J. Account.

Public Policy 29 (2010) 8295

Contents lists available at ScienceDirect

J. Account. Public Policy

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

Strategic revenue recognition to achieve earnings benchmarks

Marcus L. Caylor *
Moore School of Business, University of South Carolina, Columbia, SC 29208, United States

a r t i c l e

i n f o

Keywords:
Revenue recognition
Earnings surprises
Earnings management
Accounts receivable
Deferred revenue
SarbanesOxley Act

a b s t r a c t
I examine whether managers use discretion in revenue recognition
to avoid three earnings benchmarks. I nd that managers use discretion in both accrued revenue (i.e., accounts receivable) and
deferred revenue (i.e., advances from customers) to avoid negative
earnings surprises, but nd little evidence that discretion is used to
avoid losses or earnings decreases. For a common sample of rms
with both deferred revenue and accounts receivable, I nd evidence that managers do not prefer to exercise discretion in either
account. However, further tests show that managers preferred to
use discretion in deferred revenue before the SarbanesOxley Act
of 2002 went into effect, consistent with them choosing to manage
an account with the lowest real costs to the rm (i.e., future cash
consequences). My results suggest that the revenue recognition
joint project undertaken by the FASB and IASB to reduce managerial estimation in revenue recognition may have the unintended
consequence of leading to greater real costs imposed on shareholders as rms are likely to use even greater discretion in accounts
receivable.
2009 Elsevier Inc. All rights reserved.

1. Introduction
Revenue recognition is a timely issue as evidenced by the revenue recognition joint project currently undertaken by The Financial Accounting Standards Board (FASB) and the International Accounting Standards Board (IASB).1 Depending on the nature of a rms business, there is an accrual and a
deferral that relates to the amount of revenue recognized in an accounting period: accounts receivable

* Tel.: +1 803 777 6081; fax: +1 803 777 0712.

E-mail address: marcus.caylor@moore.sc.edu
1
See http://fasb.org/project/revenue_recognition.shtml for more details.
0278-4254/$ - see front matter 2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.jaccpubpol.2009.10.008

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

83

and deferred revenue, respectively.2 I examine both accounts to see if discretion in revenue recognition is
used to achieve three earnings benchmarks common in the literature: avoidance of losses, avoidance of
earnings decreases, and avoidance of negative earnings surprises. Accrued revenue (i.e., accounts receivable) and deferred revenue (i.e., advances from customers) present a unique opportunity to examine how
managers choose between two different types of earnings management. I test whether rms with both
accounts available to them exhibit a preference for discretion in the accrual or deferral account.
I provide an intuitive explanation for why discretion in the deferral (i.e., deferred revenue) would
be preferred. First, gross accounts receivable is managed primarily through real business activities,
such as easing credit policies. Management of deferred revenue, on the other hand, represents a situation where cash has already been received. Thus, management of deferred revenue relates to manipulation of accounting estimates. Managing gross accounts receivable is more costly to rms as it
relates to accelerating sales where cash has not yet been collected. Thus, it has future cash consequences whereas deferred revenue does not.
I construct a model for the normal change in short-term deferred revenue to determine abnormal
changes in short-term deferred revenue.3 I derive a similar model for the normal change in gross accounts receivable to determine abnormal changes in gross accounts receivable. I use gross accounts
receivable in lieu of net accounts receivable because abnormal changes in net accounts receivable could
reect changes in the allowance for bad debt and I am solely interested in discretion in revenue recognition (i.e., the revenue account).4 I estimate pre-managed earnings by removing the discretionary component related to the account in question (Dhaliwal et al., 2004; Frank and Rego, 2006), and then test
whether abnormal changes in each of these revenue accounts are greater than expected for rms with
pre-managed earnings that just miss an earnings benchmark. Next, I examine rms with both accounts
to see if managers prefer one account as a means for discretion in revenue recognition.
My results indicate that both deferred revenue and accounts receivable are managed in an attempt
to avoid negative earnings surprises. I nd little evidence that either are managed to avoid losses or
earnings decreases. In addition, I provide evidence that rms preferred to exercise discretion in deferred revenue relative to accounts receivable to avoid negative earnings surprises but that the SarbanesOxley Act of 2002 mitigated this preference.
My study makes several contributions to the literature. First, I provide the rst descriptive evidence
on deferred revenue, showing that many high technology industries have it on their balance sheets.
Second, I provide the rst comprehensive analysis of revenue manipulation in relation to all three
earnings benchmarks and show that discretion is used to avoid negative earnings surprises but nd
little evidence for the other two benchmarks. My results provide empirical support for the comprehensive revenue recognition project undertaken by the FASB and the IASB. My results suggest that
the current earnings process model for revenue recognition is subject to managerial discretion. Third,
my study is the rst to examine a common sample of rms with revenue deferrals and accruals used to
increase revenue and see whether managers exhibit a preference regarding management of these accounts. I provide evidence that deferred revenue is the preferred earnings management account prior
to the SarbanesOxley Act of 2002, indicating managers prefer to minimize real costs to the rm. My
results suggest that no such preference exists following the SarbanesOxley Act of 2002. My nding of
no preference following the SarbanesOxley Act of 2002 suggests that if the FASB and IASBs proposed
asset and liability model for revenue recognition is successful in reducing discretion in deferred revenue, it could lead to signicantly more discretionary revenue recognition through gross accounts

2
Deferred revenue goes by several other names including advances from customers, unearned revenue and revenue received in
advance. Surprisingly, little research has examined the deferred revenue account. The only study that directly examines this
account is Bauman (2005). Using a sample of 20 rms from the publishing industry, he provides evidence that stock prices treat
unearned revenue as an economic asset and that this is due to its underlying liability characteristics.
3
I use the short-term deferred revenue component and ignore the long-term component because the long-term component of
deferred revenue does not reect revenue that should have been recognized during the current period.
4
The manipulation of the allowance for doubtful accounts involves the manipulation of accounting estimates related to
receivables that will not be collected and should not be related to discretion in revenue recognition through gross accounts
receivable. I include the change in the allowance for doubtful accounts in a sensitivity analyses and nd that the results are
qualitatively unchanged.

