Accounting Conservatism in Fraud Firms: An Empirical Investigation
Accounting
Conservatism in Fraud Firms: An Empirical Investigation
Keith Jones
Assistant
Professor
George Mason
University kjonesm@gmu.edu
Gopal Krishnan
Professor
Lehigh University gkrishn1@gmu.edu
Mikhail Pevzner**
Assistant
Professor
George Mason
University mpevzner@gmu.edu
Partha Sengupta
Associate
Professor
George Mason
University psengupt@gmu.edu
We
examine whether Ball and Shivakumar (2006) and Basu (1997) models of
conservatism identify fraud firms as anti-conservative. We show that both
models do so to some extent, but Ball and Shivakumar model results are
stronger. We further show that these results are driven by firms committing
largest frauds as a percentage of firms’ assets. Our results are important to
academics who use conservatism measures in their studies, and to policy makers
who seek to assess usefulness of conservatism to capital markets.
**Corresponding
Author, 4400 University Drive, MS5F4, Fairfax, VA 22030
Accounting
Conservatism in Fraud Firms: An Empirical Investigation*
We
examine whether Ball and Shivakumar (2006) and Basu (1997) models of
conservatism identify fraud firms as anti-conservative. We show that both
models do so to some extent, but Ball and Shivakumar model results are
stronger. We further show that these results are driven by firms committing
largest frauds as a percentage of firms’ assets. Our results are important to
academics who use conservatism measures in their studies, and to policy makers
who seek to assess usefulness of conservatism to capital markets.
Key Words: Conservatism, Basu Model, Ball and
Shivakumar Model, Fraud
Data
Availability: All data employed in this study are commercially available
from sources described in the text.
1. Introduction
In this study, we examine whether 1) two
popular conditional conservatism models, namely the Basu (1997) and the Ball
and Shivakumar (2006) models correctly identify a lack of conservatism among
firms known to have committed financial statement fraud, and 2) whether these
models indicate changes in fraud firms’ conservatism policies in the years
following the discovery of the fraud.
These research questions are important for
several reasons. First, the debate on whether accounting conservatism is important
is still very much ongoing (Watts (2003)). In light of FASB’s continued
orientation towards further expansion of the role of fair value, potentially at
the expense of reducing levels of firms’ accounting conservatism, it is
important to once again identify whether the stock market participants view
conservatism as an important aspect of financial statement quality. Our paper
addresses this concern by investigating whether fraud firms change their
conservatism policies, in the years following the fraud’s discovery. Second,
within academic literature, we have seen a lively debate on how to measure
accounting conservatism correctly. Here the debate centers primarily on whether
the Basu (1997) model (hereafter the Basu model) indeed measures firms’ conservatism.
Opponents of the Basu model argue that it is econometrically biased and is also
unstable in time-series estimations (e.g. Givoly et al., 2007, and Deitrich et al, 2007). Proponents of the Basu
model argue that the Basu model has a good economic motivation and exhibits
theoretically expected properties, such as a positive correlation with
market-to-book ratio over longer time horizons (e.g. Roychowdhury and Watts,
2007, and, Ball and Kothari, 2008). We contribute to this debate by examining
whether the Basu model and its recent counterpart, Ball and Shivakumar (2006)
model (hereafter the Ball and Shivakumar model), correctly identify a lack of
conservatism among fraud firms in years when the fraud was committed. Firms that commit fraud are by definition
taking actions to boost earnings and assets aggressively and therefore we would
expect these firms to be less conservative over the fraud period. A good model of conservatism should pick up
this change in conservatism. However, if
the models fail to pick up the change in conservatism over the fraud period,
one may question the usefulness of these conservatism models to researchers and
practitioners.
Our research is based on a sample of fraud
firms with at least one year of fraudulently misstated financial statements
over the period 1987-2007. In all, we obtain 352 fraud observations. We
hand-collect pre-restatement data for these firms to measure their fraud period
conservatism levels. We also obtain financial data for a period of five years
surrounding the fraud year, with two pre-fraud years and two post-fraud years,
in order to examine if and how conservatism changes for these firms over
time. Finally we also identify a matched
sample of non-fraud firms which includes all firm-year observations that match
the fraud firms on the basis of two-digit SIC codes and year in order to
examine how conservatism in these firms compare to the fraud firms. We examine two measures of conservatism - the
Basu measure and the Ball and Shivakumar measure and examine how these measures
compare across the two samples and also over time.
