JOURNALOF

Accounting
ELSEVIER Journal of Accounting and Economics 22 (1996) 249-281
&Economids

The pricing of discretionary accruals

K.R. S u b r a m a n y a m
Leventhal School of Accounting, University of Southern California, Los Angeles, CA 90089. USA

Received May 1995; final version received August 1996)

Abstract

This paper examines if the stock market prices discretionary accruals. Evidence reveals
that, on average, the market attaches value to discretionary accruals. This evidence is
consistent with two alternative scenarios: (1) managerial discretion improves the ability
of earnings to reflect economic value, and (2) discretionary accruals are opportunistic and
wdue-irrelevant but priced by an inefficient market. Further evidence is consistent with
the former explanation. There is evidence of pervasive income smoothing, which im-
proves the persistence and predictability of reported earnings. There is also evidence that
discretionary accruals predict future profitability and dividend changes. Despite several
sensitivity checks, measurement error in the discretionary accruals proxy is an alternative
explanation for the results.

Key words: Capital markets; Valuation; Earnings management; Discretionary accruals:

Income smoothing; Dividends; Signaling

J E L class!fication: M41

1. Introduction

The role of accruals in arriving at a summarized measure of firm performance

is an important question in accounting research. Accrual earnings is regarded as
a superior measure of firm performance than cash flows because it mitigates

I thank S.P. Kothari (editor), Vic Bernard (discussant), Bill Beaver, Bob Bowen, Mark DeFond,
Ellen Engel, Frank Heflin, Bill Kross, Susan Moyer, Terry Shevlin,D. Shores, Brett Trueman, Terry
Warfield, John Wild, the workshop participants at the University of California Berkeley, the
University of Washington, the University of Wisconsin,the participants at the 1995JAE conference,
and an anonymous reviewer for their comments and suggestions.

0165-4101/96/$15.00 © 1996 Elsevier Science B.V. All rights reserved

PII S0 1 65-4 1 0 1 ( 9 6 ) 0 0 4 3 4 - X
250 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281

timing and mismatching problems inherent in measuring cash flows over short
intervals (Dechow, 1994). However, because of the flexibility accorded under the
Generally Accepted Accounting Principles (GAAP), accrual accounting is sub-
ject to managerial discretion. Managerial discretion could enhance earnings'
informativeness by allowing communication of private information (Watts and
Zimmerman, 1986; Holthausen, 1990; Healy and Palepu, 1993). On the other
hand, misalignment of managers' and shareholders' incentives could induce
managers to use the flexibility provided by GAAP to manage income oppor-
tunistically, thereby creating distortions in the reported earnings (Watts and
Zimmerman, 1986; Healy and Palepu, 1993).
Accruals, on average, have incremental information content above cash flows
(Bowen, Burghstahler, and Daley, 1987) and accrual-based earnings is a superior
measure of firm performance than various cash flow measures (Dechow, 1994).
However, this evidence does not establish if the superiority of accrual-based
earnings is because of, or despite, management's discretionary accounting
choices. The effect of discretionary accounting choices on the information
content of earnings remains relatively unexplored.
While accounting choice has been extensively examined from a contracting
perspective for a number of years (Watts and Zimmerman, 1990), research
investigating the link between accounting choice and the pricing of earnings is of
comparatively recent origin. ~ For example, Warfield, Wild, and Wild (1995)
document a negative association between incentives to manage accruals and the
information content of earnings. On the contrary, Hunt, Moyer, and Shevlin
(1995) report a positive association between the price-earnings multiplier and
the extent of smoothing. Some recent papers also examine the pricing of specific
discretionary items. For example, DeAngelo, DeAngelo, and Skinner (1993) and
Dechow (1994) document that the exclusion of special items improves the ability
of earnings to explain returns. Finally, a few papers examine the pricing of the
discretionary component in commercial banks' loan loss disclosures (Walhen,
1994; Beaver and Engel, 1996).
The evidence regarding the effects of managerial discretion on the pricing of
earnings is indirect and mixed. For example, researchers have examined the
impact of the extent of earnings management or the effect of specific discretion-
ary items on the earnings-returns relation, but extant literature is yet to
investigate the question: does the market price discretionary accruals? This paper
empirically examines the pricing of discretionary accruals. Evidence on this
issue could improve our understanding of the manner in which the capital

~A notable exception is the set of papers that examined the effectsof changes in accounting
techniqueson stockprice during the earlyyearsof positiveaccountingresearch(forexample,Ball,
1972;Kaplanand Roll, 1972;Sunder,1973).Despitethe mixedevidencefromthesepapers,this issue
(with the exceptionof probably LIFO/FIFOchoice)has not been investigatedby later researchers.
 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281 251

markets process accounting information. Evidence on this issue could also

provide some insights into the nature of discretionary accruals and the eco-
nomic incentives for discretionary accounting choice.
The empirical analysis is conducted on a sample of 21,135 firm-years compris-
ing 2,808 firms during 1973-93, Discretionary accruals are obtained by decom-
posing total accruals into discretionary and nondiscretionary components,
using a cross-sectional variation of the Jones (1991) model. The evidence
suggests that, under the maintained hypothesis that the Jones model correctly
decomposes accruals into its discretionary and nondiscretionary components,
discretionary accruals are on a v e r a g e priced by the market. Univariate regres-
sions between returns and net income, nondiscretionary income and operating
cash flow show that net income performs better than nondiscretionary income,
both in terms of coefficient value and explanatory power. While net income also
performs better than operating cash flow (a result that is consistent with
Dechow, 1994), a significant part of the improvement is attributable to the
discretionary component. Further, multivariate regressions reveal that the dis-
cretionary component of income has incremental information content beyond
the nondiscretionary component.
The pricing of discretionary accruals is a joint test of (1) the market pricing
mechanism and (2) the nature of the discretionary accruals. Hence, the pricing of
discretionary accruals is consistent with two alternative (but not necessarily
mutually exclusive) scenarios. In the first scenario, the market is efficient and
prices the discretionary component because it improves the ability of earnings to
reflect economic value of the firm. Under this hypothesis, managers improve the
value relevance of earnings by smoothing income or by communicating private
information about future profitability that is not reflected in historical cost
accounting. 2 In the second scenario, discretionary accruals distort earnings,
probably because of opportunistic earnings management. In this case, the
pricing of discretionary accruals is evidence of mispricing, i.e., the stock market
is functionally fixated on earnings (for example, Hand, 1989; Sloan, 1995).
Further tests are undertaken to discriminate between the two explanations.
The first set of tests reports evidence of pervasive income smoothing, which
reduces earnings variability and improves its persistence and predictability (see
Chaney, Jeter, and Lewis, 1996; Hunt, Moyer, and Shevlin, 1995). Additional
tests report evidence consistent with discretionary accruals conveying informa-
tion about future profitability (see DeFond and Park, 1996). Specifically, the

2The returns used in this paper are compounded over a twelve-month window. The association
between returns and discretionary accurals therefore does not necessarily indicate that there is
communication of private information through discretionary accruals. However, because the win-
dow also includes the post disclosure period, the pricing of discretionary accurals is not necessarily
inconsistent with such scenario.
252 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281

discretionary component is positively associated with future operating cash

flows, nondiscretionary income, and net income, and is also positively asso-
ciated with changes in current and future dividends.
An implicit assumption in this paper is that the cross-sectional Jones model
accurately partitions accruals into its discretionary and nondiscretionary com-
ponents. While the Jones model and its variations have been used extensively to
detect earnings management, the ability of these models to partition accruals
into the discretionary and nondiscretionary components is not without question
(see Footnote 10). Measurement error is of particular concern in this paper
because it not only introduces noise but also is an alternative explanation to the
results. Accordingly, a number of sensitivity tests are performed to verify if the
reported results are attributable to measurement error in the discretionary
accruals proxy. While the sensitivity analyses confirm the robustness of the
results, measurement error in the discretionary accruals proxy remains a plaus-
ible alternative explanation for the results.
Finally, this study uses a large sample and applies the discretionary accruals
model to test whether discretionary accruals are informative about firm value.
The evidence suggests that on average the market attaches value to discretionary
accruals, probably because the discretionary component increases the ability of
earnings to reflect fundamental value. However, the research design employed in
this paper is not best suited to identify discretionary accruals motivated by
opportunism. A carefully selected sample could reveal if the market prices
opportunistic earnings management (for example, Paek and Price, 1996, exam-
ine market pricing of accruals motivated by bonus plans and debt covenants
violations). While opportunistic earnings management does occur in specific
situations, the evidence in this paper suggests that such earnings manipulation is
not widespread, i.e., does not occur on average.
Section 2 introduces the sample and discusses the research design. Section
3 provides evidence of market pricing of discretionary accruals. Section 4 dis-
cusses alternative explanations and provides evidence that discretionary ac-
cruals are used to capture information about firm value. Section 5 concludes.

