University of Southern California Dissertations and Theses

Accrual quality and expected returns: the importance of controlling for cash flow shocks

(USC Thesis Other)

Accrual quality and expected returns: the importance of controlling for cash flow shocks

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

ACCRUAL QUALITY AND EXPECTED RETURNS:

THE IMPORTANCE OF CONTROLLING

FOR CASH FLOW SHOCKS

by

Maria Ogneva

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BUSINESS ADMINISTRATION)

May 2008

Copyright 2008 Maria Ogneva
ii

Dedication

This dissertation is dedicated to my parents, Leonid Ognev and Natalya Polovkova.

iii

Acknowledgements

I thank K.R. Subramanyam, my dissertation chair, for his continuous support and guidance in the
development of this paper. I am also thankful to other members of my dissertation committee,
including Randy Beatty, Mark DeFond, Christopher Jones, and especially Rebecca Hann. This paper
benefited from the helpful comments and suggestions from Nerissa Brown, Mei Cheng, Yaniv
Konchitchki, Siqi Li, Hai Lu, Suresh Nallareddy, Per Olsson, Joe Piotroski, Tatiana Sandino, Joe
Weber, Biqin Xie, Jieying Zhang, and the participants at the workshops in University of Southern
California, Stanford University and MIT. I am grateful to the Deloitte Foundation and the Marshall
School of Business at the University of Southern California for financial support.
iv

Table of Contents
Dedication ii

Acknowledgements iii

List of Tables vi

List of Figures vii

Abstract viii

Chapter 1: Introduction 1

Chapter 2: Motivation 8
2.1 Prior Literature 8
2.2 The DD Measure and Cash Flow Shocks 12

Chapter 3: Sample and Data 15

Chapter 4: Empirical Analyses and Results 18
4.1 The DD Measure and Future Stock Returns: Replicating CGV 18
4.1.1 Accrual Quality Characteristic and Future Returns 18
4.1.2 AQ factor Loadings and Future Stock Returns 23
4.2 The DD Measure and Cash Flow Shocks 27
4.2.3 Univariate Correlations 27
4.2.4 Asset Pricing Tests: Controlling for Cash Flow Shocks 34
4.3 The Relative DD Measure 50
4.3.5 Selecting a Relative DD Measure 50
4.3.6 Firm Characteristics Associated with the Scaled DD Measure 52
4.3.7 Validating the Scaled DD Measure 58
4.3.8 The Scaled DD Measure and Future Returns 62

Chapter 5: Additional Analyses 67
5.1 Look-Ahead Bias 67
5.2 Controlling for Signed Accruals 78

Chapter 6: Conclusion 82

Bibliography 84
v

Appendix: Cash Flow Shock Proxies 89

vi

List of Tables

Table 1. The DD Measure and Future Returns 19
Table 2. Correlations between Firm Characteristics and DD Measure 28
Table 3. Controlling for Cash Flow Shocks 36
Table 4. Cross-Sectionally Unconstrained Cash Flow Shock Controls 45
Table 5. Relative Accrual Quality Measure: Choosing a Scaling Factor 52
Table 6. Comparing DD and Scaled DD Measures: Correlations with Various
Characteristics 54
Table 7. Relative Accrual Quality Measure Validation 60
Table 8. The Scaled DD Measure and Future Returns 63
Table 9. The Lagged DD Measure and Future Returns 67
Table 10. Comparing Lagged DD and Scaled DD Measures: Correlations with
Various Characteristics 70
Table 11. Controlling for Cash Flow Shocks 71
Table 12. Cross-Sectionally Unconstrained Cash Flow Shock Controls: Lagged
Measures 74
Table 13. The Scaled Lagged DD Measure and Future Returns 77
Table 14. The Scaled DD Measure and Future Returns: Controlling for Signed
Accruals 80

vii

List of Figures

Figure 1. Cash Flow Shocks and the DD Measure 32
Figure 2. Cash Flow Shocks and the Scaled DD Measure 56

viii

Abstract

Francis, LaFond, Olsson, and Schipper (2005) document that accrual quality is
inversely related to the cost of equity capital. However, Core, Guay, and Verdi (2007) find
no association between accrual quality and future stock returns and conclude that there is
no evidence that the stock market prices accrual quality. I hypothesize that Core et al.’s
result arises because poor accrual quality firms experience negative cash flow shocks in
the future, which results in negative returns that offset the higher expected returns for
such firms. Consistent with this prediction, I find a significant negative association
between realized returns and accrual quality after controlling for cash flow shocks, either
by including proxies for future cash flow shocks in asset pricing regressions or by using
an accrual quality measure that is less correlated with future cash flow shocks. This result
is robust to properly specified and standard asset pricing tests. Overall, this paper adds to
the growing literature suggesting that accrual quality is linked to the cost of capital.
1

Chapter 1: Introduction
There is an ongoing debate as to whether the quality of accounting information
influences firms’ cost of equity capital. Much of this debate centers around the quality of
accruals. Specifically, in an influential set of papers, Francis, LaFond, Olsson, and
Schipper (2004, 2005) (henceforth, FLOS) argue that accrual quality affects the cost of
equity.
1
FLOS base their conclusions, in part, on a significant positive correlation
between the returns on the AQ factor (accrual quality factor) and contemporaneous stock
returns.
2
However, Core, Guay, and Verdi (2007) (henceforth, CGV) note that the
returns-based tests reported by FLOS are incomplete for the purpose of establishing a new
pricing factor related to accrual quality. In particular, CGV perform conventional
two-stage asset pricing tests (e.g., Fama and MacBeth 1973) and show that the loadings
on the AQ factor, although positive on average, do not explain the cross-sectional
variation in returns. In addition, CGV document that accrual quality, as a characteristic,
does not predict future (one-year-ahead) realized returns. Taken together, the findings in
CGV are inconsistent with accrual quality being priced by the market.
In this paper, I argue that CGV are unable to find an association between accrual
quality and future realized returns because their inverse measure of accrual quality—the
residual accrual volatility measure proposed by Dechow and Dichev (2002), henceforth

1
Throughout the paper, I use “cost of equity” and “expected return” on a firm’s stock as equivalent expressions.
2
The overall conclusions reached by FLOS (2004, 2005) are based on multiple tests that, in addition to realized returns,
employ earnings-to-price ratios, betas, and implied cost-of-equity estimates as alternative proxies for the cost of equity.
The results based on alternative cost of equity proxies are generally consistent with low accrual quality being penalized
by higher cost of equity.
2

the DD Measure—is negatively correlated with future cash flow shocks, i.e., firms with
low measured accrual quality experience negative cash flow shocks in the future. Due to
this correlation, the higher expected returns associated with poor accrual quality firms are
systematically offset by the negative returns arising from adverse cash flow shocks,
thereby leading to no association between future realized returns and measured accrual
quality.
Prior research provides evidence that indirectly supports the conjectures above.
Both FLOS and CGV use the DD Measure to measure accrual quality. The DD Measure is
estimated as the standard deviation of residuals from the regression of working capital
accruals on past, current, and future operating cash flows, so that higher levels of the DD
Measure represent lower accrual quality. The DD Measure is correlated with several
characteristics that are likely associated with adverse cash flow news. For example, high
DD Measure (i.e., low accrual quality) stocks experience high sales growth in the past
(Doyle et al. 2007), face higher financing constraints (Biddle and Hilary 2006), are likely
to have higher bankruptcy risk due to a high frequency of losses (Dechow and Dichev
2002), and have more volatile earnings and sales (Dechow and Dichev 2002). Each of
these characteristics has been linked to systematic future underperformance in the stock
market (Lakonishok et al. 1994; Lamont et al. 2001; Dichev 1998; Mohanram 2005),
possibly as a result of receiving adverse cash flow news.
To control for the effects of cash flow shocks, I perform two sets of analyses.
First, I include direct proxies for cash flow shocks as additional control variables in the
3

cross-sectional asset pricing regressions. Second, I indirectly control for the effects of
cash flow shocks by modifying the DD Measure to construct a measure of accrual quality
that is less correlated with future cash flow shocks. Specifically, I scale the original DD
Measure by the value of average absolute accruals estimated over the past five years.
3

This scaled measure is conceptually similar to the R
2
from the Dechow and Dichev
accrual-cash flow mapping regressions, which has been suggested as an alternative
measure of accrual quality (McNichols 2002). I then use the scaled measure as a proxy for
accrual quality in standard asset pricing tests of the type implemented by CGV .
I start my empirical analyses by replicating CGV’s findings for a sample of
88,999 firm-year observations with returns from April 1971 to March 2006.
4
First, I
confirm CGV’s “non-result” in the time period of my analysis. The DD Measure does not
predict future stock returns either at the individual stock level or at the portfolio level;
likewise, stocks with higher loadings on the AQ factor (accrual quality factor based on the
DD Measure) do not earn significantly higher returns either at the individual stock level
or at the portfolio level.
Next, I confirm that the DD Measure is correlated with several characteristics
that predict abnormal stock returns. The signs of the correlations are consistent with low

3
The idea of using a scaled measure is not original, although this paper is the first to use such a measure in the
asset-pricing context. Verdi (2006) proposes an alternative scaling specification: scaling by the standard deviation of
current accruals. I find that the measure suggested by Verdi has a lower correlation with the R
2
from individual
accrual-cash flow mapping regressions. When I replicate my analyses using this alternative specification, I find
statistically and qualitatively similar results (as described in Chapter 5).
4
This sample has three additional recent years compared to the samples used in CGV and FLOS (2005). Arguably,
including more years following the recent accounting scandals (e.g., Enron) lends more power to tests that explore the
asset-pricing implications of accounting quality.
4

accrual quality (high DD Measure) stocks experiencing more negative cash flow shocks
over the following year. The DD Measure is also negatively correlated with two direct
proxies for future cash flow shocks, namely, the earnings surprise variable based on a
simple model that assumes earnings can be predicted based on previous-year earnings and
previous-year returns, and the cash flow news proxy based on the revision in analysts’
earnings forecasts constructed as in Easton and Monahan (2005).
I first control for the effect of cash flow shocks in asset pricing tests by including
the direct cash flow shock proxies in cross-sectional asset pricing regressions. As in CGV ,
the regressions include either the DD Measure or the loading on the AQ factor (AQ factor
beta) and are estimated either at the individual stock level or at the portfolio level. After
controlling for the effects of cash flow shocks, I find a significant association between
returns and both the DD Measure (i.e., the characteristic) and the AQ factor beta (i.e., the
factor loading). However, the premium on the DD Measure, as a characteristic, is no
longer significant when I control for the three Fama-French characteristics (beta, market
value of equity, and book-to-market), whereas the premium on the AQ factor (i.e., the
regression coefficient on the AQ factor beta) remains significant after controlling for the
loadings on the three Fama-French factors.
5
At the stock level, the AQ factor premium is
equal to 0.28% (p-value of 0.05) or 0.36% (p-value of 0.28) per month, depending on the
type of cash flow shock proxy used in the analysis. At the portfolio level, the AQ factor

5
Additional stock-level analyses based on cross-sectionally unconstrained cash-flow news estimates show at least
marginally significant premiums equal to approximately 2.16% (p-value of 0.12) and 5.4% (p-value of 0.01) per annum
for the DD Measure and the AQ factor beta, respectively, even after controlling for the three Fama-French
characteristics or factors.
5

premium estimates range from 0.30% to 1.38% per month (from 3.6% to 16.56% per
annum), depending on the set of portfolios and the type of cash flow shock proxy used in
the analysis; the premiums are statistically significant at conventional levels in four out of
six specifications.
Second, I indirectly control for the effects of cash flow shocks by modifying the
DD Measure using the previously described scaling procedure. The scaling successfully
reduces the correlation of the DD Measure with cash flow shock proxies. Importantly,
however, the scaled measure (henceforth the Scaled DD Measure) retains the ability to
measure some dimension of accrual quality—I find that the scaled and the original DD
Measures are similarly associated with the exogenous indicators of accounting quality
(disclosures of material weaknesses in internal controls and accounting restatements).
When I replicate asset pricing tests using the Scaled DD Measure in place of the
original DD Measure, I find the following results. First, the Scaled DD Measure, as a
characteristic, is significantly associated with returns in the cross-sectional firm-level
regressions, both before and after controlling for the three Fama-French characteristics.
Second, the factor-mimicking portfolio that sells (buys) stocks in the two highest (lowest)
quintiles of the Scaled DD Measure (the SAQ factor) earns a statistically significant
abnormal (with respect to the three-factor model) return of 0.19% per month
(approximately 2.3% per annum).
6
Third, the SAQ factor beta is significantly associated

6
Compared to other accounting anomalies, the magnitude of the premium is relatively small, e.g. the accrual anomaly
generates an abnormal (with respect to CAPM) return of 10% per annum (Sloan 1996). However, it is comparable to the
magnitude of returns for the size anomaly – over my sample period the SMB factor earns approximately 2.3% per
annum.
6

with returns in the portfolio-level cross-sectional regressions after controlling for the
loadings on the three Fama-French factors. The SAQ factor premium (i.e., the regression
coefficient on the AQ factor beta) estimates range from 0.16% to 0.27% per month (from
1.9% to 3.2% per annum), depending on the set of portfolios used in the analysis; the
premiums are at least marginally statistically significant. I find no evidence of a
significant SAQ factor premium at the individual-firm level, possibly due to the low
power of firm-level tests.
In summary, I demonstrate that accrual quality (measured using the
Dechow-Dichev model) is associated with future expected returns after controlling for
cash flow shocks. The results are robust to multiple sensitivity tests, including the use of
lagged DD and Scaled DD Measures, controlling for the level of total accruals,
controlling for the magnitude of the scaling factor used to construct the Scaled DD
Measure, alternative scaling procedures, and controls for the effect of earnings smoothing
and the stage of the business cycle.
The contribution of the paper is two-fold. First, the paper contributes to the
recent debate on whether realized returns reflect the risk premium associated with poor
accrual quality (FLOS 2005; CGV 2007). Second, the paper underscores the importance
of controlling for cash flow shocks in asset pricing tests that use realized returns to proxy
for the cost of equity.
The rest of the paper proceeds as follows. Chapter 2 reviews the related
literature and develops the paper’s hypotheses; Chapter 3 describes the sample selection
7

procedures and the data; Chapter 4 describes empirical analyses and results; Chapter 5
reports sensitivity analyses; Chapter 6 concludes.
8

Chapter 2: Motivation
2.1 Prior Literature
The theoretical foundation for the link between accrual quality and the cost of
equity is provided by several analytical models from the estimation risk and market
microstructure literatures.
7

The estimation risk literature (e.g., Klein and Bawa 1976, Barry and Brown
1985, Coles et al. 1995) suggests that securities with less information about past return
realizations should earn higher expected returns. In the spirit of earlier estimation risk
studies, Lambert et al. (2007a, 2007b) suggest that firms with more precise information
about future cash flows have lower conditional covariances with the market, and as a
consequence, lower conditional betas and lower expected returns. Overall, the following
two-step link is suggested by the estimation risk literature: (1) firms with higher
information quality have lower forward-looking betas; and (2) lower forward-looking
betas lead to lower cost of equity. Note that the forward-looking betas cannot be directly
estimated from the past return realizations.
8

7
In this section, I discuss only models that relate the quality of information to the cost of equity under conditions that
are directly applicable to asset-pricing tests—in the presence of diversification opportunities with multiple firms and
multiple traders in the market. A number of models explore the effects of information asymmetry on liquidity but do not
have clear implications for expected returns and are directly applicable only to the setting of equity issues, such as IPOs
and SEOs, e.g. Diamond and Verrecchia (1991).
8
The traditionally estimated CAPM betas assume no differences in estimation risk among the securities and therefore
serve as an imperfect proxy for the forward-looking betas. The point is well illustrated by an example provided by Barry
and Brown (1985): Suppose there are two stocks for which betas are estimated from past returns and suppose that the
estimated betas are equal. A Bayesian investor should take into account the precision with which the betas are estimated.
Therefore, despite the equality in estimated historic betas, a stock for which beta is estimated using a shorter series of
observations should have a higher forward-looking beta and hence should be considered riskier.
9

The market microstructure literature includes several models that link
information quality to the cost of equity. Amihud and Mendelson (1986) argue that the
expected return on a stock increases with the magnitude of the bid-ask spread to
compensate for higher transaction costs. Higher information quality may decrease the
firm’s cost of equity by lowering the adverse selection component of the bid-ask spread.
More recently, Easley and O’Hara (2004) model the effect of information asymmetry on
the cost of equity. In their model, poorly informed investors demand compensation for
their informational disadvantage vis-à-vis more informed investors. As a consequence,
expected returns increase in the degree of information asymmetry about the stock’s
payoffs.
9

Overall, while analytical models suggest different mechanisms through which
information quality affects the cost of equity, they imply similar empirical predictions
regarding the association between accrual quality and cost of equity. If low accrual quality
increases either estimation risk or information asymmetry, then low accrual quality stocks
are expected to earn higher returns after controlling for historic CAPM betas. Each stream
of literature can also be used to derive the existence of a separate risk factor related to
accrual quality. For example, Kogan and Wang (2003) show that a model with estimation
risk (or parameter uncertainty) can be recast in a multifactor asset pricing framework with

9
Two recent papers dispute the conclusions reached by Easley and O’Hara (2004). Hughes et al. (2007) argue that after
controlling for factor betas, there is no cross-sectional relation between expected returns and the degree of information
asymmetry (although information asymmetry affects factor premiums). Lambert et al. (2007b) show that after
controlling for the average precision of information, the degree of information asymmetry is not related to expected
returns.
10

a separate factor related to estimation risk, while Easley et al. (2005) argue that there is a
separate risk factor related to information asymmetry.
Most empirical studies use the Dechow and Dichev (2002) measure of residual
accrual volatility (the DD Measure) to infer the association between accrual quality and
the cost of equity (expected return).
10
The DD Measure is estimated as the standard
deviation of residuals from the regression of working capital accruals on past, current, and
future operating cash flows, such that higher levels of the DD Measure indicate lower
accrual quality. FLOS (2004, 2005) form a factor-mimicking portfolio based on the DD
Measure (the AQ factor) and document that, on average, the return on the AQ factor is
positively correlated with the realized returns on individual stocks.

Nichols (2006)
documents significant abnormal daily returns on the AQ factor. In contrast, Aboody et al.
(2005) find that the hedge return earned by taking a long (short) position in the stocks
with high (low) AQ factor loadings is not statistically significant. Further, CGV point out
that asset pricing tests in FLOS (2004, 2005) are insufficient to conclude that the AQ
factor is a priced risk factor. Specifically, FLOS only show that the average loading on the
AQ factor is positive, while it is necessary to establish that stocks with higher loadings on
the AQ factor have higher average realized returns. CGV perform well-accepted
two-stage asset pricing tests (e.g., Fama and MacBeth 1973) and show that the loadings

10
Empirically, the association between accounting quality and the cost of equity has been tested using a variety of
definitions for accounting quality, including the overall quality of disclosures (e.g., Botosan 1997; Botosan and Plumlee
2002; Berger et al. 2006) and the quality of specific accounting metrics such as earnings (e.g., Barth et al. 2006;
Bhattacharya et al. 2003). Generally, these studies find that the firms with low quality accounting information are
penalized with a higher cost of equity. For the sake of brevity, I restrict this paper’s review of the literature to the studies
using accrual quality, and the DD Measure in particular, to proxy for the quality of accounting information.
11

on the AQ factor do not explain the cross-sectional variation in returns—the premium on
the AQ factor is neither economically nor statistically significant. They also document
that the DD Measure, as a characteristic, does not predict future monthly realized returns
and argue that the significant returns on the AQ factor found by Nichols (2006) result
from the daily portfolio re-balancing implicitly imposed by using equally weighted daily
portfolio returns.
11

Several studies indirectly test whether accrual quality affects realized returns by
using exposure to the AQ factor as a measure of information risk. Ecker et al. (2006)
document increases in AQ factor loadings around events signaling poor accounting
quality, including accounting restatements, lawsuits alleging accounting violations, and
bankruptcies. Chen et al. (2007) document increases (decreases) in AQ factor loadings for
firms initiating or increasing (decreasing) dividend payments. Kravet and Shevlin (2007)
document that the loadings on the AQ factor increase in the months following the
accounting restatement announcements. However, evidence on the behavior of the AQ
factor loading is insufficient to draw conclusions about whether accrual quality is
reflected in expected returns.
12

11
Frequent re-balancing leads to a biased estimate of returns due to bid-ask spread bounce (Blume and Stambaugh
1983).
12
Several studies document a higher cost of equity for firms with low accrual quality (measured by the DD Measure)
using alternative cost-of-equity measures, including: (1) implied cost of equity, calculated using either price targets and
forecasted dividends or PEG ratios (FLOS 2004; CGV 2007); (2) CAPM betas (FLOS 2004; Liu and Wysocki 2006);
and (3) industry-adjusted earnings-to-price ratios (FLOS 2004; Liu and Wysocki 2006). However, Liu and Wysocki
(2006) argue that any asset-pricing effects of DD Measure are attributable to the underlying operating risk captured by
cash flow variability and idiosyncratic return volatility.
12

To summarize, while analytical models suggest a negative link between
information quality and the cost of equity, empirical research relating accrual quality
(measured using the DD Measure) to the cost of equity (expected returns) provides mixed
evidence. Importantly, standard asset pricing tests provide little evidence that the accrual
quality risk associated with the DD Measure is priced by the stock market.