Our analysis yields a number of interesting
findings. With respect to Ball and
Shivakumar model, our results indicate that fraud firms are 1) less
conservative in the fraud period than in both pre-fraud and post-fraud period,
and 2) conservatism of fraud firms increases following the fraud’s
discovery. Furthermore, we show that
these results are driven by fraud firms which had the highest restatement level
of fraudulent net income numbers.
However, for the Basu model, our results are weaker. It fails to identify fraud firms as being
less conservative in the fraud period compared to the pre-fraud and post-fraud
periods. In comparison to the matched
sample we only find weak evidence of a slight increase in conservatism in the
Basu model in post-fraud period. Finally the Basu model picks up a decrease in conservatism in post-fraud
period, contradicting our results with respect to Ball and Shivakumar
model.
Our paper proceeds as follows. Section 2
describes our motivation and hypotheses. Section 3 describes our sample.
Section 4 discusses our results. Section 5 concludes.
2. Hypotheses
development
Conservatism is a widely-studied topic in
empirical accounting literature. This literature distinguishes between two
kinds of conservatism: unconditional and conditional (Beaver and Ryan, 2005).
Unconditional conservatism is an unconditional downward bias in accounting
earnings and book values. Expensing of R&D is an example of unconditional
conservatism. Because it reduces the value of firms’ net assets, unconditional
conservatism results in the higher market-to-book ratios. On the other hand,
conditional conservatism is characterized by the asymmetric timeliness in
recognition of economic gains and losses in accounting earnings. Asymmetric
timeliness implies that recognition of gains requires a greater degree of
verification than the recognition of losses. A good example of that is the
recognition of contingent losses versus the non-recognition of contingent
gains. Under US GAAP, contingent liabilities are recognized when probable and
can be reasonable estimated, but contingent gains are not recognized until
contingency is fully resolved, i.e. gains are no longer contingent.
Several measures of conservatism have been
widely used in the literature: Basu’s (1997) asymmetric timeliness measure,
Ball and Shivakumar’s (2006) accruals-cash flows measure, and Givoly and Hayn’s
(2000) non-operating accruals and relative skewness/variance of cash flows
measures. Our focus on this paper is on the Basu (1997) and the Ball and
Shivakumar (2006) measures because of their applicability to fraud firms.
Basu (1997) argues that the asymmetric
recognition of economic gains and losses in accounting earnings leads to an
asymmetric relation between stock returns and accounting losses and accounting
gains. Assuming market efficiency, stock
returns will incorporate both economic gains and losses, while accounting earnings
will incorporate economic losses relatively early compared to economic gains.
Hence, Basu shows that stock returns exhibit higher correlation with accounting
losses, than with accounting gains. Because of its intuitive appeal, the Basu
model has been incorporated in several studies of economic determinants of,
demand for, and economic consequences of the use of accounting conservatism
(Ball and Kothari, 2008).
Lately, the Basu model has been criticized for
being econometrically unstable. Thus, Dietrich et al. (2007) argue that the
asymmetric timeliness documented in Basu (1997) is an outcome of biased
t-statistics, and Givoly et al. (2007) document that the Basu measure
understates the degree of accounting conservatism due to temporal aggregation
of earnings and returns and due to differences in informational flow across
firms.[1]
Givoly et al. also show some results suggesting that asymmetric timeliness
documented by Basu (1997) could be driven by economic events unrelated to
accounting, such as being a target of acquisition or a lawsuit.2 Beaver et al. (2008) also show that in a
simultaneous equation setting, modeling Basu (1997) regression and Hayn (1995)
regression[2],
the Basu (1997) asymmetric timeliness coefficient becomes insignificant.
Ball and Kothari (2008) respond to the
criticisms of the Basu model by showing that the model is economically well
specified and is rooted in the research objective of identifying the
relationship between accounting earnings and stock returns. Roychowdhury and
Watts (2007) also show that another criticism of the Basu model, that
asymmetric timeliness is negatively correlated with market-to-book ratio,
disappears as the returns and earnings window is expanded. Hence, whether the
Basu (1997) measure of conservatism is a good one is still debated in the
literature.