2. Research design

2.1. Sample

The sample consists of all firms years for which necessary data is available on
the Center for Research in Security Prices (CRSP) 1992 and the Compustat 1992
databases. Financial institutions (SIC 6000 to 6999) and observations with
change in year-end are excluded. Because of the nature of some of the tests, the
sample is restricted to those firms that have a minimum of five consecutive years
 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281 253

of data on all necessary variables. The available sample comprises of 21,631 firm
years. Observations where operating cash flows, discretionary accruals, or
nondiscretionary accruals are more than three standard deviations from their
respective means are deleted. This results in a loss of 496 observations (2.5% of
the sample), reducing the final sample to 21,135 firm years representing 2,808
firms during 1973-93. The results, however, are qualitatively unchanged when
the outliers are included in the sample.

2.2. Measurement of variables

Annual stock returns are measured as compounded monthly stock returns for
a twelve-month period ending three months after the end of the fiscal year of the
firm. Net income (NI) is measured as earnings before extraordinary items and
discontinued operations (Compustat item # 18). Operating cash flows (OCF) are
as defined in Compustat (#308). 3 Accruals (ACCR) are measured as the differ-
ence between net income and operating cash flows. All three variables are
deflated by total assets at the beginning of the period.
Nondiscretionary accruals are determined using the cross-sectional variation
of the Jones model (DeFond and Jimbalvo, 1994). This method estimates
nondiscretionary accruals as a function of the level of property, plant, and
equipment, and changes in revenue by estimating the following model:

ACCRj.t/TAja- 1= ~.[1/TAj.,_ 1] -~- fi'[AREVja/TAj., 1]

+ ?.[PPEj,,/TAj.,_ 1] + ej,,, (1)

where ACCRj., are total accruals for f i r m j in year t, TAj., refers to total assets,
AREVj,, is to change in net revenue, and PPEj,t refers to property, plant, and
equipment.
This model is estimated separately for each combination of two-digit SIC
code and calendar year. A combination with less than six observations is
dropped from the sample. Nondiscretionary accruals (NDAC) is defined as the
fitted value from Eq. (1):

NDACj., = &[1/TAj.,_,] + fi.[AREVj.,/TAj., ,]

+ ~. [PPEj.,/TAj,,_ 1] (2)

3For firms that had not adopted the cash flow format, operating cash flows was determined as
follows: OCF= funds from operations (#110)-Acurrent assets (#4)+ Acurrent liabilities
( # 5) + Acash (# 1) -- Acurrent portion of long-term debt ( # 34). When item # 34 was unavailable,
it was set to zero.
254 K.R. Subramanyam / Journal of Accounting and Economics' 22 (1996) 249-281

a n d d i s c r e t i o n a r y accruals ( D A C ) is defined as the residual: 4

DACj,t = A C C R j , t / T A j , t - 1 - ~" [ I / T A i , t 1] - - fl'[AREVj,t/TAj,t-1]

- ~.[PPE~, , / T A j , t _ 1]. (3)

T h e cross-sectional version of the Jones m o d e l is chosen over its time-series
c o u n t e r p a r t for a n u m b e r of reasons. First, the cross-sectional m o d e l generates
a larger sample. A g a i n s t 21,135 valid o b s e r v a t i o n s that are available using the
cross-sectional model, only 7,345 firm-years satisfy the s a m p l e selection criteria
when the time-series Jones m o d e l is used. Second, the n u m b e r of o b s e r v a t i o n s
per m o d e l is c o n s i d e r a b l y higher for the cross-sectional model: a m e d i a n of 31
for the cross-sectional m o d e l versus only 10 for the time-series (see T a b l e 1 for
details). This increases the precision of the estimates. Third, the time-series
m o d e l is e s t i m a t e d over a p e r i o d of up to ten years. Because of the lengthy time
p e r i o d s involved, it is possible for the m o d e l to be misspecified due to non-
s t a t i o n a r i t y . Finally, use of the time-series m o d e l lowers the p o w e r of tests which
e x a m i n e time-series b e h a v i o r in d i s c r e t i o n a r y accruals, because of o v e r l a p p i n g
e s t i m a t i o n a n d t r e a t m e n t periods.
T a b l e 1 r e p o r t s details of p a r a m e t e r estimates for four a l t e r n a t i v e versions of
the Jones model: the time-series a n d cross-sectional versions of b o t h the Jones
m o d e l a n d the m o d i f i e d Jones m o d e l p r o p o s e d b y D e c h o w , Sloan, a n d Sweeney
(1995). In general, the p a r a m e t e r estimates are better specified for the cross-
sectional versions than the time-series versions. F o r example, the average
s t a n d a r d errors of the coefficients are lower, there are fewer outliers a n d
a g r e a t e r p r o p o r t i o n of the coefficients are of the p r e d i c t e d signs (positive for
c h a n g e in revenue a n d negative for p r o p e r t y , plant, a n d equipment). 5 O t h e r t h a n
the coefficient on revenue c h a n g e (which is defined differently between the Jones
a n d the m o d i f i e d versions), the Jones a n d the m o d i f i e d versions d o n o t reveal
a n y m a j o r differences.

4In the time-series version of the Jones model (used in sensitivity tests), the discretionary and
nondiscretionary components of accurals are out-of-sample predictions. Hence the nondiscretionary
and discretionary accurals are actually the predicted values and prediction errors respectively. In the
cross-sectional version, however, they refer to the fitted values and residuals respectively. The
'look-ahead' problem which arises when a within-sample prediction is made in a time-series model is
not problem when a cross-sectional estimation is undertaken.
5While the coefficients of the cross-sectional Jones model have fewer outliers and a smaller
proportion of them are of the wrong sign, even this model has a number of implausible coefficients.
In order to investigate if the results are driven by these implausible estimates, the results are
replicated after deleting those industries with coefficients in the extreme 2%, thereby considerably
reducing the outlier problem. For example the maximum/minimum coefficients on change in
revenue (property, plant, and equipment) are 0.61/- 0.50 (0.15/- 0.31) after the deletion, as
compared to 2.68/- 6.30 (1.13/- 0.90) before the deletion. The results are also replicated after
deleting industries with the wrong sign on either coefficient.The results are robust to these sensitivity
tests (see Section 3.2).
 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281 255

After separating the discretionary and nondiscretionary components of ac-

cruals, nondiscretionary income ( N D N I ) is determined as follows:

NDNIj,~ = OCFj, t + NDACj,~. (4)

Nondiscretionary income is an estimate of the net income of a firm which has
been abstracted from discretionary accounting choice. Therefore, nondiscretion-
ary income estimates accounting income which has not been subject to any
managerial discretion.