2.2 The DD Measure and Cash Flow Shocks
Standard asset pricing tests use average realized stock returns to capture the cost
of equity. However, realized returns measure cost of equity with error due to information
shocks (e.g., Elton 1999). Realized returns have three components: the expected return,
the return due to cash flow news, and the return due to discount rate news (e.g., Cambell
and Shiller 1988), of which only the expected return reflects the firm’s cost of equity. If
firms with lower (higher) levels of measured accrual quality systematically receive more
(less) negative cash flow news in the future, then asset pricing tests are biased against
finding a positive relationship between accrual quality and the cost of equity.
The evidence in the extant empirical research is (indirectly) consistent with
firms with higher DD Measures (lower accrual quality) experiencing negative cash flow
shocks in the near future. Several characteristics that are correlated with the DD Measure
are also related to future abnormal stock returns possibly arising due to cash flow shocks.
For example, Dechow and Dichev (2002) document that firms with higher DD
Measures have a higher frequency of losses, which is consistent financial distress.
13

Distressed firms earn significantly lower returns compared to the rest of the stock market
(Dichev 1998). This underperformance is not explained by the lower risk exposure, and
therefore is likely caused by systematic future negative cash flow shocks. Additionally,
poor accounting quality firms (with higher DD Measures) face higher financing
constraints (Biddle and Hilary 2006). Lamont et al. (2001) document that financing
constraints are negatively related to future realized returns, possibly due to a systematic
exposure to negative cash flow shocks.
Also, firms with higher DD Measures have higher variability of earnings and
sales (Dechow and Dichev 2002). Mohanram (2005) shows that growth firms (firms with
low book-to-market ratios) with higher earnings or sales growth variability earn
significantly lower future returns compared to their industry peers. He explains this effect
by a naïve extrapolation of good past earnings or sales performance that is likely to be less
sustainable for highly variable earnings or sales. In a similar vein, the DD Measure is
positively related to past sales growth (Doyle et al. 2007); Lakonishok et al. (1994)
document that stocks with higher past sales growth earn negative abnormal returns, which
possibly results from the market correcting the initial overpricing of these stocks.
Finally, through its correlation with losses, the DD Measure may be directly
related to the fundamental signals that predict future returns. Piotroski (2000, 2006)
demonstrates that the stock market does not fully appreciate the implications of
fundamental signals related to the financial health of the firm for the firm’s future
performance.
14

Overall, prior literature identifies several characteristics that predict future
abnormal returns and are associated with the DD Measure. To the extent that the
documented abnormal returns are related to cash flow shocks, the average realized returns
on the stocks with high DD Measures (low accrual quality) are affected by systematic
adverse cash flow news and are a downwardly biased estimate of the cost of equity.
15

Chapter 3: Sample and Data
The sample consists of 88,668 firm-years with fiscal years ending in 1970 to
2004 and with returns available from April 1971 to March 2006.
13
To avoid spuriously
inflating the returns to the trading strategies based on accrual-related variables (Beaver et
al. 2007), all asset pricing tests incorporate delisting returns for the firms that delist from
the exchanges within the estimation period. If the delisting return is missing in CRSP, I
substitute it with the average delisting return from Shumway (1997) and Shumway and
Warther (1999). Some tests require additional data causing further sample restrictions.
The DD Measure is estimated similarly to FLOS (2005). First, the following
cross-sectional regression is estimated for each year and each of the 48 Fama and French
(1997) industries (at least 20 observations are required for each regression):
TCA
it
= α
t
+ β
0t
1/ATA
it
+ β
1t
CFO
it-1
+ β
2t
CFO
it
+ β
3t
CFO
i t+1
+ β
4t
ΔREV
it
+
β
5t
PPE
it
+ ε
it
,

(1)
where
TCA
it
= total current accruals for year t (estimated using the balance
sheet approach: TCA
it
= ΔCA
it
- ΔCL
it
- ΔCash
it
+
ΔSTDEBT
it
, where ΔCA
it
is one-year change in current assets

13
The sample selection starts with 113,835 observations with fiscal years ending in 1970-2004 for which Compustat
has the data necessary to calculate the DD Measure. Each observation is then required to have sufficient data to
calculate the market value of equity, book-to-market, and CAPM beta, which leaves 89,263 observations. Finally, each
observation is required to have at least one month with a non-missing return in CRSP within the estimation period (the
twelve months starting with the fourth month after the fiscal year-end).
16

(Compustat #4), ΔCL
it
is one-year change in current liabilities
(Compustat #5), ΔCash
it
is one-year change in cash
(Compustat #1), and ΔSTDEBT
it
is one-year change in debt in
current liabilities (Compustat #34));
CFO
it
= cash flow from operations for year t estimated as NIBE
it
- TA
it
,
where NIBE
it
is net income before extraordinary items
(Compustat #18) and TA
it
is total accruals (TA
it =
TCA
it
-
DEPN
it
, where DEPN
it
is depreciation and amortization
expense in year t (Compustat #14));
ΔREV
it
= one-year change in revenues (Compustat #12);
PPE
it
= property, plant, and equipment for year t (Compustat #7);
ATA
it
= average value of total assets (Compustat #6) over years t-1, t,
and t+1.
Subscript i refers to individual firms. All variables are scaled by the average
value of total assets (Compustat #6) over years t-1, t and t+1.
The DD Measure is estimated for each firm i and each year t as the standard
deviation of residuals from the above cross-sectional regression over the period [t-4, t]. In
all asset pricing tests, the DD Measure is converted to decile ranks following FLOS
(2005). Sorting stocks into deciles is performed monthly and each stock is assigned a rank
based on the DD Measure from the most recent fiscal year that ends sixteen to four
months prior to the sorting. For example, the ranking for July 2004 is based on the DD
17

Measure from the latest fiscal year that ended within the April 2003 to March 2004
period.
18

Chapter 4: Empirical Analyses and Results
4.1 The DD Measure and Future Stock Returns: Replicating CGV
4.1.1 Accrual Quality Characteristic and Future Returns
CGV document that neither the accrual quality characteristic (the DD Measure),
nor the exposure to accrual quality risk (loading on the AQ factor) can predict future stock
returns. I replicate CGV’s results for an extended sample period that includes three
additional years (2002 to 2004).
Panel A of Table 1 replicates the characteristic-based analysis of CGV . The
following cross-sectional regression is estimated monthly using individual stock returns
(the i subscript refers to individual stocks):
R
i,t+1
–Rf
t+1
= Intercept
t
+ q
t
RDD
it
+ b
t
BETA
it
+ s
t
log(MKTV)
it

+ h
t
log(BMRATIO)
it
+ ε
it
, (3)
where
R
i,t+1
= return on stock i for month t+1;
Rf
t+1
= one-month T-bill rate for month t+1;
RDD
it
= decile rank based on the DD Measure determined at
the end of month t (the ranking procedure is described
in Chapter 3);
BETA
it
= CAPM beta estimated using 36 months of data prior
to the most recent January;
19

log(MKTV)
it
= logarithm of market value of equity at the latest fiscal
year-end preceding month t by at least three months;
log(BMRATIO)
it
= logarithm of the ratio of the book-to-market ratio at
the latest fiscal year preceding month t by at least
three months, where book value is the stockholders’
book equity plus deferred taxes minus the book value
of preferred stock, and market value of equity is as
defined previously.

Table 1. The DD Measure and Future Returns

Panel A. Returns on AQ Characteristic: Firm-Level Cross-Sectional Regressions
Intercept RDD BETA
log
(MKTV)
log
(BMRATIO)
Adjusted
R
2

Mean Coefficient 1.64 0.03 -0.13 0.31 3.33%
FM p-value (0.00) (0.83) (0.01) (0.00)
Mean Coefficient 0.79 0.03 0.90%
FM p-value (0.00) (0.39)
Mean Coefficient 1.73 -0.01 0.05 -0.14 0.30 3.48%
FM p-value (0.00) (0.38) (0.75) (0.00) (0.00)

Panel B. The AQ factor: Descriptive Statistics
Factor N Mean Std. Dev. Min Median Max t-stat p-value
AQ factor 420 0.16 3.47 -12.07 -0.21 23.39 0.95 (0.34)
MktRf 420 0.49 4.54 -23.13 0.78 16.05 2.22 (0.03)
SMB 420 0.20 3.29 -16.70 0.06 22.18 1.22 (0.22)
HML 420 0.46 3.08 -12.80 0.46 13.80 3.07 (0.00)

20

Table 1. Continued

Panel C. The AQ factor: Abnormal Returns
Dependent Variable = AQ factor
Intercept MktRf SMB HML Adjusted R
2

Coefficient 0.08 0.07 0.67 -0.17 53%
p-value (0.51) (0.01) (0.00) (0.00)

Panel D. Returns on the AQ factor Loadings: Firm-Level Cross-Sectional Regressions
Intercept
b
^
AQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

Mean Coefficient 0.48 0.37 0.16 -0.14 5.21%
FM p-value (0.00) (0.13) (0.39) (0.41)
Mean Coefficient 0.78 0.16 2.88%
FM p-value (0.00) (0.38)
Mean Coefficient 0.49 0.16 0.36 0.13 -0.12 6.39%
FM p-value (0.00) (0.38) (0.15) (0.45) (0.46)

Panel E. Returns on the AQ factor Loadings: Portfolio-Level Cross-Sectional Regressions
Independent Variables

Intercept b
^
AQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

25 B/M and Size Portfolios
Regression Coefficient 2.75 -2.16 0.22 0.64 71.66%
FM p-value (0.00) (0.00) (0.37) (0.00)
Regression Coefficient 0.64 0.26 10.56%
FM p-value (0.01) (0.16)
Regression Coefficient 2.43 0.03 -1.74 -0.14 0.70 81.31%
FM p-value (0.00) (0.89) (0.00) (0.52) (0.00)

100 DD Measure Portfolios
Regression Coefficient 0.13 0.45 0.27 0.35 28.80%
FM p-value (0.71) (0.46) (0.43) (0.39)
Regression Coefficient 0.78 0.15 22.72%
FM p-value (0.00) (0.40)
Regression Coefficient 0.12 0.15 0.45 0.31 0.34 28.11%
FM p-value (0.74) (0.41) (0.35) (0.27) (0.28)

64 B/M, Size and DD Measure Portfolios
Regression Coefficient 1.63 -1.11 0.27 0.57 58.62%
FM p-value (0.00) (0.01) (0.16) (0.00)
Regression Coefficient 0.68 0.20 5.45%
FM p-value (0.01) (0.28)
Regression Coefficient 1.72 0.10 -1.08 -0.08 0.62 66.14%
FM p-value (0.00) (0.58) (0.02) (0.73) (0.00)

The table contains results of asset pricing tests for a sample of 88,668 firm-year observations with fiscal years
ending in 1970 to 2004 and monthly returns collected from April 1971 to March 2006. All returns are reported in
percentages.

21

Table 1. Continued

Panel A reports mean coefficient estimates from the cross-sectional regression: R
i,t+1
−Rf
t+1
=Intercept
t
+ q
t

RDD
it
+ b
t
BETA
it
+ s
t
log(MKTV)
it
+ h
t
log(BMRATIO)
it
+ ε
it
, where R
i,t+1
denotes the return on stock i for month t+1;
Rf
t+1
is the one-month T-bill rate; RDD
it
is the decile rank based on the DD Measure; BETA
it
is the CAPM beta
estimated using five years of data ending the most recent December; MKTV
it
is the market value of equity;
BMRATIO
it
is the ratio of book value of equity to the market value of equity, where book value is stockholders’ book
equity plus deferred taxes minus the book value of preferred stock, and the market value of equity is defined as
previously. DD Measure, MKTV
it
, and BMRATIO
it
are estimated for the most recent fiscal year ending at least three
months prior to month t. Reported coefficients and adjusted R
2
s are the average values from 420 monthly
cross-sectional regressions. Two-sided Fama-MacBeth p-values are reported in parentheses.
Panel B reports descriptive statistics for the AQ factor and the three Fama-French factors. MktRf is the excess
return on the market portfolio. SMB is the return on the size factor-mimicking portfolio. HML is the return on the
book-to-market factor-mimicking portfolio. The AQ factor is an equally-weighted return on a hedge portfolio that is
long(short) the two top(bottom) quintiles of the DD Measure.
Panel C reports coefficient estimates for the regression: FRET
t
=α
i
+ β
i
MktRf
t
+ s
i
SMB
t
+ h
i
HML
t
+ ε
it
, where
FRET
t
is the return on the AQ factor for month t and all other variables are as defined above. The regression is based
on a series of 420 monthly returns. Two-sided p-values are reported in parentheses.
Panel D reports mean coefficient estimates from the stock-level cross-sectional regression: R
it
−Rf
it
= Intercept
t
+
b
^
i
AQ
λ
t
AQ
+ b
^
i
MktRf

λ
t
MktRft
+ b
^
i
SMB
λ
t
SMB
+ b
^
i
HML
λ
t
HML
+ ε

it
, where R

it
is return on stock i for month t; Rf
it
is the
one-month T-bill rate; and b
^
i
AQ
, b
^
i
MktRf
, b
^
i
SMB
, and b
^
i
HML
are factor betas estimated using all available returns for stock
i (with at least 18 monthly observations) from the time-series regression R
it
−Rf
it
=α
i
+ b
i
AQ
AQ
t
+ b
i
MktRfi
MktRf
t
+ b
i
SMB

SMB
t
+ b
i
HML
HML
t
+ ε
it
, where all variables are as defined above. Reported coefficients and adjusted R
2
s are the
average values from 420 monthly cross-sectional regressions. Two-sided Fama-MacBeth p-values adjusted for the
Shanken (1992) correction are reported in parentheses.
Panel E reports regression coefficients from the portfolio-level cross-sectional regressions based on the full-period
average excess returns and the full-period factor betas: PRET
i
- Rf =Intercept
t
+ b
^
i
AQ
λ
AQ
+ b
^
i
MktRf

λ
MktRft
+ b
^
i
SMB
λ

SMB
+ b
^
i
HML
λ
HML
+ ε
it
, where PRET
i
- Rf is the average excess return on portfolio i calculated using 420 monthly
portfolio returns; and b
^
i
AQ
, b
^
i
MktRf
, b
^
i
SMB
, and b
^
i
HML
are full-period factor betas estimated from the regression PRET
it

- Rf
t
=α
i
+ b
i
AQ
AQ
t
+ b
i
MktRfi
MktRf
t
+ b
i
SMB
SMB
t
+ b
i
HML
HML
t
+ ε
it
, where PRET
it
is the return on portfolio i for
month t and all variables are as defined above. Each regression is based on a series of 420 monthly returns. Two-sided
Fama-MacBeth p-values adjusted for the Shanken (1992) correction (reported in parentheses) are based on the
coefficients obtained by estimating the cross-sectional regression monthly. The panel partitions refer to three types of
portfolios: (1) “25 B/M and Size Portfolios” are based on quintiles of the book-to-market ratio and size; (2) “100 DD
Measure Portfolios” are based on percentiles of the DD Measure; and (3) “64 B/M, Size and DD Measure Portfolios”
are based on quartiles of the book-to-market ratio, size, and the DD Measure. The stocks in the portfolios are equally
weighted.
For each firm and each year t, DD Measure is calculated as the standard deviation of residuals over the [t-4, t]
period, where residuals are obtained from the following regression estimated cross-sectionally for each year and each of
the 48 Fama and French (1997) industries: TCA
it
= α
t
+ β
0t
1/ATA
it
+ β
1t
CFO
it-1
+ β
2t
CFO
it
+ β
3t
CFO
i t+1
+ β
4t

ΔREV
it
+ β
5t
PPE
it
+ ε
it,
where
it
TCA is total current accruals for year t (estimated using the balance sheet approach:
TCA
it
= ΔCA
it
- ΔCL
it
- ΔCash
it
+ ΔSTDEBT
it
, where ΔCA
it
is one-year change in current assets (#4); ΔCL
it
is
one-year change in current liabilities (#5); ΔCash
it
is one-year change in cash (#1); ΔSTDEBT
it
is one-year change in
debt in current liabilities (#34)); CFO
it
is cash flow from operations for year t estimated as NIBE
it
- TA
it
, where NIBE
it
is
net income before extraordinary items (#18); TA
it
is total accruals (TA
it =
TCA
it
- DEPN
it
, where DEPN
it
is depreciation
and amortization expense in year t (#14)); ΔREV
it
is one-year change in revenues (#12); PPE
it
is property, plant, and
equipment for year t (#7); ATA
it
is average value of total assets (#6) over years t-1, t, and t+1. All variables are scaled
by the average value of total assets. At least 20 observations are required for each regression.
All numbered items in the description of variables refer to Compustat annual data items.

The mean coefficient estimates are based on 420 monthly estimates from the
cross-sectional regressions. The p-values are based on the Fama-MacBeth t-statistics that
22

rely on the distribution of cross-sectional coefficients. The coefficient on the rank of the
DD Measure (RDD) represents an increase in the average realized return from one decile
of the DD Measure to the next and, when multiplied by nine, can be converted to the
difference in average returns between the first and the tenth deciles of the DD Measure.
The estimation results are similar to those reported in CGV . Specifically, the
coefficient on RDD is positive but not significant for the regression without controls,
while turning negative and not significant when the CAPM beta, the market value of
equity, and the book-to-market ratio are added as controls.
The above results are based on the cross-section of individual stock returns.
Next, I verify that the DD Measure does not predict future returns at the portfolio level.
As in FLOS and CGV , I form a factor-mimicking portfolio (the Accrual Quality (AQ)
factor) by taking a short (long) position in the stocks falling in the two bottom (top)
quintiles of the DD Measure in the beginning of each month.
Panel B reports the descriptive statistics for the AQ factor and for the three Fama
and French (1993) factors—market premium (MktRf), size (SMB), and book-to-market
(HML). The raw return on the AQ factor (0.16% per month) is not statistically significant.
Panel C presents the regression results corresponding to the AQ factor returns on the three
Fama-French factor returns. Collectively, the Fama-French factors explain 53% of the
variation in the AQ factor and there is no evidence of an abnormal return earned on the
AQ factor (Jensen’s alpha is equal 0.08% per month and is not statistically significant).
23

Overall, there is no evidence that lower accrual quality, as proxied by the DD
Measure, is associated with higher stock returns, which confirms CGV’s result.
4.1.2 AQ factor Loadings and Future Stock Returns
Thus far, I find no evidence that the DD Measure, as a characteristic, predicts
future returns. However, it is possible that the DD Measure is a noisy proxy for the
systematic component of accrual risk, that is, the returns’ co-variation with the AQ factor.
To establish whether the co-variation with the AQ factor is priced by the market, I
perform a two-stage analysis that is similar to that reported in CGV .
The two-stage technique is extensively used in the asset pricing literature to
verify whether the loadings on the candidate factor explain the cross-sectional variation in
expected stock returns, i.e., to verify whether stocks with higher “betas” earn higher
returns on average (e.g., Fama and MacBeth 1973; Jagannathan and Wang 1996). The
first stage involves estimating loadings on a set of factors—“factor betas”—by regressing
returns for each portfolio or stock on the contemporaneous factor returns. The first-stage
regressions are based on a time series of returns and are estimated separately for each
portfolio or each stock (subscripts i and t refer to individual stocks or portfolios and
months, respectively) as follows:
RET
it
- Rf
t
= α
i
+ b
i
AQ
AQ
t
+ b
i
MktRf
(Rm
t
−Rf
t
) + b
i
SMB
SMB
t

+ b
i
HML
HML
t
+ ε
it
, (4)
where
24

RET
it
= return on a portfolio i or stock i for month t;
Rf
t
= one-month T-bill rate;
AQ
t
= return on the AQ factor;
Rm
t
= return on a market portfolio (CRSP value-weighted
index);
SMB
t
= return on a factor-mimicking hedge portfolio for size;
HML
t
= return on a factor-mimicking hedge portfolio for
book-to-market.

For the individual stocks, factor loadings are estimated using all available
returns, with at least 18 monthly observations required for estimation. For portfolios,
factor loadings are estimated using the entire time series of portfolio returns.
The second stage involves a cross-sectional regression of realized returns on the
factor loadings estimated in the first stage (subscripts i and t refer to individual stocks or
portfolios and months, respectively):
R
it
−Rf
it
= Intercept

t
+ b
i
^ AQ
λ
t
AQ
+ b
i
^ MktRf

λ
t
MktRf
+ b
i
^ SMB
λ
t
SMB

+ b
i
^ HML
λ
t
HML
+ ε

it
, (5)
where
R
it
= return on stock i for month t, or average return on
portfolio i;
25

Rf
it
= one-month T-bill rate for month t;
b
i
^ AQ,
, b
i
^ MktRf
,
b
i
^ SMB
, b
i
^ HML

= factor betas estimated in the first stage.
14

At the portfolio level, a single cross-sectional regression is estimated using
averages of the 420 monthly portfolio returns, while at the stock level the cross-sectional
regressions are estimated monthly. Both at the portfolio level and at the stock level, the
statistical significance of regression coefficients is based on Fama-MacBeth t-statistics
and reflects the Shanken (1992) correction.
Panel D of Table 1 reports results of a second-stage estimation at the firm level.
The premium on the AQ factor (coefficient on b
AQ
) is not statistically significant. The sign
and significance of the other factors are generally consistent with what CGV document
(although CGV find that the market premium is positive and significant at the 5% level).
Panel E of Table 1 reports the results of a second-stage estimation at the
portfolio level for three sets of portfolios: (1) 25 portfolios formed by independently
sorting stocks into quintiles based on the book-to-market ratio and size; (2) 100 portfolios
based on the percentiles of the DD Measure; and (3) 64 portfolios formed by
independently sorting stocks into quartiles based on book-to-market, size, and the DD
Measure. Portfolio returns are calculated by equally weighting the individual stock

14
When the second-stage regressions include only some of the factor betas, e.g., only b
i
^
AQ
, the included factor betas are
based on the first-stage regressions that include only the returns on the relevant factors, e.g., only the return on the AQ
factor.

