payoffs and expected individual stock level to market level. As mentioned previously, we
stock returns. Given that investors are averse to losses, we expect them expect earnings downside risk to contain information on firms’ future
to require firms with higher earnings downside risk to provide higher operating performance. When a large proportion of firms in the markets
premiums as a compensation for having to bear the downside risk in have high earnings downside risk, e.g., tend to have deteriorating
accounting performance. The above analysis leads to our second operating performance, we expect economic downturns to be around the
hypothesis: corner. In such periods, higher expected stock market returns are
required to attract investors to invest in the markets. We therefore
H2. Earnings downside risk is positively associated with firms’ ex­
expect higher aggregated earnings downside risk to be accompanied
pected stock returns.
higher expected market returns, and vice versa. These considerations
Earnings consist of accruals and OCF (Dechow, 1994; Dechow, lead to our last hypothesis as follows:
Kothari, & Watts, 1998; Sloan, 1996). Accordingly, earnings downside
H4. Aggregated earnings downside risk is significantly positively
risk can be decomposed into accrual and OCF downside risks (see Sec­
associated with expected stock market returns.
tion 3.2 for further details). The properties of accrual accounting dictate
that accrual downside risk should be better than OCF downside risk in
3. Data, measures, and summary statistics
predicting firms’ future operating performance. Accruals distinguish
earnings from mere operating cash flow. One major role of accruals is to
3.1. Data and sample
match expenses with revenues such that accruals provide more accurate
forecasts of future payoffs to assets than operating cash flow per se
Our sample includes all A-share companies listed on the Shanghai
(Callen & Segal, 2004). Another role of accruals is to recognize losses
and Shenzhen Stock Exchange from 2005 to 2018. The stock return data,
due to unanticipated downward revisions of expected cash flows, and do
accounting data and macroeconomic data are obtained from the China
so in a timely manner (Dechow, 1994; Dechow et al., 1998). In partic­
Stock Market and Accounting Research Database (CSMAR). We use
ular, negative accruals, the corollary of higher accrual downside risk,
Guidelines for the Industry Classification of Listed Companies (2012) for
can timely incorporate future losses and expenses into earnings as
industry classification. Our final sample includes 23,577 firm-year ob­
managers revise their expectations downward in anticipation of risks
servations representing 2649 individual firms. We winsorize the
inherent in firm operations and business environments. Thus, it is
continuous variables at the 1% and 99% levels to mitigate the effect of
reasonable to expect that investors put more weight on accrual down­
outliers.
side risk, which is ultimately reflected in the general earnings downside
risk, while less weight is put on operating cash flow downside risk.
3.2. Measures of earnings downside risk and its components
The influence of OCF downside risk on expected stock returns can be
weak and ambiguous also because it may be related with firms’ in­
We follow Konchitchki et al. (2016) to use relative root lower partial
vestment activities, and thus upside potentials. For example, R&D in­
moment (RRLPM) as a mathematical foundation for the calculation of
vestment, customer acquisition cost and marketing expenses are
accounting-based downside risk measures. We first follow them to
immediately expensed. These expenditures have dual effects on pre­
define unexpected ROA as the residual obtained from the ROA expec­
dicting future cash flow: (1) cash flow shortage reduces OCF level, which
tation model specified below:
increases OCF downside risk; and (2) these expenditures exploit firms’
growth opportunities, which enhances expected future OCF level and ROAit = α0 + α1 ROAit− 1 + α2 SALEit− 1 + α3 SIZEit− 1 + α4 LEVERAGEit− 1
firm value. Equity investors value the potential value-enhancing effects + α5 STD ROAit− 1 + α6 OCit− 1 + εit (1)
of high OCF downside risk driven by larger opportunities. If the value
enhancing effect cancels off or dominants cash insufficiency effect, OCF where ROA is the ratio of annual earnings to total assets. A set of control
downside risk may be even negatively associated with expected stock variables are included in Eq. (1) to generate the expected level of ROA
returns. for a firm in a specific year, including ROA of the previous period, sales
Therefore, we separately examine the pricing of accrual downside (SALE), firm size (SIZE), leverage (LEVERAGE), ROA volatility over past
risk and OCF downside risk in the empirical investigation, and expect three years (STD_ROA), and operating cycle (OC). The residual of Eq. (1),
the positive relation between accrual downside risk and expected stock or εit, is the unexpected ROA. The regression is performed within in­
returns to be stronger than that between OCF downside risk and dustries over a three-year rolling window.

3
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

The earnings downside risk DR_ROA is calculated as the RRLPM of window. The εit of Eq. (3) captures unexpected accruals, and its RRLPM
unexpected ROA, which is the natural logarithm of the ratio of 1 plus is used to proxy accrual downside risk and is denoted as DR_TCA. The
RLPM2(ROAit) to 1 plus RUPM2(ROAit): calculation of the RRLPM of unexpected accruals is similar to the
( ) calculation of the RRLPM of unexpected ROA as described in Eq. (2).
1 + RLPM2 (ROAit )
DR ROAit = log (2) Lastly, we obtain unexpected OCF from the following OCF expecta­
1 + RUPM2 (ROAit )
tion model below:
[( ) ]1/2
∑ ´ OCF it = γ0 + γ 1 OCFit− 1 + γ 2 OCFit− 2 + γ 3 OCFit− 3 + γ 4 SALEit− 1 + γ5 SIZEit−
where RLPM2 (ROAit ) = 1
(εit *Iε´it ≤0 )2 , representing the root
1
5 + γ6 LEVERAGEit− 1 + γ7 STD OCF it− 1 + εit (4)
lower partial moment of ROA, and RUPM2 (ROAit ) =
[( )
∑ ´
]1/2 where OCF is the ratio of annual operating cash flow to the total assets,
1
5 (εit *Iε´it ≥0 )2 , measuring the root upper partial moment of and STD_OCF is the standard deviation of OCF over the previous three
years. Specifications of other variables are the same as in Eq. (1). The
ROA. Iε´it ≤0 is a dummy variable that equals one if ε´it < 0 and zero
regression is also performed within industries over a three-year rolling
otherwise, where ε´it is the residual from the ROA expectation regression window. The εit of Eq. (4) captures unexpected OCF, and its RRLPM is
model specified in Eq. (1). used to proxy OCF downside risk and is denoted as OCF_TCA. The
Similarly, we obtain unexpected accrual from the following accrual calculation of the RRLPM of unexpected OCF is similar to the calculation
expectation model: of the RRLPM of unexpected ROA as described in Eq. (2).
TCAit = β0 + β1 TCAit− 1 + β2 SALEit− 1 + β3 SIZEit− 1 + β4 LEVERAGEit− 1
3.3. Summary statistics
+ β5 STD TCAit− 1 + β6 OCit− 1 + εit (3)

where TCA is the ratio of total accrual to total assets and STD_TCA is the Panel A of Table 1 reports the descriptive statistics for the main
standard deviation of total accrual ratio over the previous three years. variables used in the analysis. The standard deviation of earnings
The definitions of other variables are the same as in Eq. (1). The downside risk (DR_ROA), accrual downside risk (DR_TCA), and OCF
regression is also performed within industries over a three-year rolling downside risk (DR_OCF) is 0.045, 0.057 and 0.041, respectively, indi­
cating high variation of accounting-based downside risk measures.

Table 1
Summary statistics.
Variable N Mean Min 25% Median 75% Max Std

Accounting-based downside risk

DR_ROAit 23,577 0.004 − 0.098 − 0.018 − 0.001 0.017 0.249 0.045
DR_TCAit 24,693 0.003 − 0.145 − 0.028 0.000 0.029 0.280 0.057
DR_OCFit 21,710 0.001 − 0.122 − 0.023 0.002 0.024 0.122 0.041

Operating performance measures

DLOSS1it 23,616 0.128 0.000 0.000 0.000 0.000 1.000 0.334
DLOSS2it 23,616 0.190 0.000 0.000 0.000 0.000 1.000 0.393
IBMit 23,580 0.056 − 2.636 0.021 0.066 0.143 0.851 0.331
NIMit 23,580 0.037 − 2.599 0.016 0.053 0.118 0.762 0.318
OPMit 23,580 0.029 − 2.670 0.011 0.055 0.129 0.647 0.346
GPMit 23,580 − 0.014 − 2.907 − 0.007 0.037 0.101 0.452 0.373

Expected stock returns

RETit-Rfit 369,781 0.011 − 0.326 − 0.074 0.000 0.084 0.477 0.142

Earnings Attributes
ACC_Qit 25,188 0.018 0.000 0.003 0.007 0.016 0.313 0.041
Persistenceit 25,770 − 0.229 − 3.619 − 0.537 − 0.155 0.178 1.865 0.729
Predictit 25,770 0.167 0.000 0.051 0.105 0.207 1.088 0.191
Smoothit 28,990 0.975 0.038 0.318 0.627 1.153 8.222 1.192
Relevanceit 26,372 − 0.675 − 1.000 − 0.935 − 0.755 − 0.460 − 0.023 0.291
Timelinessit 26,372 − 0.739 − 1.000 − 0.986 − 0.857 − 0.555 − 0.018 0.288
Conservatismit 26,372 0.419 − 193.666 − 1.000 − 1.000 0.097 246.087 39.446

Return downside risk (stock price crash risk)

NCSKEWit 225,939 − 0.236 − 4.563 − 0.622 − 0.212 0.178 6.153 0.741
DUVOLit 225,939 − 0.335 − 3.979 − 0.796 − 0.343 0.116 8.432 0.716

Other control variables

LEVERAGEit 31,158 0.471 0.057 0.295 0.463 0.624 1.513 0.240
SIZEit 31,013 22.192 18.307 21.510 22.198 22.902 25.079 1.174
BMit 28,622 0.434 0.035 0.220 0.363 0.572 1.422 0.290
ROAit 29,972 0.031 − 0.378 0.012 0.033 0.063 0.199 0.073
CASHit 31,158 0.174 0.005 0.084 0.139 0.227 0.619 0.128
Δ CASHit 31,158 0.007 − 0.275 − 0.033 0.005 0.045 0.323 0.090
Invest_CAPEXit 31,157 0.048 − 0.057 0.011 0.033 0.070 0.254 0.053
OOit 31,158 0.239 0.002 0.099 0.202 0.344 0.750 0.177
SIGMAit 31,010 0.030 0.013 0.023 0.029 0.036 0.059 0.010
MOMit 363,257 0.207 − 0.657 − 0.237 0.013 0.427 3.038 0.682
MKTbetait 367,561 0.955 − 2.436 0.783 0.978 1.169 3.599 0.632
SMBbetait 367,561 0.876 − 2.706 0.368 0.849 1.266 6.529 1.089
HMLbetait 367,561 − 0.135 − 5.929 − 0.687 − 0.051 0.500 4.925 1.349
UMDbetait 367,561 − 0.048 − 0.809 − 0.046 − 0.016 0.001 0.098 0.125

This table reports the summary statistics of variables used in the empirical investigation. See Appendix A for variable descriptions.

4
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

We have also reported statistics for variables use in the empirical

This table reports the correlation among major variables used in the empirical investigation. Correlations in bold are significantly different from zero at the 1% or 5% level. See Appendix A for variable descriptions.
Timeliness
investigation in Table 1. Variables related to firms’ operating perfor­

− 0.0012
mance include indicators for negative annual income before extraordi­
nary items (DLOSS1) and for negative annual income (DLOSS1), net
income scaled by total revenue (NIM), operating income after depreci­
ation scaled by total revenue (OPM), and gross profit margin defined as

0.0342
0.0091
Smooth
the difference between total revenues and cost of goods sold divided by
total revenue (GPM). The expected stock return is captured by excess
stock return, or RETit-Rfit. Variables related to earnings attributes are

Relevance
constructed following Francis, LaFond, Olsson, and Schipper (2004),

0.0414
0.3791
0.0025
including accrual quality (ACC_Q), earnings persistence, predictability,
earnings smoothness, relevance, timeliness, and conservatism. Variables
related to stock return downside risk, or stock price crash risk, include

0.0248
0.2505
0.0240
0.0056
negative coefficient of skewness (NCSKEW) and down-to-up volatility

Predict
(DUVOL). The construction of all these variables are described in details
in Appendix A.
Other firm characteristic variables include the natural logarithm of

Persistence

¡0.0225
market value of equity at the end of the fiscal year (SIZE), leverage ratio

0.0612

0.0080

0.0125
0.0031
measured by total liabilities over total assets (LEVERAGE), the ratio of
cash holdings plus cash equivalents to total assets (Cash) and the change
in cash holdings plus cash equivalents to total assets (∆Cash), the ratio of

− 0.0001
capital expenditures to total assets (Invest_CAPEX), the ratio of property,

0.0110
0.0324

0.0099
0.0017

0.0015
ACC_Q
plant, and equipment to total assets (OO), the standard deviation of daily
stock returns for the fiscal year (SIGMA), monthly return over an 11-
month period that ends one-month prior to month t (MOM), and stock

¡0.0233
¡0.0284
¡0.0173
loadings on market, size, book-to-market, and momentum factors of

0.0052
0.0051
0.0084

0.0008
Fama and French (1993) and Carhart (1997) (MKTbeta, SMBbeta,

TCA
HMLbeta, UMDbeta).
Table 2 reports the correlation matrix among main variables used in

¡0.0429

¡0.0176
¡0.0271
¡0.0225
¡0.0159
¡0.0092
¡0.0067
− 0.0010
the empirical investigation. Both earnings downside risk and its accrual
downside risk component are significantly positively associated with MOM
expected stock returns. The correlation between OCF downside risk and
expected stock returns, however, is significantly negative. The evidence
LEVERAGE

is consistent with our expectation that firms with higher earnings or

¡0.0709

¡0.0859
− 0.0039

− 0.0005
− 0.0029
0.0511

0.1404
0.0014
0.0017
accrual downside risk are likely to have higher expected returns to
compensate investors for bearing the risk the firms’ future performance
may fall short of expectation. The relation between OCF downside risk
and expected stock returns is ambiguous. Such ambiguity might result ¡0.0250
¡0.0130
¡0.0336

¡0.0773
¡0.0622
¡0.0047
0.0795
0.1489

0.0133
0.0065
from the fact that OCF downside risk contains noisy information about
BMi

firms’ future performance as OCF does not match revenues and expenses
properly. Moreover, higher OCF downside risk implies that firms’ might
have made large investments which results in large cash flow declines.
¡0.0492

¡0.0247
− 0.0041

− 0.0031
− 0.0025

− 0.0040
In such cases, higher OCF downside risk is more likely to be associated
0.0536
0.0274

0.0494

0.0070
0.0007
SIZE

with greater upside potentials for firms, and investors are likely to
require lower risk compensation for such firms. When investors feel that
the value creating effect following large investments dominants the cash
¡0.0243

¡0.0262
− 0.0009
DR_OCFi

insufficiency effect, OCF downside risk could be negatively associated

0.1086

0.3152
0.0082
0.0292
0.0115
0.0139
0.0452
0.0057
0.0005

with expected stock returns.

4. Empirical results
¡0.6126

¡0.0058

¡0.3897
¡0.0378

¡0.0280

− 0.0022
DR_TCA

0.0234
0.0169

0.0176

0.0105

0.0499
0.0435
0.0267

4.1. Earnings downside risk and subsequent operating performance

In Table 3, we examine the relation between earnings downside risk

¡0.0081

¡0.3045
¡0.0307

¡0.0293

and subsequent operating performance in both univariate and multi­

DR_ROA

0.6613
0.0388

0.0153
0.1248
0.0052

0.0481

0.0768
0.0924
0.0406
0.0003

variate settings, where firms’ operating performance is regressed on

lagged earnings downside risk. Specifically, we estimate the following
model:
¡0.0065
¡0.0190

¡0.0142

¡0.0059
¡0.0096
RETit-Rfit

− 0.0030

∑
0.0072
0.0081

0.0524
0.0190
0.0369

0.0052

0.0024

0.0015
0.0012

Performanceit = α0 + α1 DR ROAit− 1 + αk Controlsit− 1 + εit (5)

Correlation matrix.

where Performance refers to firm operating performance variables.

Following Konchitchki et al. (2016), we adopt the following firm oper­
Conservatism
LEVERAGE

Persistence

ating performance measures: DLOSS1 is an indicator variable that takes

Timeliness
Relevance
DR_OCFi
DR_ROA
DR_TCA

Smooth
Table 2

ACC_Q

Predict

a value of one if annual income before extraordinary items is negative

MOM
SIZE

TCA
BMi

and zero otherwise; DLOSS2 is an indicator variable that equals one if

5
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

Table 3
Earnings downside risk and subsequent operating performance.
DLOSS1it DLOSS2it IBMit NIMit OPMit GPMit

Panel A: Univariate analysis

DR_ROAit-1 0.133 0.837*** − 0.138 − 0.124 − 0.970*** − 1.283***
(1.604) (8.482) (− 0.910) (− 0.841) (− 5.443) (− 6.707)
Intercept 0.132*** 0.169*** − 0.002 − 0.010 0.016 − 0.000
(12.897) (15.071) (− 0.184) (− 0.824) (1.495) (− 0.026)
Year FE Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes
Adjusted R2 0.012 0.020 0.011 0.010 0.019 0.027
N 20,986 20,986 20,955 20,955 20,955 20,955

Panel B: Multivariate analysis

DR_ROAit-1 0.075 0.379*** − 0.431*** − 0.427*** − 0.562*** − 0.751***
(0.744) (3.340) (− 2.997) (− 3.010) (− 3.975) (− 4.885)
SIZEit-1 − 0.118*** − 0.133*** 0.145*** 0.129*** 0.143*** 0.143***
(− 3.420) (− 3.222) (4.767) (4.564) (4.729) (4.649)
BMit-1 0.099*** 0.119*** − 0.102*** − 0.092*** − 0.086*** − 0.052***
(5.985) (6.280) (− 6.246) (− 5.814) (− 5.367) (− 3.007)
CASHit-1 − 0.217*** − 0.180*** 0.215*** 0.206*** 0.239*** 0.268***
(− 5.222) (− 3.705) (4.756) (4.679) (5.670) (5.125)
ΔCASHit-1 0.007 − 0.076* 0.010 0.004 0.003 0.028
(0.183) (− 1.939) (0.237) (0.099) (0.080) (0.682)
Invest_CAPXit-1 − 0.233*** − 0.148** 0.082 0.080 0.063 0.080
(− 3.329) (− 1.964) (1.206) (1.211) (0.940) (1.039)
LEVERAGEit-1 0.125*** 0.236*** − 0.109*** − 0.092*** − 0.136*** − 0.129***
(4.429) (7.256) (− 3.081) (− 2.706) (− 4.217) (− 3.771)
OOit-1 0.063* 0.137*** 0.023 0.034 0.044 0.094**
(1.808) (3.284) (0.628) (0.946) (1.120) (2.400)
SIGMAit-1 − 0.967* − 1.665*** 0.238 0.226 0.661 0.685
(− 1.708) (− 2.723) (0.457) (0.447) (1.352) (1.343)
Intercept 0.034 − 0.010 0.101*** 0.072** 0.089*** 0.024
(1.035) (− 0.276) (2.935) (2.178) (2.715) (0.698)
Year FE Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes
Adjusted R2 0.024 0.036 0.024 0.022 0.028 0.032
N 19,913 19,913 19,907 19,907 19,907 19,907

This table reports the results of regressing firms’ future operating performance on earnings downside risk DR_ROA. See Appendix A for variable; descriptions. The t-
statistics are in parentheses.
*
p < 0.10.
**
p < 0.05.
***
p < 0.01.

annual net income is negative and zero otherwise; IBM is the ratio of When the control variables in Table 3 are examined, their association
annual income before extraordinary items to total revenues; NIM is the with firms’ future operating performance is as expected. For example,
ratio of annual net income to total revenues; OPM is the ratio of annual large firms, growth firms, firms with abundant cash flows, and firms
operating income after depreciation to total revenues; GPM refers to with low financial leverage tend to have better future operating per­
annual gross profit margin ratio, i.e., the difference between total rev­ formance: their tendencies of occurring losses are lower and their future
enues and cost of goods sold divided by total revenues. The control profit margins are likely to be higher.
variables are introduced in Section 3.3 and in Appendix A. We measure
all of the independent variables in year t-1, a one-year lag from the 4.2. Earnings downside risk and expected stock returns
dependent variable, allowing us to examine whether earnings downside
risk can predict future operating performance. We also include industry Next, we examine the link between earnings downside risk and ex­
and year dummy variables to control for industry and year fixed effects, pected stock returns by performing portfolio analysis and pooled cross-
and we cluster standard errors at the firm level. sectional OLS regressions. In the first approach, we compare the
The results show that earnings downside risk is positively correlated magnitude of average monthly return of portfolios constructed on the
with loss indicators and negatively correlated with profit margin vari­ basis of DR_ROA. We sort stocks into deciles based on the DR_ROA es­
ables in both univariate and multivariate regressions. Results remain the timates at the end of each year and calculate the equal-weighted port­
same when control variables are added into the regressions. The findings folio returns starting six months after the fiscal year-end for 12 months
are consistent with H1 that high earnings downside risk is indicative of a after the fiscal year-end. Panel A of Table 4 reports the monthly returns
firm’s deteriorating future operating performance. These pieces of evi­ across the DR_ROA portfolios. Firms in the high DR_ROA decile portfo­
dence confirm that even in the Chinese A-share markets, where infor­ lios tend to have higher returns relative to firms in the low DR_ROA
mation environment is less transparent and governance efficacy is lower decile portfolios. The monthly return spread between deciles with the
compared to the U.S. markets in general, earnings downside risk still largest and smallest DR_ROA values is 0.3%, which is significantly
contains information about firm’s future operating performance. different from zero at the 10% level.

6
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

Table 4
Earnings downside risk and expected stock returns.
DR_ROA Rank N Return DR_ROA

Panel A: Portfolio analysis of DR_ROA and subsequent average monthly returns

1 23,692 0.013 − 0.058

2 23,729 0.012 − 0.029
3 23,669 0.013 − 0.019
4 23,688 0.012 − 0.011
5 23,708 0.013 − 0.005
6 23,713 0.012 0.001
7 23,724 0.014 0.008
8 23,676 0.012 0.017
9 23,682 0.013 0.030
10 23,702 0.015 0.082
H-L 0.003
-statistics (1.89)

Panel B: Regression of monthly expected stock returns on DR_ROA

Model 1 Model 2 Model 3 Model 4 Model 5

DR_ROAit-1 0.054*** 0.036*** 0.039*** 0.036*** 0.036***

(7.500) (3.222) (3.463) (3.181) (3.176)
SIZEit-1 − 0.090*** − 0.089*** − 0.084*** − 0.085***
(− 15.510) (− 15.369) (− 14.483) (− 14.689)
BMit-1 0.013*** 0.014*** 0.017*** 0.016***
(8.214) (8.768) (10.250) (9.733)
LEVERAGEit-1 0.005 0.005* 0.004 0.004
(1.626) (1.706) (1.335) (1.425)
MOMit-1 − 0.035*** − 0.035*** − 0.034*** − 0.035***
(− 51.583) (− 51.735) (− 49.714) (− 50.270)
MKTbetait-1 − 0.001 − 0.001 − 0.000 − 0.000
(− 0.418) (− 0.323) (− 0.089) (− 0.194)
SMBbetait-1 0.002** 0.002** 0.002** 0.002***
(2.102) (2.009) (2.448) (2.607)
HMLbetait-1 0.000 0.000 0.000 0.000
(0.344) (0.494) (0.305) (0.201)
UMDbetait-1 − 0.045*** − 0.040*** − 0.033*** − 0.033***
(− 3.937) (− 3.493) (− 2.893) (− 2.901)
TCAit-1 − 0.025*** − 0.026*** − 0.026*** − 0.026***
(− 6.180) (− 6.489) (− 6.420) (− 6.551)
ACC_Qit-1 0.078*** 0.078*** 0.078***
(4.150) (4.205) (4.215)
Persistenceit-1 0.000 0.001 0.000
(0.819) (1.257) (1.104)
Predictit-1 − 0.010*** − 0.009*** − 0.009***
(− 3.350) (− 2.973) (− 3.129)
Relevanceit-1 0.001 0.001 0.001
(0.785) (1.131) (1.073)
Smoothit-1 − 0.001 − 0.001 − 0.001
(− 1.325) (− 1.284) (− 1.320)
Timelinessit-1 0.005*** 0.005*** 0.005***
(4.153) (4.051) (4.068)
Conservatismit-1 0.000 0.000 0.000
(0.919) (0.837) (0.847)
DUVOLit-1 0.009***
(19.066)
NCSKEWit-1 0.005***
(12.637)
Intercept − 0.011*** − 0.028*** − 0.023*** − 0.022*** − 0.023***
(− 14.981) (− 10.800) (− 8.255) (− 8.106) (− 8.441)
Year FE Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes
Adjusted R2 0.084 0.100 0.100 0.101 0.100
N 244,692 231,211 230,572 225,939 225,939

Panel A reports the average expected stock returns for deciles formed based on earnings downside risk. Panel B reports the results of regressing firms’
expected stock returns on earnings downside risk as well as a set of control variables. See Appendix A for variable descriptions. The t-statistics are in
parentheses.
*
p < 0.10.
**
p < 0.05.
***
p < 0.01.

7
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

In the second approach, we regress subsequent monthly excess Table 5

returns on current earnings downside risk: Components of earnings downside risk.
∑ DR_ROAit DR_ROAit DR_ROAit DR_ROAit DR_ROAit
RETit − Rfit = β0 + β1 DR ROAit− 1 + βk Controlsit− 1 + εit (6)
DR_TCAit 0.516*** 0.737*** 0.731*** 0.731***
(33.506) (56.150) (61.997) (61.906)
where RETit − Rfit refers to monthly excess returns. We include three sets
DR_OCFit 0.205*** 0.700*** 0.701*** 0.701***
of control variables to assess DR_ROA’s incremental ability to explain the (9.231) (44.624) (53.643) (53.600)
variation in expected stock returns. The first set includes variables CASHit-1 − 0.009*** − 0.009***
related to cross-sectional variation in returns and are common factors in (− 2.676) (− 2.712)
asset pricing tests, including firm size, book-to-market ratio, leverage, ΔCASHit-1 0.001 0.001
(0.595) (0.622)
past returns, and stocks’ loadings to the market, size, book-to-market, Invest_CAPEXit- − 0.009* − 0.009*
and momentum factors. The second set of control variables are the 1
earning attributes introduced by Francis et al. (2004), including accrual (− 1.753) (− 1.786)
quality, earnings persistence, predictability, value relevance, smooth­ BMit-1 0.003** 0.003**
(2.314) (2.323)
ness, timeliness, and conservative. The third set of control variables are
LEVERAGEit-1 0.006*** 0.006***
stock price crash risk measures NSCKEW and DUVOL. The construction (2.814) (2.798)
of these variables are described in details in Appendix A. DLOSS1it-1 − 0.001** − 0.001**
Panel B of Table 4 reports the results of pooled OLS regressions.3 (− 2.020) (− 2.017)
When monthly excess returns are regressed on DR_ROA alone in Model ROAit-1 − 0.048*** − 0.048***
(− 7.846) (− 7.839)
1, the coefficient on DR_ROA is 0.054 (t = 7.5), which is significantly OOit-1 − 0.001 − 0.001
different from zero at the 1% level. The result suggest that a one stan­ (− 0.330) (− 0.335)
dard deviation (0.045) increase in DR_ROA leads to a 0.243% increase in SIGMAit-1 0.028 0.034
monthly excess stock returns. Results are qualitatively unchanged when (0.530) (0.630)
SIZEit-1 − 0.006** − 0.006**
different sets of control variables are included, with common control
(− 2.307) (− 2.275)
variables in asset pricing tests and firm fundamentals controlled in NCSKEWit-1 0.000
Model 2, earning attributes controlled in Model 3, and additional con­ (1.312)
trols for stock price crash risk added in Models 4 and 5. The findings DUVOLit-1 0.000*
indicate that DR_ROA conveys incremental information for expected (1.734)
Intercept 0.002* 0.010*** 0.002* − 0.000 − 0.000
stock returns after controlling for conventional predictors for stock (1.660) (5.850) (1.704) (− 0.085) (− 0.126)
returns. The results render strong support for our hypothesis H2. Year FE Yes Yes Yes Yes Yes
The results of Tables 3 and 4 confirm that even in Chinese stock Firm FE Yes Yes Yes Yes Yes
markets, where information quality is lower than that in the U.S. mar­ Adjusted R2 0.451 0.040 0.751 0.752 0.751
N 23,436 20,474 20,474 19,400 19,396
kets, earnings downside risk still contains useful information that can be
used to predict firms’ future operating performance, and thus such risk is This table reports the results of regressing earnings downside risk on its accrual
priced in the Chinese stock markets as in the U.S. markets documented downside risk and OCF downside risk components with a set of control variables.
by Konchitchki et al. (2016). However, the results in Table 4 also reveal See Appendix A for variable descriptions. The t-statistics are in parentheses.
*
that the effect of earnings downside risk on expected stock returns, p < 0.10.
**
p < 0.05.
although being highly significant, is relatively smaller in magnitude ***
p < 0.01.
compared to that in the U.S. markets, which could due to the relatively
lower information quality in the Chinese markets. Result of the uni­
variate regression shown in column 1 of Table 4 reveals that a one where Controls includes CASH, ΔCASH, Invest_CAPEX, BM, LEVERAGE,
standard deviation increase in earnings downside risk, which is 0.045, DLOSS1, ROA, OO, SIGMA, SIZE, NCSKEW, and DUVOL. These variables,
will lead to an increase in monthly stock return of 0.045*0.054 = 0.24%. with definitions detailed in Section 3.2 and Appendix A, capture possible
The regression results in Konchitchki et al. (2016), in comparison, show sources of earnings downside risk that have been discussed in existing
that a one standard deviation increase in earnings downside risk (0.079) literature.
would lead to an increase in monthly stock return of 0.079*0.04088 = Table 5 provides the results of regressing DR_ROA on its two com­
0.32% in the U.S. markets, which is relatively higher than that in the ponents. The evidence shows that DR_ROA is positively correlated with
Chinese markets. both DR_TCA and DR_OCF and the regression coefficients are significant
at the 1% level across all columns, implying that DR_ROA shares great
level of overlapping risk information with DR_TCA and DR_OCF
4.3. Components of earnings downside risk
measures.
The explanatory power of independent variables in the regression
Earnings are comprised of two components, accruals and OCF
model provides a formal test of the extent to which DR_TCA and DR_OCF
(Dechow, 1994; Dechow et al., 1998; Sloan, 1996), which can contribute
can collectively explain the variation in DR_ROA. As evidenced in
differently to earnings downside risk due to the accrual accounting
Models 3–5, adjusted R2 from estimating Eq. (4) is greater than 70%,
system. Section 3.2 illustrates the methods we adopt to compute two
indicating that the risk information in DR_ROA is largely subsumed by
components of earnings downside risk. To test whether information in
DR_TCA and DR_OCF.
DR_ROA can be subsumed by DR_TCA and DR_OCF, we next examine the
contemporaneous associations of DR_ROA with DR_TCA and DR_OCF.
We estimate the following OLS regression model: 4.4. Components of earnings downside risk and expected stock returns
∑
DR ROAit = β0 + β1 DR TCAit + β2 DR OCF it + βk Controlsit− 1 + εit (7)
As discussed in Section 2, accruals distinguish earnings from mere
operating cash flow. A major role of accruals is to match expenses with
revenues when they are actually realized (Dechow, 1994, Dechow et al.,
1998). OCF, however, considers cash inflows and outflows only when
3
The results are qualitatively similar when Fama-Macbeth regressions are they occur without matching them properly to periods when revenues or
employed. expenses are recognized. In other words, although OCF is less likely to be

8
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

Table 6
Components of earnings downside risk and expected stock returns.
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

DR_ROAit-1 0.039*** 0.036***

(3.463) (3.176)
DR_TCAit-1 0.029*** 0.037*** 0.026*** 0.036***
(3.748) (3.611) (3.367) (3.515)
DR_OCFit-1 − 0.008 0.010 − 0.007 0.010
(− 0.875) (0.789) (− 0.742) (0.823)
SIZEit-1 − 0.089*** − 0.091*** − 0.086*** − 0.085*** − 0.085*** − 0.086*** − 0.081*** − 0.081***
(− 15.369) (− 15.632) (− 13.148) (− 13.071) (− 14.689) (− 14.918) (− 12.429) (− 12.357)
BMit-1 0.014*** 0.014*** 0.012*** 0.013*** 0.016*** 0.016*** 0.014*** 0.014***
(8.768) (8.916) (7.126) (7.287) (9.733) (9.859) (8.014) (8.173)
LEVERAGEit-1 0.005* 0.007** 0.008*** 0.008** 0.004 0.006** 0.007** 0.007**
(1.706) (2.464) (2.700) (2.386) (1.425) (2.122) (2.470) (2.153)
MOMit-1 − 0.035*** − 0.035*** − 0.034*** − 0.034*** − 0.035*** − 0.035*** − 0.034*** − 0.034***
(− 51.735) (− 51.636) (− 47.196) (− 47.335) (− 50.270) (− 50.171) (− 46.003) (− 46.141)
MKTbetait-1 − 0.001 − 0.000 0.001 0.001 − 0.000 − 0.000 0.001 0.001
(− 0.323) (− 0.246) (0.468) (0.409) (− 0.194) (− 0.120) (0.814) (0.755)
SMBbetait-1 0.002** 0.002** 0.001 0.001 0.002*** 0.002*** 0.002** 0.002*
(2.009) (2.150) (1.559) (1.317) (2.607) (2.747) (2.030) (1.783)
HMLbetait-1 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
(0.494) (0.408) (0.546) (0.687) (0.201) (0.114) (0.332) (0.475)
UMDbetait-1 − 0.040*** − 0.041*** − 0.053*** − 0.052*** − 0.033*** − 0.034*** − 0.047*** − 0.046***
(− 3.493) (− 3.556) (− 4.139) (− 4.059) (− 2.901) (− 2.963) (− 3.680) (− 3.596)
TCAit-1 − 0.026*** − 0.023*** − 0.026*** − 0.021*** − 0.026*** − 0.023*** − 0.026*** − 0.022***
(− 6.489) (− 5.286) (− 5.868) (− 4.553) (− 6.551) (− 5.411) (− 5.989) (− 4.701)
ACC_Qit-1 0.078*** 0.083*** 0.065*** 0.062*** 0.078*** 0.082*** 0.064*** 0.062***
(4.150) (4.379) (3.257) (3.201) (4.215) (4.406) (3.247) (3.176)
Persistenceit-1 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.001
(0.819) (0.830) (0.830) (0.780) (1.104) (1.111) (1.182) (1.134)
Predictit-1 − 0.010*** − 0.010*** − 0.008** − 0.009*** − 0.009*** − 0.010*** − 0.007** − 0.008***
(− 3.350) (− 3.450) (− 2.343) (− 2.662) (− 3.129) (− 3.247) (− 2.275) (− 2.589)
Relevanceit-1 0.001 0.001 − 0.001 − 0.001 0.001 0.001 − 0.000 − 0.000
(0.785) (0.764) (− 0.510) (− 0.597) (1.073) (1.047) (− 0.301) (− 0.389)
Smoothit-1 − 0.001 − 0.000 − 0.000 − 0.000 − 0.001 − 0.000 − 0.000 − 0.000
(− 1.325) (− 1.068) (− 0.807) (− 1.129) (− 1.320) (− 1.085) (− 0.852) (− 1.165)
Timelinessit-1 0.005*** 0.005*** 0.004*** 0.004*** 0.005*** 0.005*** 0.004*** 0.004***
(4.153) (4.169) (3.288) (3.311) (4.068) (4.084) (3.125) (3.150)
Conservatismit-1 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
(0.919) (0.893) (1.064) (1.076) (0.847) (0.822) (0.909) (0.916)
NCSKEWit-1 0.005*** 0.005*** 0.004*** 0.004***
(12.637) (12.584) (10.664) (10.717)
Intercept − 0.023*** − 0.024*** − 0.006** − 0.006* − 0.023*** − 0.024*** − 0.007** − 0.006*
(− 8.255) (− 8.837) (− 1.966) (− 1.747) (− 8.441) (− 9.008) (− 2.154) (− 1.931)
Year FE Yes Yes Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes Yes Yes
Adjusted R2 0.100 0.100 0.100 0.100 0.100 0.100 0.100 0.100
N 230,572 230,118 191,837 191,837 225,939 225,494 188,096 188,096

This table reports the results of regressing monthly expected stock returns on accrual and OCF downside risks with a set of control variables. See Appendix A for
variable descriptions. The t-statistics are in parentheses.
*
p < 0.10.
**
p < 0.05.
***
p < 0.01.

manipulated by corporate managers, its information content on firms’ that the relationship between DR_ROA and expected returns are largely
future operating performance may be nosier than that of accruals due to driven by DR_TCA. One possible explanation for these results is that
potential mismatching problems. Therefore, we expect that accruals and accruals improve earnings’ ability to reflect firms’ operating perfor­
OCF downside risk to have different pricing effects. Examining the im­ mance by altering the time of cash flow recognition in earnings.
plications of these two components for predicting expected stock returns Compared with OCF, accruals incorporate future losses and expenses
will advance our understanding of the drivers of earnings downside into earnings much more rapidly and thus provide accurate information
risk’s effect on expected stock returns. Therefore, in this section, we for future downward earnings. Thus, investors put more weight on
regress monthly excess returns on accrual and OCF downside risk DR_TCA than DR_OCF when assessing firms’ future risk and returns
separately, and the results are reported in Table 6. (Sloan, 1996). Overall, the results provide support for our conjecture
Results in Table 6 show that monthly excess returns is significantly that the pricing effect of DR_ROA is largely attributable to DR_TCA.
related to lagged DR_TCA and DR_ROA, but its relationship with DR_OCF
is insignificant. Specifically, in Models 2 and 4, after controlling for
4.5. Cross-sectional variation
conventional return predictors, the coefficients on DR_TCA are 0.029
and 0.0037, with t-values of 3.748 and 3.611, respectively, which are
If earnings downside risk is priced in stock markets as it contains
economically and statistically significant. When stock price crash risk is
information on firms’ future operating performance, as discussed in
also added in Models 5–8, results are qualitatively similar to those in
Section 4.1 and shown in Table 3, we expect such a pricing effect to be
Models 1–4, suggesting that investors demand a higher risk premium for
affected by firms’ information environment and governance efficacy.
holding stocks with higher DR_ROA or DR_TCA. The results also indicate
Among firms with more transparent information environment and more

9
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

effective governance systems, the released financial information is more Table 7

reliable. In such firms, the earnings downside risk measure constructed Cross-sectional variation.
based on publicly disclosed firm financial information contains more Panel A: Subsamples formed based on firm size
precise information about firms’ future performance, and thus its rela­
Model 1 Model 2 Model 3 Model 4
tion with expected stock returns should be stronger.
We therefore perform subsample tests in this subsection. We parti­ Size ≤ Median Size > Median

tion our sample firms based on firm size, industry, firm age, or number DR_ROAit-1 0.011 0.013 0.024** 0.0245**
of analysts following. Firms with size larger (smaller) than sample me­ (1.028) (1.191) (2.042) (2.067)
NCSKEWit-1 − 0.001* − 0.004***
dian are viewed as large (small) firms. We also divided sample firms
(− 1.951) (− 6.311)
based on whether they belong to high-tech industries or not. High-tech Intercept 0.054*** 0.053*** − 0.018*** − 0.019***
firms include those in cutting-edge technology industries such as (11.334) (11.301) (− 4.760) (− 5.032)
biomedicine, chemical industry, machinery, automobile, electric Controls as in Table 4 Yes Yes Yes Yes
equipment, new energy, electronics, computer science, electric appli­ Firm FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
ance, and communication engineering. All other firms are classified as in Adjusted R2 0.019 0.019 0.019 0.019
non-high-tech industries. Firms in high-tech industries are likely to have N 122,020 112,664 122,672 113,275
more intangible assets and are more difficult to value based on financial
statements than firms in non-high-tech industries. Further, we classify Panel B: Subsamples formed based on industries
firms into old and young ones conditional on whether they have been
Model 1 Model 2 Model 3 Model 4
listed for more than 5 years. Lastly, we categorize sample firms condi­
tional on whether the number of analysts following is below or above Firms in high-tech Firms in non-high-tech
industries industries
sample median.
We expect larger firms, firms belong to non-high-tech industries, DR_ROAit-1 0.019 0.019 0.024*** 0.028***
older firms, and firms with more analysts following to have more (1.358) (1.320) (2.776) (3.126)
NCSKEWit-1 − 0.002*** − 0.002***
transparent information environment and more effective governance
(− 2.846) (− 4.180)
systems than their counterparts. The positive relation between earnings Intercept 0.007* 0.007 0.015*** 0.015***
downside risk and expected stock returns, therefore, is expected to be (1.755) (1.625) (4.837) (4.689)
stronger among such firms. The results are shown in Table 7. Controls as in Table 4 Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Panels A to D of Table 7 report the results for subsamples constructed
Year FE Yes Yes Yes Yes
based size, industry, age, and analyst coverage, respectively. Consistent Adjusted R2 0.018 0.018 0.015 0.016
with our expectation, the relation between earnings downside risk and N 91,074 89,252 139,498 136,687
expected stock returns remain significantly positive in subsamples of
larger firms, non-high-tech firms, older firms, and firms with more an­ Panel C: Subsamples formed based on firm age
alysts following. In smaller firms, high-tech firms, younger firms, and
Model 1 Model 2 Model 3 Model 4
firms with fewer analysts following, however, the relation between
earnings downside risk and expected returns is either weakly signifi­ Age ≤ 5 Age > 5
cantly positive or turns to be insignificant. The results are supportive of DR_ROAit-1 − 2.869 1.375 0.045* 0.045*
our Hypothesis H3 that the pricing effect of earnings downside risk is (− 0.353) (0.129) (1.903) (1.904)
affected by firms’ information environment and governance efficacy. NCSKEWit-1 − 0.095 − 0.001
(− 0.652) (− 0.879)
When information is more transparent and governance efficacy is Intercept − 1.197 − 1.061 − 0.049*** − 0.049***
stronger, the quality of financial information disclosed by firms is (− 1.498) (− 1.235) (− 5.945) (− 5.975)
higher, which will increase the information precision of earnings Controls as in Table 4 Yes Yes Yes Yes
downside risk measure and its influence over expected stock returns. Firm FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Adjusted R2 0.711 0.730 0.007 0.007
4.6. Aggregated earnings downside risk and stock market returns N 49 49 19,261 19,259

In previous sections, we focus on the pricing of accounting-based Panel D: Subsamples formed based on the number of analyst following
downside risk at the individual firm level. In this section, we test Model 1 Model 2 Model 3 Model 4
whether firm-level accounting-based downside risk effect could be
Number of analysts Number of analysts
extended to the aggregated market level. We construct aggregated
following ≤Median following > Median
measures of accounting-based downside risk and empirically explore the
DR_ROAit-1 0.017* 0.021** 0.052*** 0.051***
ability of the aggregated measures to forecast excess stock market
(1.891) (2.229) (3.428) (3.365)
returns by performing the following model: NCSKEWit-1 − 0.000 − 0.005***
∑ (− 0.659) (− 7.234)
Rmt − Rft = β0 + β1 ADRit− 1 + βk Controlsit− 1 + εit (8) Intercept 0.004 0.004 0.014*** 0.014***
(1.0317) (1.012) (3.639) (3.464)
where Rmt − Rft refers to one-quarter-ahead quarterly excess stock Controls as in Table 4 Yes Yes Yes Yes
market returns, with Rm indicating quarterly stock market returns and Firm FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Rf indicating quarterly risk-free returns. ADR refers to aggregated Adjusted R2 0.020 0.020 0.018 0.019
DR_ROA, DR_TCA or DR_OCF, computed as the value-weighted average N 124,436 121,931 106,136 104,008
of firm-level downside risk using market capitalization at the end of the
This table reports the results of regressing monthly expected stock returns on
fiscal quarter. This bottom-up aggregation approach is similar to that
earnings downside risk in subsamples formed base on firm characteristics.
used in Hirshleifer et al. (2009) and Kang et al. (2010). To ensure that Panels A, B, C, and D report results for subsamples formed based on firm size,
the accounting information is known to investors at the beginning of the industry, firm age, and the number of analyst following, respectively. See Ap­
return quarter, we match returns in quarter t to accounting information pendix A for variable descriptions. The t-statistics are in parentheses.
in quarter t-1. Following a vast body of literature on market return *
p < 0.10.
predictability, we also include commonly used predictive variables to

10
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

**
p < 0.05. zero ROA. Accordingly, we calculate alternative unexpected ROA and
***
p < 0.01. the associated earnings downside risk measures. All these alternative
earnings downside risk measures are significantly positively associated
compare against aggregated accounting-based downside risk. EP is the subsequent monthly returns with and without controlling for other
value-weighted natural logarithm of earnings-to-price ratio. DP is the variables. These results are consistent with our conjectures, suggesting
value-weighted natural logarithm of dividend-to-price ratio. BM is the that the positive link between earnings downside risk and expected stock
value-weighted book-to-market ratio. TERM is the yield spread of 10- returns is robust to different ROA expectation models used to construct
year Treasury bonds over a one-month Treasury bill. TBILL is the yield the DR_ROA measure.
of a one-month Treasury bill. To interpret the economic significant of We also use alternative accrual downside risk measures to mitigate
the return predictability, the independent variable is standardized to the potential concern that the results may be driven by our accrual
have zero mean and unit variance. The t-statistics are computed using expectation model in Eq. (3). Following the same spirit of constructing
Newey-West standard errors. DR_TCA, we calculate alternative DR_TCA measures using residuals
Table 8 presents the results of the predictive regressions of quarterly derived from the different accrual expectation models, including those
excess stock market returns on DR_ROA, DR_TCA and DR_OCF. From of Jones (1991) and Dechow and Dichev (2002). Results based on these
2005 to 2018, ADR_ROA, ADR_TCA and ADR_OCF are strong positive alterative accrual downside risk measures are qualitatively similar to
predictors of market returns. Specifically, the predictive coefficients on our findings: all three alterative alternative accrual downside risk
ADR_ROA are 0.041, 0.043 and 0.037 in univariate regression with t- measures are significantly positively associated with expected stock
statistics of 1.999, 1.711 and 1.783, respectively. The magnitudes are returns in various model specifications.
economically large, for example, a one standard deviation increase in The results of these robustness checks are not reported for brevity,
ADR_ROA is associated with a 4.1% increase in one-quarter-ahead but are available upon request.
market returns. More importantly, the relation between aggregated
accounting-based downside risk and expected stock returns remains 5. Conclusion
statistically significant when other economic predictive variables are
included in the regression. The adjusted R2 varies from 2.8% to 32.3% in In this study, we test the validity of accounting-based downside risk
all specifications. in forecasting firms’ future operating performance and expected stock
Overall, Table 8 shows that firm-level downside risk effects can be returns in Chinese A-share markets. Based on a sample of Chinese listed
extended to the aggregated stock market. The evidence indicates firms over the period from 2005 to 2018, we show that firms with higher
accounting-based downside risk captures information on macroeco­ earnings downside risk are more likely to have deteriorating operating
nomic conditions, and may be used as a predictor for stock market performance in subsequent periods, confirming that the measure con­
returns. tains unique risk information. Further tests show that earnings downside
risk is significantly and positively related to expected stock returns. Such
4.7. Robustness checks findings add to Konchitchki et al. (2016) by showing that even in
emerging markets like China, where information environment trans­
In unreported tests, we have conducted various sensitivity tests to parency and governance efficacy is generally lower than that in the U.S
ensure that the findings are robust to various variable construction markets, accounting-based downside risk measures constructed based
methods. For instance, we have calculated alternative earnings down­ on publicly disclosed firm financial information still conveys informa­
side risk measures where expected ROA is captured by previous-year tion for firms’ future performance and is properly priced in the stock
industry average, previous five-year median of firm-specific ROA and markets.

Table 8
Aggregated earnings downside risk and stock market returns.
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

ADR_ROAt-1 0.041** 0.025

(1.999) (1.054)
ADR_TCAt-1 0.043* 0.072**
(1.711) (2.373)
ADR_OCFt-1 0.037* 0.058***
(1.783) (2.665)
EPt-1 − 2.635 3.687 − 4.852*
(− 0.855) (0.457) (− 1.726)
BMt-1 0.588* 1.190* 0.765**
(1.799) (1.955) (2.465)
DPt-1 − 0.000** − 33.865*** − 0.000***
(− 2.245) (− 5.023) (− 2.895)
TBILLt-1 − 4.410 − 2.480 − 6.116*
(− 1.089) (− 0.609) (− 1.708)
TERMt-1 − 6.776 1.292 − 10.061*
(− 1.125) (0.235) (− 1.836)
Intercept 0.028 0.038 0.020 0.113 − 0.055 0.226
(1.316) (1.529) (0.982) (0.647) (− 0.326) (1.553)
2
Adjusted R 0.045 0.028 0.031 0.105 0.323 0.166
N 64 69 69 64 66 66

This table reports predictive regressions of quarterly stock market returns on aggregated earnings downside risk (ADR_ROA), aggregated accrual downside risk
(ADR_TCA), and aggregated operating cash flow (ADR_OCF). The dependent variable is value-weighted excess market returns. The independent variable is value-
weighted average of firm-level earnings downside risk. The independent variable is standardized to have zero mean and unit variance. See Appendix A for vari­
able descriptions. The t-statistics are in parentheses.
*
p < 0.10.
**
p < 0.05.
***
p < 0.01.

11
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

We take one step further to decompose earnings downside risk into GPM: Annual gross profit margin ratio, the difference between total
accrual and OCF downside risks, and show that accrual and OCF revenues and cost of goods sold divided by total revenues.
downside risks play different roles in affected expected stock returns.
Accrual downside risk has a strong and positive association with ex­ A.3. Cost of capital
pected stock returns, while the relation between OCF downside risk and
stock returns is ambiguous. The findings indicate that the return pre­ RET − Rf: Proxy for monthly excess return and is measured as raw
dictability of earnings downside risk mainly origins from its accrual monthly return RET minus risk-free rate Rf.
downside risk component. Rm − Rf: Proxy for quarterly excess market return and is calculated
Moreover, we show that the pricing effect of earning downside risk is quarterly stock market return Rm minus quarterly risk-free return Rf.
affected by the information environment and governance efficacy at the CostDebt1: Proxy for cost of debt capital and is measured as the ratio
firm level. The positive relation between earnings downside risk and of interest expenses to the average of total short-and long-term debts.
expected stock returns concentrates among larger firms, non-high-tech CostDebt2: Proxy for cost of debt capital and is measured as the
firms, older firms, and firms with more analysts following. We ranked ratio of interest expenses to the average of total short-and long-
contend that in such firms, where information quality is higher, the term debts.
earnings downside risk measure constructed based on financial infor­
mation disclosed by firms contains more precise information about A.4. Earnings attributes
firms’ future performance, thus has greater influence over expected
stock returns. ACC_Q: The accrual quality measure of Dechow and Dichev (2002)
Finally, we show that aggregated accounting-based downside risk at calculated based on data over past 5 years.
the market level exhibits predictive power for stock market returns. This Persistence: Proxy for earnings persistence and is measured as nega­
evidence confirms that accounting-based downside risk aggregated at tive one times the slope of coefficient from an autoregressive model of
the market level contains information about macroeconomic conditions, order one (AR1) for the ratio of annual earnings to beginning total assets
and thus could be used as a predictor for future stock market returns. using a five year rolling window.
Predict: Proxy for predictability and is calculated as the square root of
Acknowledgements the error variance estimated from an autoregressive model of order one
(AR1) for the ratio of annual earnings to beginning total assets using a
Yan Luo acknowledges the financial support from National Natural five year rolling window.
Science Foundation of China (NSFC Project Numbers 72072034, Relevance: Proxy for value relevance and is calculated as negative one
71772049) and Wei Huang acknowledges the financial support from times R-square from the OLS regression of 12-month returns on the level
National Social Science Fund of China (Project Number 19BJY252). and change in earnings to market value of assets, estimated over a five-
year rolling window.
Appendix A. Variable definitions Smooth: Proxy for earnings smoothing and is calculated as standard
deviation of net income divided to beginning total assets, to its standard
A.1. Accounting-based downside risk measures deviation of cash flows from operations divided by beginning total as­
sets. Standard deviations are calculated over a five-year rolling window.
DR_ROA: Proxy for earnings downside risk, the probability that Timeliness: Proxy for timeliness and is calculated as negative one
earnings fall below its expected level. DR_ROA is measured as the log­ times R-square from the reverse earnings return coefficient model in
arithm of the ratio of one plus root lower partial moment of earnings Basu (1997), estimated over a five-year rolling window.
RLPM2(ROA) to one plus root upper partial moment of earnings Conservatism: Proxy for conservatism and is measured as the negative
RUPM2(ROA). of the ratio of the coefficient on bad news to the coefficient on good
DR_TCA: Proxy for accrual downside risk, the probability that news, following Basu (1997) and Walker (1999).
accrual falls below its expected level. DR_TCA is measured as the loga­
rithm of the ratio of one plus root lower partial moment of accrual A.5. Return downside risk (stock price crash risk)
RLPM2(TCAit) to one plus root upper partial moment of accrual
RUPM2(OCF). NCSKEW: negative one times the third moment of residual returns
DR_OCF: Proxy for operating cash flow downside risk, the probability divided by the standard deviation of the residual returns raised to the
that operating cash flow falls below its expected level. DR_OCF is third power, estimated using daily data over a horizon of one year.
measured as the logarithm of the ratio of one plus root lower partial DUVOL: the logarithm of the ratio of the standard deviation of re­
moment of operating cash flow RLPM2(OCF) to one plus root upper sidual returns below the mean to standard deviation of residual returns
partial moment of operating cash flow RUPM2(OCF). above the mean, estimated using daily data over a horizon of one year.
EDR: Earnings downside risk and its two components-accrual and
OCF downside risks. A.6. Control variables
ADR: Proxy for aggregated earnings downside risk and is computed
as the value- or equal-weighted average of firm-level earnings downside ROA: The ratio of annual earnings to total assets.
risk using market capitalization at the end of the fiscal quarter. SALE: The ratio of total sales to total assets.
SIZE: The natural logarithm of market value of equity at the end of
A.2. Performance measures the fiscal year.
LEVERAGE: The ratio of total liabilities to total assets.
DLOSS1: An indicator variable that takes a value of one if annual STD_ROA: The standard deviation of ROA over the previous three
income before extraordinary items is negative and zero otherwise. years.
DLOSS2: An indicator variable that equals one if annual net income is OC: The operating cycle and is measured as the natural logarithm of
negative and zero otherwise. IBM: The ratio of annual income before 360 days’ worth of accounts receivable scaled by total revenues plus
extraordinary items to total revenues. inventory scaled by cost of goods sold.
NIM: The ratio of annual net income to total revenues. TCA: The ratio of total accrual to total assets.
OPM: The ratio of annual operating income after depreciation to total STD_TCA: The standard deviation of total accrual ratio over the
revenues. previous three years.

12
Y. Luo et al. International Review of Financial Analysis 78 (2021) 101920

OCF: The ratio of annual operating cash flow to the total assets. Dechow, P. M., & Dichev, I. D. (2002). The quality of accruals and earnings: The role of
accrual estimation errors. The Accounting Review, 77(s-1), 35–59.
STD_OCF: The standard deviation of OCF over the previous three
Dechow, P. M., Kothari, S. P., & Watts, R. L. (1998). The relation between earnings and
years. cash flows. Journal of Accounting and Economics, 25(2), 133–168.
BM: Book-to-market ratio and is calculated as book value of equity Eliwa, Y., Haslam, J., & Abraham, S. (2016). The association between earnings quality
divided by market capitalization at the end of the fiscal year. and the expected stock returns capital: Evidence from the UK. International Review of
Financial Analysis, 48, 125–139.
Cash: The ratio of cash holdings plus cash equivalents to total assets. Fama, E. F., & French, K. R. (1993). Common risk factors in the returns on stocks and
∆Cash: The ratio of changes in cash holdings and cash equivalent to bonds. Journal of Financial Economics, 33(1), 3–56.
total assets. Fishburn, P. C. (1977). Mean-risk analysis with risk associated with below-target returns.
American Economic Review, 67, 116–126.
Invest_CAPEX: The ratio of capital expenditures to total assets. Francis, J., LaFond, R., Olsson, P. M., & Schipper, K. (2004). Costs of equity and earnings
OO: The ratio of property, plant, and equipment to total assets. attributes. The Accounting Review, 79(4), 967–1010.
SIGMA: The standard deviation of daily stock returns for the fiscal Gul, F. (1991). A theory of disappointment aversion. Econometrica, 667–686.
Hirshleifer, D., Hou, K., & Teoh, S. H. (2009). Accruals, cash flows, and aggregate stock
year. returns. Journal of Financial Economics, 91(3), 389–406.
MOM: The buy and hold returns over the 11-month period that ends Hutton, A. P., Marcus, A. J., & Tehranian, H. (2009). Opaque financial reports, R2, and
one-month prior to month t. crash risk. Journal of Financial Economics, 94(1), 67–86.
Jin, L., & Myers, S. C. (2006). R2 around the world: New theory and new tests. Journal of
MKTbeta: Stock return sensitivity to the market factor of Fama and Financial Economics, 79(2), 257–292.
French (1993) based on data over the past 60 months ending in month t. Jones, J. J. (1991). Earnings management during import relief investigations. Journal of
SMBbeta: Stock return sensitivity to the size factor of Fama and Accounting Research, 193–228.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk.
French (1993) based on data over the past 60 months ending in month t.
Econometrica, 47(2), 263–291.
HMLbeta: Stock return sensitivity to the book-market factor of Fama Kang, Q., Liu, Q., & Qi, R. (2010). Predicting stock market returns with aggregate
and French (1993) based on data over the past 60 months ending in discretionary accruals. Journal of Accounting Research, 48(4), 815–858.
month t. Kim, J. B., Li, Y., & Zhang, L. (2011a). Corporate tax avoidance and stock price crash risk:
Firm-level analysis. Journal of Financial Economics, 100(3), 639–662.
UMDbeta: Stock return sensitivity to the momentum factor of Car­ Kim, J. B., Li, Y., & Zhang, L. (2011b). CFOs versus CEOs: Equity incentives and crashes.
hart’s (1997) based on data over the past 60 months ending in month t. Journal of Financial Economics, 101(3), 713–730.
Kim, J. B., & Zhang, L. (2016). Accounting conservatism and stock price crash risk: Firm-
level evidence. Contemporary Accounting Research, 33(1), 412–441.
References Konchitchki, Y., Luo, Y., Ma, M. L., & Wu, F. (2016). Accounting-based downside risk,
cost of capital, and the macroeconomy. Review of Accounting Studies, 21(1), 1–36.
Ang, A., Chen, J., & Xing, Y. (2006). Downside risk. Review of Financial Studies, 19(4), Koonce, L., McAnally, M. L., & Mercer, M. (2005). How do investors judge the risk of
1191–1239. financial items? The Accounting Review, 80(1), 221–241.
Basu, S. (1997). The conservatism principle and asymmetric timeliness of earnings. Lugo, S. (2019). Insider ownership and the cost of debt capital: Evidence from bank
Journal of Accounting and Economics, 24(1), 3–37. loans. International Review of Financial Analysis, 63, 357–368.
Callen, J. L., & Segal, D. (2004). Do accruals drive firm-level stock returns? A variance Roy, A. D. (1952). Safety first and the holding of assets. Econometrica, 431–449.
decomposition analysis. Journal of Accounting Research, 42(3), 527–560. Sloan, R. G. (1996). Do stock prices fully reflect information in accruals and cash flows
Carhart, M. M. (1997). On persistence in mutual fund performance. Journal of Finance, 52 about future earnings? Accounting Review, 289–315.
(1), 57–82. Walker, P. M. (1999). International differences in the timeliness, conservatism, and
Dechow, P. M. (1994). Accounting earnings and cash flows as measures of firm classification of earnings. Journal of Accounting Research, 37, 53–87.
performance: The role of accounting accruals. Journal of Accounting and Economics,
18(1), 3–42.

13