Table 1
C o m p a r i s o n o f a l t e r n a t i v e e x p e c t a t i o n m o d e l s for a c c r u a l s

%
Model a Mean Std.dev. Median Maximum Minimum positive

Number of observations per regression

TS- Jones 9.02 1.32 10 10 6
TS Modified 9.02 1.32 10 10 6
CS Jones 59.87 76.73 31 455 6
CS Modified 59.57 76.42 31 455 6

Estimate of
TS - J o n e s - 23,23 415.32 - 0.04 7356.9 - 14474.4 49
( - 0.03) ( -- 0.04)

TS - Modified - 20.28 394.26 0.01 7394.3 13065.8 50

(0.04) (0.01)

CS Jones 0.12 4.52 0.00 103.59 - 43.76 51

( - 0.47) (0.04)

CS Modified 0.15 4.26 0.01 97.91 - 30.06 52

( - 0.29) (0.10)

Estimate of r, Predicted sign: +

TS Jones 0.09 1.16 0.09 54.05 - 89.05 70
(0.88) (0.71)

TS Modified 0.04 0.83 0.05 52,66 - 76,25 61

(0.468) (0.388)

CS Jones 0.06 0.37 0.07 2.68 - 6.30 73

(2.04) (1.37)

CS Modified 0.03 0.35 0.04 2.42 - 5.55 65

(1.36) (0.86)

aTS = time-series, C S = c r o s s - s e c t i o n a l .
256 K.R. Subramanyam / Journal o f Accounting and Economics 22 (1996) 249-281

Table 1 (continued)

%
Model a Mean Std.dev. Median Maximum Minimum positive

Estimate ofT, Predicted sign: -

TS - Jones - 0.04 5.09 - 0.06 419.20 - 210.83 30
( - 0.91) ( - 0.68)
TS Modified - 0.02 4.86 - 0.06 410.28 - 183.47 32
( - 0.839) ( - 0.63)
CS Jones - 0.07 0.10 - 0.07 1.13 - 1.29 7
( - 3.07) ( - 2.46)
CS - Modified - 0.06 0.09 - 0.06 1.16 - 0.90 9
( - 2.83) ( - 2.24)

The sample consists of all available data on the 1992 Standard and Poor's Compustat Industrial
Annual and Full Coverage Annual databases. The time-series Jones model is the original
model proposed by Jones (1991): ACCRj.t/TA~.,_ 1 = ct. [1/TAj,t_ 1] + fl'[AREVj,t/TA~.t- 1] +
y'[PPEj.t/TAj,~_ 1] + ej.t, where A C C R refers to total accruals, TA refers to total assets, R E V refers
to sales revenue, and PPE refers to property, plant, and equipment. Thej subscript denotes the firm
and the t subscript denotes the year. The time-series Jones model is estimated from the past
time-series for each firm. The modified time-series model is as proposed by Dechow, Sloan, and
Sweeney (1994). It is identical to the time-series Jones except that the sales revenue is adjusted for
change in accounts receivable. The cross-sectional Jones and modified Jones models are similar in
form to their time-series counterparts. However, they are estimated separately for each combination
of calendar year and two-digit SIC code. Figures in parentheses denote means/medians of individual
regression t-statistics.

2.3. D e s c r i p t i v e s t a t i s t i c s

Table 2 reports descriptive information about the variables. Net income and
operating cash flows are positive in over 80% of the sample. There appears to be
a slight bias towards profitable firms, probably because of the sample selection
criteria. Total accruals and nondiscretionary accruals are negative on an aver-
age because of depreciation. Mean and median discretionary accruals are close
to zero. This occurs by construction. However, discretionary accruals are more
variable than the nondiscretionary component, but are understandably less
variable than total accruals. 6

6The standard deviation of discretionary accurals (0.11) is marginally higher than that of net income
(0.10). While the lower variance of net income is possibly due to the negative correlation across
earnings' components, an alternative explanation is variance inflation caused by measurement error
in estimating discretionary accurals. Therefore sensitivity tests are conducted after eliminating firms
with extreme 50% variance of discretionary accurals, reducing the standard deviations of discretion-
ary accurals and net income to 0.059 and 0.064 respectively. The results in this paper are robust to
this sensitivity.
K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281 257

I I I I 1 [ I
"- ~ ~b~ =

..~ ~-o~ ~ _ _ ~
0 0 0 0 0 ~

~ m
I I I

•,~ ~ o~ :=

~ ~ ~0 ~ ~ o ~_
I I I

w
-m~ ~ .- ~ = - ~ o ~

s.N ~-- e u
°~
> o~~
258 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281

3. Discretionary accruals and stock returns

This section addresses the central issue of the paper: does the stock market
price the discretionary component of earnings? This issue is investigated by
regressing the discretionary and nondiscretionary components of net income
with contemporaneous stock returns, and examining the effect of discretionary
accruals on both the coefficient and the explanatory power. Section 3.1 presents
results of models where returns are regressed on earnings and its components.
Section 3.2 examines the robustness of the results to misspecification of the
cross-sectional Jones model.

3.1. Regression o f returns on earnings and its components

To assess the pricing of discretionary accruals, returns are regressed on the

levels of earnings and its components. This is consistent with Dechow (1994),
who also uses levels of earnings and cash flows when examining the value
relevance of accrual accounting. 7
First, univariate regressions individually using three alternative firm per-
formance measures - operating cash flows (OCF), nondiscretionary income
(NDNI), and net income (NI) - are estimated. The explanatory power and the
slope coefficients of these three models are compared with the objective of
assessing the relative information content of the three measures. This is similar
to the approach adopted by Dechow (1994). Dechow individually regresses net
income and two measures of cash flow with returns and compares the relative
magnitudes of the coefficient and the explanatory power of these univariate
regressions to investigate the value relevance of accruals. In a similar manner,
the value relevance of discretionary and nondiscretionary accruals is examined
by comparing univariate regressions of returns on net income, operating cash
flow, and nondiscretionary income. If accrual accounting increases value rel-
evance by reducing timing and mismatching problems as argued by Dechow
(1994), moving from operating cash flows to nondiscretionary income should
show an improvement in the coefficient and explanatory power. If discretionary
accounting choice introduces noise to the reported income as argued by analysts
(Graham, Dodd, and Cottle, 1962) and also some academics (Revsine, 1991),
then no increase (or even a decline) in the coefficient and explanatory power
should be observed when moving from nondiscretionary earnings to net income.

7The use of earnings levels in a regression of returns and earnings has theoretical and empirical
support. Ohlson and Shroff (1992) and Kothari (1992) describe the conditions under which the
earnings level is superior to earnings change as a surrogate for unexpected earnings. Easton and
Harris (1991) provide empirical evidenceconsistent with this claim.
 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281 259

Panel A of Table 3 reports results of the three univariate regressions. The first
row reports results of the regression using operating cash flows (OCF). The
adjusted R-square is 2.93% and the regression coefficient is 0.80 (which is
significant at the 0.01 level). The second row reports results of regression using
net income before extraordinary items (NI). The adjusted R-square is 5.44% and
the coefficient is 1.36 (significant at the 0.01 level). There is a substantial increase
in explanatory power and in the coefficient. These results are consistent with
Dechow (1994) and suggest that the market attaches value to (total) accruals.
The third model reports results of regressing returns with nondiscretionary
income (NDNI). The adjusted R-square is 4.33% and the coefficient is 0.89
(significant at the 0.01 level). These results indicate that nondiscretionary income
is more value-relevant than operating cash flows, but less than net income. This
is consistent with both nondiscretionary and discretionary accruals contributing
to the value relevance of earnings.
Panel A provides evidence consistent with the hypothesis that net income
explains a greater portion of contemporaneous returns than either nondis-
cretionary income or operating cash flow. However, mere comparison of
R-squares does not provide a statistical test of the hypotheses. Panel C provides
results of a likelihood ratio test suggested by Vuong (1989) that allows the re-
searcher to determine which model explains more of the dependent variable. ~ All
the comparisons, in particular the increased explanatory power of net income over
nondiscretionary income, are statistically significant at the 0.01 level or better.
The next stage of the analysis examines the incremental information content
of the discretionary and nondiscretionary components of net income by regress-
ing returns on earnings' components in multivariate models, Panel B of
Table 3 reports results of these models. In model 4, earnings is decomposed
into two components: operating cash flows and accruals. The R-square of the
model is 6.15%, which is higher than regressing returns on operating cash
flows alone (the Vuong statistic is significant at the 0.01 level). The coefficients
on operating cash flows and accruals are 1.45 and 1.15 respectively (both are
significant at the 0.01 level), indicating that both components have incremental
information content (this result is consistent with earlier research such as
Bowen, Burghstahler, and Daley, 1987). Model 5 reports results of decomposing
nondiscretionary income into cash flow and nondiscretionary accruals. The
R-square is 4.33% and the coefficients on cash flow and nondiscretionary
accruals are 0.89 and 0.89 respectively. The results indicate that nondiscretion-
ary accruals has incremental information content over operating cash flow,
although the increase in explanatory power is not as much as that of total
accruals (the Vuong statistic comparing model 5 with model 1, however, is

8 For a succinct discussion of the merits and some technical details regarding the Vuong's likelihood
ratio test, see Dechow (1994).
260 K.R. Subramanyam / Journal o f Accounting and Economics 22 (1996) 249-281

<

c~

0 ~rq ~o6
E

&
E

e.
el
&
e~
..~
Fm
0 8
,~ o ~ c5 ~ c5 ~ c5 o6

o o 0
~ N N
 K.R, Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281 261

o~ =~o

.B = ~

~ ~'
.o ~ca = ~~ ' ~~ . ~,
- o=
,--
c,io6

c4o6

tr~

o-- 4 ~ o ~
0 eq
rO 0
,,¢m.