26

returns.
15
For each portfolio set, I estimate three types of factor models: (1) the
benchmark model that includes only the three Fama-French factors; (2) the benchmark
model that includes only the AQ factor; and (3) the full model, which includes the three
Fama-French factors plus the AQ factor. The results are consistent with the findings in
CGV—the premium accorded to the AQ factor (i.e., the regression coefficient on b
i
AQ
) is
not statistically significant for any set of portfolios used in analysis.
The magnitudes and significance of premiums on all factors are mostly
consistent with those reported by CGV . For the 25 and the 64 portfolios, only the premium
on the HML factor is significantly positive; the premium on the market factor is negative
and significant and the premium on the SMB factor is not significant. For the 100
portfolios based on the DD Measure, none of the factors has a significantly positive risk
premium.
Overall, CGV’s results are confirmed in the extended sample of this
paper—there is no evidence that higher exposure to accrual quality risk, as proxied by the
loading on the AQ factor, is associated with higher stock returns.

15
The weighting procedure is different from that in CGV, who use value-weighting to calculate portfolio returns. While
value-weighting has been frequently used in prior literature (e.g., Fama and French 1993), equal-weighting is also
acceptable (e.g., Carhart 1997). To detect the effect of accrual quality it may be preferable to use equal-weighting since
it is likely that the quality of accruals (or accounting information in general) is more important for smaller firms with
less saturated information environments.
27

4.2 The DD Measure and Cash Flow Shocks
4.2.3 Univariate Correlations
Table 2 reports correlations between the DD Measure and the stock
characteristics linked to the abnormal stock returns: (1) past sales growth (SALEGRW); (2)
the Altman (1968) Z-score (ALTM), which is inversely related to the probability of
bankruptcy; (3) the Piotroski (2000) F-score (PIOTR), which is based on eight signals of
financial health;
16
(4) the Kaplan and Zingales (1997) score measuring the degree of
financing constraints (KZ_INDEX); (5) an indicator variable for past earnings variability
exceeding the industry median (EVAR); and (6) an indicator variable for past sales
variability exceeding the industry median (SVAR). All reported correlations represent
averages of the yearly coefficients; the reported p-values are based on the Fama-MacBeth
(1973) t-statistics and are adjusted for serial correlation using the Newey-West (1987)
method.
17

16
I measure the Piotroski (2000) F-score as a sum of eight indicators: (1) positive earnings before extraordinary items
(EBEI); (2) positive operating cash flow; (3) positive change in EBEI; (4) positive total accruals; (5) negative change in
leverage; (6) positive change in current ratio; (7) positive change in gross margin ratio; and (8) positive change in asset
turnover. Piotroski (2000) also includes an indicator for recent equity issues.
17
This method of calculating correlations assures that the documented dependencies are cross-sectional in nature and
do not arise due to possible co-variation among the variables over time. Controlling for the time trends is important
given the pronounced increasing trend in the DD Measure documented by Rajgopal and Venkatachalam (2007) over
1962 to 2002.
28

Table 2. Correlations between Firm Characteristics and DD Measure

DD MEASURE
Variable # Obs. PEARSON SPEARMAN
Firm Characteristics:
|ACC| 88,668 0.36 (0.00) 0.27 (0.00)
ACC 88,668 -0.02 (0.09) -0.01 (0.22)
SALEGRW 79,581 0.10 (0.97) 0.01 (0.00)
ALTM 88,601 -0.16 (0.00) -0.05 (0.00)
PIOTR 88,668 -0.12 (0.01) -0.10 (0.00)
KZ_INDEX 88,387 -0.04 (0.15) 0.05 (0.20)
EVAR 78,194 0.35 (0.00) 0.35 (0.00)
SVAR 78,230 0.26 (0.00) 0.27 (0.00)

Cash Flow Shock Proxies:
SURP 87,828 -0.08 (0.01) -0.04 (0.00)
CN_EM 31,221 -0.16 (0.05) -0.15 (0.00)

Risk Proxies:
BETA 88,668 0.15 (0.01) 0.19 (0.01)
MKTV 88,668 -0.19 (0.00) -0.46 (0.00)
BMRATIO 88,668 -0.03 (0.19) -0.03 (0.02)

The table reports Pearson and Spearman correlations between various firm characteristics and the DD Measure for
a sample of 88,668 firm-year observations with fiscal years ending in 1970 to 2004. The exact number of observations
varies depending on the variable.
The variables are defined as follows: |ACC| is absolute value of total accrual; ACC is total accrual estimated using
balance-sheet method as ΔCA
it
- ΔCL
it
- ΔCash
it
+ ΔSTDEBT
it
- DEPN
it
,, where ΔCA
it
is one-year change in current
assets (#4); ΔCL
it
is one-year change in current liabilities (#5); ΔCash
it
is one-year change in cash (#1); ΔSTDEBT
it
is
one-year change in debt in current liabilities (#34)); DEPN
it
is depreciation and amortization expense in year t
(#14)SALEGRW is average rate of growth in sales (#12) over the past five years; ALTM is Altman probability of
bankruptcy; PIOTR is the eight-signal financial health F-score from Piotroski (2000) ( the score is a sum of eight
indicators for: (1) positive earnings before extraordinary items (EBEI); (2) positive operating cash flow; (3) positive
change in EBEI; (4) positive total accruals; (5) negative change in leverage; (6) positive change in current ratio; (7)
positive change in gross margin ratio; and (8) positive change in assets turnover); KZ_INDEX is degree of financing
constraints using the Lamont et al. (2001) specification of the Kaplan and Zingales (1997) model (KZ INDEX =
−1.002× CashFlow + 0.283 × Q + 3.139 × Leverage −39.368 × Dividends − 1.315 × CashHoldings, where CashFlow =
(#18 + #14)/#8; Q = (#6 + CRSP December market value – #60 – #74) / #6; Leverage = (#9 + #34) / (#9 + #34 +
#216); Dividends = (#21 + #19) / #8; Cash Holdings = #1 / #8; where #8 is taken from the beginning of fiscal year);
EVAR is an indicator variable equal to one if the firm’s variability of earnings exceeds the industry median, where
earnings is defined as earnings before extraordinary items (#18) scaled by the beginning-of-year total assets (#6); SVAR
is an indicator variable equal to one if the firm’s variability of sales (#12) scaled by the beginning-of-year total assets
(#6) exceeds the industry median; for both EVAR and SVAR the variability is computed over the past ten years and
industry is defined as one of the 48 industries in Fama and French (1997) classification; BETA is CAPM beta estimated
using five years of data ending the most recent December; MKTV is the market value of equity from Compustat for the
fiscal year-end; BMRATIO is the ratio of book value of equity to the market value of equity, where book value is
stockholders’ book equity plus deferred taxes minus the book value of preferred stock, and market value of equity is
defined as previously; SURP is the earnings surprise based on the statistical earnings prediction model described in the
Appendix; and CN_EM is the cash flow news proxy estimated as in Easton and Monahan (2005), where the details of
the estimation are described in the Appendix.
29

Table 2. Continued

Correlations are computed by year. All variables are winsorized at the top and bottom 1% within each year.
Reported correlation coefficients are average values for 35 years (correlation coefficients for CN_EM are based on 24
years). Two-sided Fama-MacBeth p-values are reported in parentheses. The p-values are computed with a Newey-West
(1987) correction with six lags.
All numbered items in the description of variables refer to Compustat annual data items.

Out of the six analyzed characteristics, four are negatively related to future
returns, including past sales growth (SALEGRW), the Kaplan and Zingales
degree-of-financing-constraints score (KZ_INDEX), and the two variability indicators
(EVAR and SVAR), and two are positively related to future returns, including the Altman
Z-score (ALTM) and the Piotroski F-score (PIOTR). The correlations between the
characteristics and the DD Measure are generally consistent with the presence of
underperformance for the lower accounting quality stocks. Poor accounting quality (i.e.,
high DD Measure) stocks have higher average past growth in sales (SALEGRW), a higher
probability of bankruptcy (ALTM), a lower Piotroski F-score (PIOTR), and higher sales
(SVAR) and earnings (EVAR) variability compared to their respective industry averages.
18

I also investigate the degree of correlation between the DD Measure and the
signed and unsigned total accruals. Sloan (1996) documents that signed total accruals

18
To provide evidence on whether the previously documented abnormal returns related to these characteristics are
caused by the reaction to cash flow news, I investigate whether the characteristics also predict future cash flow shocks
measured by the two cash flow news proxies discussed in Appendix. The results (unreported) provide evidence
consistent with some of the abnormal returns being caused by the cash flow shocks. The correlations with the elevated
variability indicators (EVAR and SVAR) are significantly negative for the Pearson correlations and the Spearman
correlations with the analyst forecasts-based proxy. For all other characteristics, the Spearman correlations with each
cash flow shock proxy have the expected signs and are statistically significant. Specifically, the correlations are negative
for past sales growth and the degree of the Kaplan-Zingales financing constraints score, and they are positive for the
Piotroski F-score and the Altman Z-score. The Pearson correlations, although generally of the hypothesized signs, are
significant only for some variables and some cash flow news proxies.

30

predict future abnormal returns. Although the DD Measure depends on the variance of
current accrual, to the extent that accrual distribution is skewed the variance and the mean
of accruals may be negatively or positively correlated. If the correlation between the DD
Measure and the signed accruals is high, the predictive ability of the DD Measure with
respect to future returns needs to be distinguished from the predictive ability of total
accruals. According to Table 2, the correlation between the absolute value of total accruals
and the DD Measure is high, 0.36/0.27 Pearson/Spearman. However, the correlation with
signed accruals is only -0.02/-0.01 Pearson/Spearman.
Next I correlate the DD Measure with the two direct proxies for cash flow
shocks. The first proxy is an earnings surprise measure, SURP (henceforth, earnings
surprise). The earnings surprise measure is based on the expectation of earnings derived
from a simple statistical earnings prediction model. The model assumes that earnings can
be predicted using the previous year’s earnings and returns. The model is estimated
cross-sectionally for the hold-out period of one year by regressing earnings on the
previous year’s earnings and returns. The resulting coefficients are used to construct the
next year’s predicted earnings. The second cash flow news proxy is constructed using
I/B/E/S analysts’ EPS forecasts, as suggested by Easton and Monahan (2005) (henceforth,
E&M cash flow news proxy), and is labeled CN_EM.
19
The estimation details of the two
proxies are reported in the Appendix.

19
Among other studies that use earnings and earnings forecasts to proxy for cash flow news are Barth et al. (2006) and
Campbell et al. (2005). Barth et al. use analyst forecast revisions to measure cash flow news in the context of credit risk
evaluation. Campbell et al. use changes in discounted sums of ROE to proxy for the cash flow shock portion of realized
stock returns.
31

Table 2 reports correlations between the DD Measure and the direct proxies for
cash flow news. As expected, there is a negative correlation between each cash flow news
proxy and the DD Measure. Panels A and B of Figure 1 plots the mean and the median
values of the earnings surprise (Panel A) and the E&M cash flow news proxy (Panel B) by
the DD Measure decile. While the magnitudes of the correlation coefficients do not
appear large in absolute values and range from -0.16 to -0.04, there is a strong monotonic
decreasing pattern for the cash flow news proxies across the deciles of the DD Measure.
To provide evidence on the shape of the distribution of each cash flow news proxy across
the DD Measure deciles, I also plot the mean plus or minus one standard deviation in
Panels C and D of Figure 1. Interestingly, the variance of each cash flow news proxy
increases with the magnitude of the DD Measure. This pattern is consistent with lower
predictability of earnings for the stocks with lower accrual quality.
32

Figure 1. Cash Flow Shocks and the DD Measure

Panel A. Earnings Surprise by Decile of the DD Measure
-0.04
-0.02
0
0.02
0.04
0.06
0.08
1 2 3 4 5 6 7 8 9 10
SURP (Mean)
SURP (Median)

Panel B. E&M Cash Flow News Proxy by Decile of the DD Measure
-0.14
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0
1 2 3 4 5 6 7 8 9 10
CN_EM (Mean)
CN_EM (Median)

33

Figure 1. Continued.

Panel C. Earnings Surprise Distribution by Decile of the DD Measure
-0.7
-0.5
-0.3
-0.1
0.1
0.3
0.5
1 2 3 4 5 6 7 8 9 10
SURP (Mean)
+ 1 stddev
- 1 stddev

Panel D. E&M Cash Flow News Proxy Distribution by Decile of the DD Measure
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
1 2 3 4 5 6 7 8 9 10
CN_EM (Mean)
+ 1 stddev
- 1 stddev

The figure plots the univariate statistics for cash flow news proxies by decile of the DD Measure for a sample of
88,668 firm-year observations with fiscal years ending in 1970 to 2004. The exact number of observations varies
depending on the availability of data for measuring each cash flow news proxy. Panels A and C contain mean/median
and mean/standard deviation statistics for earnings surprises (SURP) for a sample of 87,828 firm-years. Panels B and D
contain mean/median and mean/standard deviation statistics for the Easton and Monahan (2005) cash flow news proxy
(CN_EM) for a sample of 31,221 firm-years with fiscal years in 1981 to 2004.
34

Table 2 also reports the correlations between the DD Measure and several
risk-related characteristics. High DD Measure (low accrual quality) stocks have higher
CAPM betas (BETA), lower market capitalizations (MKTV), and lower book-to-market
ratios (BMRATIO).
To summarize, firms with higher levels of the DD Measure (lower accrual
quality) tend to have characteristics associated with negative future abnormal returns.
More importantly, there is a significant negative correlation between the DD Measure and
the two direct cash flow news proxies that are based on a simple statistical earnings
prediction model and analysts’ earnings forecast revisions. Overall, this evidence is
consistent with firms with high DD Measures (low accrual quality) being systematically
exposed to more negative cash flow shocks in the near future.
4.2.4 Asset Pricing Tests: Controlling for Cash Flow Shocks
Next I examine whether the DD Measure explains the cross-sectional variation
in returns after controlling for the returns generated by cash flow shocks. To do so, I
include the earnings surprise (SURP) or the E&M cash flow news proxy (CN_EM) in each
cross-sectional asset pricing regression described in Section 4.1. Requiring that each
observation have a non-missing value for the earnings surprise (E&M cash flow news)
proxy constrains the sample to 87,828 observations within the 1970 to 2004 period
(31,221 observations within the 1981 to 2004 period).
35

I first verify that CGV’s results are present in these restricted samples. Each
panel of Table 3 replicates one of CGV’s cross-sectional regression tests: Panel A contains
stock-level regressions of future returns on the ranked DD Measure (RDD), Panel B
contains stock-level regressions of returns on the AQ factor beta, and Panel C contains
portfolio-level regressions of returns on the AQ factor beta. The partitions of Panels A and
B pertain to the samples constrained by the availability of each cash flow shock proxy.
Panel C is partitioned by the type of cash flow shock proxy and the type of portfolios used
in analysis. The first line of each partition reports the results of the replication. The
premiums on RDD and on the AQ factor (coefficient on the AQ factor beta) are not
significant for either restricted sample in the stock-level regressions (Panels A and B). At
the portfolio level, the premiums on the AQ factor (coefficients on the AQ factor beta
reported in Panel C) are not significant for any portfolio partition and any restricted
sample, with the exception of 25 size and book-to-market portfolios in a sample
constrained by the availability of the E&M cash flow news proxy (CN_EM), where the
premium is equal to 1.09% per month (p-value of 0.01).

36

Table 3. Controlling for Cash Flow Shocks

Panel A. Returns on AQ Characteristic: Firm-Level Cross-Sectional Regressions
Intercept RDD
CF Shock
Proxy BETA
log
(MKTV)
log
(BM
RATIO)
Adjusted
R
2

CF Shock Proxy = SURP
Mean Coefficient 0.79 0.03 0.90%
FM p-value (0.00) (0.38)
Mean Coefficient 0.43 0.06 6.06 2.16%
FM p-value (0.05) (0.05) (0.00)
Mean Coefficient 1.50 0.00 6.54 0.12 -0.17 0.48 4.72%
FM p-value (0.00) (0.90) (0.00) (0.41) (0.00) (0.00)
CF Shock Proxy = CN_EM
Mean Coefficient 0.93 0.00 1.45%
FM p-value (0.00) (0.91)
Mean Coefficient 0.99 0.06 6.14 3.20%
FM p-value (0.00) (0.07) (0.00)
Mean Coefficient 2.46 0.01 6.13 0.14 -0.21 -0.02 6.32%
FM p-value (0.00) (0.72) (0.00) (0.50) (0.00) (0.86)

Panel B. Returns on the AQ factor Loadings: Firm-Level Cross-Sectional Regressions
Intercept
b
^
AQ

CF Shock
Proxy
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

CF Shock Proxy = SURP
Mean Coefficient 0.78 0.16 2.89%
FM p-value (0.00) (0.39)
Mean Coefficient 0.36 0.37 6.10 3.96%
FM p-value (0.09) (0.05) (0.00)
Mean Coefficient 0.23 0.36 5.90 0.27 0.24 -0.08 7.36%
FM p-value (0.04) (0.05) (0.00) (0.26) (0.17) (0.61)
CF Shock Proxy = CN_EM
Mean Coefficient 0.92 0.06 4.03%
FM p-value (0.00) (0.81)
Mean Coefficient 1.04 0.42 6.14 5.46%
FM p-value (0.00) (0.08) (0.00)
Mean Coefficient 0.72 0.28 5.89 0.54 0.22 -0.26 9.51%
FM p-value (0.00) (0.22) (0.10) (0.10) (0.26) (0.20)
37

Table 3. Continued

Panel C. Returns on the AQ factor Loadings: Portfolio-Level Cross-Sectional Regressions
Independent Variables
Adjusted
R
2

Intercept
b
^
AQ

CF Shock
Proxy
b
^
MktRf
b
^
SMB
b
^
HML

CF Shock Proxy = SURP
25 B/M and Size Portfolios
Regression Coeff. 2.44 0.00 -1.76 -0.11 0.70 79.44%
FM p-value (0.00) (0.99) (0.00) (0.62) (0.00)
Regression Coeff. 0.70 0.41 6.35 -0.67 0.10 1.01 86.18%
FM p-value (0.01) (0.25) (0.00) (0.04) (0.83) (0.00)
100 DD Measure Portfolios
Regression Coeff. 0.23 0.15 0.31 0.38 0.28 28.06%
FM p-value (0.53) (0.38) (0.51) (0.13) (0.33)
Regression Coeff. 0.07 0.41 5.04 0.22 0.34 0.24 38.57%
FM p-value (0.85) (0.08) (0.00) (0.43) (0.09) (0.48)
64 B/M, Size and DD Measure Portfolios
Regression Coeff. 1.70 0.09 -1.06 -0.07 0.63 68.47%
FM p-value (0.00) (0.63) (0.02) (0.76) (0.00)
Regression Coeff. 1.00 0.31 4.45 -0.77 0.07 0.84 75.31%
FM p-value (0.06) (0.17) (0.00) (0.21) (0.69) (0.00)
CF Shock Proxy = CN_EM
25 B/M and Size Portfolios
Regression Coeff. 1.02 1.09 -0.25 0.15 0.40 86.15%
FM p-value (0.17) (0.01) (0.33) (0.41) (0.14)
Regression Coeff. 0.73 1.38 3.38 0.20 0.26 0.26 86.89%
FM p-value (0.22) (0.00) (0.00) (0.47) (0.17) (0.65)
100 DD Measure Portfolios
Regression Coeff. 0.89 0.16 0.05 0.26 -0.33 20.65%
FM p-value (0.49) (0.28) (0.33) (0.52) (0.99)
Regression Coeff. 0.87 0.30 1.72 0.12 0.33 -0.31 20.58%
FM p-value (0.52) (0.02) (0.00) (0.32) (0.10) (0.86)
64 B/M, Size and DD Measure Portfolios
Regression Coeff. 0.88 0.41 -0.19 0.35 0.31 29.90%
FM p-value (0.02) (0.23) (0.28) (0.19) (0.15)
Regression Coeff. 0.85 0.85 7.54 0.15 0.73 0.02 44.35%
FM p-value (0.02) (0.00) (0.00) (0.47) (0.17) (0.65)

The table reports results of asset pricing tests that include controls for cash flow shocks: either the earnings
surprise (SURP) or the Easton and Monahan cash flow news measure (CN_EM). The tests using SURP are based on a
sample of 87,828 observations with fiscal years ending in 1970 to 2004 (with returns earned from April 1971 to March
2006). The tests using CN_EM measure are based on a sample of 31,221 firm-years with fiscal years in 1981 to 2004
(with returns earned from May 1981 to March 2006).
Panel A reports mean coefficient estimates from cross-sectional regressions: R
i,t+1
−Rf
t+1
=Intercept
t
+ q
t
RDD
it
+
c
t
CF_SHOCK
it
+ b
t
BETA
it
+ s
t
log(MKTV)
it
+ h
t
log(BMRATIO)
it
+ ε
it
, where R
i,t+1
denotes the return on stock i for
month t+1; Rf
t+1
is the one-month T-bill rate; RDD is the decile rank based on the DD Measure; CF_SHOCK
it
is either
earnings surprise (SURP) or Easton&Monahan cash flow news measure (CN_EM). All other variables are defined in
Table 2.
38