84

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

receivable. This would impose more real costs on the rm. Fourth, I derive a discretionary model of
deferred revenue.
The remainder of my paper is organized as follows. Section 2 reviews the relevant literature and
develops hypotheses. Section 3 introduces the research design. Section 4 provides results of my study,
and Section 5 contains implications of my study.
2. Hypotheses development
Burgstahler and Dichev (1997) and Degeorge et al. (1999) introduce three earnings benchmarks.
Prior earnings management studies have provided evidence related to earnings benchmarks and specic accruals and reserves related to expense recognition (e.g., Frank and Rego, 2006).5 Some of these
studies have examined reserves specically related to receivables. McNichols and Wilson (1988) provide
evidence that the allowance for doubtful accounts is manipulated for rms with unusually high or low
earnings. Teoh et al. (1998) show that the allowance for doubtful accounts is manipulated around initial
public offerings. Jackson and Liu (2008) nd the allowance for uncollectible accounts has become increasingly conservative over time, and that rms use over-accruals of bad debt expense that have accumulated
on the balance sheet to manage earnings. Beatty et al. (2002) nd evidence that loan loss reserves associated with loan receivables are manipulated to avoid small earnings decreases for publicly traded banks.6
Two studies have considered revenue recognition and do not nd evidence that discretion is used
in gross accounts receivable to avoid losses and earnings decreases (Marquardt and Wiedman, 2004;
Roychowdhury, 2006).7 Such a nding is surprising given that SEC restatements most often result from
improperly recognized revenues (Dechow et al., 1996).8 Recent evidence suggests that the analyst benchmark is the most important benchmark sought by managers (Dechow et al., 2003; Brown and Caylor,
2005) indicating that rms may be more likely to manage earnings to achieve this benchmark than
the others. Firms can recognize more revenue to meet or just beat earnings benchmarks via discretion
in accounts receivable, deferred revenue or a combination of both.
I hypothesize that managers do use discretion in revenue recognition to avoid negative earnings
surprises. More formally, my rst two hypotheses are:
H1. Firms with pre-managed earnings that just miss the analyst benchmark experience an abnormal
increase in gross accounts receivable.

H2. Firms with pre-managed earnings that just miss the analyst benchmark experience an abnormal
decrease in short-term deferred revenue.
I might not nd results for the analyst benchmark for three reasons. First, I examine a post-SAB 101
environment where revenue recognition regulations are designed to prevent aggressive recognition
(see SAB 101).9 Rountree (2006) nds that deferred revenue was more likely to be targeted by SAB
101, suggesting that deferred revenue might more likely be affected than accounts receivable. Second,
managers use other (non-revenue-based) accruals for avoiding negative surprises (Moehrle, 2002; Frank
and Rego, 2006) potentially mitigating the use of revenue-based accruals and deferrals. Third, managers
have another mechanism, expectations management, to achieve the analyst benchmark.
5
McNichols (2000) points out that a key advantage of looking at specic accruals is that researchers can rely on generally
accepted accounting principles to guide them in specifying the non-discretionary component.
6
Prior research has found that managers have incentives to manage earnings (e.g., Park and Park, 2004) and that directors have
incentives to allow it (e.g., Ronen et al., 2006).
7
Marquardt and Wiedman (2004) nd that managers do not exercise discretion in gross accounts receivable to avoid an
earnings decrease. Roychowdhury (2006) nds no signicant evidence that gross accounts receivable are managed to avoid a loss
(see footnote 25 of his study).
8
In addition, Stubben (2008) provides evidence suggesting that discretionary models of accounts receivable are better than
discretionary models of aggregate accruals at detecting earnings management.
9
In 1999, the Securities Exchange Commission (SEC) issued Staff Accounting Bulletin (SAB) No. 101, which states that revenue
can be recognized only when the following four criteria are met: (1) persuasive evidence of an arrangement exists, (2) delivery has
occurred or services have been rendered, (3) the sellers price to the buyer is xed or determinable, and (4) collectibility is
reasonably assured.

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

85

I do not expect to nd discretion related to the avoidance of losses or earnings decreases based on
the two aforementioned studies. Therefore, I do not state these as formal hypotheses. However, it is
unclear whether the results found by Marquardt and Wiedman (2004) and Roychowdhury (2006) extend to a more recent time period or whether they extend to deferred revenue given the relatively
lower real costs that discretion imposes on the rm.
Major differences exist between manipulation of accounts receivable and deferred revenue. First,
with deferred revenue, cash has already been received. Three of the four criteria from SAB 101 for revenue recognition have usually been met with the recording of deferred revenue (i.e., persuasive evidence of an arrangement exists, the sellers price to the buyer is xed or determinable, and
collectibility is reasonably assured). Thus, discretion in deferred revenue arises as to when delivery
has occurred or services are rendered.
Second, with deferred revenue, managers are less able to manipulate through real activities. For
gross accounts receivable, managers accelerate the recognition of revenue through real activities
manipulation, such as providing favorable credit terms, easing creditworthiness restrictions, and
speeding up the shipment of goods.10 In contrast to gross accounts receivable, managers accelerate recognition of deferred revenue by increasing estimates of services provided.11 This causes manipulation of
deferred revenue to be less costly vis--vis accounts receivable in terms of its future cash consequences.
For example, providing favorable credit terms to speed up the recognition of a receivable has future cash
consequences, as a higher percentage of customers may default in the future. In addition, managing gross
accounts receivable has long-term reputation effects with customers, even if it has no direct future cash
consequences. For instance, a large supplier may push unwanted merchandise on small retailers to speed
up recognition of a receivable (e.g., channel stufng).
I expect managers to prefer discretion in deferred revenue relative to accounts receivable as a result of the real costs imposed by accounts receivable. However, a preference may not exist if managers
are near-sighted regarding the extent of long-term costs. In addition, my sample period includes the
impact of the SarbanesOxley Act of 2002, hereafter SOX, which Cohen et al. (2008) suggests reduces
accrual-based earnings management and increases the incidence of real earnings management. My
third hypothesis is:
H3. Firms with deferred revenue will use less discretion in gross accounts receivable, relative to
deferred revenue, to avoid negative earnings surprises.
3. Sample selection and research design
3.1. Sample selection
I obtain annual earnings, short-term deferred revenue, accounts receivable, sales, total assets, cash
ow from operations and other nancial statement data from the 2007 Compustat Industrial File. I obtain annual analyst earnings forecast data and reported annual earnings for computing earnings surprises from the split-unadjusted I/B/E/S Detail File. To be consistent with prior literature on earnings
management (e.g., Burgstahler and Dichev, 1997), I exclude regulated and nancial rms (i.e., SIC
codes between 4400 and 5000 and SIC codes between 6000 and 6500). I also exclude any rms related
to public administration (i.e., SIC codes of 9000 or higher).
3.2. Modeling normal changes in gross accounts receivable
To derive a model for the expected amount of gross accounts receivable in time t, I make a couple of
simplifying assumptions. I assume that gross accounts receivable are related to current periods sales,
10
Gross accounts receivable can be managed through subjective estimates of how much revenue has been earned, but only apply
to a very few industries where long-term construction projects exist.
11
While managers can use real activities manipulation to determine when a credit sale is recorded, with deferred revenue the
use of real activities manipulation is less likely. For instance, it is unlikely that managers would withhold services to customers to
reduce the amount of revenue recognized. Deferred revenue could be manipulated through altering contract terms, however,
because a customer has to agree to the new terms, such an action is less likely to occur.