,.~ 0 0

' ~ o---- o,.,.

o o "~
g
E ~ ~ oa ©

04
cq..,'q
t'~ 0",

¢-q ¢,%

•~ =~

~.m. ,- .ca

e'q
., ,.r.-. ca . ~

m.m.
"" ~.m z g> ~u

~N
C, m

-{
262 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281

significant at 0.01 level). Model 6 decomposes net income into nondiscretionary

income and discretionary accruals. The R-square is 6.48%, compared to 4.33%
for nondiscretionary income alone (the Vuong statistic is significant at the 0.01
level) and the coefficient on discretionary accruals is 1.00 (significant at the 0.01
level). This indicates that discretionary accruals has incremental explanatory
power over nondiscretionary income.
Finally, model 7 decomposes net income into its three component parts:
operating cash flows, nondiscretionary income, and discretionary accruals. The
R-square is 6.48 % which is significantly better than that of regressing returns on
operating cash flow and nondiscretionary income (Vuong statistic is significant
at the 0.01 level). The coefficients on operating cash flow, discretionary accruals,
and nondiscretionary accruals are 1.42, 1.47, and 1.00 respectively (all are
significant at the 0.01 level). These results indicate that each component has
incremental information content. However, the weight attached to the dis-
cretionary component is lower than the weight attached to the nondiscretionary
component. 9
In summary, these results reveal that, under the maintained assumption that
the cross-sectional Jones model is able to correctly decompose total accruals
into the discretionary and nondiscretionary components, discretionary accruals
have incremental information content and improve earnings' ability to explain
returns. This evidence is consistent with the market attaching value to dis-
cretionary accounting choice.

3.2. Sensitivity analysis

The maintained hypothesis of the paper is that the cross-sectional Jones

model correctly decomposes earnings into the discretionary and nondiscre-
tionary components. While the Jones model and its subsequent variations are
state-of-art techniques for determining discretionary accruals, these models
are simplistic and their ability to accurately decompose accruals into discre-
tionary and nondiscretionary components is not without question.l° Therefore,

9This difference is statistically significant at the 0.01 level. While this result indicates that the market
prices the discretionary and nondiscretionary components differently, such an inference needs to be
made cautiously. This is because the lower coefficient on discretionary accurals is subject to
alternative interpretations and is not robust to alternative computations of discretionary accruals.
See Section 3.2 for details.
1oA number of papers have used the Jones model (or its variations) to discern accounting choice (for
example, Pourciau, 1993; Perry and Williams, 1994; Gaver, Gaver, and Austin, 1995; DeFond and
Jimbalvo, 1994). On the other hand, certain researchers are critical of the efficacy of these techniques
(Holthausen, Larcker, and Sloan, 1995). Recently, a few papers have critically examined the ability of
these techniques to partition accruals into discretionary and nondiscretionary components. The
results are mixed. While Dechow, Sloan, and Sweeney (1995) conclude that these techniques are well
specified in general but generate low power tests, Hansen (1995) and Guay, Kothari, and Watts
(1996) question the ability of these models to partition accruals in a meaningful manner.
 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281 263

measurement error arising from misspecification of the cross-sectional Jones

model is a plausible alternative explanation for the results in Section 3.1, i.e., the
discretionary accruals proxy could correlate with returns even when not priced
by the market because it is contaminated with the nondiscretionary component.
Accordingly a number of sensitivity tests are conducted to examine the robust-
ness of the results to this possibility, both by excluding subsamples where there
is a greater possibility of measurement error and by using alternative accrual
expectation models.
While the cross-sectional Jones model appears to be better specified than
alternative models, it is not free of measurement problems. For example, the
coefficients in the estimation model are often extreme values or not of the
predicted sign (see Table 1). Further, the variance of discretionary accruals is
greater than that of net income (see Table 2), suggesting the presence of
measurement error. The robustness of the results to these problems is investi-
gated by estimating multivariate regressions of returns on earnings components
after deleting subsamples where these problems are likely to be pronounced.
Panel A of Table 4 reports these results: model 1 reports results after eliminating
industries with extreme 2% of the coefficients in the estimation model (see
Footnote 5), model 2 reports results after including only those industries with
the predicted signs in the estimation model (positive for change in revenue and
negative for property, plant, and equipment), and model 3 reports results after
eliminating firms with extreme 50% of variance of discretionary accruals (see
Footnote 6). The results of all three models are qualitatively similar to that
reported in Section 3.1: the coefficient on discretionary accruals is positive
but lower than that of nondiscretionary accruals and there is an increase in
explanatory power with the inclusion of the discretionary component. These
results confirm the robustness of the findings in Section 3.1 to these estimation
problems.
Panel B of Table 4 reports results of multivariate regressions of returns on
earnings components using alternative models for estimation of discretionary
accruals. Model 4 reports results using the cross-sectional variation of the
modified Jones model proposed by Dechow, Sloan. and Sweeney (1995).
Dechow, Sloan, and Sweeney argue that this is a more powerful model since it
incorporates revenue-based earnings management which is not captured in
Jones' original formulation. The results of this model are qualitatively similar to
those using the cross-sectional Jones.
In this paper, the cross-sectional model is estimated separately for each two-
digit SIC code. While the use of two-digit SIC codes ensures a larger number of
observations per model, it is possible that significant diversity could exist across
firms in each model. This diversity could result in firm-specific economic
transactions not being captured by the model, and therefore being classified as
discretionary. In order to mitigate this problem, the regression is estimated by
using the cross-sectional Jones model estimated by four-digit SIC codes and by
using the time-series Jones model. Model 5 reports results of the four-digit SIC
264 K,R. Subramanyam / Journal o f Accounting and Economics 22 (1996) 249 281

,- <
0

e
¢)

e.

.=.

e~

~ c5 c5 ~ ~5 cq ~5 ¢q c5 ~ c5 o6

e.

o
e~

0 .~
K,R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281 265

'.., "" -- ~.~ o0. ' ~ ~ O'--

,.=~ c . h = o ~ ,~ .-..E ,-, ,,.-

.-,~_ ~

~',= ~ ~ P ' - ' = " g'= 6

~.~ =-

~-'~ o o.~ 6 , - , - ~ o .

.~-, ~ ~" ~ e.., ,~ ~ ~ ~.. " 0 "~

~-'~ ;.~ ~ -.~ ,.~ ,.- .-~ ~ .