Table 3. Continued

Panel B reports mean coefficient estimates from cross-sectional regressions: R
it
−Rf
it
= Intercept
t
+ b
^
i
AQ
λ
t
AQ
+ c
t
CF_SHOCK
it
+ b
^
i
MktRf

λ
t
MktRft
+ b
^
i
SMB
λ
t
SMB
+ b
^
i
HML
λ
t
HML
+ b
^
i
AQ
λ
t
AQ
+ ε

it
, where R

it
is return on stock i for month t; Rf
it
is the one-month T-bill rate; CF_SHOCK
it
is either earnings surprise (SURP) or Easton and Monahan cash flow-news
measure (CN_EM); and b
^
i
AQ
, b
^
i
MktRf
, b
^
i
SMB
, and b
^
i
HML
are factor betas estimated using all available returns for stock i
(with at least 18 monthly observations) from the time-series regression R
it
−Rf
it
=α
i
+ b
i
AQ
AQ
t
+ b
i
MktRfi
MktRf
t
+ b
i
SMB

SMB
t
+ b
i
HML
HML
t
+ ε
it
, where all variables are as defined above.
Panel C reports regression coefficients from the cross-sectional regressions based on the full-period average
excess returns and the full-period factor betas: PRET
i
- Rf =Intercept

+ b
^
i
AQ
λ
AQ
+ c
t
CF_SHOCK
i
+ b
^
i
MktRf

λ
MktRft
+ b
^
i
SMB
λ

SMB
+ b
^
i
HML
λ
HML
+ ε
it
, where PRET
i
- Rf is the average excess return on portfolio i calculated using
420 monthly portfolio returns; CF_SHOCK
i
is the time-series average of portfolio-level cash flow shock proxy,
which is calculated as an average of either earnings surprise (SURP) or Easton and Monahan cash flow news measure
(CN_EM) across all stocks comprising portfolio i; and b
^
i
AQ
, b
^
i
MktRf
, b
^
i
SMB
, and b
^
i
HML
are full period factor betas. For
each portfolio, the full-period factor betas are estimated from a regression PRET
it
- Rf
t
=α
i
+ b
AQ
i
AQ
t
+ b
MktRf
i
MktRf
t

+ b
SMB
i
SMB
t
+ b
HML
i
HML
t
+ ε
it
, where all variables are defined above. Each regression is based on a series of 420
monthly returns. The panel partitions refer to three types of portfolios: (1) “25 B/M and Size Portfolios” are based on
quintiles of the book-to-market ratio and size; (2) “100 DD Measure Portfolios” are based on percentiles of the DD
Measure; and (3) “64 B/M, Size and DD Measure Portfolios” are based on quartiles of the book-to-market ratio, size,
and the DD Measure. The stocks in the portfolios are equally weighted.
Coefficients and adjusted R
2
s reported in Panels A and B are the average values from 420 (299) monthly
cross-sectional regressions when SURP (CN_EM) is used as a cash flow shock proxy. Coefficients and adjusted R
2
s
reported in Panel C are from a single cross-sectional regression that uses average portfolio returns. All panels report
two-sided Fama-MacBeth p-values that are based on the coefficients obtained by estimating the cross-sectional
regression monthly. P-values reported in Panels B and C are adjusted for the Shanken (1992) correction.

Next, I control for the effects of cash flow shocks by adding the cash flow shock
proxies as additional independent variables in each of the cross-sectional asset pricing
regressions. Panel A of Table 3 reports results of various versions of the following
cross-sectional regression testing whether the DD Measure as a characteristic predicts
future stock returns after controlling for cash flow shocks:
R
i,t+1
−Rf
t+1
=Intercept
t
+ q
t
RDD
it
+ c
t
CF_SHOCK
it

+ b
t
BETA
it
+ s
t
log(MKTV)
it
+ h
t
log(BMRATIO)
it
+ ε
it
, (6)
where
RDD
it

=
decile rank based on the DD Measure;
CF_SHOCK
it

=
either earnings surprise (SURP) or the E&M cash
39

flow news measure (CN_EM);
and all other variables are as defined earlier.
After controlling for either cash flow shock proxy, the premium on the DD
Measure (coefficient on RDD) increases to 0.06% monthly and becomes statistically
significant (p-value = 0.06). The premium is also economically significant and can be
interpreted as the difference between the highest and the lowest deciles of accrual quality
of approximately 6.5% per annum (0.06%*12*9). Both cash flow shock proxies are
significantly correlated with contemporaneous returns indicating that they do indeed
capture new information.
However, the premium on RDD is reduced in magnitude and is no longer
statistically significant after the three Fama-French factors—the book-to-market ratio
(log(BMRATIO)), the CAPM beta (BETA), and the market value of equity
(log(MKTV))—are included as additional controls.
20
This result is consistent with either
accounting risk being subsumed by other known risk factors or the DD Measure being a
noisy proxy for the exposure to systematic accrual risk. I further distinguish between the
two explanations by controlling for cash flow shocks in the second stage of the
regressions that use AQ factor betas as measures of systematic accrual quality risk (the
two-stage estimation is described earlier in Section 4.1.2).
Panel B of Table 3 reports results of the second-stage regressions at the
individual stock level. After controlling for earnings surprise and the three Fama-French

20
Additional tests (not reported) indicate that of the three characteristics, only including the market value of equity
leads to an insignificant premium on the DD Measure.
40

factor betas, the premium on the AQ factor becomes statistically significant and is equal
to 0.37% per month (approximately 4.4% per annum). After controlling for the E&M cash
flow news proxy and the three Fama-French factor betas, the premium is 0.28% per
month (approximately 3.4% per annum), but is not statistically significant.
The stock-level estimation may lack power in detecting the premium associated
with the AQ factor. Panel C of Table 3 reports the results of the portfolio-level analysis.
The portfolio-level cash flow shock proxies are obtained by averaging across all stocks
included in each portfolio. When the regressions include earnings surprise (SURP) as a
cash flow shock proxy, the premium on the AQ factor becomes statistically significant (at
the 10% level) for the 100 portfolios and marginally significant (p-value of 0.17) for the
64 portfolios. For the 25 portfolios, the premium is large in magnitude (0.41% per month),
but not statistically significant (p-value of 0.25). When the E&M cash flow news proxy
(CN_EM) is included in the regressions, the premium on the AQ factor is statistically
significant, equal to 0.30% and 0.85% per month for the 100 portfolios and the 64
portfolios, respectively. For the 25 portfolios, the premium is increased from 1.09% to
41

1.38% per month after controlling for the E&M cash flow news proxy and remains
statistically significant.
21

The purpose of including cash flow shock proxies in asset pricing regressions is
to control for the returns associated with unexpected cash flows (or earnings). The
magnitude of the return response to an unexpected earnings shock depends not only on the
magnitude of the shock, but also on its persistence (e.g., Kormendi and Lipe 1987). The
literature on the determinants of Earnings Response Coefficients (ERCs) suggests that
both losses (Hayn 1995) and extreme earnings (Freman and Tse 1992) have lower
persistence. To account for differences in the persistence of cash flow shocks across
deciles of accrual quality, I estimate cross-sectional asset pricing regressions which allow
the coefficient on the earnings surprise to change depending on whether the reported
earnings is a loss or an extreme earnings realization. Specifically, instead of controlling
for a single earnings surprise variable, I control for the change in earnings, the level of
one-period-ahead earnings, indicators for losses and the extreme 20% of earnings, and the
interaction between one-period-ahead earnings with losses and extreme earnings
indicators. The change in earnings and the level of earnings are included to account for
average earnings persistence and the magnitude of earnings surprises (Ali and Zarowin

21
CGV caution against using equal portfolio weights when calculating daily returns for the AQ factor due to the bid-ask
bounce bias associated with frequent re-balancing (Blume and Stambaugh 1983). Although equal weighting is unlikely
to cause a bid-ask bounce bias when monthly returns are used in the analysis, I verify whether my results are robust to
estimating annual buy-and-hold returns on the SAQ factor and using annual re-balancing. Under the alternative return
estimation procedure, I sort stocks into portfolios once a year in the beginning of each April and calculate portfolio
returns by equally weighting the annual buy-and-hold returns on the individual stocks. This alternative procedure yields
an average abnormal (with respect to the three-factor model) return to the SAQ factor of 1.65% per annum that is
significant at the 10% level.

42

1992). I also include the return earned on the firm’s stock over the previous fiscal year (as
in the model used to generate SURP described in the Appendix). Three other
specifications also allow a different coefficient on extreme earnings changes or
decompose earnings and earnings changes into cash flow and accrual components. The
four regression specifications are:
R
i,t+1
−Rf
t+1
=Intercept
t
+ q
t
RDD
it
+ b
t
BETA
it
+ s
t
log(MKTV)
it

+ h
t
log(BMRATIO)
it
+ c
0
t
LAGRET
it
+ c
1
t
ROE
it+1
+ c
2
t
CHROE
it+1
+ c
3
t
LOSS
it+1
+
c
4
t
ROE*LOSS
it+1
+ c
5
t
EXTR
it+1
+ c
6
t
ROE*EXTR
it+1
+ ε
it
, (7a)

R
i,t+1
−Rf
t+1
=Intercept
t
+ q
t
RDD
it
+ b
t
BETA
it
+ s
t
log(MKTV)
it

+ h
t
log(BMRATIO)
it
+ c
0
t
LAGRET
it
+ c
1
t
CFO
it+1
+ c
2
t
ACC
it+1
+ c
3
t
CHCFO
it+1

+ c
4
t
CHACC
it+1
+ c
5
t
LOSS
it+1
+ c
6
t
CFO*LOSS
it+1
+ c
7
t
ACC*LOSS
it+1
+ c
8
t
EXTR
it+1
+ c
9
t
CFO*EXTR
it+1
+ c
10
t
ACC*EXTR
it+1
+ ε
it
, (7b)

R
i,t+1
−Rf
t+1
=Intercept
t
+ q
t
RDD
it
+ b
t
BETA
it
+ s
t
log(MKTV)
it

+ h
t
log(BMRATIO)
it
+ c
0
t
LAGRET
it
+ c
1
t
ROE
it+1
+ c
2
t
CHROE
it+1
+ c
3
t
LOSS
it+1

+ c
4
t
ROE*LOSS
it+1
+ c
5
t
EXTR
it+1
+ c
6
t
ROE*EXTR
it+1
+ c
7
t
EXTRCH
it+1

+ c
8
t
CHROE*EXTRCH
it+1
+ ε
it
, (7c)

R
i,t+1
−Rf
t+1
=Intercept
t
+ q
t
RDD
it
+ b
t
BETA
it
+ s
t
log(MKTV)
it

+ h
t
log(BMRATIO)
it
+ c
0
t
LAGRET
it
+ c
1
t
CFO
it+1
+ c
2
t
ACC
it+1
+ c
3
t
CHCFO
it+1

43

+ c
4
t
CHACC
it+1
+ c
5
t
LOSS
it+1
+ c
6
t
CFO*LOSS
it+1
+ c
7
t
ACC*LOSS
it+1

+ c
8
t
EXTR
it+1
+ c
9
t
CFO*EXTR
it+1
+ c
10
t
ACC*EXTR
it+1
+ c
11
t
EXTR
it+1

+ c
12
t
CFO*EXTR
it+1
+ c
13
t
ACC*EXTR
it+1
+ ε
it
, (7d)
where
RDD
it

=
decile rank based on the DD Measure;
ROE
it+1
= return on equity during fiscal year t+1, i.e.,
earnings before extraordinary items (Compustat
#18) over the value of equity at the beginning of
fiscal year t+1 (Compustat #60);
CFO
it+1
= cash flow from operations for year t estimated as
ROE
it+1
- ACC
it+1
;
ACC
it+1
= is total accruals scaled by average value of total
assets (#6) over years t-1, t, and t+1 (as defined in
Table 1);
CHROE
it+1
= change in ROE from year t to year t+1;
CHCFO
it+1
= change in CFO from year t to year t+1;
CHACC
it+1
= change in ACC from year t to year t+1;
LOSS
it+1
= indicator variable equal to one if FROE
it
is
negative;
EXTR
it+1
= indicator variable equal to one if FROE
it
falls into
one of the extreme deciles for a given year;
44

EXTRCH
it+1
= indicator variable equal to one if CHROE
it
falls
into one of the extreme deciles for a given year; all
other variables are as defined earlier;
LAGRET
it
= the raw return on the stock of the firm computed
over fiscal year t. Year t is the most recent fiscal
year ending at least three months prior to month t;
and all other variables are as defined earlier.
Panels A and B of Table 4 contains the results of the cross-sectionally
unconstrained stock-level estimation. In Specifications I to IV , the premium on the RDD
characteristic is equal to 0.02% per month (with p-value ranging from 0.08 to 0.13) after
controlling for the three Fama-French characteristics, and a premium on the AQ factor
loading ranges between 0.49% to 0.50% per month (p-value = 0.01) after controlling for
the three Fama-French factors. The coefficients on the control variables are consistent
with lower persistence of losses and extreme earnings – the coefficients on the interaction
variables between future ROE and loss/extreme earnings indicators are significantly
negative. When earnings are decomposed into accruals and cash flows, the accrual part is
more persistent in the full specification that identifies both the extreme level and extreme
changes in earnings. Also, cash flows and accruals are less persistent when combined with
extreme or negative earnings.
22

22
Note that the coefficient on earnings level (earnings change) decreases (increases) with the degree of earnings
persistence, see Ohlson (1995) for the theoretical derivation.
45

Table 4. Cross-Sectionally Unconstrained Cash Flow Shock Controls

Panel A. Returns on AQ Characteristic: Firm-Level Analysis
Independent Variables Coef.
FM
p-valu
e Coef.
FM
p-valu
e Coef.
FM
p-valu
e Coef.
FM
p-valu
e
Intercept 0.73 (0.00) 1.41 (0.00) 0.84 (0.00) 1.43 (0.00)
RDD 0.02 (0.13) 0.02 (0.11) 0.02 (0.12) 0.02 (0.08)
BETA 0.25 (0.08) 0.24 (0.09) 0.25 (0.08) 0.24 (0.09)
log(MKTV) -0.20 (0.00) -0.21 (0.00) -0.20 (0.00) -0.21 (0.00)
log(BMRATIO) 0.82 (0.00) 0.65 (0.00) 0.75 (0.00) 0.60 (0.00)
LAGRET -0.24 (0.00) -0.23 (0.00) -0.25 (0.00) -0.24 (0.00)
ROE 3.39 (0.00) 10.43 (0.00)
CFO 2.58 (0.00) 9.24 (0.00)
ACC 3.01 (0.00) 9.85 (0.00)
CHROE 10.91 (0.00) 9.78 (0.00)
CHCFO 5.01 (0.00) 4.64 (0.00)
CHACC 4.01 (0.00) 3.67 (0.00)
LOSS -0.65 (0.00) -1.43 (0.00) -0.48 (0.00) -1.22 (0.00)
ROE*LOSS -9.91 (0.00) -8.95 (0.00)
CFO*LOSS -6.11 (0.00) -5.23 (0.00)
ACC*LOSS -6.04 (0.00) -5.09 (0.00)
EXTR 0.09 (0.36) 0.43 (0.00) 0.01 (0.89) 0.42 (0.00)
ROE*EXTR -2.98 (0.00) -2.11 (0.00)
CFO*EXTR 0.29 (0.37) 0.44 (0.18)
ACC*EXTR 2.40 (0.00) 2.58 (0.00)
EXTRCH 0.00 (0.95) 0.05 (0.42)
CHROE*EXTRCH -7.58 (0.00)
CHCFO*EXTRCH -7.13 (0.00)
CHACC*EXTRCH -7.36 (0.00)
Adjusted R
2
5.68% 5.72% 5.87% 5.96%

46

Table 4. Continued

Panel B. Returns on AQ Factor Loadings: Firm-Level Analysis
Independent Variables Coef.
FM
p-valu
e Coef.
FM
p-valu
e Coef.
FM
p-valu
e Coef.
FM
p-valu
e
Intercept -0.25 (0.05) 0.19 (0.12) -0.13 (0.32) 0.23 (0.06)
b
^
AQ
0.50 (0.01) 0.50 (0.01) 0.49 (0.01) 0.50 (0.01)
b
^
MktRf
0.32 (0.18) 0.34 (0.16) 0.32 (0.18) 0.34 (0.15)
b
^
SMB
0.35 (0.04) 0.34 (0.04) 0.34 (0.05) 0.34 (0.04)
b
^
HML
-0.12 (0.46) -0.15 (0.35) -0.13 (0.43) -0.15 (0.35)
LAGRET -0.47 (0.00) -0.45 (0.00) -0.46 (0.00) -0.44 (0.00)
ROE 3.66 (0.00) 11.80 (0.00)
CFO 2.82 (0.00) 10.16 (0.00)
ACC 3.28 (0.00) 10.76 (0.00)
CHROE 7.30 (0.00) 6.32 (0.00)
CHCFO 3.30 (0.00) 3.00 (0.00)
CHACC 2.46 (0.00) 2.19 (0.00)
LOSS -0.71 (0.00) -1.36 (0.00) -0.51 (0.00) -1.15 (0.00)
ROE*LOSS -6.75 (0.00) -5.92 (0.00)
CFO*LOSS -4.25 (0.00) -3.40 (0.00)
ACC*LOSS -4.23 (0.00) -3.33 (0.00)
EXTR -0.06 (0.52) 0.25 (0.00) -0.13 (0.13) 0.25 (0.00)
ROE*EXTR -2.08 (0.00) -1.15 (0.00)
CFO*EXTR 0.18 (0.54) 0.38 (0.20)
ACC*EXTR 1.85 (0.00) 2.07 (0.00)
EXTRCH 0.00 (0.95) 0.04 (0.47)
CHROE*EXTRCH -8.82 (0.00)
CHCFO*EXTRCH -7.89 (0.00)
CHACC*EXTRCH -8.07 (0.00)
Adjusted R
2
8.09% 8.17% 8.31% 8.42%

47

Table 4. Continued

Panel C. Returns on the AQ Factor Loadings: Portfolio-Level Cross-Sectional Regressions
Independent Variables

Intercept b
^
AQ
b
^
MktRf
b
^
SMB
b
^
HML

AdjustedR
2

Using Residual Returns from Cross-Sectional Regression I
25 B/M and Size Portfolios
Regression Coefficient -0.01 -0.05 -2.24 0.08 1.94 72.85%
FM p-value (0.02) (0.79) (0.00) (0.74) (0.00)
100 DD Measure Portfolios
Regression Coefficient 0.00 0.63 0.75 0.58 0.00 70.33%
FM p-value (0.98) (0.00) (0.10) (0.02) (0.99)
64 B/M, Size and DD Measure Portfolios
Regression Coefficient -0.03 0.49 -0.24 0.07 1.52 59.82%
FM p-value (0.09) (0.01) (0.50) (0.69) (0.00)
Using Residual Returns from Cross-Sectional Regression II
25 B/M and Size Portfolios
Regression Coefficient -0.02 0.30 -1.67 0.08 2.05 83.40%
FM p-value (0.00) (0.16) (0.00) (0.70) (0.00)
100 DD Measure Portfolios
Regression Coefficient 0.00 0.84 0.50 0.56 0.08 79.94%
FM p-value (0.81) (0.00) (0.27) (0.02) (0.78)
64 B/M, Size and DD Measure Portfolios
Regression Coefficient -0.03 0.64 -0.31 0.13 1.67 70.89%
FM p-value (0.06) (0.00) (0.42) (0.49) (0.00)
Using Residual Returns from Cross-Sectional Regression III
25 B/M and Size Portfolios
Regression Coefficient -0.01 0.08 -1.82 0.09 1.76 73.48%
FM p-value (0.01) (0.71) (0.00) (0.67) (0.00)
100 DD Measure Portfolios
Regression Coefficient 0.00 0.65 0.80 0.55 0.00 69.29%
FM p-value (0.95) (0.00) (0.08) (0.02) (0.99)
64 B/M, Size and DD Measure Portfolios
Regression Coefficient -0.02 0.53 -0.17 0.08 1.37 60.76%
FM p-value (0.16) (0.00) (0.63) (0.62) (0.00)
Using Residual Returns from Cross-Sectional Regression IV
25 B/M and Size Portfolios
Regression Coefficient -0.02 0.27 -1.60 0.06 2.06 81.48%
FM p-value (0.00) (0.22) (0.00) (0.78) (0.00)
100 DD Measure Portfolios
Regression Coefficient 0.00 0.80 0.46 0.53 0.07 75.72%
FM p-value (0.92) (0.00) (0.31) (0.03) (0.79)
64 B/M, Size and DD Measure Portfolios
Regression Coefficient -0.03 0.59 -0.38 0.11 1.67 69.78%
FM p-value (0.09) (0.00) (0.33) (0.55) (0.00)

The table reports results of asset pricing tests that include cross-sectionally unconstrained controls for cash flow
shocks. The tests are based on a sample of 87,828 observations with fiscal years ending in 1970 to 2004 (with returns
earned from April 1971 to March 2006).
48