86

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

as accounts receivable are sales accrued in the current period.12 I further assume that gross accounts
receivable are related to next periods cash ow from operations, since the receivable amounts will be
collected in this next period. This implies that changes in gross accounts receivable should be positively
related to contemporaneous changes in sales and future changes in cash ow from operations. I include
both of these variables in my model to capture any non-discretionary component that is not captured by
the other. Based on these assumptions, I estimate abnormal changes in gross accounts receivable by running linear regressions by industry (2-digit SIC code) and scal year using all available rms with the
requisite data13:

DGross A=Rt =At1 a0 a1  1=At1 b1  DSt =At1 b2  DCFOt1 =At1 et

where DGross A/Rt is the change in gross accounts receivable during year t (change in Compustat Annual Data Item 2 plus Compustat Annual Data Item 67), DSt is the change in sales during year t (change
in Compustat Annual Data Item 12), DCFOt+1 is the change in cash ow from operations during year
t + 1 (change in Compustat Annual Data Item 308), and At1 is the beginning of the year total assets
(Compustat Annual Data Item 6).14
In addition to the scaled intercept term found in prior discretionary accrual studies, I also include a
constant term based on Kothari et al. (2005) who nd that it results in better-specied discretionary
models.15
I use the estimated coefcients from these industry-year regressions to compute a normal or predicted change in gross accounts receivable. Abnormal changes in gross accounts receivable is the difference between the actual change in gross accounts receivable and these predicted values. An
abnormal increase in gross accounts receivable occurs when the actual change exceeds the predicted
value. Abnormally low growth in gross accounts receivable occurs when the actual change is less than
the predicted value.
3.3. Modeling normal changes in deferred revenue
To derive a model for expected short-term deferred revenue, I make a similar set of assumptions
as those made for expected accounts receivable. I assume that short-term deferred revenue is related to next periods sales (deferred revenues are amounts deferred to next period) and the cash
ow from operations in the current period (the cash associated with the deferred revenue was received in the current period). This implies that changes in short-term deferred revenue should be
positively related to contemporaneous changes in cash ow from operations and future changes
in sales. Based on these assumptions, I estimate abnormal changes in deferred revenue by running
linear regressions by industry (2-digit SIC code) and scal year using all available rms with the requisite data16:

DDef Revt =At1 a0 a1  1=At1 b1  DSt1 =At1 b2  DCFOt =At1 et

where DDef Revt is the change in short-term deferred revenue (Change in Compustat Annual Data Item
356) during year t, DSt+1 is the change in sales (Change in Compustat Annual Data Item 12) during year
t + 1, DCFOt is the change in cash ow from operations (Change in Compustat Annual Data Item 308)
during year t, and At1 is the beginning of the year total assets (Compustat Annual Data Item 6).
12

This assumption is adopted from Dechow et al. (1998).

I estimate at the 2-digit level to be consistent with prior literature. I also winsorize all variables entering both of my
discretionary models at the extreme (1st and 99th) percentiles of their respective distributions to be consistent with prior
literature. In addition, I require at least eight industry-year observations to estimate the model.
14
I choose a cross-sectional industry discretionary model in lieu of a rm-specic model due to data restrictions (i.e., a small
time-series of data). Bartov et al. (2000) provide evidence that suggests a cross-sectional discretionary model is superior to a timeseries model.
15
Kothari et al. (2005) include both a scaled and unscaled intercept term in their discretionary accrual models.
16
I require at least eight industry-year observations to estimate this model similar to the constraint for the gross accounts
receivable model.
13

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

87

Similar to my estimation of abnormal changes in accounts receivable, I use the estimated coefcients from these industry-year regressions to compute a normal or predicted change in deferred revenue.17 I use the short-term component and ignore the long-term component because the latter does not
reect revenue that could be recognized during the current period.18
3.4. Models to test hypotheses
Pre-managed earnings are obtained by removing the discretionary component of earnings
(Dhaliwal et al., 2004; Frank and Rego, 2006; among others). Pre-managed earnings are used in lieu
of post-managed earnings because this better reects ex-ante behavior. To compute pre-managed
earnings for gross accounts receivable, I subtract the abnormal change in gross accounts receivable
from reported earnings (Compustat Annual Data Item 18). To compute pre-managed earnings for deferred revenue, I add the abnormal change in deferred revenue to reported earnings (Compustat Annual Data Item 18). I dene a pre-managed earnings change in year t as pre-managed annual earnings
(Compustat Data Item 18) in year t minus annual earnings in year t  1. For tests involving I/B/E/S
data, I convert the abnormal change of the revenue account to an undeated amount by multiplying
by lagged total assets and then scaling by common shares outstanding used to calculate EPS (Compustat Annual Data Item 54) in order to properly adjust I/B/E/S reported earnings per share. I dene a
pre-managed earnings surprise in year t as pre-managed I/B/E/S reported earnings in year t minus
the consensus analyst forecast of earnings in year t.19
To test my rst two hypotheses, I examine how abnormal changes in gross accounts receivable (deferred revenue) are related to instances where a rms pre-managed earnings just misses the analysts
forecast. I estimate the regression:

AbnormalDGross A=Rt or AbnormalDDef Revt

a0 a1  PRE-MANAGED JUSTMISSt a2  PRE-MANAGED MEETJUSTBEATt
b1  SIZEt1 b2  BMt1 et

where AbnormalDGross A/R (AbnormalDDef Rev) is the abnormal change in gross accounts receivable
(deferred revenue). PRE-MANAGED_JUSTMISS is dened as an indicator variable equal to 1 if a rm
reports a pre-managed negative earnings surprise in year t of no more than 0.2% of the end of the prior
scal years stock price. PRE-MANAGED_MEETJUSTBEAT is dened as an indicator variable equal to 1 if
a rm reports a pre-managed non-negative earnings surprise in year t of less than 0.2% of the end of
the prior scal years stock price.20
I also estimate the regression for the loss and earnings decrease benchmarks to see if discretion exists. PRE-MANAGED_JUSTMISS is dened as an indicator variable equal to 1 if a rm reports a premanaged loss (earnings decrease) in year t of no more than 0.5% (0.25%) of the end of the prior scal
17
In an attempt to provide some evidence on my discretionary deferred revenue model relative to prior discretionary models, I
formed deciles based on the level of discretionary accruals for all rms in the Compustat universe for the same time period using a
modied-Jones model and examined the mean abnormal change in deferred revenue within these deciles. An interesting feature of
this comparison is that deferred revenue is likely to be only a small proportion of aggregate accruals on average and discretionary
deferred revenue will move in an opposite direction to aggregate discretionary accruals, thus any negative relation between the
two will provide comfort that my model is effectively picking up discretionary behavior. Untabulated analyses reveal that for every
decile the sign of mean abnormal changes in deferred revenue is opposite to the sign of mean discretionary accruals.
18
The abnormal change in short-term deferred revenue could partially be the result of long-term deferred revenue being
reclassied as short-term deferred revenue. I add the change in long-term deferred revenue as an additional control to all of my
tests. The results were qualitatively similar for all three earnings benchmarks (untabulated).
19
I calculate the consensus annual earnings forecast based on the median of the last individual earnings forecasts made by all
analysts in the 90-day period preceding the end of the scal year. This has the advantage over using I/B/E/S summary forecasts
because it avoids the stale forecast problem.
20
Any denition of small miss or small beat is arbitrary. My choice is based on prior research that has examined the avoidance of
negative earnings surprises. I use a 0.2% interval width for earnings surprises consistent with Burgstahler and Eames (2006). An
additional advantage of this choice is that it represents the best trade-off between the smallest interval width and the most
observations to make reliable statistical inferences. However, my results are qualitatively similar using other interval widths (e.g.,
0.3%).

88

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

years stock price.21 PRE-MANAGED_MEETJUSTBEAT is dened as an indicator variable equal to 1 if a

rm reports pre-managed non-negative earnings (earnings increase) in year t of less than 0.5% (0.25%)
of the end of the prior scal years stock price.
PRE-MANAGED_MEETJUSTBEAT is used as a reference group since these rms should have no exante motives to manage revenue upwards given they were able to achieve the benchmark before discretion in revenue is considered. I expect to nd a signicant and positive (negative) coefcient on
PRE-MANAGED_JUSTMISS for accounts receivable (deferred revenue). An F-test is conducted between
PRE-MANAGED_JUSTMISS and PRE-MANAGED_MEETJUSTBEAT when both are signicantly different
from zero. To control for systematic differences in abnormal changes in gross accounts receivable (deferred revenue), I include SIZE, the natural logarithm of a rms beginning of the year market value of
equity. To control for any effect growth opportunities have on abnormal changes in gross accounts
receivable (deferred revenue), I include the book-to-market (BM) ratio.22
The sample to test the rst hypothesis includes 15,193 (7284) rm-year observations for the loss
and earnings decrease (earnings surprise) benchmarks for scal years 20012005. The sample to test
the second hypothesis is 4846 (2664) rm-year observations for the loss and earnings decrease (earnings surprise) benchmarks for scal years 20012005. Fiscal years before 2001 are not used because
deferred revenue data coverage in Compustat is very sparse. Fiscal years after 2005 are not used because I require one-year ahead variables for the discretionary models.
To test my third hypothesis, I require rms to have both accounts receivable and deferred revenue.
This reduces my sample to 1807 rm-year observations. I re-estimate model 3 for this common sample. To the extent that one account affects the other, I include the abnormal change of the other revenue account as an additional control for this analysis. If my third hypothesis is correct, I should nd
signicance on PRE-MANAGED_JUSTMISS for deferred revenue but not for gross accounts receivable.

4. Results
4.1. Descriptive evidence
Table 1 provides descriptive information for the FamaFrench (FF) industry groups for scal years
20012005, ranked in ascending order by percentage of rms in the industry with non-zero short-term
deferred revenue (Fama and French, 1997). For scal year 2005, 30.7% of all rms reported non-zero
short-term deferred revenue.23 All 48 FF industry groups have some rms with short-term deferred revenue on their balance sheets. For ease of exposition, I only include the 25 FF industry groups with the highest proportion of rms with deferred revenue in Table 1. Table 1 indicates that the top industry groups in
terms of percent of rms with short-term deferred revenue include such industries as Printing and Publishing (56.96%), Computers (49.60%), Telecommunications (40.47%), Pharmaceutical Products (36.49%),
and Defense (29.85%). With the exceptions of Printing and Publishing and Personal Services, the top 10
industry groups are high technology sectors that relate to medical or computer/electronics technology.
Table 2 reports the mean coefcient estimates from estimating the models for the normal change in
gross accounts receivable and the normal change in deferred revenue.24 I compute t-statistics by
dividing the mean of the distribution across all industry-year observations for each of the variables
in the model by the standard error of this distribution. The model for gross accounts receivable has
an adjusted R2 of over 40%. Consistent with my predictions, there is a signicant and positive relationship between changes in gross accounts receivable and changes in current sales (coefcient = 0.1076;
t-statistic = 18.69) and future cash ow from operations (coefcient = 0.0508; t-statistic = 4.03). The
model for deferred revenue has an adjusted R2 of 30%. Consistent with my predictions, there is also
21

These intervals are consistent with Burgstahler and Dichev (1997).

I winsorize this variable at the top and bottom 1% of its distribution.
Firms with missing assets were excluded. I do not exclude utilities and nancial rms for purposes of providing descriptive
evidence in Table 1. A similar proportion is found for other years in my sample.
24
Pearson and Spearman correlations between changes in gross accounts receivable and changes in future cash ow and changes
in current sales, and between changes in short-term deferred revenue and changes in cash ow and changes in future sales are all
signicant at the 1% level.
22
23

89

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

Table 1
Short-term deferred revenue by FamaFrench industry group.
FF industry

Proportion of rms with

deferred revenue (%)

Mean deferred
revenue-to-assets (%)

Median deferred
revenue-to-assets (%)

Business Supplies
Chemicals
Trading
Recreational Products
Healthcare
Tobacco Products
Electrical Equipment
Real Estate
Machinery
Transportation
Coal
Retail
Insurance
Restaurants, Hotel, Motel
Entertainment
Defense
Medical Equipment
Electronic Equipment
Pharmaceutical Products
Personal Services
Telecommunications
Measuring and Control Equip
Business Services
Computers
Printing and Publishing

51
105
309
46
90
9
93
75
247
240
18
385
308
187
205
20
392
824
976
150
728
347
2480
804
180

13.60
14.02
14.19
14.24
14.85
15.52
17.58
17.90
22.03
22.37
23.08
23.25
24.68
26.71
28.87
29.85
29.88
34.38
36.49
38.17
40.47
43.70
44.63
49.60
56.96

0.98
2.30
2.99
8.05
3.89
0.05
2.76
1.55
5.77
2.96
2.18
2.21
1.79
2.52
5.49
1.59
6.15
4.27
5.57
19.66
4.35
37.19
13.37
9.05
6.54

0.87
0.38
0.39
2.47
1.35
<0.01
1.10
0.43
1.83
0.85
0.42
1.15
<0.01
1.60
1.64
0.53
2.20
1.80
1.60
8.15
1.27
1.55
6.21
4.40
2.68

This table reports the Fama and French (1997) industry name, total number of non-missing and non-zero observations for the
ratio of short-term deferred revenue-to-total assets (Compustat Annual Data Item 356 divided by Compustat Annual Data Item
6), percentage of rms in that industry with non-missing and non-zero short-term deferred revenue, as well as the mean and
median of the ratio of short-term deferred revenue-to-total assets. For ease of exposition, I only include the 25 industry groups
with the highest proportion of rms with deferred revenue. I multiply ratios by 100 to convert to percentages for expositional
purposes.

a signicant and positive relationship between changes in short-term deferred revenue and changes in
future sales (coefcient = 0.0221; t-statistic = 4.86) and current cash ow from operations (coefcient = 0.0288; t-statistic = 2.05).25
Table 3 provides descriptive statistics for all rms with either accounts receivable or deferred revenue for scal years 20012005. Gross accounts receivable has a mean of nearly 345 million dollars and
a mean change of approximately 1.2% of beginning assets. Deferred revenue has a mean of nearly 47
million dollars and a mean change in deferred revenue of approximately 1.1% of beginning assets. By
denition, the mean abnormal change in short-term deferred revenue and the mean abnormal change
in gross accounts receivable are zero since I include a constant term in my discretionary models. Table 3
also provides information pertaining to the control variables used to test my hypotheses, SIZE and BM.
4.2. Earnings benchmark results
Table 4 reports the results of OLS regressions examining my rst hypothesis related to abnormal
changes in gross accounts receivable.26 Consistent with my rst hypothesis that managers exercise
25
I removed one industry-year observation from computing the means for the discretionary deferred revenue model because it
appeared to be an outlier (i.e., it is the only observation to change the statistical signicance of one of the coefcients if it is
removed). If I do not remove this observation the t-statistic on changes in current cash ow from operations is 0.92. All results
reported in the paper are qualitatively similar if the 13 rms in this industry-year grouping are removed.
26
I use the NeweyWest standard error correction for autocorrelation and heteroskedasticity (Newey and West, 1987). I also
examined my results using the Rogers (1993) standard error correction that adjusts for clustering and obtain qualitatively similar
results.

90

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

Table 2
Model parameters for normal change models.

DGross A=Rt =At1 a0 a1  1=At1 b1  DSt =At1 b2  DCFOt1 =At1 et

DDef Revt =At1 a0 a1  1=At1 b1  DSt1 =At1 b2  DCFOt =At1 et
Independent variables

Expected sign

Dependent variable: DGross A/Rt/At1

Dependent variable: DDef Revt/At1

Intercept

1/At1

0.0026***
(3.20)
0.0453
(0.54)

DSt/At1

0.0009
(0.88)
0.1585*
(1.83)
0.1076***
(18.69)

DSt+1/At1

DCFOt/At1

DCFOt+1/At1

Adjusted R2

0.0221***
(4.86)
0.0288**
(2.05)
0.0508***
(4.03)
40.6%

29.9%

This table provides parameter estimates for the normal change models of gross accounts receivable and short-term deferred
revenue. I require at least eight non-missing observations within an industry-year for estimation. To be consistent with prior
literature on earnings management (e.g., Burgstahler and Dichev, 1997), I exclude regulated and nancial rms (i.e., SIC codes
between 4400 and 5000 and SIC codes between 6000 and 6500). I also exclude any rms related to public administration (i.e.,
SIC codes of 9000 or higher). I winsorize all variables that enter the models at the top and bottom 1% of their respective
distributions. The coefcient estimates are based on means of industry-years and t-statistics are based on the standard error of
those means. The coefcient estimates for the abnormal change in gross accounts receivable model is based on 49 industries
and 230 industry-years over 20012005, and the coefcient estimates for the abnormal change in deferred revenue is based on
24 industries and 85 industry-years over 20012005. I also report the associated mean of the adjusted R2s across these
industry-years. The dependent variables are DGross A/Rt, dened as the change in gross accounts receivable (change in Compustat Annual Data Item 2 plus Compustat Annual Data Item 67), and DDef Revt, dened as the change in short-term deferred
revenue (change in Compustat Annual Data Item 356). The independent variables include a constant term, an intercept scaled
by lagged total assets, 1/At1 (Compustat Annual Data Item 6), change in sales for year t, DSt (change in Compustat Annual Data
Item 12), change in sales in year t + 1, DSt+1, change in cash ow from operations during year t, DCFOt (change in Compustat
Annual Data Item 308), and change in cash ow from operations during year t + 1, DCFOt+1.
*
Denotes statistical signicance at the 10% two-tailed level.
**
Denotes statistical signicance at the 5% two-tailed level.
***
Denotes statistical signicance at the 1% two-tailed level.

discretion in accounts receivable to avoid negative earnings surprises, I nd that abnormal changes in
accounts receivable are more positive for PRE-MANAGED_JUSTMISS (coefcient = 0.0024; t-statistic = 2.97).27 There is also a signicant negative coefcient on PRE-MANAGED_MEETJUSTBEAT. One
explanation for this coefcient is that rms use gross accounts receivable to reduce earnings to smooth
future earnings if they have achieved the analyst benchmark. I examine the proportion of rms that had
pre-managed earnings meeting or just beating analyst forecasts that as a result of revenue-decreasing
discretion ultimately fell short of the benchmark. I nd that only 24% of these rms fell short of the
benchmark because of revenue-decreasing discretion suggesting that rms were primarily smoothing
earnings. I nd no evidence for the avoidance of losses conrming prior literature and showing that discretion is not used in a more recent time period.28 I nd marginal evidence that gross accounts receivable
is managed to avoid earnings decreases and further tests reveal that this nding only holds post-SOX.29
The evidence for deferred revenue is provided in Table 5. There is a negative and signicant coefcient on PRE-MANAGED_JUSTMISS for the analyst benchmark (0.0055; t-statistic = 4.94) for
27
All coefcient estimates are presented in decimal form. For instance, this coefcient translates into an abnormal increase in
gross accounts receivable that was 0.24% of beginning total assets.
28
Although I nd a signicant and positive coefcient on PRE-MANAGED_JUSTMISS for the loss avoidance benchmark, an F-test
reveals that it is not signicantly different from that of PRE-MANAGED_MEETJUSTBEAT (F-test = 0.15).
29
I interpret this as marginal evidence given a sample size of over 15,000 observations and a two-tailed signicance level of only
5%. Observations in scal years following 2002 are dened as post-SOX.