,~_
~ ,"~~ . ~ - ~ = E . ~ ~.. ~~',z
~ ~:~ [.,...
~ ~~ ~= ._ ~ o

_ . ~ '~ ~,~ o~,-~- = ~- ~-~ ,

266 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281

cross-sectional model. The results are qualitatively similar to that using the
two-digit grouping. Model 6 reports results implementing the time-series Jones
model. While the discretionary component is priced in this model, the weights
attached to the discretionary and nondiscretionary components are similar
(unlike the cross-sectional models which attach a lower weight to the discretion-
ary component).
Finally, Dechow, Sloan, and Sweeney (1995) show a significant negative
association between operating cash flows and total accruals. While part of this
association may be due to discretionary income smoothing, it is possible that
accruals are mechanically associated with operating cash flows. ~1 The Jones
model does not capture this negative association when partitioning accruals
into the discretionary and nondiscretionary components. Accordingly, as pro-
posed by Hunt, Moyer, and Shevlin (1995), the cross-sectional accruals' expecta-
tion model is estimated after including operating cash flows as an additional
explanatory variable (in addition to change in revenue and property, plant, and
equipment). The results of this model (model 7) are qualitatively similar to that
of the cross-sectional Jones model.
In summary, the results in Section 3.1 are robust to various sensitivity checks,
indicating that it is less likely that the results are driven by measurement
problems with the cross-sectional Jones model. However, despite the robustness
of the results, it is not possible to rule out measurement error as an alternative
explanation.

4. Alternative explanations

The evidence in the previous section indicates that discretionary accounting

choices are priced by the market. Since the pricing of discretionary accruals is
a joint test of the market pricing mechanism and the nature of the discretionary
accruals, this evidence is consistent with two alternative scenarios. In the first
scenario, the market prices the discretionary component because it captures
value-relevant information. Under this hypothesis, managers use their discre-
tion to improve the ability of reported earnings to refect fundamental value of
the firm (Watts and Zimmerman, 1986; Holthausen, 1990). For example, man-
agers may smooth reported earnings to counteract the effect of transitory
movements in profitability. Managers may also use their discretion to commun-
icate their knowledge about firm profitability which is yet to be reflected in the

l~Certain transactions, such as cash purchase of inventory, affect operating cash flows without
affectingnet income.These transactions, by definition,affecttotal accruals in the opposite direction
as operating cash flows, thereby inducing a negative correlation between the two.
 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281 267

historical cost-based earnings. 12 In the alternative scenario, managers use their

discretion opportunistically (Watts and Zimmerman, 1986), thereby garbling the
reported earnings. In such a scenario, the pricing of discretionary accruals is
evidence of market mispricing, probably because of functional fixation on the
bottom line earnings number (Sloan, 1995; Hand, 1989).
The analysis in this section attempts to discriminate between the two alterna-
tive explanations by examining if discretionary accruals improve the ability of
earnings to reflect future profitability. Section 4.1 explores the income smooth-
ing hypothesis by investigating whether firms smooth earnings on average
and if this smoothing improves the value relevance of earnings. Section 4.1.1
presents evidence of income smoothing while Section 4.1.2 examines its
implications. Section 4.2 examines whether discretionary accruals communi-
cate information about future profitability both directly, by examining the
association between current discretionary accruals and future measures of
profitability (Section 4.2.1), and indirectly, by examining the relation between
discretionary accruals and changes in current and future dividends (Section
4.2.2).

4. l. Discretionary accruals and income smoothing

Previous literature has indicated that one motivation of earnings manage-

ment is to smooth earnings (Ronen and Sadan, 1981). Smoothing is defined as
a form of earnings management with the objective to 'reduce the divergence of
reported earnings from an earnings number that is 'normal' or 'expected' for the
firm' (Moses, 1987). Opportunistic smoothing may be undertaken to maximize
benefits from bonus plans (Healy, 1985) or to signal lower risk (Trueman and
Titman, 1988). Hunt, Moyer, and Shevlin (1995) report that market value is
positively associated with the magnitude of reduction in earnings volatility
through discretionary income smoothing.
While income smoothing has an opportunistic connotation, not all smooth-
ing is necessarily opportunistic. Hand (1989) observes that managers may
smooth earnings to align expectations with that of the market and even to
increase the persistence of earnings. If earnings are smoothed to mitigate
the effects of transitory cash flows and adjust reported earnings towards a
more stable trend, then income smoothing can enhance the value relevance of
earnings (also see Chaney, Jeter, and Lewis, 1996).

12This 'recognition lag' in historical cost-based earnings is a well-documented phenomenon

IBeaver, Lambert, and Morse, 1980; Warfield and Wild, 1992; Collins, Kothari, Shanken, and
Sloan, 1994).
268 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281

4.1.1. Evidence of income smoothin9

Panel A of Table 5 reports contemporaneous correlations between various
components of net income. The panel presents a matrix of mean (median)
firm-specific Pearson correlation coefficients. It can be seen that net income is
positively correlated with each of its components. This is not surprising since net
income is merely an aggregation of its components. Operating cash flows (OCF)
and total accruals (ACCR) are negatively correlated, with mean (median) cor-
relation of -0.63 (-0.80). This is consistent with previous evidence such as in
Dechow (1994) and Dechow, Sloan, and Sweeney (1995). While this negative
correlation could arise because of accrual accounting, it could also arise from
income smoothing.
Discretionary accruals (DAC) is negatively correlated with nondiscretionary
accruals (NDAC): the mean (median) correlation is -0.26 ( - 0 . 3 3 ) . 13 This
evidence is consistent with income smoothing. Further evidence is obtained by
examining the correlations between discretionary accruals, operating cash flows,
and nondiscretionary income (NDNI). The mean (median) correlation between
operating cash flows and discretionary accruals is -0.55 ( - 0 . 6 6 ) , in compari-
son to the mean (median) correlation of -0.07 ( - 0 . 1 0 ) between nondiscretion-
ary accruals and operating cash flows. This evidence is consistent with dis-
cretionary accounting choices explaining a larger proportion of the negative
correlation between accruals and operating cash flows than accrual accounting.
Nondiscretionary income is also negatively correlated with discretionary ac-
cruals: the mean (median) correlation is - 0.67 ( - 0.81). However, inferences
from this result must be made with caution because the negative correlation
could be spuriously induced by measurement error.
To further explore the income smoothing hypothesis, the effect of discretion-
ary accruals on the volatility of earnings is examined by comparing the standard
deviation of net income with those of operating cash flows and nondiscretionary
income. Panel B of Table 5 reports mean and median values of firm-specific
means and standard deviations of net income, operating cash flows, and nondis-
cretionary income. The median standard deviation of net income is 4% of assets
which is substantially lower than that of nondiscretionary income, which is 9%,
or that of operating cash flow, which is 7%. These results indicates that net
income is less volatile than either nondiscretionary income or operating cash
flows.
Finally, panel C of Table 5 reports the autocorrelation structure of the first-
differences in net income, operating cash flows, and nondiscretionary income.

13This result may appear puzzlingbecause the two variables are orthogonal by construction. While
it is true that discretionaryand nondiscretionaryaccuralsare orthogonal by constructionfor a given
combinationof industry and calendaryear,the reported negativecorrelationsrepresentsthe average
of the firm-specificcorrelations over time.
 K.R. Subramanyam / Journal of Accounting and Economies 22 (1996) 249 281 269

Table 5
Evidence of i n c o m e s m o o t h i n g t h r o u g h d i s c r e t i o n a r y accruals

Panel A: Contemporaneous correlations

OCF NDNI ACCR NDAC DAC

Net i n c o m e (NI) 0.31 0.38 0.22 0.19 0.11

(0.40) (0.48) (0.30) (0.30) (0.14)

O p e r a t i n g cash flow (OCF) 0.75 - 0.63 - 0.07 0.55

(0.85) ( 0.80t (-0.10) - 0.66)

N o n d i s c r e t i o n a r y i n c o m e (NDNI) - 0.42 0.36 0.67

( - 0.51 ~ (0.46) 0.81)

Accruals (ACCR) 0.21 0.73

(0.29) (0.82)
N o n d i s c r e t i o n a r y accruals (NDAC) 0.26
- 0.33)

Panel B: Time-series mean and standard deviation

Mean Std. dev.