Table 4. Continued

Panel A reports mean coefficient estimates from cross-sectional regressions: R
i,t+1
−Rf
t+1
=Intercept
t
+ q
t
RDD
it
+
c
t
CF_SHOCK
it
+ b
t
BETA
it
+ s
t
log(MKTV)
it
+ h
t
log(BMRATIO)
it
+ ε
it
, where R
i,t+1
denotes the return on stock i for
month t+1; Rf
t+1
is the one-month T-bill rate; RDD is the decile rank based on the DD Measure; CF_SHOCK
it
is the
cash-flow proxy represented by several variables. ROE
it
denotes return on equity during fiscal year t+1, i.e., earnings
before extraordinary items (Compustat #18) over the value of equity at the beginning of fiscal year t+1 (Compustat #60);
CFO
it
is cash flow from operations for year t estimated as ROE
it+1
- ACC
it+1
, where ACC
it+1
is total accruals scaled by
average value of total assets (#6) over years t-1, t, and t+1 (as defined in Table 1); CHROE
it+1
– change in ROE from
year t to year t+1; CHCFO
it+1
– change in CFO from year t to year t+1; CHACC
it+1
– change in ACC from year t to year
t+1; LOSS
it+1
– indicator variable equal to one if ROE
it+1
is negative; EXTR
it+1
– indicator variable equal to one if
ROE
it+1
falls into one of the extreme deciles for a given year; EXTRCH
it+1
– indicator variable equal to one if CHROE
it
falls into one of the extreme deciles for a given year; all other variables are as defined earlier; and LAGRET
it
denotes the
raw return on the stock of the firm computed over fiscal year t. Year t is the most recent fiscal year ending at least three
months prior to month t. All other variables are defined in Table 2.
Panel B reports mean coefficient estimates from cross-sectional regressions: R
it
−Rf
it
= Intercept
t
+ b
^
i
AQ
λ
t
AQ
+ c
t
CF_SHOCK
it
+ b
^
i
MktRft

λ
t
MktRft
+ b
^
i
SMB
λ
t
SMB
+ b
^
i
HML
λ
t
HML
+ b
^
i
AQ
λ
t
AQ
+ ε

it
, where R

it
is return on stock i for month
t; Rf
it
is the one-month T-bill rate; CF_SHOCK
it
variables are as defined in Panel A.
Panel C reports regression coefficients from the cross-sectional regressions based on the full-period average
excess returns and the full-period factor betas: RES
i
=Intercept

+ b
^
i
AQ
λ
AQ
+ b
^
i
MktRft

λ
MktRft
+ b
^
i
SMB
λ

SMB
+ b
^
i
HML
λ
HML
+ ε
it
, where RES
i
is the average residual return on portfolio i calculated using 420 monthly portfolio
returns; the residual returns are from the cross-sectional regressions of the form R
i,t+1
−Rf
t+1
=Intercept
t
+ c
t
CF_SHOCK
it
+ ε
it
, where R
i,t+1
denotes the return on stock i for month t+1; Rf
t+1
is the one-month T-bill rate;
CF_SHOCK
it
is the cash-flow proxy represented by several variables. Cash-flow shock variables differ across the
partitions: Specification I includes ROE
it+1
, CHROE
it+1
, LOSS
it+1
, EXTR
it+1
, ROE
it+1
*LOSS
it+1
, ROE
it+1
*EXTR
it+1
.
Specification II includes CFO
it+1
, ACC
it+1
, CHCFO
it+1
, CHACC
it+1
, LOSS
it+1
, EXTR
it+1
, CFO
it+1
*LOSS
it+1
,
ACC
it+1
*LOSS
it+1
, CFO
it+1
*EXTR
it+1
, ACC
it+1
*EXTR
it+1
. Specification III includes ROE
it+1
, CHROE
it+1
, LOSS
it+1
,
EXTR
it+1
, ROE
it+1
*LOSS
it+1
, ROE
it+1
*EXTR
it+1
, EXTRCH
it+1
, CHROE
it+1
*EXTRCH
it+1
. Specification IV includes
CFO
it+1
, ACC
it+1
, CHCFO
it+1
, CHACC
it+1
, LOSS
it+1
, EXTR
it+1
, CFO
it+1
*LOSS
it+1
, ACC
it+1
*LOSS
it+1
, CFO
it+1
*EXTR
it+1
,
ACC
it+1
*EXTR
it+1
, EXTRCH
it+1
, CHCFO
it+1
*EXTRCH
it+1
, CHACC
it+1
*EXTRCH
it+1
, where all variables are as defined
earlier. Each regression is based on a series of 420 monthly returns. The panel partitions refer to three types of
portfolios: (1) “25 B/M and Size Portfolios” are based on quintiles of the book-to-market ratio and size; (2) “100 DD
Measure Portfolios” are based on percentiles of the DD Measure; and (3) “64 B/M, Size and DD Measure Portfolios”
are based on quartiles of the book-to-market ratio, size, and the DD Measure. The stocks in the portfolios are equally
weighted.
Coefficients and adjusted R
2
s reported in Panels A and B are the average values from 420 monthly cross-sectional
regressions. Coefficients and adjusted R
2
s reported in Panel C are from a single cross-sectional regression that uses
average portfolio returns. All panels report two-sided Fama-MacBeth p-values that are based on the coefficients
obtained by estimating the cross-sectional regression monthly.

Panel C of Table 4 reports the results of portfolio-level cross-sectionally
unconstrained estimation. The tests use residual returns that are purged from the effects of
cash-flow shocks by regressing individual stock returns on the combinations of variables
proxying for cash-flow shocks. The four variable combinations that are used at this stage
correspond to those previously used in the firm-level analyses. Specification I includes
ROE
it+1
, CHROE
it+1
, LOSS
it+1
, EXTR
it+1
, ROE
it+1
*LOSS
it+1
, ROE
it+1
*EXTR
it+1
.
49

Specification II includes CFO
it+1
, ACC
it+1
, CHCFO
it+1
, CHACC
it+1
, LOSS
it+1
, EXTR
it+1
,
CFO
it+1
*LOSS
it+1
, ACC
it+1
*LOSS
it+1
, CFO
it+1
*EXTR
it+1
, ACC
it+1
*EXTR
it+1
. Specification
III includes ROE
it+1
, CHROE
it+1
, LOSS
it+1
, EXTR
it+1
, ROE
it+1
*LOSS
it+1
,
ROE
it+1
*EXTR
it+1
, EXTRCH
it+1
, CHROE
it+1
*EXTRCH
it+1
. Specification IV includes
CFO
it+1
, ACC
it+1
, CHCFO
it+1
, CHACC
it+1
, LOSS
it+1
, EXTR
it+1
, CFO
it+1
*LOSS
it+1
,
ACC
it+1
*LOSS
it+1
, CFO
it+1
*EXTR
it+1
, ACC
it+1
*EXTR
it+1
, EXTRCH
it+1
,
CHCFO
it+1
*EXTRCH
it+1
, CHACC
it+1
*EXTRCH
it+1
. The resulting residual returns are then
used to calculate portfolio returns. The portfolio returns are used in the two-stage
Fama-MacBeth type tests. The resulting premiums on the AQ factor beta are significant
for all specifications except for the tests that use 25 Fama-French portfolios based on size
and book-to-market ratio.
Overall, asset pricing tests presented in this section provide evidence suggesting
that accrual quality, proxied by the DD Measure, is negatively associated with the cost of
equity. After controlling for cash flow news effects, accrual quality (as a characteristic) is
significantly negatively related to expected returns, but only before controlling for the
three Fama-French characteristics. The AQ factor betas (i.e., loadings on the AQ factor)
are significantly positively related to expected returns even after controlling for the
loadings on the three Fama-French factors. However, the AQ factor premiums are not
significant after controlling for the E&M cash flow news proxy in stock-level regressions,
as well as after controlling for earnings surprise in two out of six portfolio-level
regressions.
50

4.3 The Relative DD Measure
The tests conducted in the previous section have two disadvantages. First, using
direct proxies for cash flow shocks introduces significant sample restrictions due to data
availability constraints. Second, the cash flow shock proxies are constructed using
forward-looking information and therefore introduce a look-ahead bias. In this section, I
introduce a direct adjustment to the DD Measure that reduces its correlation with future
cash flow shocks and obviates the need for directly controlling for returns associated with
cash flow shocks.
4.3.5 Selecting a Relative DD Measure
A negative association between the DD Measure and future cash flow shocks
arises, in part, due to the DD Measure being highly correlated with the variability of
earnings and sales. As noted in McNichols (2002), the DD Measure is likely to be
mechanically correlated with earnings and sales variability because, by construction, the
DD Measure represents the absolute variation in residuals from accruals-cash flow
mapping regressions. To remove the mechanical correlation with the variability measures,
McNichols proposes using a relative measure of accrual quality, namely, the R
2
from the
accrual-cash flow mapping regressions. Because the DD Measure is derived from the
cross-sectional regressions, the R
2
from the accrual-cash flow mapping regressions cannot
51

be used directly.
23
Accordingly, I scale the DD Measure by three alternative scaling
factors and select a scaled measure that is the closest approximation to the R
2
from the
individual regressions. The three scaling factors are: (1) the absolute value of average
current accruals; (2) the mean of absolute current accruals; and (3) the standard deviation
of current accruals. All three factors are estimated using data from the five previous fiscal
years, i.e., the period used to estimate the original DD Measure.
Table 5 contains correlations between the R
2
s from the individual by-firm
regressions (By-Firm RSQ), the original DD Measure, and the three cross-sectional scaled
measures, denoted as DD Measure, Scaled I, Scaled II, and Scaled III, respectively. The
By-Firm RSQ is the R
2
from regression (1) based on the prior 12 years of data. Two scaled
variables—Scaled II and Scaled III—demonstrate the highest (negative) correlations with
the By-Firm RSQ. Since the Scaled II correlation coefficient is slightly larger than the
Scaled III correlation coefficient, I adopt the former as a relative measure of accrual
quality (henceforth, the Scaled DD Measure).
24

23
Although the R
2
measure intuitively follows from the Dechow-Dichev model, it requires estimating individual
by-firm regressions, while a large-scale asset-pricing study requires a cross-sectional measure as in FLOS (2005).
24
The type of adjustment used in the Scaled III measure has been previously suggested by Verdi (2006). I report the
results of a sensitivity analysis using Scaled III as a relative measure of accrual quality in the additional analysis chapter
(Chapter Chapter 5:).
52

Table 5. Relative Accrual Quality Measure: Choosing a Scaling Factor

Spearman\Pearson
Correlation
By-Firm RSQ DD Measure Scaled I
Scaled II
(Scaled
DD Measure)
Scaled III
By-Firm RSQ 1 -0.152 -0.106 -0.363 -0.332
DD Measure -0.094 1 0.113 0.269 0.191
Scaled I -0.224 0.289 1 0.257 0.161
Scaled II
(Scaled DD Measure)
-0.346 0.344 0.651 1 0.879
Scaled III -0.337 0.265 0.382 0.863 1

The table reports correlation coefficients between various accrual quality measure specifications for a sample of
88,668 firm-year observations with fiscal years ending in 1970 to 2004. By-firm estimates are available for 50,016
firm-years.
The variables are defined as follows: By-Firm RSQ denotes R
2
from the following by-firm regression estimated
using the last twelve years of data: TCA
it
= α
t
+ β
0t
1/ATA
it
+ β
1t
CFO
it-1
+ β
2t
CFO
it
+ β
3t
CFO
i t+1
+ β
4t
ΔREV
it
+ β
5t

PPE
it
+ ε
it,
where TCA
it
are the total current accruals for year t; CFO
it
is the cash flow from operations for year t (total
current accruals and cash flow from operations are estimated using balance sheet approach); ΔREV
it
is the one-year
change in revenues (Compustat #12); PPE
it
is property, plant and equipment for year t (Compustat #7); ATA
it
is the
average value of total assets (Compustat #6) over years t-1, t and t+1. All variables are scaled by the average value of
total assets and all twelve observations are required for the estimation; DD Measure is the standard deviation of
residuals from the cross-sectional Dechow-Dichev regressions, as defined previously in Table 1; Scaled I is the DD
Measure scaled by the absolute value of average current accruals, estimated over the five previous years; Scaled II
(Scaled DD Measure) is the DD Measure scaled by the average value of absolute current accruals, estimated over the
five previous years; Scaled III is the DD Measure scaled by the standard deviation in current accruals, estimated over
the five previous years. The current accruals used for deflation are scaled by the average value of total assets.
Correlations are computed by year. All variables are winsorized at top and bottom 1% within each year. Reported
correlation coefficients are average values for 35 years.
4.3.6 Firm Characteristics Associated with the Scaled DD Measure
Next, I investigate whether the scaling procedure decreases the systematic
association between accrual quality and future cash flow shocks. Table 6 reports
univariate correlations for the original DD Measure and the Scaled DD Measure with
various firm characteristics and the direct cash flow shock proxies. When the Scaled DD
Measure is used to measure accrual quality, the correlations with firm characteristics no
53

longer uniformly indicate that low accrual quality firms are likely to underperform in the
future. While the correlation with the Altman score (ALTM) remains negative and
increases in magnitude and the correlation with the financing constraints index
(KZ_INDEX) becomes significantly positive, the correlation with past sales growth
(SALEGRW) changes sign and becomes negative and the correlation with the sales
variability indicator (SVAR) turns statistically insignificant. The correlations with the
other characteristics, while still consistent with the future underperformance of low
accrual quality firms, are reduced in magnitude. For example, the correlation with
earnings variability indicator (EVAR) is reduced from 0.35 (0.35) to 0.12 (0.13) based on
the Pearson (Spearman) test. Similarly, the correlation with the Piotroski F-score (PIOTR)
decreases in magnitude from -0.12 (-0.10) to -0.06 (-0.06) using the Pearson (Spearman)
test.

54

Table 6. Comparing DD and Scaled DD Measures: Correlations with Various Characteristics

DD MEASURE SCALED DD MEASURE
Variable # Obs. PEARSON SPEARMAN PEARSON SPEARMAN
Firm Characteristics:
|ACC| 88,668 0.36 (0.00) 0.27 (0.00) -0.06 (0.00) -0.01 (0.08)
ACC 88,668 -0.02 (0.09) -0.01 (0.22) -0.15 (0.00) -0.21 (0.00)
SALEGRW 79,581 0.10 (0.97) 0.01 (0.00) -0.01 (0.75) -0.11 (0.00)
ALTM 88,601 -0.16 (0.00) -0.05 (0.00) -0.19 (0.00) -0.23 (0.00)
PIOTR 88,668 -0.12 (0.01) -0.10 (0.00) -0.06 (0.00) -0.06 (0.00)
KZ_INDEX 88,387 -0.04 (0.15) 0.05 (0.20) 0.03 (0.32) 0.13 (0.00)
EVAR 78,194 0.35 (0.00) 0.35 (0.00) 0.12 (0.00) 0.13 (0.00)
SVAR 78,230 0.26 (0.00) 0.27 (0.00) -0.01 (0.57) 0.00 (0.92)

Cash Flow Shock Proxies:
SURP 87,828 -0.08 (0.01) -0.04 (0.00) -0.02 (0.02) -0.01 (0.01)
CN_EM 31,221 -0.16 (0.05) -0.15 (0.00) 0.00 (0.55) 0.02 (0.87)

Risk Proxies:
BETA 88,668 0.15 (0.01) 0.19 (0.01) -0.03 (0.03) -0.03 (0.04)
MKTV 88,668 -0.19 (0.00) -0.46 (0.00) 0.05 (0.00) -0.01 (0.59)
BMRATIO 88,668 -0.03 (0.19) -0.03 (0.02) 0.00 (0.24) -0.02 (0.14)

The table reports Pearson and Spearman correlations between various firm characteristics and the DD Measure or Scaled DD Measure for a sample of
88,668 firm-year observations with fiscal years ending in 1970 to 2004. The exact number of observations varies depending on the variable. All variables are
defined in Table 2.
Correlations are computed by year. All variables are winsorized at the top and bottom 1% within each year. Reported correlation coefficients are average
values for 35 years (correlation coefficients for CN_EM are based on 24 years). Two-sided Fama-MacBeth p-values are reported in parentheses. The p-values are
computed with the Newey-West (1987) correction with six lags.
55

In general, the correlations with the direct proxies for the future cash flow
shocks suggest a significantly less pronounced association between the new Scaled DD
Measure and the cash flow shocks. The Pearson (Spearman) correlation between the
earnings surprise (SURP) and the Scaled DD Measure is reduced from -0.08 (-0.04) to
-0.02 (-0.01). The magnitude of the Pearson (Spearman) correlation between the E&M
cash flow news proxy (CN_EM) and the Scaled DD Measure is only 0.00 (0.02) and no
longer statistically significant.
To provide visual evidence on the association between cash flow shock proxies
and the new Scaled DD Measure, I plot the mean and median values for SURP and
CN_EM by deciles of the Scaled DD Measure. An informal inspection of the resulting
plots suggests a weakly decreasing pattern for the earnings surprises (Figure 2, Panel A)
and no systematic association with the proxy based on analyst forecast revisions (Figure 2,
Panel B). Panels C and D of Figure 2 contain the mean and standard deviation plots for
each cash flow news measure. Compared to corresponding plots of Figure 1, the increase
in the standard deviations of cash flow news measures across the deciles of the accrual
quality measure is far less pronounced. That is consistent with a reduced association
between the Scaled DD Measure and the degree of earnings predictability.
56

Figure 2. Cash Flow Shocks and the Scaled DD Measure

Panel A. Earnings Surprise by Decile of the Scaled DD Measure
-0.04
-0.02
0
0.02
0.04
0.06
0.08
1 2 3 4 5 6 7 8 9 10
SURP (Mean)
SURP (Median)

Panel B. E&M Cash Flow News Proxy by Decile of the Scaled DD Measure
-0.14
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0
1 2 3 4 5 6 7 8 9 10
CN_EM (Mean)
CN_EM (Median)

57

Figure 2. Continued

Panel C. Earnings Surprise Distribution by Decile of the DD Measure
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
1 2 3 4 5 6 7 8 9 10
SURP (Mean)
+ 1 stddev
- 1 stddev

Panel D. E&M Cash Flow News Proxy Distribution by Decile of the DD Measure
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
1 2 3 4 5 6 7 8 9 10
CN_EM (Mean)
+ 1 stddev
- 1 stddev

The figure plots the univariate statistics for cash flow news proxies by decile of the Scaled DD Measure for a
sample of 88,668 firm-year observations with fiscal years ending in 1970 to 2004. The exact number of observations
varies depending on the availability of data for measuring each cash flow news proxy. Panels A and C contain
mean/median and mean/standard deviation statistics for earnings surprises (SURP) for a sample of 87,828 firm-years.
58

Figure 2. Continued

Panels B and D contain mean/median and mean/standard deviation statistics for the Easton and Monahan (2005)
cash flow news proxy (CN_EM) for a sample of 31,221 firm-years with fiscal years in 1981 to 2004. The details of the
estimation of both cash flow news proxies are described in the Appendix.

Overall, the evidence presented in Table 6 and Figure 2 suggests that the new
relative Scaled DD Measure is less strongly related to future cash flow shocks. The
correlations with various firm characteristics no longer uniformly point to future
underperformance of low accrual quality (high Scaled DD Measure) firms. More
importantly, the correlations with direct cash flow shock proxies are either insignificant or
smaller in magnitude.
4.3.7 Validating the Scaled DD Measure
One potential problem with the deflated measure is its construct validity, i.e., it
is unclear whether the deflated measure still captures a dimension of accounting quality
that is similar to that captured by the original (unscaled) measure. Accordingly, I compare
the relative association of the two measures (i.e., the DD Measure and the Scaled DD
Measure) with two ex post (and unambiguous) measures of accounting quality: (1)
accounting restatements, and (2) disclosures of weaknesses in internal controls with
respect to financial reporting.
25

25
Prior literature links both events to earnings quality. For example, Richardson et al. (2003) find a positive association
between accounting restatements and earnings management; Doyle et al. (2007) report a positive association between
weaknesses in internal controls reported under Sections 302 and 404 of the Sarbanes-Oxley Act (2002) and the original
DD Measure.
59

The first part of the analysis is based on a sample of 1,210 firms disclosing
material weaknesses under Sections 302 and 404 of the Sarbanes-Oxley Act from August
2002 to November 2005.
26
A firm-year is classified as having an internal control
weakness (ICW) if an ICW was reported within a year after the fiscal year-end.
Following Doyle et al. (2007), I distinguish between account-specific and
company-level weaknesses, where account-specific weaknesses relate to controls over
specific account balances or transaction-level processes and company-level weaknesses
relate to problems in the control environment or the overall financial reporting process.
Table 7, Panel A reports the mean values of the DD and the Scaled DD Measures for the
firms reporting each type of ICW and the control sample of firms that reported no ICWs.
Both the DD and the Scaled DD Measure are higher for each type of ICW firm compared
to the control sample. However, the differences for the company-level (account-specific)
ICWs are statistically significant only for the DD (Scaled DD) Measure.
27

26
The sample is described in Doyle et al. (2007). I thank Sarah McVay for making the sample publicly available.
27
Finding that the DD Measure is significantly higher only for the company-level weaknesses is consistent with Doyle
et al. (2007).
60

Table 7. Relative Accrual Quality Measure Validation

Panel A. Ex-post Events: Internal Control Weaknesses Announcement
Sample # Obs. DD Measure
Scaled
DD Measure
Account-Specific ICW 281 0.062 1.353
Company-Level ICW 142 0.074 1.259
Control Sample 12,656 0.058 1.244
Account-Specific ICW - Control 0.004 0.109
(0.22) (0.01)
Company-Level ICW - Control 0.015 0.015
(0.00) (0.80)

Panel B. Ex-post Events: Accounting Restatements
Sample # Obs. DD Measure
Scaled
DD Measure
Core Restatements 508 0.056 1.212
Other Restatements 167 0.062 1.220
Control Sample 23,783 0.056 1.130
Core Restatements - Control 0.000 0.082
(0.86) (0.00)
Other Restatements - Control 0.006 0.091
(0.08) (0.07)

The table reports average values for the DD Measure and the Scaled DD Measure for the sample partitions based
on the presence of ex-post accounting events. Sample size varies depending on the type of analysis. The DD Measure
and the Scaled DD Measure are defined previously.
Panel A reports the average values for the sub-samples partitioned by the type of internal control weakness
disclosure. The analysis is restricted to fiscal years 2002 to 2004. A firm-year is classified as having an internal control
weakness if the internal control weakness is announced within one year from the fiscal year-end. “Account-specific”
internal control weaknesses relate to controls over specific account balances or transaction-level processes.
“Company-level” internal control weaknesses relate to fundamental problems in the control environment or the overall
financial reporting. The internal control weakness announcement dates and the classification of weaknesses are from the
sample collected by Doyle et al. (2007).
Panel B reports the average values for the sub-samples partitioned by the type of accounting restatement. The
analysis is restricted to fiscal years 1997 to 2004. A firm-year is classified as having a restatement if restatement is
announced within one year from the fiscal year-end. “Core Restatements” refer to restatements in “core” accounts: cost
or expense accounting, assets or inventory accounts or revenue recognition. All other restatements are designated as
“Other Restatements”. The restatement announcement dates are taken from GAO reports.
Two-sided p-values for comparisons with control sample are reported in parentheses. The DD and the Scaled DD
Measures are winsorized at the top and bottom 1%.