91

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

Table 3
Descriptive statistics.
Mean

Std. dev.

25%

75%

Panel A: rms with accounts receivable

Gross A/R (in $ mil)
DGross A/R
SIZE
BM

344.513
0.012
4.889
0.538

3592.539
0.090
2.501
1.557

3.755
0.017
3.162
0.227

102.336
0.032
6.606
0.879

Panel B: rms with deferred revenue

Deferred revenue (in $ mil)
DDef Rev
SIZE
BM

46.773
0.011
5.219
0.401

341.469
0.056
2.352
0.694

0.242
0.002
3.612
0.164

13.166
0.014
6.737
0.619

This table provides descriptive statistics. All variables are scaled by lagged total assets, except for the raw values of gross
accounts receivable and deferred revenue, book-to-market ratio, and log of size. DGross A/R is dened as the change in gross
accounts receivable (change in Compustat Annual Data Item 2 plus Compustat Annual Data Item 67). DDeferred revenue is
dened as the change in short-term deferred revenue (change in Compustat Annual Data Item 356). SIZE is the natural
logarithm of a rms size using beginning of the year market value of equity (Compustat Annual Data Item 25  Compustat
Annual Data Item 199). BM is the beginning of the year book-to-market ratio ((Compustat Annual Data Item 60 + Compustat
Annual Data Item 74)/(Compustat Annual Data Item 25  Compustat Annual Data Item 199)).

abnormal changes in deferred revenue. This supports my second hypothesis that managers exercise
discretion in deferred revenue to avoid negative earnings surprises as well.30 I do not nd signicance
for the other two benchmarks suggesting that managers do not use discretion in deferred revenue to
avoid these benchmarks.31
To provide some evidence on the prevalence of revenue recognition to avoid negative earnings surprises, I also conduct an analysis similar to that of Frank and Rego (2006). I examine the proportion of
rms that use discretion in revenue accounts to cross over the analyst forecast. I nd that 70.7%
(71.4%) of rms that had pre-managed earnings just missing analyst forecasts were able to use discretion in gross accounts receivable (short-term deferred revenue) to meet or beat the benchmark.
4.3. Do managers express a preference for revenue management?
Table 6 provides results regarding my third hypothesis. In panel A, I fail to nd evidence consistent
with my third hypothesis. I nd that that rms with deferred revenue use discretion in gross accounts
receivable (coefcient = 0.0036; t-statistic = 2.11). I continue to nd a signicant and negative coefcient on abnormal changes in deferred revenue (coefcient = 0.0046; t-statistic = 3.41). These results suggest that managers do not care about the real costs imposed on shareholders from their
choice of revenue management. However, my results are subject to the effects of a large exogenous
shock that occurred during my sample period. This shock was SOX in which reported earnings were
subject to added auditor scrutiny. To examine whether this legislation impacted managerial preference, I estimate a regression splitting the coefcients into pre- and post-SOX periods. Because SOX
was not signed into law until July 30, 2002, I dene scal years following 2002 as post-SOX years.
In panel B, I nd that SOX did not affect the coefcient on deferred revenue, but it did cause an increase in management of gross accounts receivable. More importantly, the results suggest that in a
30
To the extent that my proxy for discretionary revenue recognition contains measurement error, a correlation may be induced
between pre-managed earnings and the abnormal change in revenue account (Leone and Rock, 2002). However, it is unclear why
such a relation would exist for only the PRE-MANAGED_JUSTMISS interval in relation to the other pre-managed intervals.
Nonetheless, I perform two additional analyses. In the rst, I regress the abnormal change in revenue account on pre-managed
earnings and nd insignicant coefcients for both revenue measures. I also include PRE-MANAGED_EARNINGS, the magnitude of
pre-managed earnings, in the regressions reported in Tables 4 and 5 to control for this correlation if it exists. I obtain qualitatively
similar results.
31
Although I nd a signicant and negative coefcient on PRE-MANAGED_JUSTMISS for the loss and earnings decrease avoidance
benchmarks, F-tests reveal that neither is signicantly different from that of PRE-MANAGED_MEETJUSTBEAT (F-test = 0.16 and Ftest = 0.03, respectively).

92

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

Table 4
Abnormal changes in gross accounts receivable to avoid missing earnings benchmarks.

AbnormalDGrossA=Rt a0 a1  PRE-MANAGED JUSTMISSt a2  PRE-MANAGED MEETJUSTBEATt b  CONTROLS et

a1
a2
a 1 = a2

Avoid loss

Avoid earnings decrease

Avoid negative earnings surprise

0.0100***
(3.55)
0.0086***
(3.50)
0.15

0.0038**
(2.06)
0.0002
(0.10)
N/A

0.0024***
(2.97)
0.0038***
(4.73)
87.15***

This table provides regression results for my rst hypothesis. AbnormalDGross A/R is the abnormal change in gross accounts
receivable dened using the model in the text. The primary independent variable is PRE-MANAGED_JUSTMISS corresponding to
the range in which a rm just misses an earnings benchmark using a distribution based on pre-managed earnings. PREMANAGED_MEETJUSTBEAT is included as a reference group, in which I conduct an F-test between this coefcient and that of the
primary variable. The control variables are SIZE, dened as the natural logarithm of a rms size using beginning of the year
market value of equity and BM, dened as the beginning of the year book-to-market ratio. t-Statistics are reported in parentheses under the coefcient estimates based on the NeweyWest standard error correction for autocorrelation and heteroskedasticity (Newey and West, 1987). Coefcient estimates are reported in decimal form. For ease of exposition, I exclude
coefcient estimates for the intercept and control variables.
**
Denotes statistical signicance at the 5% two-tailed level (except for F-tests which are based on one-tailed signicance level).
***
Denotes statistical signicance at the 1% two-tailed level (except for F-tests which are based on one-tailed signicance
level).