Mean Median Mean Median

Net i n c o m e (NI) 0.04 0.05 0.06 0.04

O p e r a t i n g cash flow (OCF) 0.07 0.08 0.09 0.07
N o n d i s c r e t i o n a r y i n c o m e (NDNI) 0.05 0.05 0.10 0.09

Panel C: Autocorrelation of first differences

ANI AOCF ANDNI

Mean Median Mean Median Mean Median

Lag 1 0.18 - 0.20 - 0.39 - 0.46 - 0.41 - 0.48

Lag 2 - 0.13 - 0.16 - 0.03 - 0.04 0.02 - 0.03
Lag 3 - 0.06 - 0.10 - 0.01 - 0.01 - 0.01 - 0.02

The original s a m p l e consists of 21,631 firm-years d u r i n g the period 1973-93 for which a m i n i m u m of
five consecutive years of d a t a are available. O b s e r v a t i o n s that are m o r e t h a n three s t a n d a r d
d e v i a t i o n s from the m e a n for o p e r a t i n g cash flows, n o n d i s c r e t i o n a r y accruals, a n d d i s c r e t i o n a r y
accruals are excluded. This results in a loss of 496 o b s e r v a t i o n s reducing the final s a m p l e to 21,135
firm-years.
Panel A presents m e a n (median) firm-specific c o n t e m p o r a n e o u s P e a r s o n c o r r e l a t i o n s between net
i n c o m e (NI), o p e r a t i n g cash flow (OCF), n o n d i s c r e t i o n a r y i n c o m e (NDNI), total accruals (ACCR),
n o n d i s c r e t i o n a r y accruals (NDAC), a n d d i s c r e t i o n a r y accruals (DAC). Figures in parentheses
indicate m e d i a n values. P a n e l B presents m e a n s a n d m e d i a n s of the firm-specific time-series m e a n s
a n d s t a n d a r d d e v i a t i o n s for net i n c o m e (NI), o p e r a t i n g cash flow (OCF), a n d n o n d i s c r e t i o n a r y
i n c o m e (NDNI). Panel C presents a u t o c o r r e l a t i o n of the first differences (changes) in net income
(ANI), o p e r a t i n g cash flow (AOCF), and n o n d i s c r e t i o n a r y i n c o m e (ANDNI).
270 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281

All three variables are negatively autocorrelated for up to three years, although
the autocorrelation dampens out with time. First-differences of both operating
cash flows and nondiscretionary income display a similar autocorrelation pat-
tern: a high negative autocorrelation at lag 1 that dampens fairly rapidly.
Specifically, the median autocorrelation for nondiscretionary income (operating
cash flows) is -0.48 (-0.46) at lag 1, but drops to -0.03 (-0.04) and -0.02
( - 0.01) at lags 2 and 3 respectively. This pattern is consistent with mean
reversion, suggesting the presence of large transitory components in both
operating cash flows and nondiscretionary income. In contrast, net income has
a relatively flatter autocorrelation structure. Specifically, the median autocorre-
lations in the first-difference of net income are -0.20, -0.16, and -0.10 at lags
1, 2, and 3 respectively. While the extent of (negative) autocorrelation is rela-
tively lower at lag 1, it persists for a longer time. Thus, while there is some mean
reversion in the net income series, it is much lower than that in either operating
cash flows or nondiscretionary income. This evidence is consistent with net
income being a smoother series than either nondiscretionary income or operat-
ing cash flows.
Collectively, the above evidence is consistent with income smoothing. How-
ever, caution must be exercised in drawing inferences from the results using
nondiscretionary income because measurement error in the estimate of dis-
cretionary accruals could spuriously indicate the presence of income smoothing
by inflating the variance of discretionary accruals.

4.1.2. Implications of income smoothing

This section examines the implications of income smoothing for the persist-
ence and predictability of earnings, and the pricing of earnings and its
components. This is investigated by examining the autocorrelations and cross-
correlations between the levels of current and future net income, operating cash
flows, and nondiscretionary income. The higher these autocorrelations, the
greater the persistence (Collins and Kothari, 1989). The matrix of correlations is
presented in panel A of Table 6. Two empirical regularities can be observed from
the correlations. The first regularity pertains to the correlations between current
net income and future values of all three variables. It can be seen that current net
income is more positively correlated with future values of all three variables
(with the exception of three-year-ahead net income). For example the median
correlation between current net income and one-year-ahead net income (operat-
ing cash flows, nondiscretionary income) is 0.43 (0.08, 0.08), which is higher than
the median correlation between operating cash flows and one-year-ahead net
income (operating cash flows, nondiscretionary income) of 0.27 ( - 0.01, - 0.02)
or the median correlation between current nondiscretionary income and one-
year-ahead net income (operating cash flows, nondiscretionary income) of
0.31 (0.02, - 0.03). Qualitatively similar results are observed at the two- and
K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281 271

k.)
II II

o "~

S
b
o 0

~b
:B 8~
0
8
I I I I
s=
©

8 li li ~:~ '4:::I

o
d
E
8
=_

II i

,=
".5

~p

o
8
8~
P

8
©

8
0

Z 0 Z
272 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281

three-year intervals. These results are consistent with the greater persistence of
net income vis-a-vis operating cash flow or nondiscretionary income.
The second regularity is that future levels of net income generally correlate
more highly with current levels of all three variables. For example, the median
correlation between one-year-ahead net income and current net income (operat-
ing cash flow, nondiscretionary income) is 0.43 (0.27, 0.31), which is higher than
the correlation between one-year-ahead operating cash flow and current net
income (operating cash flow, nondiscretionary income) of 0.08 ( - 0.01, 0.02), or
the correlation between one-year-ahead nondiscretionary income and current
net income (operating cash flow, nondiscretionary income) of 0.08 ( - 0.01,
- 0.03). A similar result is observed at the two- and three-year intervals. This
result is consistent with higher predictability of net income.

4.2. Discretionary accruals and communication of private information

Prior literature suggests that managers use the discretion provided by GAAP
to convey private information to the market. For example, Healy and Palepu
(1993) remark that 'disclosure s t r a t e g i e s . . , provide a potentially important
means for corporate managers to impart their knowledge to outside investors,
even if capital markets are efficient'. While the pricing of discretionary accruals
is not compelling evidence of private information communication, it is not
necessarily inconsistent with this scenario. A formal test of this hypothesis is
beyond the scope of this paper. However, evidence consistent with this scenario
is presented in this section as one plausible explanation for the pricing of
discretionary accruals.
This section examines whether discretionary accruals communicate informa-
tion about future profitability through two different types of tests. The first set of
tests (Section 4.2.1) directly examines the ability of discretionary accruals to
predict future profitability, where profitability is measured in terms of operating
cash flow, nondiscretionary income, and net income. The second set of tests
(Section 4.2.2) examines the association between discretionary accruals and
current and future dividend changes.

4.2.1. Discretionary accruals and future profitability

This section presents results of OLS regressions between levels of various
future profitability measures and the discretionary and nondiscretionary com-
ponents of current income. Future profitability is measured as one-, two-, and
three-year-ahead operating cash flows, nondiscretionary income, and net income.
Panel A of Table 7 reports results pertaining to one-year-ahead profitability. All
three components of net income (operating cash flows, nondiscretionary ac-
cruals, and discretionary accruals) predict future levels of all three profitability
 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281 273

indicators (all coefficients are significant at the 0.01 level). Collectively, the three
components explain 16.68%, 5.01%, and 37.64% of one-year-ahead operating
cash flow, nondiscretionary income, and net income, respectively. More perti-
nent, discretionary accruals are significantly and positively associated with
future levels of profitability after controlling for the nondiscretionary compo-
nents (the coefficients are 0.23, 0.27, and 0.46 respectively for operating cash
flows, nondiscretionary income, and net income, and all are significant at the
0.01 level). The coefficient on DAC is significantly higher than that on N D A C
when the dependent variable is one-year-ahead OCF, but is significantly lower
when the dependent variables are one-year-ahead N I and N D N I . 14 Finally,
column 7 reports the incremental increase in explanatory power of these regres-
sions through the inclusion of discretionary accruals. It can be seen that the
inclusion of discretionary accruals significantly increases the explanatory power
in every case (the Vuong statistic is significant at the 0.01 level in each model),
suggesting that discretionary accruals significantly improve the ability of in-
come to explain future profitability.
Panels B and C report results pertaining to two- and three-year-ahead
profitability levels. The results are qualitatively similar, albeit somewhat weaker,
than that pertaining to one-year-ahead profitability. Again, discretionary ac-
cruals have incremental explanatory power above the nondiscretionary compo-
nents (the coefficient on D A C is positive and significant and there is a significant
increase in the explanatory power with the inclusion of DAC in every single
case).
In summary, the results in Table 7 indicate that discretionary accruals predict
levels of future profitability after controlling for current levels of operating cash
flow and nondiscretionary accruals. The results are consistent with managers
using discretionary accruals to communicate information regarding future
profitability.