The second part of the analysis is based on a sample of accounting restatements
from the two Government Accountability Office (GAO) reports, which include
restatements due to accounting irregularities made by public firms between January 1,
1997 and September 30, 2005 (GAO 2002, 2006). The reports cover restatements in
61

different accounts, including those that may not reflect persistently bad quality of
accounting accruals, as in the case of restatements related to mergers and acquisitions.
Therefore, I distinguish between the restatements in core items, as defined in the prior
literature (Palmrose and Scholts 2004; Palmrose et al. 2004), and other restatements. The
“core restatements” category includes the correction of errors in cost or expense
accounting, assets or inventory accounts, or revenue recognition.
28
The “other
restatements” category includes non-operating or merger-related items, such as
accounting of related-party transactions, acquisition accounting for goodwill or
convertible securities accounting. The final sample comprises 675 restatements. A
firm-year is classified as a restatement if a restatement was reported within a year after the
fiscal year-end.
Panel B of Table 7 compares the average magnitudes of the DD and the Scaled
DD Measures for the restating and the control sub-samples. The Scaled DD Measure is
higher for both the “Other Restatements” and the “Core Restatements” sub-samples
compared to the control sample of firms that did not issue a restatement, while the DD
Measure is significantly higher only for the “Other Restatements” sub-sample.
Overall, I find that the construct validity of the Scaled DD Measure is at least as
good as that of the original DD Measure. While the DD and the Scaled DD Measures
display a similarly mixed association with material weaknesses in internal controls, the

28
Palmrose et al. (2004) exclude asset impairments from the core items. I classify these restatements as “core” because
inventory and PPE are included in the calculation of the DD Measure.
62

Scaled DD Measure appears to be more strongly associated with accounting
restatements.
29

4.3.8 The Scaled DD Measure and Future Returns
In this section I replicate the asset pricing tests reported in Section 4.1 using the
Scaled DD Measure in place of the original DD Measure. All portfolio formation and
variable measurement procedures are exactly as described in Section 4.1.
Table 8, Panel A reports the average coefficients from monthly regressions of
future individual stock returns on the decile rank of the Scaled DD Measure (RDD_Scaled)
and the three Fama-French characteristics (beta, market value, and book-to-market). The
premium on RDD_Scaled is statistically significant both before and after the three
Fama-French factors are included in the regressions. The coefficient on RDD_Scaled can
be interpreted as approximately a 3.2% per annum difference in the returns between the
top and the bottom deciles of the Scaled DD Measure.

29
Additional (unreported) analysis compares the average values of the DD and the Scaled DD Measures for the firms
subject to Accounting and Auditing Enforcement Releases (AAERs) issued in 1997 to 2006. I find that neither measure
indicates significantly lower quality of accounting accruals for the AAER firms. However, these results may be due to
the limited power of statistical tests since the sample of AAERs that satisfies all data availability requirements is limited
to 83 firm-year observations.
63

Table 8. The Scaled DD Measure and Future Returns

Panel A. Returns on AQ Characteristic: Firm-Level Cross-Sectional Regressions
Intercept
RDD_
Scaled
BETA
log
(MKTV)
log
(BMRATIO)
Adjusted
R
2

Mean Coefficient 1.64 0.03 -0.13 0.31 3.33%
FM p-value (0.00) (0.83) (0.01) (0.00)
Mean Coefficient 0.78 0.03 0.11%
FM p-value (0.01) (0.00)
Mean Coefficient 1.48 0.03 0.04 -0.13 0.32 3.41%
FM p-value (0.00) (0.00) (0.80) (0.01) (0.00)

Panel B. The SAQ factor: Descriptive Statistics
Factor N Mean Std. Dev. Min Median Max t-stat p-value
SAQ factor 420 0.17 1.20 -4.41 0.15 6.67 2.86 (0.00)
AQ factor 420 0.16 3.47 -12.07 -0.21 23.39 0.95 (0.34)
MktRf 420 0.49 4.54 -23.13 0.78 16.05 2.22 (0.03)
SMB 420 0.20 3.29 -16.70 0.06 22.18 1.22 (0.22)
HML 420 0.46 3.08 -12.80 0.46 13.80 3.07 (0.00)

Panel C. The SAQ factor: Abnormal Returns
Dependent Variable = SAQ factor
Intercept MktRf SMB HML Adjusted R
2

Coefficient 0.19 -0.04 0.07 -0.03 5%
p-value (0.00) (0.01) (0.00) (0.11)

Panel D. Returns on the SAQ factor Loadings: Firm-Level Cross-Sectional Regressions
Intercept
b
^
SAQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

Mean Coefficient 0.48 0.37 0.16 -0.14 5.21%
FM P-Value (0.00) (0.13) (0.39) (0.41)
Mean Coefficient 0.89 0.05 2.16%
FM p-value (0.00) (0.54)
Mean Coefficient 0.48 0.02 0.39 0.15 -0.16 6.75%
FM p-value (0.00) (0.79) (0.11) (0.42) (0.34)

64

Table 8. Continued

Panel E. Returns on the SAQ factor Loadings: Portfolio-Level Cross-Sectional Regressions
Independent Variables

Intercept b
^
SAQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

25 B/M and Size Portfolios
Regression Coefficient 2.75 -2.16 0.22 0.64 71.66%
FM p-value (0.00) (0.00) (0.37) (0.00)
Regression Coefficient 0.85 0.19 14.33%
FM p-value (0.00) (0.15)
Regression Coefficient 2.73 0.27 -2.03 -0.02 0.73 78.56%
FM p-value (0.00) (0.04) (0.00) (0.94) (0.00)
100 Scaled DD Measure Portfolios
Regression Coefficient -0.03 0.28 0.99 -0.39 23.54%
FM p-value (0.96) (0.57) (0.00) (0.21)
Regression Coefficient 0.87 0.17 23.30%
FM p-value (0.00) (0.01)
Regression Coefficient -0.28 0.16 0.67 0.75 -0.33 32.02%
FM p-value (0.58) (0.01) (0.19) (0.02) (0.33)
64 B/M, Size and Scaled DD Measure Portfolios
Regression Coefficient 2.72 -2.12 0.22 0.62 63.98%
FM p-value (0.00) (0.00) (0.39) (0.00)
Regression Coefficient 0.85 0.19 17.00%
FM p-value (0.00) (0.10)
Regression Coefficient 2.64 0.25 -1.95 0.00 0.71 71.28%
FM p-value (0.00) (0.01) (0.00) (0.99) (0.00)

The table contains results of asset pricing tests for a sample of 88,668 firm-year observations with fiscal years
ending in 1970 to 2004 and monthly returns collected from April 1971 to March 2006. All returns are reported in
percentages.
Panel A reports mean coefficient estimates from the cross-sectional regression: R
i,t+1
−Rf
t+1
=Intercept
t
+ q
t

RDD_Scaled
it
+ b
t
BETA
it
+ s
t
log(MKTV)
it
+ h
t
log(BMRATIO)
it
+ ε
it
, where R
i,t+1
denotes the return on stock i for
month t+1; Rf
t+1
is the one-month T-bill rate; RDD_Scaled
it
is the decile rank based on the Scaled DD Measure; all
other variables are as defined earlier. Reported coefficients and adjusted R
2
s are the average values from 420 monthly
cross-sectional regressions. Two-sided Fama-MacBeth p-values are reported in parentheses.
Panel B reports descriptive statistics for the SAQ factor, the AQ factor and the three Fama-French factors. The
SAQ factor is an equally-weighted return on a hedge portfolio that is long(short) the two top(bottom) quintiles of the
Scaled DD Measure; all other factors are as defined earlier.
Panel C reports coefficient estimates for the regression: FRET
t
=α
i
+ β
i
MktRf
t
+ s
i
SMB
t
+ h
i
HML
t
+ ε
it
, where
FRET
t
is the return on the SAQ factor for month t and all other variables are as defined above. The regression is based
on a series of 420 monthly returns. Two-sided p-values are reported in parentheses.
Panel D reports mean coefficient estimates from the stock-level cross-sectional regression: R
it
−Rf
it
= Intercept
t
+
b
^
i
SAQ
λ
t
SAQ
+ b
^
i
MktRf

λ
t
MktRft
+ b
^
i
SMB
λ
t
SMB
+ b
^
i
HML
λ
t
HML
+ ε

it
, where R

it
is return on stock i for month t; Rf
it
is the
one-month T-bill rate; and b
^
i
SAQ
, b
^
i
MktRf
, b
^
i
SMB
, and b
^
i
HML
are factor betas estimated using all available returns for
stock i (with at least 18 monthly observations) from the time-series regression R
it
−Rf
it
=α
i
+ b
i
SAQ
SAQ
t
+ b
i
MktRfi

MktRf
t
+ b
i
SMB
SMB
t
+ b
i
HML
HML
t
+ ε
it
, where all variables are as defined above. Reported coefficients and adjusted
R
2
s are the average values from 420 monthly cross-sectional regressions. Two-sided Fama-MacBeth p-values adjusted
for the Shanken (1992) correction are reported in parentheses.
Panel E reports regression coefficients from the portfolio-level cross-sectional regressions based on the full-period
average excess returns and the full-period factor betas: PRET
i
- Rf =Intercept
t
+ b
^
i
SAQ
λ
SAQ
+ b
^
i
MktRf

λ
MktRft
+ b
^
i
SMB
λ

SMB
+ b
^
i
HML
λ
HML
+ ε
it
, where PRET
i
- Rf is the average excess return on portfolio i calculated using 420
65

Table 8. Continued

monthly portfolio returns; and b
^
i
SAQ
, b
^
i
MktRf
, b
^
i
SMB
, and b
^
i
HML
are full-period factor betas estimated from the regression
PRET
it
- Rf
t
=α
i
+ b
i
SAQ
AQ
t
+ b
i
MktRfi
MktRf
t
+ b
i
SMB
SMB
t
+ b
i
HML
HML
t
+ ε
it
, where PRET
it
is the return on
portfolio i for month t and all variables are as defined above. Each regression is based on a series of 420 monthly
returns. Two-sided Fama-MacBeth p-values adjusted for the Shanken (1992) correction (reported in parentheses) are
based on the coefficients obtained by estimating the cross-sectional regression monthly. The panel partitions refer to
three types of portfolios: (1) “25 B/M and Size Portfolios” are based on quintiles of the book-to-market ratio and size;
(2) “100 Scaled DD Measure Portfolios” are based on percentiles of the Scaled DD Measure; and (3) “64 B/M, Size and
Scaled DD Measure Portfolios” are based on quartiles of the book-to-market ratio, size, and the Scaled DD Measure.
The stocks in the portfolios are equally weighted.

Next, I verify whether the predictive ability of the Scaled DD Measure is
retained at a portfolio level. For this purpose, I form a factor-mimicking portfolio based
on the Scaled DD Measure, namely, the SAQ factor, following the same procedures that
are used to create the AQ factor. The descriptive statistics for the SAQ factor, the AQ
factor, and the three Fama and French (1993) factors are presented in Panel B of Table 8.
The mean returns earned on the AQ and SAQ factors are similar (0.16% and 0.17% per
month, respectively). However, the variability of returns over time is considerably lower
for the SAQ factor, resulting in the statistical significance only for the SAQ factor. Next, I
regress the SAQ factor returns on the three Fama-French factor returns. Panel C of Table 8
presents the regression results. Collectively, the Fama-French factors explain only 5% of
the variation in the SAQ factor (recall that the Fama-French factors explain 53% of the
variation in the AQ factor). Also, unlike the AQ factor, the SAQ factor earns significantly
positive abnormal returns as reflected in the regression intercept.
A significant abnormal return on the SAQ factor indicates that the Scaled DD
Measure, as a characteristic, is able to predict future realized returns and its predictive
ability is not explained by the exposure to the known risk factors. Next, I verify that the
66

co-variability with the SAQ factor explains the cross-section of returns (i.e., that firms
with higher SAQ factor betas earn on average higher returns compared to firms with lower
SAQ factor betas), for which purpose I perform the two-stage estimation described in
detail in Section 4.1.2.
Panel D of Table 8 reports the results of the stock-level two-stage estimation.
The premiums on the SAQ factor betas are small and not significant. However, this result
may be due to the low power of asset pricing regressions at the individual-firm level.
Panel E of Table 8 reports the results of the portfolio-level two-stage estimation. Across
all types of portfolio partitions the SAQ factor commands a premium that is statistically
significant at the 10% level after controlling for the three Fama-French factors, and the
magnitude of the premium ranges from 0.16% to 0.27% per month (approximately 1.9%
to 3.2% per annum). Overall, the two-stage analysis suggests that the factor-mimicking
portfolio based on the Scaled DD Measure represents a priced risk factor.
30

30
I replicate all analyses using the Scaled III measure (described in Section 4.3.5) as a relative measure of accrual
quality. All tests are robust to this alternative specification, although the magnitude of the premium assigned to the
alternative measure is smaller compared to the Scaled DD Measure.

For example, in the cross-sectional regression that
controls for size, book-to-market, and CAPM beta, the premium on the rank of the Scaled III measure is 0.02%
compared to 0.03% reported earlier for the rank of the Scaled DD Measure.

67

Chapter 5: Additional Analyses
5.1 Look-Ahead Bias
The estimation of the Dechow-Dichev model requires information on next-
period cash flows, which introduces a look-ahead bias in asset pricing tests. I replicate all
tests using the Scaled DD Measure and the DD Measure that are lagged by one year,
hereafter labeled as Lagged Scaled DD Measure and Lagged DD Measure, respectively.
Table 9 reports replication of the results presented in Table 1 using the Lagged
DD Measure. The results are qualitatively similar to those reported in Table 1 – the
Lagged DD Measure is not associated with future abnormal returns neither as a
characteristic nor as a risk factor.

Table 9. The Lagged DD Measure and Future Returns

Panel A. Returns on LAQ Characteristic: Firm-Level Cross-Sectional Regressions
Intercept RLDD BETA
log
(MKTV)
log
(BMRATIO)
Adjusted
R
2

Mean Coefficient 0.77 0.03 0.86%
FM p-value (0.00) (0.28)
Mean Coefficient 1.66 0.00 0.04 -0.13 0.29 3.50%
FM p-value (0.00) (0.70) (0.81) (0.00) (0.00)

Panel B. The LAQ factor: Descriptive Statistics
Factor N Mean Std. Dev. Min Median Max t-stat p-value
LAQ factor 420 0.18 3.31 -11.60 -0.13 19.99 1.15 (0.25)

Panel C. The LAQ factor: Abnormal Returns
Dependent Variable = LAQ factor
Intercept MktRf SMB HML Adjusted R
2

Coefficient 0.0793 0.0761 0.6622 -0.1334 55.26%
p-value (0.48) (0.00) (0.00) (0.00)

68

Table 9. Continued

Panel D. Returns on the LAQ factor Loadings: Firm-Level Cross-Sectional Regressions
Intercept
b
^
LAQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

Mean Coefficient 0.77 0.16 2.91%
FM p-value (0.00) (0.37)
Mean Coefficient 0.49 0.16 0.36 0.13 -0.12 6.36%
FM p-value (0.00) (0.37) (0.15) (0.47) (0.48)

Panel E. Returns on the LAQ factor Loadings: Portfolio-Level Cross-Sectional Regressions
Independent Variables

Intercept b
^
LAQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

25 B/M and Size Portfolios
Regression Coefficient 2.75 . -2.16 0.22 0.64 71.66%
FM p-value (0.00) (0.00) (0.27) (0.00)
Regression Coefficient 0.63 0.25 10.85%
FM p-value (0.01) (0.16)
Regression Coefficient 2.40 0.03 -1.72 -0.14 0.70 81.09%
FM p-value (0.00) (0.90) (0.00) (0.53) (0.00)

100 DD Measure Portfolios
Regression Coefficient 0.91 -0.21 0.33 -0.06 30.57%
FM p-value (0.00) (0.60) (0.18) (0.81)
Regression Coefficient 0.70 0.21 33.00%
FM p-value (0.00) (0.20)
Regression Coefficient 1.07 0.21 -0.23 0.08 -0.08 32.06%
FM p-value (0.00) (0.20) (0.57) (0.75) (0.77)

64 B/M, Size and DD Measure Portfolios
Regression Coefficient 1.17 -0.73 0.29 0.70 46.85%
FM p-value (0.00) (0.06) (0.21) (0.00)
Regression Coefficient 0.71 0.16 3.54%
FM p-value (0.01) (0.33)
Regression Coefficient 1.23 0.13 -0.65 -0.10 0.78 55.58%
FM p-value (0.00) (0.48) (0.09) (0.61) (0.00)

The table contains results of asset pricing tests for a sample of 81,340 firm-year observations with fiscal years
ending in 1970 to 2004 and monthly returns collected from April 1971 to March 2006. All returns are reported in
percentages. The Lagged DD Measure is estimated as the standard deviation of residuals over the [t-5, t-1] period,
where all calculations exactly follow procedures described in Table 1. The LAQ factor is constructed based on the
Lagged DD Measure using procedures described in Table 1, Panel B. All the tests are equivalent to those presented in
Table 1.

Table 10 reports correlations between the Lagged (Scaled) DD Measure and the
characteristics contained in Table 2. High correlations with corresponding non-lagged
69

measures suggest that the accrual quality measures are highly persistent, which is
expected given the four-year overlap in estimation periods. All other correlations are
qualitatively similar to those reported in Table 2. Importantly, the correlations with cash
flow news proxies are of similar magnitudes and signs to those reported for the
non-lagged DD Measures. This similarity suggests that the negative correlation between
the DD Measure and future cash flow news does not result from the fact that future
one-period cash flows are included in the estimation of the DD Measure.

70

Table 10. Comparing Lagged DD and Scaled DD Measures: Correlations with Various Characteristics

LDD MEASURE SCALED LDD MEASURE
Variable # Obs. PEARSON SPEARMAN PEARSON SPEARMAN
DD MEASURE 81,340 0.87 (0.00) 0.88 (0.00) 0.20 (0.00) 0.25 (0.00)
SCALED DD MEASURE 81,340 0.20 (0.00) 0.24 (0.00) 0.80 (0.00) 0.79 (0.00)

Firm Characteristics:
|ACC| 81,340 0.26 (0.00) 0.20 (0.00) 0.01 (0.22) 0.03 (0.00)
ACC 81,340 0.03 (0.00) 0.02 (0.01) -0.10 (0.00) -0.14 (0.00)
SALEGRW 76,186 0.10 (0.00) 0.01 (0.47) -0.01 (0.75) -0.10 (0.00)
ALTM 81,277 -0.12 (0.00) -0.02 (0.34) -0.18 (0.00) -0.21 (0.00)
PIOTR 81,340 -0.07 (0.00) -0.06 (0.00) -0.03 (0.00) -0.03 (0.00)
KZ_INDEX 81,110 -0.05 (0.09) 0.03 (0.41) 0.03 (0.33) 0.12 (0.01)
EVAR 81,340 0.52 (0.00) 0.60 (0.00) 0.12 (0.00) 0.12 (0.00)
SVAR 81,340 0.35 (0.00) 0.46 (0.00) -0.04 (0.01) -0.09 (0.00)

Cash Flow Shock Proxies:
SURP 80,610 -0.08 (0.01) -0.05 (0.00) -0.02 (0.02) -0.02 (0.00)
CN_EM 29,111 -0.16 (0.00) -0.14 (0.00) -0.01 (0.55) 0.00 (0.88)

Risk Proxies:
BETA 81,340 0.14 (0.00) 0.19 (0.00) -0.03 (0.03) -0.03 (0.04)
MKTV 81,340 -0.19 (0.00) -0.46 (0.00) 0.05 (0.00) 0.00 (0.95)
BMRATIO 81,340 -0.04 (0.02) -0.03 (0.18) -0.02 (0.24) -0.04 (0.14)

The table reports Pearson and Spearman correlations between various firm characteristics and the Lagged DD Measure or Scaled Lagged DD Measure
for a sample of 81,340 firm-year observations with fiscal years ending in 1970 to 2004. The exact number of observations varies depending on the variable. All
variables are defined in Table 2.
Correlations are computed by year. All variables are winsorized at the top and bottom 1% within each year. Reported correlation coefficients are average
values for 35 years (correlation coefficients for CN_EM are based on 24 years). Two-sided Fama-MacBeth p-values are reported in parentheses. The p-values
are computed with the Newey-West (1987) correction with six lags.
71

Table 11 replicates the tests presented in Table 3 with cash flow shock proxies
included as the control variables in the cross-sectional asset-pricing regressions. Overall,
the results are qualitatively similar to those obtained for the original DD Measure. In case
of unconstrained cash flow news estimation reported in Panel D, the coefficient on the
ranked Lagged DD Measure becomes statistically significant, although similar in
magnitude to the corresponding coefficient in Table 3.