Ball and Shivakumar (2006) suggest another
measure of conservatism. They show that when cash flows are negative, accruals
and cash flows have higher correlation because accruals capture expectations of
future economic losses. This model has become recently more widely used because
of its intuitive appeal. This model is particularly appealing to use in firms
where stock returns data are not readily available, such as privately-held
firms, to which the Ball and Shivakumar apply the model.[3] In contrast to the Basu (1997) model whose
validity has been widely examined, to our knowledge, the validity of the Ball
and Shivakumar model has not yet been extensively tested.[4]
Firms that commit
financial statement fraud represent a particularly good setting to analyze the
validity of the Basu and the Ball and Shivakumar measures. We focus solely on
frauds that are income increasing because firms committing these frauds are least
likely to be conservative. If the Basu and the Ball and Shivakumar models are
well-specified, then we should expect these models to show that fraud firms are
less conservative in years of the fraud. We choose to focus on these two
measures of conservatism rather than include certain other measures such as
those based on cumulative operating accruals or time series measures such as
relative skewness and variances in cash flows or earnings because fraud often
occurs over a relatively short period and the effects reverse subsequently so
that its impact would be “washed out” over longer horizons. Also most frauds affect revenue recognition
and asset capitalization (Beasley et al. 2009), and thus measures based on
non-operating accruals may not be able to pick up the effects of fraud.ing
accruals. Finally, the market-to-book ratio is another potential measure of
conservatism but since many fraud firms are also growth firms and this has been
used extensively in the literature as a important red flag to indicate fraud
(Loebbecke et al., 1989; and Beneish, 1999) the market-to-book ratio would not
be a good method to measure conservatism in a fraud setting.
Hence, our first hypothesis is as
follows:
H1a: Fraud firms are less
conservative than non-fraud firms according to the Ball and Shivakumar measure.
H1b: Fraud firms are less
conservative than non-fraud firms according to the Basu measure.
We also expect that upon discovering of the
fraud, fraud firms adopt more conservative accounting policies in the post-fraud
period. Demand for conservatism arises from increased public scrutiny and
increases in firms’ litigation risk and information asymmetry, and prior
research shows that firms experiencing such increases respond by becoming more
conservative (Khan and Watts, 2007; Ball and Shivakumar, 2006). Moreover, firms
that experience accounting frauds generally replace their managers, and it is
likely that for reputational reasons these managers become more conservative.
Moreover, fraud firms improve their corporate governance structures in
post-fraud periods (Farber, 2005). Higher levels of corporate governance are
associated with higher levels of accounting conservatism (Garcia Lara et
al., 2007), further suggesting that
conservatism levels should improve in the post-fraud years. This leads us to
the next two hypotheses:
H2A: Fraud firms are more conservative during the
post-fraud periods in comparison to the prefraud period.
H2B: Fraud firms are more conservative than
non-fraud firms during post-fraud periods.
We further expect that our findings in tests
of Hypotheses 1 and 2 are affected by the magnitude of the fraud committed by a
particular firm. If conservatism models we examine in our paper are effective
in picking up the lack of conservatism of firms committing frauds of highest
magnitudes should be deemed to be more anti-conservative in pre-fraud periods
than firms that commit small frauds. A magnitude of ex-post earnings
restatement is a measure of the level of fraud committed by a firm. Thus, we
predict:
H3: Fraud firms with highest earnings restatement levels are less
conservative than fraud firms with lower restatement levels.
In addition, if conservatism is a good way to
address agency problems between managers and investors (Khan and Watts (2007),
LaFond and Watts (2008)) and between managers and debtholders (Moerman (2008),
Watts (2003)), then firms committing greatest levels of fraud should experience
higher demand for ex-post conservatism than firms that commit smaller frauds.
Thus, we expect:
H4: Firms with highest earnings
restatement levels experience stronger increases in conservatism in post-fraud
periods, than firms with lower levels of earnings restatements.