4.2.2. Discretionary accruals and current and future dividend changes

The signaling role of dividends is a well-researched issue. Dividends are
argued to signal management's private information regarding the economic
value of the firm (Bhattacharya, 1979; Miller and Rock, 1985; John and
Williams, 1985). Empirical evidence is consistent with this hypothesis: a number

~4Note that the coefficienton discretionaryaccurals is more than twice that of non discretionary
accurals when the dependent variable is future operating cash flow.This result is difficultto explain
under the measurementerror explanation (i.e., the positivecoefficienton the discretionaryaccurals
proxy simply reflectscontamination from the nondiscretionarycomponent rather than the value
relevance of discretionaryaccurals).
274 K.R. Subramanyam / Journal o f Accounting and Economics 22 (1996) 2 4 9 - 2 8 1

eq ~ c5 ~ o6 ~

.=.
~ ez

=o

L~
c-

e.

e3
L tt~ ~

k) ~ ~ ~ eq cS rq :~ ~ e.i c5 ~ c5 ~

z
0

• c5 ~ o ~i c5 ~6

e.
0

+
r~

,.o
¢.
 KR. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281 275

~
.~' ~ . ~ = ~ E . ~

r~ o

~';nn .--

~ m

2 ~N
~.~

~,' oo ¢'q oo ~ ¢'.10

.~,--~ ,-~ r~

e'. ~ ~ t,...

~ E.--. ," o

q~
g

~" ,x:z
q

G a2 z ?
276 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281

of event studies d o c u m e n t stock price effects of dividend a n n o u n c e m e n t s (for

example, Asquith and Mullins, 1983; Woolridge, 1983). Therefore, dividend
changes is used to test the hypothesis that managers convey value-relevant
information t h r o u g h their discretionary accounting choices. In particular, a pos-
itive association between c o n t e m p o r a n e o u s (and future) dividend changes and
discretionary accruals would be consistent with discretionary accruals convey-
ing information a b o u t the e c o n o m i c value of the firm.
Table 8 reports results of logit regressions between discretionary and nondis-
cretionary c o m p o n e n t s of income and d i c h o t o m o u s transformations of contem-
poraneous (and one-year-ahead) dividend changes.15 Panel A (models 1 and 2)
presents regressions with the change in current dividends as the dependent
variable, while panel B (models 3 and 4) presents regressions with the change in
one-year-ahead dividends as the dependent variable. The coefficients on D A C in
model 1 and model 2 are 3.52 and 4.15 respectively, and both are significant at
the 0.01 level. The Pearson Chi-squared (which measures the goodness of fit) is
also significant for both models. The coefficients on D A C in model 3 and model
4 are 4.55 and 4.73 respectively, and both are significant at the 0.01 level. The
Pearson Chi-squared for both models is also significant at the 0.01 level.
The results in Table 8 d o c u m e n t a positive association between discretionary
accruals and current (and one-year-ahead) dividend changes. This evidence is
consistent with both earnings and dividends signaling information a b o u t future
profitability, i.e., managers use both accounting and dividend choices as
multiple signals a b o u t firm value. 16 This evidence also introduces a different
perspective to an a n o m a l o u s empirical regularity. F a m a and Baibak (1968) and
Watts (1973) report that dividend changes often respond to earnings with a lag,
rather than forecast future earnings; a result that appears to contradict the
dividend signaling hypothesis. The evidence in this section suggests that this
a n o m a l y can at least partly be explained by realizing managers signal private
information regarding future profitability t h r o u g h both dividends and earnings.

15Because of its irregular distribution, dividend changes are transformed in a dichotomous manner.
Two different dichotomous variables are defined: DIV1 (models 1 and 3 in Table 8) takes a value of
1 for an increase in dividend and 0 otherwise, while DIV2 (models 2 and 4 in Table 8) takes the value
of 1 for an increase in dividend and 0 for a decrease in dividend. The key difference between the two
variables is that, while DIV1 includes dividend maintenance along with dividend reduction, DIV2
excludes dividend maintenance from the sample.
16An important issue that arises at this juncture is why discretionary accruals do not preempt
dividends as a signal, i.e., why managers use dividends as signals after they have signaled through
earnings and why does the stock market respond to dividends that succeed earnings announce-
ments. There are two possible explanations. First, while dividend changes and discretionary accruals
are correlated, it is possible that each has incremental information content above the other. Second,
it is possible that dividends are more credible signals because of higher dissipative costs.
 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281 277

.E
E

5
•~ . . . '~ .~ ~

0
o ~.~ "- ~ = ~

©
e~

~, ~ ~ 0~ "~

=, ~.~ . ~ "~ ~.~

~~ ~.~ ~ ~ ~
:E

?
e.
©

?.
2~
.E
g .o~= o~ ~ ~ ~
.~ ~ ~ ~_ ~ ~ . ~ o = ~ E

P o ~
,2 .2 ~ ~'~.-~ ~ ,- ~ ~ .~:
.~e'~ .~ .~ ~ . ~.~
b- .a
278 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281

5. Summary and conclusion

This paper shows that discretionary accruals (measured using a variation of

the Jones model) is priced by the stock market. While this result is consistent
with the pricing of opportunistic earnings manipulation by an inefficient market,
further tests suggest that the pricing of discretionary accruals arises because
mangers use their discretion to improve the ability of earnings to reflect
fundamental value. There is evidence of pervasive income smoothing that
improves the persistence and predictability of earnings. There is also evidence
that discretionary accruals communicate information about future firm profit-
ability. While the results are robust to various sensitivity checks, measurement
error in the discretionary accruals proxy is a plausible alternative explanation
for the results.
The results in this paper have implications for both accounting research and
practice. First, the results are inconsistent with pervasive accrual manipulation
that distorts reported earnings (while there is evidence of income smoothing, the
smoothing appears to improve rather than diminish the value relevance of
reported earnings). While opportunistic accrual manipulation undoubtedly oc-
curs in specific instances, it is less certain that it occurs on a v e r a g e in the
population. This is not surprising given the existence of labor markets and
reputation effects (Fama, 1980). The results in this paper reiterate the superiority
of accrual-based earnings over cash flow measures. Accrual accounting is
superior to cash flows not only because it reduces timing and mismatching
problems (Dechow, 1994) but also because the flexibility accorded by GAAP
allows managers to reflect value-relevant information not captured in the
nondiscretionary component.
Second, the evidence that the discretionary component of earnings is priced
by the market suggests another motive for earnings management. The literature
hitherto has documented contracting motives, such as bonus plans and debt
covenants (Watts and Zimmerman, 1990). Healy and Palepu (1993) suggest that
influencing the price of a security in an asymmetrically informed market could
be one motivation for accounting choice. The evidence in this paper is consistent
with their assertion.
Finally, the evidence in this paper contributes to the ongoing debate on
'uniformity versus flexibility' in accounting standard setting. For example, Dye
and Verrecchia (1995) remark that ' . . . whether expanding discretion in ac-
counting choice is desirable appears to depend on whether the prospects for
improved communication of the firm's financial condition are more than offset
by the effects of managerial opportunism'. The results in this paper are consis-
tent with accounting choice improving the quality of the communication
through earnings reports. Therefore, in addition to the benefits of efficient
contracting (Watts and Zimmerman, 1990), improved communication appears
to be another reason to allow for flexibility in accounting standards.
 K.R. Subramanyam / Journal of Accounting and Economics' 22 (1996) 249-281 279

References

Asquith, P. and D. Mullins, 1983, The impact of initiating dividend payments on shareholders
wealth, Journal of Business 56, 77-96.
Ball, R., 1972, Changes in accounting techniques and stock prices, Journal of Accounting Research
10 (Supplement), 1-38.
Beaver, W. and E.E. Engel, 1996, Discretionary behavior with respect to allowance for loan losses
and the behavior of security prices, Journal of Accounting and Economics, forthcoming.
Beaver, W., Lambert, and D. Morse, 1980, The information content of security prices, Journal of
Accounting and Economics 2, 3-28.
Bhattacharya, S., 1979, Imperfect information, dividend policy and the bird in the hand fallacy, Bell
Journal of Economics 10, 259 270.
Bowen, R.M., D. Burghstahler, and L.A. Daley, 1987, The incremental information content of
accrual versus cash flows, The Accounting Review 62, 723-747.
Chancy, P.K., D.C. Jeter, and C.M. Lewis, 1996, The use of accruals in income smoothing:
Permanent earnings hypothesis, Working paper (Vanderbilt University, Nashville, TN).
Collins, D. and S.P. Kothari, 1989, An analysis of intertemporal and cross-sectional determinants of
the earnings response coefficients, Journal of Accounting and Economics 11, 143 181.
Collins, D., S.P. Kothari, J. Shanken, and R. Sloan, 1994, Lack of timeliness and noise as
explanations for the low contemporaneous returns-earnings relation, Journal of Accounting and
Economics 18, 289-324.
DeAngelo, H., L. DeAngelo, and D.J. Skinner, 1993, Accounting choice in troubled companies,
Journal of Accounting and Economics 17, 113-144.
Dechow, P., 1994, Accounting earnings and cash flows as measures of firm performance: The role of
accounting accruals, Journal of Accounting and Economics 17, 3-42.
Dechow, P., S.P. Kothari, and R.L Watts, 1995, The correlation structure of earnings, cash flows
and accruals, Working paper (University of Rochester, Rochester, NY).
Dechow, P., R. Sloan, and A. Sweeney, 1995, Detecting earnings management, The Accounting
Review 70, 193 226.
DeFond, M. and J. Jimbalvo, 1994, Debt covenant violation and manipulation of accruals:
Accounting choice in troubled companies, Journal of Accounting and Economics 17, 145 176.
DeFond, M. and C.W. Park, 1996, Smoothing income in anticipation of future earnings, Working
paper (The Hong Kong University of Science and Technology, Hong Kong).
Dye, R.A. and R.E. Verrecchia, 1995, Discretion vs uniformity: Choices among GAAP, The
Accounting Review 70, 389-416.
Easton, P.D. and T.S. Harris, 1991, Earnings as an explanatory variable for returns, Journal of
Accounting Research 29, 19-36.
Fama, E., 1980, Agency problems and the theory of the firm, Journal of Political Economy 88,
288 307.
Fama, E. and H. Baibalk, 1968, Dividend policy: An empirical analysis, American Statistical
Association Journal 63.
Gaver, J., K. Gaver, and J. Austin, 1995, Additional evidence on bonus plans and income manage-
ment, Journal of Accounting and Economics 19, 3-28.
Graham, B.~ D. Dodd, and S. Cottle, 1962, Security analysis: Principles and technique (Irwin,
Homewood, IL).
Guay, W. P., S.P. Kothari, and R.L. Watts, 1996, A market-based evaluation of discretionary
accruals models, Working paper (University of Rochester, Rochester, NY).
Hand, J., 1989, Did firms undertake debt~equity swaps for an accounting paper profit or true
financial gain?, The Accounting Review 64, 587-623.
Hansen, G.A., 1995, Do discretionary accruals proxies measure earnings management?, Working
paper (University of Rochester, Rochester, NY).
280 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249 281

Healy, P., 1985, The impact of bonus schemes on the selection of accounting principles, Journal of
Accounting and Economics 7, 85-107.
Healy, P.M. and K.G. Palepu, 1993, The effect of firms' financial disclosure policies on stock prices,
Accounting Horizons 7, 1-11.
Holthausen, R.W., 1990, Accounting method choice: opportunistic behavior, efficient contracting
and information perspectives, Journal of Accounting and Economics 12, 207 218.
Holthausen, R.W., D. Larcker, and R. Sloan, 1995, Annual bonus schemes and the manipulation of
earnings, Journal of Accounting and Economics 19, 29-74.
Hunt, A., S.E. Moyer, and T. Shevlin, 1995, Earnings smoothing and equity value, Working paper
(University of Washington, Seattle, WA).
John, K. and J. Williams, 1985, Dividend, dilution and taxes: A signaling equilibrium, Journal of
Finance 40, 1053-1070.
Jones, J., 1991, Earnings management during import relief investigations, Journal of Accounting
Research, Autumn, 193-228.
Kaplan, R.S. and R. Roll, 1972, Investor evaluation of accounting information: Some empirical
evidence, Journal of Business 45, 225-257.
Kothari, S.P., 1992, Price earnings regressions in the presence of prices leading earnings: Implica-
tions for earnings response coefficients, Journal of Accounting and Economics 15, 173 202.
Miller, M.H. and K.Rock, 1985, Dividend policy under asymmetric information, Journal of Finance
40, 1031-1051.
Moses, O.D., 1987, Income smoothing and incentives: Empirical tests using accounting changes, The
Accounting Review 62, 358-377.
Ohlson, J. and P. Shroff, 1992, Changes versus levels in earnings as explanatory variables for returns:
Some theoretical considerations, Journal of Accounting Research 30, 210-226.
Paek, W.D. and E. Press, 1996, Discretionary accruals, manager's incentives and stock market
responses, Working paper (Temple University, Philadelphia, PA).
Perry, S. and T. Williams, 1994, Earnings management preceding management buyout offers,
Journal of Accounting and Economics 18, 157-180.
Pourciau, S., 1993, Earnings management and non-routine executive changes, Journal of Ac-
counting and Economics 16, 317-348.
Revsine, L., 1991, The selective financial misrepresentation hypothesis, Accounting Horizons 5,
16-27.
Ronen, J. and S. Sadan, 1981, Smoothing income numbers: Objectives and implications (Addison
Wellesley, Reading, MA).
Schipper, K., 1989, Commentary on earnings management, Accounting Horizons 3, 91-102.
Sloan, R.G., 1995, Do stock prices impound information in accruals about future earnings?,
Working paper (University of Pennsylvania, Philadelphia, PA).
Sunder, S., 1973, Relationship between accounting changes and stock prices: Problems of
measurement and some empirical evidence, Journal of Accounting Research 11 (Supplement),
1-45.
Trueman, B. and S. Titman, 1988, An explanation for accounting income smoothing, Journal of
Accounting Research 26 (Supplement), 127-144.
Vuong, Q.H., 1989, Likelihood ratio tests for model selection and non-nested hypotheses, Econo-
metrica 57, 307-333.
Walhen, J.M., 1994, The nature of information in commercial bank loan loss disclosures, The
Accounting Review 69, 455-478.
Warfield, T. and J.J. Wild, 1992, Accounting recognition and the relevance of earnings as an
explanatory variable for returns, The Accounting Review 67, 821-842.
Warfield, T., J.J. Wild, and K.J. Wild, 1995, Managerial ownership and the informativeness of
earnings, Journal of Accounting and Economics 20, 61-92.
Watts, R., 1973, The information content of dividends, Journal of Business 46, 191-211.
 K.R. Subramanyam / Journal of Accounting and Economics 22 (1996) 249-281 281

Watts, R. and J. Zimmerman, 1986, Positive accounting theory (Prentice-Hall, Englewood Cliffs,
N J).
Watts, R. and J. Zimmerman, 1990, Positive accounting theory: A ten year perspective, The
Accounting Review 65, 131-156.
White, H., 1980, A heteroskedasticity-consistent covariance matrix estimator and a direct test for
heteroskedasticity, Econometrica 48, 817-838.
Woolridge, J.R., 1983, Dividend changes and security prices, Journal of Finance 38, 1607 1615.