Table 11. Controlling for Cash Flow Shocks

Panel A. Returns on Lagged AQ Characteristic: Firm-Level Cross-Sectional Regressions
Intercept RLDD
CF Shock
Proxy BETA
log
(MKTV)
log
(BM
RATIO)
Adjusted
R
2

CF Shock Proxy = SURP
Mean Coefficient 0.76 0.03 0.86%
FM p-value (0.00) (0.26)
Mean Coefficient 0.39 0.06 6.15 2.16%
FM p-value (0.06) (0.02) (0.00)
Mean Coefficient 1.41 0.01 6.66 0.11 -0.15 0.49 4.78%
FM p-value (0.00) (0.37) (0.00) (0.46) (0.00) (0.00)
CF Shock Proxy = CN_EM
Mean Coefficient 0.97 -0.01 1.46%
FM p-value (0.00) (0.86)
Mean Coefficient 1.05 0.05 6.30 3.36%
FM p-value (0.00) (0.14) (0.00)
Mean Coefficient 2.35 0.00 6.36 0.19 -0.19 -0.01 6.50%
FM p-value (0.00) (0.83) (0.00) (0.37) (0.00) (0.95)

Panel B. Returns on the LAQ factor Loadings: Firm-Level Cross-Sectional Regressions
Intercept
b
^
LAQ

CF Shock
Proxy
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

CF Shock Proxy = SURP
Mean Coefficient 0.77 0.16 2.91%
FM p-value
(0.00) (0.38)
Mean Coefficient 0.34 0.35 6.10 3.99%
FM p-value
(0.10) (0.05) (0.00)
Mean Coefficient 0.23 0.35 5.91 0.27 0.23 -0.08 7.34%
FM p-value
(0.04) (0.05) (0.00) (0.27) (0.19) (0.63)
72

Table 11. Continued

CF Shock Proxy = CN_EM
Mean Coefficient 0.91 0.06 4.08%
FM p-value
(0.00) (0.79)
Mean Coefficient 1.03 0.40 6.14 5.52%
FM p-value
(0.00) (0.08) (0.00)
Mean Coefficient 0.73 0.27 5.90 0.49 0.25 -0.21 9.53%
FM p-value
(0.00) (0.20) (0.00) (0.13) (0.20) (0.28)

Panel C. Returns on the LAQ factor Loadings: Portfolio-Level Cross-Sectional Regressions
Independent Variables
Adjusted
R
2

Intercept
b
^
LAQ

CF Shock
Proxy
b
^
MktRf
b
^
SMB
b
^
HML

CF Shock Proxy = SURP
25 B/M and Size Portfolios
Regression Coeff. 2.41 0.00 -1.74 -0.11 0.70 79.24%
FM p-value (0.00) (0.99) (0.00) (0.62) (0.00)
Regression Coeff. 0.71 0.41 6.18 -0.66 0.10 1.00 85.64%
FM p-value (0.01) (0.22) (0.00) (0.05) (0.81) (0.00)
100 DD Measure Portfolios
Regression Coeff. 0.78 0.20 0.20 -0.12 -0.14 32.69%
FM p-value (0.01) (0.24) (0.60) (0.64) (0.58)
Regression Coeff. 0.71 0.34 3.00 0.14 -0.15 -0.22 35.09%
FM p-value (0.05) (0.03) (0.00) (0.86) (0.79) (0.54)
64 B/M, Size and DD Measure Portfolios
Regression Coeff. 1.26 0.11 -0.66 -0.11 0.76 54.03%
FM p-value (0.00) (0.55) (0.08) (0.57) (0.00)
Regression Coeff. 0.58 0.31 3.62 -0.33 0.02 0.90 58.76%
FM p-value (0.10) (0.13) (0.00) (0.36) (0.90) (0.00)
CF Shock Proxy = CN_EM
25 B/M and Size Portfolios
Regression Coeff. 1.00 0.99 -0.24 0.15 0.39 85.30%
FM p-value (0.15) (0.01) (0.31) (0.40) (0.15)
Regression Coeff. 0.73 1.25 3.24 0.19 0.26 0.26 85.87%
FM p-value (0.21) (0.00) (0.00) (0.37) (0.08) (0.60)
73

Table 11. Continued

100 Lagged DD Measure Portfolios
Regression Coeff. 0.86 0.18 -0.21 0.47 0.30 10.44%
FM p-value (0.01) (0.46) (0.15) (0.12) (0.63)
Regression Coeff. 0.88 0.50 4.55 -0.03 0.66 0.16 16.29%
FM p-value (0.01) (0.00) (0.00) (0.26) (0.01) (0.99)
64 B/M, Size and Lagged DD Measure Portfolios
Regression Coeff. 1.41 0.46 -0.70 0.34 0.32 32.24%
FM p-value (0.00) (0.15) (0.09) (0.25) (0.12)
Regression Coeff. 1.14 0.80 6.01 -0.21 0.62 0.15 41.02%
FM p-value (0.01) (0.00) (0.00) (0.30) (0.01) (0.48)

The table reports results of asset pricing tests that include controls for cash flow shocks: either the earnings
surprise (SURP) or the Easton and Monahan cash flow news measure (CN_EM). The tests using SURP are based on a
sample of 81,340 observations with fiscal years ending in 1970 to 2004 (with returns earned from April 1971 to March
2006). The tests using CN_EM measure are based on a sample of 31,221 firm-years with fiscal years in 1981 to 2004
(with returns earned from May 1981 to March 2006).
All tests are described in Table 3.

Table 12 contains replication of the tests in Table 4 using the lagged measures.
The results are generally qualitatively similar, with the premiums on the accrual quality
factors and characteristics somewhat larger and more significant.

74

Table 12. Cross-Sectionally Unconstrained Cash Flow Shock Controls: Lagged Measures

Panel A. Returns on Lagged AQ Characteristic: Firm-Level Analysis
Independent Variables Coef.
FM
p-valu
e Coef.
FM
p-valu
e Coef.
FM
p-valu
e Coef.
FM
p-valu
e
Intercept 0.63 (0.01) 1.31 (0.00) 0.74 (0.00) 1.32 (0.00)
RLDD 0.02 (0.05) 0.02 (0.04) 0.02 (0.03) 0.03 (0.02)
BETA 0.24 (0.08) 0.23 (0.10) 0.24 (0.09) 0.23 (0.10)
log(MKTV) -0.19 (0.00) -0.20 (0.00) -0.18 (0.00) -0.19 (0.00)
log(BMRATIO) 0.84 (0.00) 0.66 (0.00) 0.77 (0.00) 0.61 (0.00)
LAGRET -0.24 (0.00) -0.23 (0.00) -0.26 (0.00) -0.25 (0.00)
ROE 3.50 (0.00) 10.69 (0.00)
CFO 2.64 (0.00) 9.59 (0.00)
ACC 3.12 (0.00) 10.23 (0.00)
CHROE 10.95 (0.00) 9.76 (0.00)
CHCFO 5.01 (0.00) 4.62 (0.00)
CHACC 3.94 (0.00) 3.57 (0.00)
LOSS -0.65 (0.00) -1.42 (0.00) -0.47 (0.00) -1.19 (0.00)
ROE*LOSS -10.16 (0.00) -9.15 (0.00)
CFO*LOSS -6.21 (0.00) -5.26 (0.00)
ACC*LOSS -6.12 (0.00) -5.11 (0.00)
EXTR 0.08 (0.43) 0.41 (0.00) -0.01 (0.89) 0.38 (0.00)
ROE*EXTR -2.99 (0.00) -2.04 (0.00)
CFO*EXTR 0.34 (0.34) 0.57 (0.11)
ACC*EXTR 2.40 (0.00) 2.61 (0.00)
EXTRCH 0.00 (0.94) 0.05 (0.41)
CHROE*EXTRCH -7.80 (0.00)
CHCFO*EXTRCH -7.48 (0.00)
CHACC*EXTRCH -7.68 (0.00)
Adjusted R
2
5.82% 5.88% 6.01% 6.14%

75

Table 12. Continued

Panel B. Returns on LAQ Factor Loadings: Firm-Level Analysis
Independent Variables Coef.
FM
p-valu
e Coef.
FM
p-valu
e Coef.
FM
p-valu
e Coef.
FM
p-valu
e
Intercept -0.25 (0.05) 0.19 (0.11) -0.13 (0.33) 0.24 (0.05)
b
^
LAQ
0.49 (0.01) 0.49 (0.01) 0.48 (0.01) 0.49 (0.01)
b
^
MktRf
0.31 (0.19) 0.34 (0.16) 0.32 (0.18) 0.34 (0.15)
b
^
SMB
0.34 (0.05) 0.34 (0.05) 0.33 (0.05) 0.34 (0.05)
b
^
HML
-0.12 (0.46) -0.15 (0.36) -0.12 (0.44) -0.15 (0.35)
LAGRET -0.47 (0.00) -0.45 (0.00) -0.46 (0.00) -0.44 (0.00)
ROE 3.66 (0.00) 11.80 (0.00)
CFO 2.82 (0.00) 10.16 (0.00)
ACC 3.28 (0.00) 10.77 (0.00)
CHROE 7.30 (0.00) 6.32 (0.00)
CHCFO 3.30 (0.00) 3.00 (0.00)
CHACC 2.45 (0.00) 2.18 (0.00)
LOSS -0.71 (0.00) -1.36 (0.00) -0.51 (0.00) -1.15 (0.00)
ROE*LOSS -6.74 (0.00) -5.91 (0.00)
CFO*LOSS -4.24 (0.00) -3.39 (0.00)
ACC*LOSS -4.22 (0.00) -3.32 (0.00)
EXTR -0.06 (0.49) 0.25 (0.00) -0.14 (0.12) 0.25 (0.00)
ROE*EXTR -2.07 (0.00) -1.15 (0.00)
CFO*EXTR 0.19 (0.51) 0.40 (0.19)
ACC*EXTR 1.86 (0.00) 2.08 (0.00)
EXTRCH -0.01 (0.93) 0.04 (0.49)
CHROE*EXTRCH -8.82 (0.00)
CHCFO*EXTRCH -7.90 (0.00)
CHACC*EXTRCH -8.07 (0.00)
Adjusted R
2
8.08% 8.16% 8.30% 8.41%

76

Table 12. Continued

Panel C. Returns on the LAQ Factor Loadings: Portfolio-Level Cross-Sectional Regressions
Independent Variables

Intercept b
^
LAQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

Using Residual Returns from Cross-Sectional Regression I
25 B/M and Size Portfolios
Regression Coefficient -0.01 -0.05 -2.24 0.08 1.94 72.85%
FM p-value (0.05) (0.80) (0.00) (0.67) (0.00)
100 LDD Measure Portfolios
Regression Coefficient 0.00 0.68 0.44 0.35 -0.16 71.74%
FM p-value (0.87) (0.00) (0.27) (0.17) (0.55)
64 B/M, Size and LDD Measure Portfolios
Regression Coefficient -0.02 0.44 -0.19 0.07 1.57 52.09%
FM p-value (0.27) (0.01) (0.54) (0.73) (0.00)
Using Residual Returns from Cross-Sectional Regression II
25 B/M and Size Portfolios
Regression Coefficient -0.02 0.30 -1.67 0.08 2.05 83.40%
FM p-value (0.00) (0.14) (0.00) (0.65) (0.00)
100 LDD Measure Portfolios
Regression Coefficient 0.00 0.82 0.44 0.34 -0.15 78.21%
FM p-value (0.68) (0.00) (0.27) (0.18) (0.57)
64 B/M, Size and LDD Measure Portfolios
Regression Coefficient -0.01 0.60 -0.32 0.08 1.66 64.26%
FM p-value (0.50) (0.00) (0.35) (0.69) (0.00)
Using Residual Returns from Cross-Sectional Regression III
25 B/M and Size Portfolios
Regression Coefficient -0.01 0.08 -1.82 0.09 1.76 73.48%
FM p-value (0.02) (0.71) (0.00) (0.62) (0.00)
100 LDD Measure Portfolios
Regression Coefficient 0.00 0.70 0.53 0.31 -0.17 71.87%
FM p-value (0.89) (0.00) (0.18) (0.23) (0.52)
64 B/M, Size and LDD Measure Portfolios
Regression Coefficient -0.01 0.48 -0.08 0.06 1.39 50.30%
FM p-value (0.40) (0.01) (0.75) (0.77) (0.00)
Using Residual Returns from Cross-Sectional Regression IV
25 B/M and Size Portfolios
Regression Coefficient -0.02 0.27 -1.60 0.06 2.06 81.48%
FM p-value (0.00) (0.20) (0.00) (0.73) (0.00)
100 LDD Measure Portfolios
Regression Coefficient 0.00 0.79 0.43 0.30 -0.14 75.22%
FM p-value (0.75) (0.00) (0.28) (0.24) (0.59)
64 B/M, Size and LDD Measure Portfolios
Regression Coefficient -0.01 0.56 -0.34 0.05 1.65 61.96%
FM p-value (0.45) (0.00) (0.34) (0.78) (0.00)

The table reports results of asset pricing tests that include cross-sectionally unconstrained controls for cash flow
shocks. The tests are based on a sample of 81,340 observations with fiscal years ending in 1970 to 2004 (with returns
earned from April 1971 to March 2006). All tests are described in Table 4.
77

Finally, Table 13 contains replication of the tests in Table 8 using the Lagged
Scaled DD Measure. Although the results are generally qualitatively similar, the
premiums on the accrual quality factors and characteristics are somewhat smaller. In case
of 100 Lagged Scaled DD Measure Portfolios, the loading on the Lagged SAQ factor
becomes only marginally significant with p-value of 0.12.

Table 13. The Scaled Lagged DD Measure and Future Returns

Panel A. Returns on Lagged AQ Characteristic: Firm-Level Cross-Sectional Regressions
Intercept RLDD_Scaled BETA
log
(MKTV)
log
(BMRATIO)
Adjusted
R
2

Mean Coefficient 0.83 0.02 0.10%
FM p-value (0.00) (0.04)
Mean Coefficient 1.50 0.02 0.04 -0.13 0.29 3.45%
FM p-value (0.00) (0.00) (0.82) (0.01) (0.00)

Panel B. The SLAQ factor: Descriptive Statistics
Factor N Mean Std. Dev. Min Median Max t-stat p-value
SLAQ factor 420 0.11 1.11 -3.62 0.07 4.08 2.04 (0.04)
LAQ factor 420 0.18 3.31 -11.60 -0.13 19.99 1.15 (0.25)
MktRf 420 0.49 4.54 -23.13 0.78 16.05 2.22 (0.03)
SMB 420 0.20 3.29 -16.70 0.06 22.18 1.22 (0.22)
HML 420 0.46 3.08 -12.80 0.46 13.80 3.07 (0.00)

Panel C. The SLAQ factor: Abnormal Returns
Dependent Variable = SLAQ factor
Intercept MktRf SMB HML Adjusted R
2

Coefficient 0.1183 -0.0211 0.0441 -0.0137 1.29%
p-value (0.03) (0.11) (0.01) (0.49)

Panel D. Returns on the SLAQ factor Loadings: Firm-Level Cross-Sectional Regressions
Intercept
b
^
SLAQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

Mean Coefficient 0.48 0.02 0.40 0.15 -0.16 6.78%
FM p-value (0.00) (0.72) (0.11) (0.42) (0.33)
Mean Coefficient 0.91 0.04 1.99%
FM p-value (0.00) (0.54)

78

Panel E. Returns on the SLAQ factor Loadings: Portfolio-Level Cross-Sectional Regressions
Independent Variables

Intercept b
^
SLAQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

25 B/M and Size Portfolios
Regression Coefficient 2.75 . -2.16 0.22 0.64 71.66%
FM p-value (0.00) (0.00) (0.27) (0.00)
Regression Coefficient 0.86 0.22 13.40%
FM p-value (0.00) (0.16)
Regression Coefficient 2.77 0.27 -2.08 0.02 0.72 77.57%
FM p-value (0.00) (0.07) (0.00) (0.95) (0.00)
100 Lagged Scaled DD Measure Portfolios
Regression Coefficient 1.32 -0.94 0.54 0.40 17.40%
FM p-value (0.00) (0.05) (0.08) (0.16)
Regression Coefficient 0.92 0.10 9.55%
FM p-value (0.00) (0.10)
Regression Coefficient 1.22 0.09 -0.78 0.46 0.40 20.35%
FM P-Value (0.00) (0.12) (0.10) (0.12) (0.15)
64 B/M, Size and Lagged Scaled DD Measure Portfolios
Regression Coefficient 1.86 -1.38 0.27 0.73 52.31%
FM p-value (0.00) (0.00) (0.29) (0.00)
Regression Coefficient 0.86 0.20 10.89%
FM p-value (0.00) (0.12)
Regression Coefficient 1.87 0.22 -1.30 0.10 0.79 56.93%
FM p-value (0.00) (0.01) (0.00) (0.68) (0.00)

The table contains results of asset pricing tests for a sample of 81,340 firm-year observations with fiscal years
ending in 1970 to 2004 and monthly returns collected from April 1971 to March 2006. All returns are reported in
percentages.
All tests are described in Table 5.

Overall, the cross-sectional regression tests are largely robust to the lagged
specification with the DD Measure producing slightly stronger evidence, and the Scaled
DD Measure producing slightly weaker evidence in favor of the accrual quality being
priced by the market.
5.2 Controlling for Signed Accruals
According to Sloan (1996) the stock market over-weights information contained
in accruals due to overestimating their persistence compared to cash flows. As a result,
79

firms with lower (higher) accrual levels exhibit abnormally positive (negative) returns in
the future. Table 6 reports a negative association between the Scaled DD Measure and
signed total accruals, which could contribute to the positive relation between future
returns and the Scaled DD Measure. To verify that the results obtained for the Scaled
DD Measure are not explained by the accrual anomaly (Sloan 1996), I control for the
effects of signed accruals in two ways. First, I include an accrual decile variable in the
firm-level characteristic-based regressions. Second, I construct an “orthogonalized”
accrual quality factor by first sorting stocks into decile portfolios based on the accrual
magnitude and then within each decile sorting into quintiles based on the Scaled DD
Measure. The results of the tests are reported in Table 14. The characteristic-based results
suggest that the accrual anomaly explains between one-half and two-thirds of the
premium on the Scaled DD Measure, although the premiums remain statistically
significant at the 10% level. However, the portfolio-level factor loading-based results
remain qualitatively similar both in terms of the premium magnitudes and the levels of
statistical significance. Overall, there is some evidence that the pricing effects of the
Scaled DD Measure and the level of accruals are not independent of each other. Given
that signed accruals may serve as an alternative way of operationalizing accrual quality,

80

further analysis is required to establish the net effect of accrual quality on future expected
returns.
31, 32

Table 14. The Scaled DD Measure and Future Returns: Controlling for Signed Accruals

Panel A. Returns on AQ Characteristic: Firm-Level Cross-Sectional Regressions with
Signed Accruals Controls
Intercept
RDD_
Scaled
BETA
log
(MKTV)
log
(BM
RATIO)
RACC
Adjusted
R
2

Mean Coefficient 1.33 0.01 -0.08 0.21%
FM p-value (0.00) (0.25) (0.00)
Mean Coefficient 1.98 0.01 0.04 -0.13 0.30 -0.08 3.49%
FM p-value (0.00) (0.08) (0.81) (0.01) (0.00) (0.00)

Panel B. The Orthogonalized SAQ factor: Descriptive Statistics
Factor N Mean Std. Dev. Min Median Max t-stat p-value
OSAQ factor 420 0.08 1.03 -3.34 0.04 5.14 1.53 (0.13)
MktRf 420 0.49 4.54 -23.13 0.78 16.05 2.22 (0.03)
SMB 420 0.20 3.29 -16.70 0.06 22.18 1.22 (0.22)
HML 420 0.46 3.08 -12.80 0.46 13.80 3.07 (0.00)

Panel C. The Orthogonalized SAQ factor: Abnormal Returns
Dependent Variable = OSAQ factor
Intercept MktRf SMB HML Adjusted R
2

Coefficient 0.08 -0.00 0.04 -0.03 2.59%
p-value (0.10) (0.82) (0.01) (0.09)

31
Additionally, I verify that the significant premium on the Scaled DD Measure is not entirely attributable to the
scaling factor used to construct the Scaled DD Measure (average value of absolute current accruals). When I include the
decile rank based on the value of the scaling factor in cross-sectional asset pricing regressions, the premium on the
Scaled DD Measure remains statistically and economically significant.
32
Several studies criticize the Dechow-Dichev model on a conceptual basis. For example, Wysocki (2005) argues that
the DD Measure spuriously attributes higher earnings quality to firms that engage in opportunistic earnings smoothing.
Further, Liu (2006) finds that firms in the later stages of the business cycle appear to have lower accrual quality
compared to young and growing firms. While ex-ante it is unlikely that younger firms or firms engaged in opportunistic
earnings management should have lower cost of capital, I control for these biases by including decile ranks based on the
business cycle stage (as measured in Liu 2006) and earnings smoothness (as measured in FLOS 2004) as additional
variables in the cross-sectional asset pricing regressions. The results for both the original DD Measure and the Scaled
DD Measure are robust to the inclusion of these additional control variables.