3. Research
design
Our research design expands the
Ball and Shivakumar model and the Basu model to
incorporate the effects of
fraud. We are interested in two types of
comparisons: (i) a comparison of conservatism of fraud firms with that of
non-fraud firms, and (ii) a comparison of conservatism across periods
surrounding the fraud. We achieve this
by combining the fraud firms with all other firms that have the same two-digit
SIC codes as the fraud firms. This
combined sample is explored over three different (but overlapping)
sub-periods. The first subperiod
includes the five years prior to the fraud and the fraud period. The second sub-period includes the fraud
period and a post-fraud period of five years.
Finally, the third sub-period includes the pre-fraud period and the
post-fraud periods. Separate analysis of
conservatism is conducted for each of these three sub-periods. The research design for tests of the Ball and
Shivakumar model and the Basu model are described
separately below:
3.1. Tests using the Ball and Shivakumar
Model
The Ball and Shivakumar (2005) model’s
accrual model is of the following form:
TA=β0 +β1DCF+β2CF+β3DCF*CF+γ
(1)
where,
|
TA=
|
total accruals, defined as earnings before
extraordinary items (data123) minus
firm’s cash flows (data308) for firm years after 1988 and later; for
1987 and earlier it is calculated using the Sloan (1996) approach.
|
|
CF =
|
cash flows, defined as the difference
between earnings before extraordinary items (data123) and a firm’s
accruals.
|
|
DCF =
|
an indicator variable that equals 1 if CFt <0; 0
otherwise.
|
TA=α0 +α1DCF+α2CF+α3DCF*CF+α4FRAUDPRD+α5FRAUDFRM+α6CF*DCF*FRAUDPRD
+α7CF*DCF*FRAUDFRM+α8CF*DCF*FRAUDPRD*FRAUDFRM+γ (2)
where,
FRAUDPRD =
an indicator variable that equals 1 if the year represents a pre-fraud year, 0
if the
year is a fraud year and 2 if the year is after the
fraud year.
FRAUDFRM = an indicator variable that equals 1 if the
firm is a fraud firm; 0 otherwise.
We estimate equation (2) to examine potential
differences in conservatism between the fraud and non-fraud firms and
differences in conservatism over three periods: the pre-fraud period, the fraud
period and the post-fraud period. In
each test we compare two periods at a time.
Therefore we use three panels of observations: (i) fraud firms and
non-fraud firms in the pre-fraud and fraud periods, (ii) fraud firms and
non-fraud firms in the fraud and post-fraud periods, and (iii) fraud and
non-fraud firms in the pre-fraud and post-fraud periods. If fraud firms are less conservative than the
non-fraud firms over the fraud period α8 should be negative.
To test Hypothesis 2, we re-run equation (2)
including only the observations for Pre-fraud and post-fraud firm-years.
Hypothesis 2 predicts that α8>0
in this sub-sample.
Hypotheses 3 and 4 involve examining the effects of the magnitude of the
restatement.
In order to conduct these tests we modify equation
(2) to take the form:
TA=α0 +α1DCF+α2CF+α3DCF*CF+α4FRAUDPRD+α6RESTATEH+α6DCF*CF*FRAUDPRD
+α7CF*DCF*FRAUDPRD*RESTATEH+γ
(3) Where,
RESTATEH
= an indicator variable that equals 1 if
the amount of income restatement (i.e.,
difference between originally
reported earnings and restated earnings), deflated by total assets, is higher
than the median value; 0 otherwise.
For Hypothesis 3, we restrict our sample to
two subsamples: firm-years in pre-fraud and fraud periods, and firm years in
fraud and post-fraud periods. For Hypothesis 4, we focus on firm-years in
pre-fraud and post-fraud periods. For Hypothesis 3, our prediction is that α7<0. For
Hypothesis 4, our prediction is that α7>0.
For Hypothesis 3, we restrict our sample to
two subsamples: firm-years in pre-fraud and fraud periods, and firm years in
fraud and post-fraud periods. For Hypothesis 4, we focus on firm-years in
pre-fraud and post-fraud periods. For Hypothesis 3, our prediction is that α7<0.
For Hypothesis 4, our prediction is that α7>0.
3.1. Tests using the Basu Model:
The standard Basu model estimates the reverse
regression of earnings on market returns as follows:
EARN =α0 +α1RET+α2D+a3RET*D+ε (4) where,
EARN = earnings
before extraordinary items (data18), deflated by the prior period total
assets,
RET = buy
and hold stock return, cumulated starting three months after the beginning of
a firm’s fiscal year, and ending
three months after the end of the company’s fiscal year,
D = is an indicator variable that
equals 1 when RET <0; 0
otherwise.
We modify this model to include the effect of fraud
as follows
EARN =α0 +α1RET+α2D+a3RET*D+α4FRAUDPRD+α5FRAUDFRM+α6RET*D*FRAUDPRD
+α7RET*D*FRAUDFRM+α8RET*D*FRAUDPRD*FRAUDFRM+ε (5)
As before, hypothesis 1 suggests that α8 should be negative.
For hypothesis 2 we restrict the sample to
fraud firms only and re-estimate equation (5) including only pre-fraud and
post-fraud periods. Our hypothesis 2
predicts that α8>0 in this subsample.
Hypotheses 3 and 4 involve examining the effects of the magnitude of the
restatement. In order to conduct these
tests we modify equation (5) to take the form:
EARN =α0 +α1RET+α2D+a3RET*D+α4FRAUDPRD+α5RESTATEH+α6RET*D*FRAUDPRD
+α7RET*D*FRAUDORD*RESTATEH+ε
(6)
For Hypothesis 3, we restrict our sample to
two subsamples: firm-years in pre-fraud and fraud periods, and firm years in
fraud and post-fraud periods. For Hypothesis 4, we focus on firm-years in
pre-fraud and post-fraud periods. For Hypothesis 3, our prediction is that α7<0. For Hypothesis 4, our prediction is that α7>0.
4. Sample
Selection
Our fraud sample includes firms that fraudulently overstated annual
earnings (i.e., the firm misstated earnings on at least one 10-K filing). While
there are other types of fraud, overstatement of earnings is the most common
type of fraud and is relates most clearly to a lack of conservatism. We did not
include frauds that misstated quarterly data because the conservatism models in
our study are designed to detect conservatism using annual data. We identified our fraud sample from three
sources. The first source is the COSO published report: “Fraudulent Financial
Reporting: 1987-1997 - An Analysis of U.S. Public Companies”. The COSO study investigated frauds that were
identified in SEC’s Accounting and Auditing Enforcement Releases (AAERs) issued
during the period of 1987-1997. COSO identifies 204 fraud firms. Second, we
performed our own search of AAERs issued during 1998-2007. We used “fraud” as a
search term and identified an additional 268 fraud firms. Third, we identified
another six firms by searching the popular press and the American Accounting
Association Monograph on litigation involving Big4 auditors and their
predecessor firms. From this combined sample of 478 observations, we excluded
those firms that (i) didn’t misreport at least one 10-K (e.g. fraudulent
manipulated quarterly data only), (ii) committed non-financial frauds (e.g.
insider trading, omitted disclosures, backdated options), (iii) did not manage
earnings (e.g. reported sales on a gross rather than a net basis which
increased sales and cost of sales by the same amount), and (iv) didn’t have
financial data available in Compustat
or CRSP, or we are unable to locate company data (e.g. small
firms or foreign companies). Our final fraud sample consists of 187 fraud
firms.[5] The procedures for identifying fraud firms
are summarized in Appendix A. We then collect Compustat data for these firms
for the fraud years we identify and 6 years before and after the fraud period.
We use firm-years with all available Compustat data in our analysis. These data
are further described in Table 1.
To test Hypotheses 1a, 1b and 2, we construct
a matched sample of non-fraud firms by including all firms that did not commit
fraud, with available data in Compustat in our control sample. We match these
non-fraud firms with fraud firm sample on a firm year and 2 digit SIC code, and
include all non-fraud firm-year observations with available data in our
analysis. These data are further described in Table 1.
To test Hypotheses 3 and 4, we hand-collect the ex-post income
restatement amounts for fraud firms, in addition to the data already used to
test Hypotheses 1a, 1b and 2. Because restatement amounts are not available for
all firms, our sub-samples are much smaller in size.
5. Results
5.1. Descriptive
statistics
Table 1 summarizes the descriptive
statistics of the sample of fraud and non-fraud firms during pre-fraud period,
fraud period, and post-fraud period. Several interesting patterns emerge.
Consistent with expectations, earnings performance of fraud firms declines over
time: median net income (EARN)
changes from 0.06 to 0.04 to -0.03 between pre-fraud, fraud and post-fraud
periods.[6]
This result is consistent with the other evidence in the literature suggesting
that fraud firms overstate their earnings and experience declines in
performance, subsequent to the fraud’s discovery (e.g. Rosner, 2003). A similar
pattern could be observed with respect to fraud firms’ raw and abnormal stock
returns (variables RET and BHAR respectively). Fraud firms’ total accruals also experience
declines through time, most notably in the post-fraud period, consistent with
our expectation that fraud firms should become more conservative after the
fraud’s discovery. However, consistent with the possible accrual manipulation
during fraud period, Fraud firms’ cash flows remain similar through time.
Interestingly, no such consistency could be observed for non-fraud firms:
earnings, accruals and cash flows follow the same declining temporal patterns
for these firms. Neither could we see a discernable pattern for stock returns
of non-fraud firms. The analysis of the other variables reveals that fraud
firms in our sample are somewhat different from the control non-fraud firms:
fraud firms are slightly larger and have slightly less debt[7],
but are not significantly different with respect to their price-earnings ratio
(PE). Since these differences are not
substantial, sample selection biases seem to be less of an
issue here.
5.2 Tests using the Ball and Shivakumar
model
Table 2 summarizes our tests of whether
the Ball and Shivakumar model detects differences in conservatism across fraud
and non-fraud firms as well as across fraud and nonfraud periods. Our model allows the general coefficient of
conservatism DCF*CF to vary both
across firm types (fraud vs. non-fraud firms) and across time (pre-fraud,
fraud, and post-fraud periods). The
coefficient for DCF*CF*FRAUDFRM
captures the association between fraud firms and conservatism. Results reported in panels A, B and C, show
that the fraud firms are less conservative than the non-fraud firms in all
sub-periods as the coefficient for DCF*CF*FRAUDFRM is negative and statistically significant in
all regressions. More importantly, we
are interested in understanding whether fraud firms are relatively less
conservative than the non-fraud firms over the fraud period and this is
examined by looking at the coefficient on variable DCF*CF* FRAUDPRD*FRAUDFRM. We find the coefficient for this variable to
be negative and statistically significant at the 0.01 level in panel A and
positive and statistically significant in Panel B, suggesting that fraud firms
are indeed less conservative than non-fraud firms in the fraud period compared
to the pre or post fraud periods.
Moreover, the results of panel C suggest that in the post-fraud period,
fraud firms increase their conservatism substantially compared to the pre-fraud
period, as the coefficient for DCF*CF*
FRAUDPRD*FRAUDFRM is positive and statistically significant in panel
C..
We examine next whether the
Ball and Shivakumar model detects differences in
conservatism across high and low
restatement firms. The results of this
analysis are presented in Panel B of table 2.
We only include fraud firms in this analysis (no control sample), as we
are interested in how conservatism varies with restatement size among fraud
firms. Our primary variable of interest is DCF*CF*FRAUDPRD*RESTATEH. Our results, suggest that
conservatism indeed varies with the
size of the restatement. In our
comparison of the pre-fraud period with the fraud period, the coefficient on DCF*CF*FRAUDPRD*RESTATEH is negative and
significant at the 0.01 level showing that high restatement firms are
significantly less conservative than the other fraud firms in the fraud
period. This result suggests that the
corresponding result in Panel A of Table 2 is driven by high restatement
firms. Furthermore, the results indicate
that there is an improvement in conservatism of fraud firms in post-fraud
period
that
is driven
by the
high restatement
firms as the coefficient for
DCF*CF*FRAUDPRD*RESTATEH is positive and statistically
significant.
These results, taken together, suggest that:
(i) the Ball and Shivakumar model correctly identifies the lack of conservatism
among fraud firms during the fraud period, compared to the general population;
(ii) the lack of conservatism is primarily driven by high restatement firms,
and (iii) the fraud firms increase their conservatism in post-fraud period, in
response to the discovered fraud. The latter result also appears to be driven
by high restatement firms.
5.3 Tests using the Basu Model
Next we go on to examine the degree to which
the Basu model picks up differences in conservatism across fraud and non-fraud
firms and over time. The research design
mimics the tests used for the Ball and Shivakumar model. Panel A of table 3 reports the results of
tests examining whether the fraud firms show a lack of conservatism over the
fraud period. Compared to the results
for the Ball and Shivakumar model the results using the Basu model are weaker. First, the coefficient for D*RET*FRAUDFRM is not statistically
significant in panel A suggesting no difference in conservatism in the fraud
and non-fraud firms (the coefficient is statistically significant in pabnels B
and C though). In order to examine
whether fraud firms are less conservative than non-fraud firms in the fraud
period we examine the coefficient for D*RET*FRAUDPRD*FRAUDFRM and as in the case of the Ball and
Shivakumar model, the
coefficient for this variable is
positive and statistically significant suggesting that that fraud firms are
indeed less conservative than the non-fraud firms in the fraud period. Finally with respect to the comparison of
conservatism of the fraud firms over time, the coefficient on variable D*RET*FRAUDPRD*FRAUDFRM is positive with a t-value of 1.63, which is marginally
significant in a one-tail test showing weak evidence of an improvement in
conservatism in the post-fraud period.
We repeat our analysis of high restatement
firms for the Basu model and the results are reported in Panel B of Table 3.
Again, our primary variable of interest is the regression coefficient on
variable D*RET*FRAUDPRD*RESTATEH. We find that for high restatement fraud firms
are significantly less conservative than the other fraud firms in fraud period
vs. prefraud period (the coefficient is negative and statistically
significant). However, with respect to pre-fraud and post-fraud period
comparison, we find that the Basu model picks up a decrease in conservatism for high restatement firms, a result
opposite to that of the Ball and Shivakumar model (see Panel B of Table
2).
6. Conclusion
We examine whether
the Basu (1997) model and the Ball and Shivakumar (2006) model identifies
accounting firms committing fraud as anti-conservative. We find that the Ball
and Shivakumar model shows that fraud firms are less conservative than
non-fraud firms in fraud period as compared to pre-fraud period. Moreover, the
Ball and Shivakumar model shows that fraud firms’ conservatism increases in
post-fraud period as compared to non-fraud firms. For the Ball and Shivakumar
model, we find that these results, to a large degree, are driven by firms
committing the largest frauds, where the degree of fraud is measured by the
size of the subsequent earnings restatement deflated by the firm’s assets. Our
results for the Basu model are weaker. We find that according to the Basu
model, fraud firms become more conservative than non-fraud firms in post-fraud
period. However, when we also split our
sample by magnitude of fraud-related restatement, we find that the Basu model
identifies fraud firms as anti-conservative as compared to pre-fraud period.
However, with respect to post-fraud period, we find that the Basu model shows a
decline in conservatism of fraud firm, which is inconsistent with our
expectations and the results of Ball and Shivakumar model. Thus, these results
suggest that the Ball and Shivakumar model is a more powerful test of firm’s
conservatism, at least when it comes to distinguishing fraud firms from
non-fraud firms.
Our results are important
for the overall debate on measurement of accounting
conservatism which has risen more
actively in the recent years. Furthermore, our results are also important for
policy-makers who are debating the merits of greater conservatism for capital
markets (Watts, 2003).
References:
Ball, R. and S.P. Kothari (2008).
Econometrics of the Basu Asymetric Timeliness Coefficient and
Accounting Conservatism. Working Paper, U of Chicago
and MIT
Ball, R. and L. Shivakumar (2007).
Earnings Quality at Initial Public Offerings. Journal of Accounting and Economics, forthcoming.
Ball, R., R. Bushman and F.
Varvari (2007). The Debt-Contracting Value of Accounting Information and Loan
Syndicate Structure. Working Paper, University of North Carolina.
Bartov, E., D. Givoly and C. Hayn
(2002). The rewards to meeting or beating earnings expectations.
Journal of Accounting
and Economics, 33(2), 173-204
Basu, S. (1997). The conservatism
principle and the asymmetric timeliness of earnings. Journal of Accounting and Economics, 24(1), 3-37
Beasley, M., J. Carcello, and D. Hermanson. (1999). Fraudulent Financial Reporting: 1987 1997 An
Analysis of U.S. Public Companies. Committee of Sponsoring Organizations of
the Treadway Commission.
Beaver, W. and S. Ryan (2005). Conditional and
Unconditional Conservatism: Concepts and Modeling. Review of Accounting Studies, 10, 269-209
Beneish.M.
1999. The detection of earnings
manipulation. Financial Analyst Journal 24-36.
Dietrich, R., K. Mueller, E. Riedl
(2007). Asymmetric timeliness tests of accounting conservatism. Review of Accounting Studies, 12(1),
95-124
Farber, D.B.(2005). Restoring
trust after fraud: Does corporate governance matter? The Accounting Review 80, 539–561.
García Lara, J.M., García Osma, B., and Penalva, F.
(2007). Accounting Conservatism and
Corporate Governance. Review of Accounting Studies
Givoly, D. and C. Hayn (2000). The
changing time-series properties of earnings, cash flows and accruals: Has financial
reporting become more conservative? Journal
of Accounting and Economics, 29, 287-320
Givoly, D., C. Hayn, A. Natarajan
(2007). Measuring reporting conservatism. The
Accounting Review, 82(1), 65-106
Hayn, C. (1995). The information
content of losses. Journal of Accounting
and Economics, 20(2), 125-153
Jones, K., G. Krishnan and K.
Melendrez (2008). Do Models of
Discretionary Accruals Detect Actual Cases of Fraudulent and Restated Earnings?
An Empirical Evaluation. Forthcoming in Contemporary
Accounting Research.
Khan, M. and R. Watts (2007).
Estimation and Validation of a Firm-Year Measure of Conservatism. Working
Paper. MIT.
Loebbecke, J., M. Eining, and J. Willingham. 1989.
Auditors' experience with material irregularities: Frequency, nature,
anddetectability. Auditing: A Journal of
Practice & Theory 9 (Fall):
1-28.
Moerman, R. (2006). The role of
information asymmetry and financial reporting quality in debt contracting:
Evidence from the secondary loan market. Working Paper. University of
Pennsylvania.
Petersen (2007). Estimating standard errors in finance
panel data sets: comparing approaches. Forthcoming, Review of Financial Studies.
Rosner, R. (2003). Earnings
manipulation in failing firms. Contemporary
Accounting Research, 20(2), 361-408
Roychowdhury, S. and R. Watts
(2007). Asymmetric timeliness of earnings, market-to-book and conservatism in
financial reporting. Journal of
Accounting and Economics, 44(1), 2-32.
Watts, R. (2003). Conservatism in Accounting Part II:
Evidence and Research Opportunities.
Accounting
Horizons 17 No. 4 (2003): 287-301
[1] In particular, for firms
with significant number of economic events, where the amount of information
arrival is more gradual, conservatism appears to be under-stated. This suggests
that Basu measure is biased against finding conservatism amongst larger firms.
[2] Basu (1997) is a reverse
earnings returns regression and can be interpreted as just the outcome of the
stock market assigning lower earnings multiple on less persistent accounting
losses (Hayn (1995)).
[3] Moerman (2006), Jones et
al. (2008), Ball, Bushman and Vasvari (2006), Pae (2007)
[4] Jones et al. (2008)
investigate the relative model of discretionary accruals estimates produced by
Ball and Shivakumar (2006) model.
[5]
In addition, we found that Compustat
does not consistently report restatement data. It appears that if the restated
data is available when Compustat
personnel enter the data in their
database, the restated data is entered and the fraudulent numbers are
discarded. It does not appear that Compustat
changes data upon restatements several years after the original data is entered
in their database. Therefore, we compared Compustat
data with the original 10-K filing to verify that the data in Compustat is the fraudulently reported
numbers and not the restated data. We found that Compustat reports restated data for 36 of the 361 fraud firm-years
in our fraud sample. We hand-collect the original fraudulent data for those 36
firm years. SEC filings are available on EDGAR beginning in 1994. SEC filings for
selected companies are available on Lexis/Nexis
for years prior to 1994 and we were able to locate data for several firms prior
to 1994.
[6] Significant at 0.01 level
using Wilcoxon test.
[7]
Significant at 0.01 level using Wilcoxon test.
No comments