Table 5
Abnormal changes in deferred revenue to avoid missing earnings benchmarks.

AbnormalDDef Revt a0 a1  PRE-MANAGED JUSTMISSt a2  PRE-MANAGED MEETJUSTBEATt b  CONTROLS et

a1
a2
a 1 = a2

Avoid loss

Avoid earnings decrease

Avoid negative earnings surprise

0.0095***
(2.60)
0.0113***
(4.45)
0.16

0.0054***
(3.16)
0.0058***
(3.19)
0.03

0.0055***
(4.94)
0.0010
(1.01)
N/A

This table provides regression results for my second hypothesis. AbnormalDDef Rev is the abnormal change in short-term
deferred revenue dened using the model developed in the text. The primary independent variable is PRE-MANAGED_JUSTMISS
corresponding to the range in which a rm just misses an earnings benchmark using a distribution based on pre-managed
earnings. PRE-MANAGED_MEETJUSTBEAT is included as a reference group, in which I conduct an F-test between this coefcient
and that of the primary variable. The control variables are SIZE, dened as the natural logarithm of a rms size using beginning
of the year market value of equity and BM, dened as the beginning of the year book-to-market ratio. t-Statistics are reported in
parentheses under the coefcient estimates based on the NeweyWest standard error correction for autocorrelation and
heteroskedasticity (Newey and West, 1987). Coefcient estimates are reported in decimal form. For ease of exposition, I exclude
coefcient estimates for the intercept and control variables.
***
Denotes statistical signicance at the 1% two-tailed level (except for F-tests which are based on one-tailed signicance
levels).

pre-SOX environment, managers did express a preference for deferred revenue, the revenue management tool that imposed the least real costs on the rm.32
5. Conclusions and Implications
I examine whether managers use accounting discretion in two accounts related to revenue recognition, short-term deferred revenue and gross accounts receivable, to avoid missing three common
32
An alternative explanation could be that these rms have less receivables than deferred revenue so these rms would not nd
discretion in accounts receivable to be as economically feasible. However, I nd that these rms actually have much higher mean
accounts receivable than deferred revenue by a factor of 11 indicating that such an alternative explanation is not plausible.

93

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

Table 6
Abnormal changes in gross accounts receivable (deferred revenue) to avoid negative earnings surprises using a common sample.

AbnormalDGrossA=Rt AbnormalDDef Revt a0 a1  PRE-MANAGED JUSTMISSt a2

 PRE-MANAGED MEETJUSTBEATt b  CONTROLS et
AbnormalDGross A/R

AbnormalDDef Rev

0.0036**
(2.11)
0.0043**
(2.39)

0.0046***
(3.41)
0.0007
(0.61)

6.54**

N/A

0.0018
(0.64)
0.0062*
(1.84)
0.0041**
(2.45)
0.0039**
(2.10)

0.0061***
(3.30)
0.0016
(0.99)
0.0044***
(3.06)
0.0006
(0.44)

N/A
32.84***

N/A
N/A

Panel A: full sample

a1
a2
F-test

a 1 = a2
Panel B: pre- and post-SarbanesOxley periods
a1a (pre-SOX)

a2a (pre-SOX)
a1b (post-SOX)
a2b (post-SOX)
F-test

a1a = a2a
a1b = a2b

This table provides regression results for my third hypothesis using a common sample of rms with both accounts receivable
and short-term deferred revenue. Panel A provides results for the full sample period. Panel B partitions the sample into pre- and
post-SOX periods. AbnormalDGross A/R (AbnormalDDef Rev) is the abnormal change in gross accounts receivable (short-term
deferred revenue) dened using the models developed in the text. The primary independent variable is PRE-MANAGED_JUSTMISS corresponding to the range in which a rm just misses analysts forecasts using a distribution based on premanaged earnings. PRE-MANAGED_MEETJUSTBEAT is included as a reference group, in which I conduct an F-test between this
coefcient and that of the primary variable. The control variables are SIZE, dened as the natural logarithm of a rms size using
beginning of the year market value of equity and BM, dened as the beginning of the year book-to-market ratio. I include the
abnormal change of the other account as an additional control. t-Statistics are reported in parentheses under the coefcient
estimates based on the NeweyWest standard error correction for autocorrelation and heteroskedasticity (Newey and West,
1987). Coefcient estimates are reported in decimal form. For ease of exposition, I exclude coefcient estimates for the intercept
and control variables.
*
Denotes statistical signicance at the 10% two-tailed level (except for F-tests which are based on one-tailed signicance level).
**
Denotes statistical signicance at the 5% two-tailed level (except for F-tests which are based on one-tailed signicance level).
***
Denotes statistical signicance at the 1% two-tailed level (except for F-tests which are based on one-tailed signicance
level).

earnings benchmarks. My evidence suggests that managers accelerate the recognition of revenue
using both accounts when pre-managed earnings miss the analyst benchmark by a small amount.
Using a common sample, I nd that managers preferred to exercise discretion in deferred revenue
as opposed to accounts receivable to avoid negative earnings surprises prior to the passage of SOX.
I offer an explanation why deferred revenue would be the preferred account. Managing deferred revenue has lower real costs relative to accounts receivable. However, if managers do not have a choice
they will choose a mechanism that does have negative future consequences in order to avoid negative
surprises in the current period. Finally, I introduce a discretionary model for deferred revenue that future researchers can use when studying deferrals.
My results have implications for policy makers and auditors. First, although not the focus of my
study, my results suggest that signicant discretion in revenue recognition remains post-SAB 101. This
nding provides empirical support for the comprehensive revenue recognition project currently in
progress at the FASB and the IASB.33 The project proposes the adoption of an asset and liability model
33

See http://fasb.org/board_handouts/11-20-07_RR_Educ_Session.pdf for more details.

94

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

for recognizing revenue. The asset and liability model would replace the current earnings process model
that is often subject to managerial estimation error. Under the current earnings process model, estimates
have to be made by managers as to when revenue has been recognized and changes to assets and liabilities are residuals of this revenue. In contrast, the asset and liability model recognizes revenue based on
changes in the underlying assets and liabilities and the recognized revenue is the residual. This model
takes a contract-based view, which considers when revenue should be recognized based on a customer
perspective, not based on a managers perspective. Standards setters should also examine my evidence
regarding whether a preference exists for discretion in revenue recognition. While a common allegation
is that managers are near-sighted at the expense of long-term value creation, my results suggest that
managers expressed a preference for the revenue recognition mechanism that had the least long-term
consequences before the passage of SOX.
Results in this study suggest that future policy should consider trade-offs between alternate revenue recognition mechanisms and consider that some managerial discretion can be more harmful in
the long-term than others. While the asset and liability model to revenue recognition may represent
the solution to reducing discretion in deferred revenue, it could also have the undesirable effect of
leading to signicantly more discretionary revenue recognition through gross accounts receivable.
My results suggest that rms had a preference for deferred revenue in the pre-SarbanesOxley period.
However, this preference no longer exists in a post-SarbanesOxley period. Thus, such a shift to an asset and liability model could ultimately result in more real costs being imposed on shareholders. Finally, my results provide evidence suggesting that auditors and nancial statement users should be
cognizant of discretion in deferred revenue.
Acknowledgements
This study is based in part on my doctoral dissertation at Georgia State University. I am grateful for
helpful comments and suggestions from my dissertation committee: Larry Brown (chair), Lynn Hannan, Jayant Kale, and Siva Nathan. This paper has also beneted from the helpful comments and suggestions of Ashiq Ali, Tony Chen, Bill Cready, Robert Freeman, Artur Hugon, Scott Jackson, Ross
Jennings, Steve Kachelmeier, Bill Kinney, Krishna Kumar, Yen Lee, Andrew Leone, Tom Lopez, Arianna
Pinello, Suresh Radhakrishnan, Galen Sevcik, Scott Vandervelde, Rich White and workshop participants at the 2006 American Accounting Association Annual Meetings, Georgia State University, the
University of South Carolina, the University of Texas at Austin and the University of Texas at Dallas.
I also appreciate the efforts of the journals editors and reviewers. I am grateful to Thomson Financial/I/B/E/S for providing data on analysts earnings forecasts. All data are available from public databases identied in the paper.
References
Bartov, E., Gul, F., Tsui, J., 2000. Discretionary accruals models and audit qualications. Journal of Accounting and Economics 30,
421452.
Bauman, M., 2005. A market-based examination of revenue and liability recognition: evidence from the publishing industry.
Review of Accounting & Finance, 5263.
Beatty, A., Ke, B., Petroni, K., 2002. Earnings management to avoid earnings declines across publicly and privately held banks.
The Accounting Review 77, 547570.
Brown, L., Caylor, M., 2005. A temporal analysis of quarterly earnings thresholds: propensities and valuation consequences. The
Accounting Review 80, 423440.
Burgstahler, D., Dichev, I., 1997. Earnings management to avoid earnings decreases and losses. Journal of Accounting and
Economics 24, 99126.
Burgstahler, D., Eames, M., 2006. Management of earnings and analysts forecasts to achieve zero and small positive earnings
surprises. Journal of Business Finance & Accounting 33, 633652.
Cohen, D., Dey, A., Lys, T., 2008. Real and accrual-based earnings management in the pre- and post-SarbanesOxley periods. The
Accounting Review 83, 757787.
Dechow, P., Kothari, S., Watts, R., 1998. The relation between earnings and cash ows. Journal of Accounting and Economics 25,
133168.
Dechow, P., Richardson, S., Tuna, I., 2003. Why are earnings kinky? An examination of the earnings management explanation.
Review of Accounting Studies 8, 355384.
Dechow, P., Sloan, R., Sweeney, A., 1996. Causes and consequences of earnings manipulation: an analysis of rms subject to
enforcement actions by the SEC. Contemporary Accounting Research 13, 136.
Degeorge, F., Patel, J., Zeckhauser, R., 1999. Earnings management to exceed thresholds. Journal of Business 72, 133.

M.L. Caylor / J. Account. Public Policy 29 (2010) 8295

95

Dhaliwal, D., Gleason, C., Mills, L., 2004. Last-chance earnings management: using the tax expense to meet analysts forecasts.
Contemporary Accounting Research 21, 431459.
Fama, E., French, K., 1997. Industry costs of equity. Journal of Financial Economics 43, 153193.
Frank, M., Rego, S., 2006. Do managers use the valuation allowance account to manage earnings around certain earnings
targets? Journal of the American Taxation Association 28, 4365.
Jackson, S., Liu, K., 2008. The allowance for uncollectible accounts, conservatism, and earnings management. Working paper,
University of South Carolina.
Kothari, S., Leone, A., Wasley, C., 2005. Performance matched discretionary accrual measures. Journal of Accounting and
Economics 39, 163197.
Leone, A., Rock, S., 2002. Empirical tests of budget ratcheting and its effect on managers discretionary accrual choices. Journal of
Accounting and Economics 33, 4367.
Marquardt, C., Wiedman, C., 2004. How are earnings managed? An examination of specic accruals. Contemporary Accounting
Research 21, 461491.
McNichols, M., 2000. Research design issues in earnings management studies. Journal of Accounting and Public Policy 19, 313
345.
McNichols, M., Wilson, P., 1988. Evidence of earnings management from the provision for bad debts. Journal of Accounting
Research 26, 131.
Moehrle, S., 2002. Do rms use restructuring charge reversals to meet earnings targets? The Accounting Review 77, 397413.
Newey, W., West, K., 1987. A simple, positive semi-denite, heteroskedasticity and autocorrelation consistent covariance
matrix. Econometrica 55, 703708.
Park, M., Park, T., 2004. Insider sales and earnings management. Journal of Accounting and Public Policy 23, 381411.
Rogers, W., 1993. Regression standard errors in clustered samples. Stata Technical Bulletin 13, 1923.
Ronen, J., Tzur, J., Yaari, V., 2006. The effect of directors equity incentives on earnings management. Journal of Accounting and
Public Policy 25, 359389.
Rountree, B., 2006. Mandatory accounting changes and rms nancial reporting environments. Working paper, Rice University.
Roychowdhury, S., 2006. Earnings management through real activities manipulation. Journal of Accounting and Economics 42,
335370.
Securities and Exchange Commission (SEC), 1999. Revenue Recognition in Financial Statements. Staff Accounting Bulletin No.
101, Government Printing Ofce, Washington, DC.
Stubben, S., 2008. Discretionary revenues as a measure of earnings management. Working paper, University of North Carolina.
Teoh, S., Wong, T., Rao, G., 1998. Are accruals during initial public offerings opportunistic? Review of Accounting Studies 3, 175
208.