81

Table 14. Continued

Panel D. Returns on the OSAQ factor Loadings: Firm-Level Cross-Sectional Regressions
Intercept
b
^
OSAQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

Mean Coefficient 0.49 0.00 0.38 0.16 -0.16 6.32%
FM p-value
(0.00) (0.99) (0.13) (0.39) (0.35)

Mean Coefficient 0.89 0.04 1.75%
FM p-value
(0.00) (0.56)

Panel E. Returns on the OSAQ factor Loadings: Portfolio-Level Cross-Sectional Regressions
Independent Variables

Intercept b
^
OSAQ
b
^
MktRf
b
^
SMB
b
^
HML

Adjusted
R
2

25 B/M and Size Portfolios
Regression Coefficient 2.75 -2.16 0.22 0.64 71.66%
FM p-value (0.00) (0.00) (0.27) (0.00)
Regression Coefficient 0.77 0.17 8.37%
FM p-value (0.00) (0.18)
Regression Coefficient 2.80 0.25 -2.10 -0.04 0.71 77.89%
FM p-value (0.00) (0.04) (0.00) (0.81) (0.00)
100 Scaled DD Measure Portfolios
Regression Coefficient -0.03 0.28 0.99 -0.39 23.54%
FM p-value (0.96) (0.57) (0.00) (0.21)
Regression Coefficient 0.76 0.21 21.34%
FM p-value (0.01) (0.00)
Regression Coefficient -0.14 0.18 0.44 0.85 -0.30 31.82%
FM p-value (0.77) (0.01) (0.36) (0.00) (0.32)
64 B/M, Size and Scaled DD Measure Portfolios
Regression Coefficient 2.72 -2.12 0.22 0.62 63.98%
FM p-value (0.00) (0.00) (0.23) (0.00)
Regression Coefficient 0.80 0.15 7.09%
FM p-value (0.00) (0.19)
Regression Coefficient 2.90 0.18 -2.20 0.00 0.68 68.52%
FM p-value (0.00) (0.03) (0.00) (0.98) (0.00)

The table contains results of asset pricing tests for a sample of 88,668 firm-year observations with fiscal years
ending in 1970 to 2004 and monthly returns collected from April 1971 to March 2006. All returns are reported in
percentages.
The orthogonalized SAQ factor (OSAQ factor) is constructing by first sorting stocks into deciles by the signed
total accrual magnitude. Then within each total accrual decile the stocks are sorted into quintiles by the DD Measure
magnitude. The OSAQ factor is an equally-weighted return on a hedge portfolio that is long(short) the two
top(bottom) quintiles of the DD Measure.
The description of tests can be found in Table 8.
82

Chapter 6: Conclusion
In this paper, I document that poor accrual quality firms (i.e., firms with high
DD Measures) systematically experience more negative cash flow shocks compared to
good accrual quality firms (with low DD Measures). After controlling for the effect of
cash flow shocks by either including proxies for future cash flow shocks in asset pricing
regressions or using a relative accrual quality measure that is less correlated with future
cash flow shocks, I find that the low accrual quality stocks (or stocks with higher accrual
quality factor betas) are accorded significantly higher expected returns compared to the
high accrual quality stocks (or stocks with lower accrual quality factor betas). These
results are robust to a number of sensitivity checks.
Several concurrent papers explore the cost-of-equity effects of alternative
accounting quality measures (e.g., Barth et al. 2006, Berger et al. 2006). While there are
multiple ways to proxy for the quality of accounting information, documenting that the
accrual quality measure based on the Dechow and Dichev (2002) model is priced by the
market is important due to both the theoretical appeal of the model and the extent to
which the model is used in accounting research. The model is especially appealing in the
asset pricing context because it captures an important feature of accruals (and earnings)
quality, namely, errors in accrual estimates and subsequent reversals in these errors, which
is related to the concept of accounting information precision and therefore maps more
readily into theoretical work on the link between the cost of capital and accounting
83

information quality (e.g., Lambert et al. 2007b). Further, the estimates derived from the
model do not rely on stock market variables in measuring accounting quality and thus
avoid possible mechanical associations with returns (in contrast, the magnitude of the
earnings response coefficient depends on the firm’s discount rate, e.g. Kormendi and Lipe
1987). Overall, documenting that the accrual quality measure based on the Dechow and
Dichev (2002) model is priced by the stock market represents an important contribution to
the literature on the cost of equity effects of accounting quality.

84

Bibliography
Aboody, D., J. Hughes, and J. Liu. 2005. Earnings quality, insider trading, and cost of
capital. Journal of Accounting Research 43 : 651-673.
Ali, A., and P. Zarowin. 1992. The role of earnings levels in annual earnings-returns
studies. Journal of Accounting Research 30: 286–296.
Altman, E. 1968. Financial ratios, discriminant analysis and the prediction of corporate
bankruptcy. Journal of Finance 23: 589–609.
Amihud, Y ., and H. Mendelson. 1986. Asset-pricing and the bid-ask spread. Journal of
Financial Economics 17: 223-249.
Barry, C., and S. Brown, 1985. Differential information and security market equilibrium.
Journal of Financial and Quantitative Analysis 20: 407-422.
Barth, M., Y . Konchitchki, and W. Landsman. 2006. Cost of capital and financial
statement transparency. Working paper. Stanford University and University of
North Carolina.
Barth, M., L. Hodder and S. Stubben. 2008. Fair Value Accounting for Liabilities and
Own Credit Risk” forthcoming, The Accounting Review.
Beaver, W., M. McNichols, and R. Price. 2007. Delisting returns and th-eir effect on
accounting-based market anomalies. Journal of Accounting and Economics 43:
341-368.
Berger, P., Chen, H., Li, F., 2006. Firm specific information and cost of equity capital.
Working paper. University of Chicago.
Bhattacharya, U., H. Daouk, and M. Welker. 2003. The world price of earnings opacity.
The Accounting Review 78: 641-678.
Biddle, G., and G. Hilary. 2006. Accounting quality and firm-level capital investment, The
Accounting Review 81: 963-982.
Blume, M., and R. F. Stambaugh. 1983. Biases in computed returns: An application to the
size effect. Journal of Financial Economics 12: 387-404.
Botosan, C.A. 1997. Disclosure level and the cost of equity capital. The Accounting
Review 72: 323-350.
85

Botosan, C., and M. Plumlee. 2002. A re-examination of disclosure level and the expected
cost of equity capital. Journal of Accounting Research 40: 21-40.
Campbell, J.Y ., and R. J. Shiller. 1988. The dividend-price ratio and expectations of future
dividends and discount factors. Review of Financial Studies 1: 195-228.
Campbell, J.Y ., C. Polk, and T. Vuolteenaho. 2007. Growth or Glamour? Fundamentals
and Systematic Risk in Stock Returns. Working paper. Harvard University,
London School of Economics.
Carhart, M. M. 1997. On persistence in mutual fund performance. Journal of Finance 52:
57-82.
Chen, S., T. Shevlin, and Y .H. Tong. 2007. Does the pricing of financial reporting quality
change around dividend changes? Journal of Accounting Research 45: 1–40.
Coles, J., U. Loewenstein, and J. Suay. 1995. On equilibrium pricing under parameter
uncertainty. The Journal of Financial and Quantitative Analysis 30: 347-374.
Collins, D. W., S.P. Kothari, J. Shanken, and R.G. Sloan. 1994. Lack of timeliness and
noise as explanations for the low contemporaneous return-earnings association.
Journal of Accounting and Economics 18: 289-324.
Core, J., W. Guay, and R. Verdi. 2007. Is accruals quality a priced risk factor?
Forthcoming in Journal of Accounting and Economics.
Dechow, P., and I. Dichev. 2002. The quality of accruals and earnings: the role of accrual
estimation errors. The Accounting Review 77: 35-59.
Diamond, D., and R. Verrecchia. 1991. Disclosure, liquidity and cost of capital. Journal of
Finance 46: 1325-59.
Dichev, I. D. 1998. Is the risk of bankruptcy a systematic risk? Journal of Finance 53:
1131-1147.
Doyle, J.T., W. Ge, and S.E. McVay. 2007. Accruals quality and internal control over
financial reporting. Forthcoming, The Accounting Review.
Easley, D., and M. O’Hara. 2004. Information and the cost of capital. Journal of Finance
59: 1553–1583.
Easley, D., S. Hvidkjaer, and M. O’Hara. 2005. Factoring information into returns.
Working paper. Cornell University. University of Maryland.
86

Easton, P., and S. Monahan. 2005. An evaluation of accounting based measures of
expected returns. The Accounting Review 80: 501-538.
Ecker, F., J. Francis, I. Kim, P. Olsson, and K. Schipper. 2006. A returns-based
representation of earnings quality. The Accounting Review 81: 749-780.
Elton, J. 1999. Expected return, realized return and asset-pricing tests. Journal of Finance
54: 1199-1220.
Fama, E., and J. MacBeth. 1973. Risk, return, and equilibrium: empirical tests. Journal of
Political Economy 81: 607–636.
Fama, E., and K. French. 1993. Common risk factors in the returns on bonds and stocks.
Journal of Financial Economics 33: 3-56.
Fama, E., and K. French. 1997. Industry costs of equity. Journal of Financial Economics
43: 153-194.
Francis, J., R. LaFond, P. Olsson, and K. Schipper. 2004. Costs of equity and earnings
attributes. The Accounting Review 79: 967-1010.
Francis, J., R. LaFond, P. Olsson, and K. Schipper. 2005. The market pricing of accruals
quality. Journal of Accounting and Economics 39: 295-327.
Freeman, R., and S.Y . Tse. 1992. A non-linear model of security price responses to
unexpected earnings. Journal of Accounting Research 30: 185–209.
General Accounting Office (GAO). 2002. Financial statement restatements: Trends,
market impacts, regulatory responses, and remaining challenges. GAO-03-138
(October).
General Accounting Office (GAO). 2006. Financial Restatements. Update of public
company trends, market impacts, and regulatory enforcement activities.
GAO-06-678 (July).
Hayn, C.. 1995. The information content of losses. Journal of Accounting and Economics
20: 125–153.
Hughes, J. S., J. Liu, and J. Liu. 2007. Information asymmetry, diversification, and cost of
capital. The Accounting Review 82: 705-729.
Jagannathan, R., and Z. Wang. 1996. The conditional CAPM and the cross-section of
expected returns. Journal of Finance 51: 3–53.
87

Kaplan, S., and L. Zingales, 1997. Do investment-cash flow sensitivities provide useful
measures of financing constraints? The Quarterly Journal of Economics 112:
169-215.
Klein, R., and V . Bawa. 1976. The Effect of estimation risk on optimal portfolio choice.
Journal of Financial Economics 3: 215-231.
Kogan, L., and T. Wang. 2003. A Simple theory of asset pricing under model uncertainty.
Working paper. MIT and University of British Columbia.
Kormendi, R., and R. Lipe. 1987. Earnings innovations, earnings persistence, and stock
returns. Journal of Business 60: 323-346.
Kravet, T. D., and T. J. Shevlin. 2007. Accounting restatements and information risk.
Working Paper. University of Washington.
Lakonishok, J., A. Shleifer, and R. Vishny. 1994. Contrarian investment, extrapolation,
and risk. Journal of Finance 49: 1541–1578.
Lambert, R., C. Leuz, and R. Verrecchia. 2007a. Accounting information, disclosure, and
the cost of capital. Journal of Accounting Research 45: 385-420.
Lambert, R., C. Leuz, and R. Verrecchia. 2007b. Information asymmetry, information
precision, and the cost of capital. Working Paper. University of Pennsylvania and
University of Chicago.
Lamont, O., C. Polk, and J. Saa-Requejo. 2001. Financial constraints and stock returns.
Review of Financial Studies 14: 529-554.
Liu, M., 2006. Accruals and managerial operating decisions over the firm life cycle.
Working Paper. Pennsylvania State University.
Liu, M., and P. Wysocki. 2006. Operating risk, information risk, and cost of capital.
Working Paper. MIT.
McNichols, M. 2002. Discussion of ‘The quality of accruals and earnings: the role of
accrual estimation errors.’ The Accounting Review 77: 61-69.
Monraham, P. 2005. Separating winners from losers among low book-to-market stocks
using financial statement analysis. Review of Accounting Studies 10: 133-170
Newey, W., and K. West. 1987. A simple, positive semi-definite heteroscedasticity and
autocorrelation consistent covariance matrix. Econometrica 55: 703-708.
88

Nichols, D.C. 2006. Fundamental risk or information risk? An analysis of the residual
accrual volatility factor. Working Paper. Cornell University.
Ohlson, J.A. 1995. Earnings, book values and dividends in security valuation.
Contemporary Accounting Research 11: 661-687
Palmrose, Z-V ., and S. Scholz. 2004. The circumstances and legal consequences of
non-GAAP reporting: Evidence from restatements. Contemporary Accounting
Research 21: 139–80.
Palmrose, Z-V ., V . J. Richardson, and S. Scholz. 2004. Determinants of market reactions
to restatement announcements. Journal of Accounting and Economics 37: 59–89.
Piotroski, J.D. 2000. Value investing: the use of financial statement information to
separate winners from losers. Journal of Accounting Research 38: 1-41.
Piotroski, J.D. 2006. Further evidence on the relation between historical changes in
financial condition, future stock returns and the value/glamour effect. Working
Paper. University of Chicago.
Rajgopal, S., and M. Venkatachalam. 2007. Financial reporting quality and idiosyncratic
return volatility over the last four decades. Working Paper. University of
Washington, Duke University.
Richardson, S., I. Tuna, and M. Wu. 2003. Capital market pressures and earnings
management: The case of earnings restatements. Working paper. University of
Pennsylvania.
Shanken, J.. 1992. On the estimation of beta-pricing models. Review of Financial Studies
5: 1-34.
Shumway, T. 1997. The delisting bias in CRSP data. The Journal of Finance 52: 327 –
340.
Shumway, T., and V .A. Warther. 1999. The delisting bias in CRSP’s Nasdaq data and its
implications for the size effect. The Journal of Finance 59: 2361-2379.
Sloan, R. G. 1996. Do stock prices fully reflect information in accruals and cash flows
about future earnings? The Accounting Review 71: 289–315.
Verdi, R. S. 2006. Financial reporting quality and investment efficiency. Working Paper.
University of Pennsylvania.
Wysocki, P., 2005. Assessing earnings and accruals quality. Working paper. MIT.
89

Appendix: Cash Flow Shock Proxies

Earnings Surprise (SURP)

The earnings surprise variable (SURP) is calculated as actual ROE (return on equity)
minus expected ROE (E
t
(ROE
it+1
)). Expected ROE is estimated from a simple statistical
model that is based on two assumptions: (1) annual earnings follow a first-order
autoregressive process (AR1), and (2) returns lead earnings in capturing value-relevant
information, and thus current returns reflect information about future earnings (e.g.
Collins et al. 1994):

ROE
it+1
= β
0
+ β
1
ROE
it
+ β
2
RET
it
+ ε
it+1
,

where ROE
it
denotes return on equity for the firm i during fiscal year t, i.e., earnings
before extraordinary items (Compustat #18) over the value of equity at the beginning of
fiscal year t (Compustat #60), and RET
it
denotes the raw return on the stock of the firm
computed over fiscal year t.

Estimation of E
t
(ROE
it+1
) is performed in two stages: (1) the predictive coefficients β
0

^
,
β
1

^
, and β
2

^
are obtained by estimating the model cross-sectionally over the previous
year (e.g., using ROE
it-1
, ROE
it
,

and RET
it-1
); (2) earnings prediction is calculated by
combining the estimated coefficients with variable values: E
t
(ROE
it+1
)= β
0

^

+ β
1

^

ROE
it
+ β
2

^

RET
it
. For example, the earnings predicted at the end of 2004 and expected
to be earned over the course of 2005 are calculated using the values of predictive
variables for fiscal 2004 and coefficients estimated by regressing ROE from fiscal 2004
on ROE and returns from fiscal 2003.
33

Easton and Monahan Cash Flow News Proxy (CN_EM)

The cash flow news proxy based on analyst forecast revisions (CN_EM) is measured as an
exponent of a continuously compounded cash flow news measure from Easton and
Monahan (2005). The cash flow news measure is a sum of the revision in one-year-ahead
forecasted ROE and the capitalized revision in the two-year-ahead forecasted ROE:

LOG_CN
i,t+1
=ΔLOG_FROE
i,t+1
+ρ/(1-ρω)*ΔLOG_FROE
i,t+2
,

33
All results are robust to using an alternative specification of earnings surprise, where earnings are scaled by the
beginning-of-year market value of equity in place of the book value of equity.
90

where LOG_FROE
i,t+n
= ln(1 + FEPS
it,t+n
/BPS
it+n-1
), FEPS
it,t+n
denotes the n-years- ahead
consensus EPS forecast from I/B/E/S made in June after the fiscal year-end, and BPS
it

denotes book value of equity at the end of the fiscal year (Compustat #60) divided by the
number of shares outstanding (Compustat #54). Revisions refer to changes in forecasts
from June t to June t+1.

I use the ρ estimates reported in Easton and Monahan (2005). Persistence coefficients ω
t
are estimated through a pooled time-series cross-sectional regression for each of the 48
Fama and French (1997) industries: LOG_ROE
i,t-τ
= ω
0t
+ ω
t
× LOG_ROE
i,t-(τ-1),
where τ
is a number between zero and nine, and ROE is return on equity.

Abstract (if available)

Abstract Francis, LaFond, Olsson, and Schipper (2005) document that accrual quality is inversely related to the cost of equity capital. However, Core, Guay, and Verdi (2007) find no association between accrual quality and future stock returns and conclude that there is no evidence that the stock market prices accrual quality. I hypothesize that Core et al. 's result arises because poor accrual quality firms experience negative cash flow shocks in the future, which results in negative returns that offset the higher expected returns for such firms. Consistent with this prediction, I find a significant negative association between realized returns and accrual quality after controlling for cash flow shocks, either by including proxies for future cash flow shocks in asset pricing regressions or by using an accrual quality measure that is less correlated with future cash flow shocks. This result is robust to properly specified and standard asset pricing tests. Overall, this paper adds to the growing literature suggesting that accrual quality is linked to the cost of capital.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Who manages the firm matters: the incremental effect of individual managers on accounting quality

PDF

Does differential sensitivity to aggregate earnings shocks drive post-earnings-announcement drift?

PDF

Share repurchases: how important is market timing?

PDF

Expectation dynamics and stock returns

PDF

Real asset liquidity and asset impairments

PDF

Can I borrow your footnotes? Learning and network benefits of footnote similarity

PDF

Evolution of returns to scale and investor flows during the life cycle of active asset management

PDF

Three essays in derivatives, trading and liquidity

PDF

Private disclosure contracts within supply chains and managerial learning from stock prices

PDF

Two essays on major macroeconomic shocks in the Japanese economy: demographic shocks and financial shocks

PDF

Essays in asset pricing

PDF

Essays on real options

PDF

Changing fundamental analysis in the new economy: the case of DuPont analysis and STEM firms

PDF

Does mandatory adoption of international financial reporting standards decrease the voting premium for dual-class shares: theory and evidence

PDF

Essays on the role of entry strategy and quality strategy in market and consumer response

PDF

Essays on the effect of cognitive constraints on financial decision-making

PDF

“What you see is all there is”: The effects of media co‐coverage on investors’ peer selection

PDF

Essays on information, incentives and operational strategies

PDF

The changing policy environment and banks' financial decisions

PDF

Essays on monetary policy and international spillovers

Asset Metadata

Creator Ogneva, Maria (author)

Core Title Accrual quality and expected returns: the importance of controlling for cash flow shocks

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Business Administration

Degree Conferral Date 2008-05

Publication Date 05/02/2008

Defense Date 03/10/2008

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag accounting quality,cash flow shocks,OAI-PMH Harvest

Language English

Advisor Subramanyam, K.R. (committee chair), Beatty, Randolph (committee member), DeFond, Mark (committee member), Hann, Rebecca (committee member), Jones, Christopher S. (committee member)

Creator Email ogneva@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m1219

Unique identifier UC178254

Identifier etd-Ogneva-22080502 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-70048 (legacy record id),usctheses-m1219 (legacy record id)

Legacy Identifier etd-Ogneva-22080502.pdf

Dmrecord 70048

Document Type Dissertation

Rights Ogneva, Maria

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu