Documente Academic
Documente Profesional
Documente Cultură
DOI 10.1007/s11142-011-9164-5
Abstract I examine whether accounting complexity in the area of revenue recognition increases the probability of restating reported revenue. I measure revenue
recognition complexity using the number of words and recognition methods from
the revenue recognition disclosure in the 10-K and a factor score based on the
number of words and methods. Tests reveal that revenue recognition complexity
increases the probability of revenue restatements, and these restatements are the
result of both intentional and unintentional misreporting. Furthermore, complexity
moderates the consequences of restatementlower incidence of AAERs, less
negative restatement announcement returns, and lower subsequent CEO turnover
suggesting that stakeholders of the firm consider accounting complexity when
responding to misreporting.
Keywords Misreporting Restatement Revenue recognition Accounting
complexity Restatement consequences
JEL Classification
G38 M41
1 Introduction
I examine how accounting complexity affects both the incidence and consequences of
misreporting, an important issue as standard setters are confronted with how to
account for increasingly complex transactions. Since the Financial Accounting
Standards Board (FASB) is not interested in reducing representational faithfulness in
favor of simplicity (FASB 2008, QC24), understanding the effects of complexity is
important in evaluating the costs and benefits of the FASBs approach. I focus on one
K. Peterson (&)
University of Oregon, Lundquist College of Business, Eugene, OR 97403, USA
e-mail: kylepete@lcbmail.uoregon.edu
123
73
effect of accounting complexity, misreporting, because the FASB and Securities and
Exchange Commission (SEC) have both suggested complexity is a major contributor
to the increased incidence of financial statement misreporting (Cox 2005; Herz 2005).
I focus on revenue recognition specifically because (1) revenue recognition applies to
all firms; (2) revenue misreporting is a common type of restatement (Palmrose et al.
2004; GAO 2002, 2006); and (3) anecdotal evidence suggests that revenue
recognition can be complex (Sondhi and Taub 2006; Herz 2007; Turner 2001).
Complexity is the state of being difficult to understand and apply (SEC 2008).
Complex accounting specifically pertains to the difficulty in understanding the
mapping of transactions (or potential transactions) and standards into financial
statements.1 My three empirical proxies capture complexity by using the firms
revenue recognition disclosures: (1) the number of words in the revenue recognition
policy description in the notes to the financial statements; (2) the number of methods
listed in that same description; and (3) a factor score using both the words and
methods. Due to the inherent difficulty in disentangling complexity resulting from
transactions and standards, my proxies capture overall complexity, but still allow me
to provide evidence on the effects of accounting complexity on misreporting and
various stakeholders reactions to misreporting when accounting is complex.
I hypothesize that revenue recognition complexity increases the likelihood of
revenue restatements due to two competing (but not exclusive) effects. First, when
accounting is complex, managers are more likely to err when applying standards to
transactions, increasing the likelihood of unintentional misreporting due to
mistakes. Second, complex accounting may allow managers to manipulate financial
statements as suggested by Picconi (2006) and Bergstresser et al. (2006). To
determine whether complexity is associated more with mistakes or manipulation, I
test whether my complexity proxies are more associated with revenue restatements
being an irregularity or an error as defined in Hennes et al. (2008). I also examine
the effect of complexity on the consequences of restatement, including the
likelihood of an SEC Accounting and Auditing Enforcement Release (AAER),
restatement announcement returns, and CEO turnover following the restatement.
I conduct my tests on a sample of 333 revenue restatements from 1997 through
2005. To test whether revenue recognition complexity increases the probability of
restating revenue, I compare firms restating revenue with two sets of control firms:
(1) firms that had a restatement during the sample period but did not restate revenue
(hereafter referred to as nonrevenue restatements) and (2) a matched sample of firms
that do not have any kind of restatement during the sample period.
Results show that firms with complex revenue recognition are more likely to
restate revenue. Depending on the model and the proxy of complexity, a one
standard deviation increase in revenue recognition complexity increases the
probability of revenue misreporting between 7.6 and 13.0 percent relative to a
matched sample. Compared with other determinants in the models, this suggests
complexity is an important determinant of revenue misreporting. However, I find
1
Prior literature has not developed a definition of accounting complexity. The SECs Advisory
Committee on Improvements to Financial Reporting (ACIFR) provides a definition in its final
recommendation report (SEC 2008), and it is similar to the one presented in this paper. Accounting
complexity is described more thoroughly in Sect. 2.
123
74
K. Peterson
The results in this paper do not address whether accounting complexity is pareto optimal or should be
reduced. Without an examination of all costs and benefits of complexity, it is not feasible to make any case
for social welfare. For example, potential benefits of accounting complexity relative to simpler accounting
could be reduced earnings management or better comparability, which are not examined in this study.
My findings suggest that firms that restate revenue have more complex revenue recognition; however,
from Plumlee and Yohn (2009), it seems that firms do not necessarily highlight complexity as a reason for
the restatement but are more likely to use vague descriptions like internal error. What managers say
about the causes of misreporting does not likely include a description of all relevant factors that led to the
restatement (e.g., undue pressure to meet targets, executive compensation, governance failures,
complexity), but examining all these factors in a multivariate setting provides a better understanding
of all these effects.
123
75
2 Hypothesis development
2.1 Accounting complexity
I define accounting complexity as the amount of uncertainty related to the mapping
of transactions or potential transactions and standards into the financial statements.4
This definition is intended to apply to both preparers and users of financial
statements and is similar to the one proposed by the SECs Committee on
Improvements to Financial Reporting (ACIFR) in its recommendation report issued
in August 2008.5
Accounting complexity stems from a combination of transactions and financial
reporting standards. With regard to transactions, the increasingly sophisticated
nature of business transactions can be difficult to understand (SEC 2008). For
example, uncertainty about transactions increases for firms with numerous
customer-specific contracts or agreements documented by multiple contracts.
Reporting standards can also increase complexity, including the following
characteristics highlighted by the ACIFR (SEC 2008):
While it may be useful to separate the source of complexity into transactions and
standards, isolating these two sources empirically is difficult (or impossible)
because standards are written with transactions in mind. For example, basic
transactions generally only need simple guidance. But for complex transactions,
standard setters could provide simple guidance (for example, for revenue, waiting
for the collection of cash or full performance on a contract before recognition), or
more complex guidance (for example, recognition depending on more complex
4
No formal definition of accounting complexity exists in the academic literature. Prior research has
examined firm or organization complexity (Bushman et al. 2004), information complexity (Plumlee
2003), and information overload (Schick et al. 1990 for a review), concepts not wholly unrelated to
accounting complexity.
The ACIFR define financial reporting complexity for preparers as the difficulty to properly apply [US
GAAP] and communicate the economic substance of a transaction and for investors as the difficulty in
understanding the economic substance of a transaction or event and the overall financial position and
results of a company (SEC 2008).
123
76
K. Peterson
Although this theory suggests managers take advantage of complex accounting by managing the
financial statements, complexity is not a necessary condition for manipulation. Many fraudulent practices
are implemented using simple accounting settings (e.g., fictitious sales, bill-and-hold transactions, and
capitalizing expenses).
123
77
the misreporting is similar across both theories, I propose the following hypothesis,
stated in alternate form:
H1 Managers of firms with more complex revenue recognition are more likely to
misreport revenue than managers of firms with less complex revenue recognition.
Even though both the mistake and manipulation theories lead to the prediction in
H1, the null hypothesis of no result could occur if the effect of complexity on
revenue misreporting were small or if misreporting is solely driven by managerial
incentives and governance, as hypothesized in prior literature (Palmrose et al. 2004;
Burns and Kedia 2006; Zhang 2006; Callen et al. 2009). More importantly, these
tests allow me to quantify the economic importance of the effect of complexity on
the likelihood of misstating revenue. Prior literature (Hennes et al. 2008, Palmrose
et al. 2004) has shown that intentional misreporting has more severe consequences
than unintentional misreporting. Since complexity could lead to intentional or
unintentional misreporting, I examine whether my complexity proxies are
associated with the probability that the restatement is an irregularity versus an
error as defined in Hennes et al. (2008). I also examine the effect of complexity on
the consequences of misreporting. I do not make formal hypotheses for these tests
due to the competing effects of the mistake and manipulation theories on the
predictions. I discuss these tests in Sect. 5.
I consider SAB 101 and EITF restatements as mandatory restatements caused by a change in
accounting standard. During the sample period, the Emerging Issues Task Force issued EITFs 99-19,
00-10, 00-14, 00-22, 00-25 to clarify revenue recognition issues such as recognizing gross v. net, shipping
and handling costs, sales incentives, and other consideration from a vendor. Including the SAB and EITF
firms in testing H1 provides similar results.
123
78
K. Peterson
I do not match on industry because it likely introduces a noisy sort on revenue recognition complexity,
potentially controlling for the effect being tested. However, I do control for industry in the regression
analysis.
123
79
the Appendix, which contains four sample revenue recognition disclosures and
their complexity proxies to illustrate how these disclosures capture complexity. I
collect revenue recognition disclosures contained in the summary of significant
accounting policies from firms most recent 10-K prior to the restatement
announcement using the SEC EDGAR Database. I measure revenue recognition
complexity using the number of words (WORDS), a proxy for the number of
methods (METHODS), and a factor score (RRC SCORE) based on WORDS and
METHODS from the disclosures.9,10 METHODS is measured as the number of
occurrences of the word-stems recogn and record found in the disclosure.
Bushman et al. (2004) and others have used measures of general organizational
complexity relying on proxies such as firm size and operating or geographic
segments. My measures of revenue recognition complexity could be associated with
these traditional measures of organizational complexity, so for the tests described in
the next section, I include controls for operating complexity and firm size.
Table 1 Panel A provides revenue recognition complexity statistics for the
revenue restatement sample and both comparison samples and tests for differences
in means and medians. RRC SCORE is calculated for each combined sample and
produces a score that is mean zero. The tests reveal that revenue restaters have
significantly higher mean and median WORDS, METHODS, and RRC SCORE than
both sets of comparison firms (p values \ 0.001).
Because managers have discretion with their revenue recognition disclosures,
they could alter their disclosures to appear more or less complex. To alleviate
concerns that managers may be manipulating revenue recognition disclosures prior
to the restatement, I also collect the revenue recognition disclosures just following
the restatement announcement. If discretion of the disclosure exists, it should be
reduced following the restatement due to auditor scrutiny accompanying the
restatement. Table 1 Panel B shows that revenue restatement firms have more
WORDS and METHODS and higher RRC SCORE than non-revenue restatement
firms in both the pre- and post-periods, suggesting that the higher complexity for
revenue restatement firms is not driven by managerial discretion and still exists
post-restatement.11
9
Certain practices or factors could also lead to increased complexity and risk of misreporting beyond
what may be captured by disclosure length (see AICPA Practice Alert 98-3, 1998). In unreported analysis,
I measure RRC SCORE where I also include the total number of counts in the disclosure (by using keyword searches) for the following revenue recognition practices: the percentage of completion method,
multiple deliverables, vendor-specific objective evidence, barter or nonmonetary exchange revenue, or
fair valuing aspects of the contract. Results using this measure are consistent with the results reported in
the tables for RRC SCORE.
10
As a test of validity of my complexity measures, I examine whether my measures are associated with
the variation and error in analysts forecasts of revenue, an indication that complexity increases
uncertainty. Results (untabulated) indicate all three proxies are positively related to both the error and
variation in analysts revenue forecasts with p values less than 10 percent after controlling for analyst
following, size, and book-to-market.
11
It is also interesting to note that for both the revenue restaters and non-revenue restaters, the number of
WORDS and METHODS increased in the post period, but the increase was greater for the revenue
restaters (91.3 and 1.57 for revenue restaters; 38.0 and 0.53 for non-revenue restaters). The greater
increase in post-restatement disclosures for revenue restaters could be an attempt to resolve confusion
over already complex revenue recognition.
123
123
333
RRC SCORE
-0.30
4.02
169.3
-0.18
270.5
5.86
322
322
RRC SCORE
0.56
0.46
1.57***
6.86
Difference
7.43
Test
322
322
METHODS
7.06
POST METHODS
91.3***
Test
361.9
Difference
322
322
WORDS
POST WORDS
0.50
0.35
7.51
1.00***
6.00
5.00
9.39
80.0***
273.5
193.5
817
817
817
817
817
817
-0.22
-0.19
5.35
0.53***
4.48
3.95
6.86
38.0***
221.1
183.2
Mean
-0.31
-0.30
7.50
0.00***
3.00
3.00
9.72
34.0***
136.0
102.0
Median
-0.39
3.00
113.0
-0.28
3.00
103.0
Mean
Median
324
324
324
859
4.00
186.5
0.20
5.00
859
859
Non-revenue restatements
0.30
5.81
184.5
0.36
0.46
264.8
5.00
192.0
5.88
268.6
Median
Revenue restatements
324
RRC SCORE
Variable
324
324
WORDS
METHODS
333
333
WORDS
METHODS
Mean
Median
Mean
Comparison group
Revenue restatements
0.78***
0.65***
1.04***
2.96***
1.92***
53.4***
140.7***
87.4***
Diff.
Mean test
0.61***
1.79***
95.5***
0.63***
1.88***
82.1***
Diff.
Mean tests
9.41
7.81
4.87
9.56
7.31
4.45
7.98
5.81
t test
6.11
5.73
5.89
7.70
7.23
5.50
t test
0.82***
0.65***
1.00***
3.00***
2.00***
46.0***
137.5***
91.5***
Diff.
Median test
0.59***
2.00***
71.5***
0.64***
2.00***
89.0***
Diff.
Median tests
64.7
45.3
12.3
63.9
49.4
25.0
62.6
56.5
v2
28.5
32.9
26.1
47.7
51.1
56.7
v2
80
K. Peterson
0.15*
1.78
0.91
-0.03
3.94
-0.02***
Median
0.13**
Diff.
Mean test
2.04
t test
0.17***
Diff.
Median test
7.0
v2
This table contains revenue recognition disclosure statistics for a sample of revenue restatements and two comparison samples, a sample of non-revenue restatements and a
matched sample. Panel A presents descriptive statistics for the revenue recognition complexity proxies by sample. Panel B contains comparisons of revenue recognition
disclosure statistics pre- and post-restatement for both revenue and non-revenue restatement firms. The number of firms in Panel B differs from those presented in Panel A
due to the requirement to have post-restatement revenue recognition disclosures. Bolded statistics are significant at the 10 percent level. WORDS (POST WORDS) is the
number of words in the revenue recognition footnote disclosure in the most recent 10-K filing prior to (following) the restatement announcement. METHODS (POST
METHODS) is the number of times recogn or record is used in the revenue recognition footnote disclosure in the most recent 10-K filing prior to (following) the
restatement announcement. RRC SCORE (POST RRC SCORE) is the factor score of (POST) WORDS and (POST) METHODS using the principal components methods.
WORDS and METHODS are winsorized at the 1 and 99 percentiles. T tests and Wilcoxon sign-ranked tests (both 2-tailed) are calculated for the difference in means and
medians, respectively. *, **, and *** Indicate significance at 10%, 5%, and 1%
0.10*
1.77
Test
Mean
Median
Mean
Non-revenue restatements
Revenue restatements
Difference
Variable
Table 1 continued
123
82
K. Peterson
123
83
123
84
K. Peterson
Table 2 Logistic regression estimates on the relation between revenue recognition complexity and
restatements using a restatement design
Pred.
WORDS
Coeff
METHODS
Z-stat
Coeff
RRC SCORE
Z-stat
Coeff
Z-stat
Complexity
0.330***
4.36
0.082***
3.71
0.344***
OPCOMPLEX
0.317**
2.17
0.316**
2.16
0.306**
BTM
-0.198
LOSSPER
0.438
1.56
0.387
1.39
0.392
1.40
SALESFCST
0.495***
2.61
0.533***
2.88
0.506***
2.70
EARNVOL
0.005
0.99
0.006
1.04
0.005
1.01
-1.29
-0.261*
-1.74
-0.231
4.60
2.08
-1.52
CHSALES
-0.279*
-1.65
-0.295*
-1.71
-0.288*
-1.67
PRERET
-0.262**
-2.31
-0.259**
-2.32
-0.257**
-2.32
BIGN
-0.407
-1.54
-0.346
-1.32
-0.398
-1.51
LOGMVE
0.034
0.64
0.009
0.19
0.018
0.35
AUDITOR
0.295
1.32
0.326
1.45
0.307
1.37
AR ACCRUAL
4.217***
2.60
5.003***
3.12
4.493***
2.81
1,192
1,192
1,192
Pseudo R2
0.161
0.155
0.161
This table presents estimates of a logistic regression model where the dependent variable is one if the firm
restated revenue and zero if the firm had a restatement but restated something other than revenue.
Complexity is a placeholder for each of the three complexity proxies (WORDS, METHODS, RRC
SCORE) as described in Table 1. OPCOMPLEX is the log(GEOSEG ? OPSEG), where GEOSEG
(OPSEG) are the number of geographic (operating) segments reported for the firm in COMPUSTAT.
BTM is the firms book-to-market ratio at the end of the fiscal year just prior to the restatement
announcement. LOSSPER is the percentage of firm years with negative income prior to the restatement
announcement. SALEFCST is an indicator equal to one if the firm had an analyst sales forecast any time
prior to the restatement and zero otherwise. EARNVOL is the standard deviation of earnings scaled by the
absolute mean value of earnings for the five fiscal years prior to the restatement. CHSALES is the average
change in annual net sales for the 2 years prior to the restatement. PRERET is the 12-month stock returns
(including delisting returns) for the firm prior to the restatement. BIGN is an indicator equal to one if the
firm was audited by a large accounting firm and zero otherwise. LOGMVE is the log market value of
equity at the fiscal year end prior to the restatement. AR ACCRUAL is the 5 year average A/R accrual
scaled by sales prior to the restatement. Z-statistics are calculated using Huber/White robust standard
errors with firm-level clustering to adjust standard errors for multiple restatements from the same firm.
Results for the intercept, industry, and year indicators are not shown but are included in the model. *, **,
and *** Indicate significance at 10, 5, and 1%
123
85
Table 3 Conditional logistic regression estimates on the relation between revenue recognition complexity and restatements using a matched sample design
Predict
Without PPS
Coeff
RRC SCORE
OPCOMPLEX
0.491***
-0.115
Including PPS
Z-stat
4.48
1.252***
2.92
0.220
0.42
-8.551
-1.05
BTM
1.609
1.55
0.716
1.47
SALESFCST
-0.483*
EARNVOL
0.012
CHSALES
1.092***
-0.137
Z-stat
-0.60
LOSSPER
PRERET
Coeff
4.832*
1.65
-1.68
1.763
1.08
1.54
-0.001
-0.25
2.96
-0.96
6.364**
2.14
0.735
1.26
BIGN
-1.348***
-3.08
3.975*
1.85
LOGMVE
-0.618***
-3.02
-2.462*
-1.75
-1.010
-0.45
AR ACCRUAL
-0.58
-9.440
DEBT ISSUE
0.874*
1.66
2.684
0.86
EQUITY ISSUE
0.845*
1.90
6.972
0.81
-0.74
LEVERAGE
-1.103
-1.59
-2.922
EP
0.165
0.44
0.317
0.12
OP ACC
3.741***
3.34
-0.209
-0.03
LOG PPS
0.357
1.01
648
186
Pseudo R2
0.339
0.636
This table presents estimates of a conditional logistic regression model using a matched sample where the
dependent variable is one if the firm restated revenue and zero if the firm is a matched firm. DEBT ISSUE
is the sum of long- and short-term debt issued (dltis ? dltr) divided by average total assets for the fiscal
year prior to the restatement. EQUITY ISSUE is equal to common and preferred stock issued (sstk)
divided by average total assets for the fiscal year prior to the restatement. LEVERAGE is the ratio of
short- and long-term debt (dltt ? bast) divided by total assets for the fiscal year prior to the restatement.
EP is the ratio of earnings per share (epspx) to price (prcc_f) at the end of the fiscal year prior to
restatement. OP ACC is operating accruals (oiadp-oancf) divided by average total assets for the fiscal year
prior to the restatement. LOG PPS is the change in the value of stock options held for a percentage change
in the value of the firm as outlined in Core and Guay (2002) and Burns and Kedia (2006). The first three
specifications exclude LOG PPS from the model because it reduces the sample size considerably. All
other variables are defined in Tables 1 and 2. Z-statistics are presented using Huber/White robust standard
errors with firm-level clustering to adjust standard errors for multiple restatements from the same firm.
Results for industry and year indicators are not shown but are included in the model. *, **, and ***
Indicate significance at 10, 5, and 1%
are positive and significant in both specifications, consistent with the findings in
Table 2 and H1. Although many of the coefficients are similar to those in Table 2,
the coefficient on CHSALES is now positive and significant, indicating that restating
firms in general have higher sales growth than non-restating firms. Table 3 results
also suggest that revenue restatement firms are more likely to access the equity and
debt markets and have higher operating accruals prior to a restatement announcement compared to matched-sample control firms. Many of the significant results for
123
86
K. Peterson
control variables disappear when LOG PPS is added to the model, most likely an
indication of losing power due to the sample being restricted.
I estimate marginal effects for the regressions in Table 3 as the average change in
the predicted probability as the variable for the treatment observation moves one
standard deviation centered on the observed value, holding all the other variables
constant at their observed values (see Greene 1997). A one standard deviation
increase in revenue complexity increases the probability of misreporting between
7.6 and 13.0 percent relative to the matched sample firms, which is similar to the
marginal effects of complexity in Table 2.
I include three variables that provide some indication of intent: (1) whether the
firm restated more than just revenue (MULTIPLE); (2) whether the restatement is
credited to the firms auditor (AUDITOR); and (3) a dummy equal to one if the
restatement caused the firm to miss the sales forecast for the first period of the
restatement and zero otherwise (MISS FCST). I also include three measures of
the magnitude of the restatement: (1) the number of periods the company is restating
in quarters (RESTLEN); (2) the percentage change in revenue over all periods of the
misreporting due to the restatement (CHREV); and (3) the percentage change in net
income over all periods of the misreporting due to the restatement (CHNI). Finally, I
include controls for size (LOGMVE), whether the firm was audited by a large
accounting firm (BIGN), and industry indicators.
5.2 Irregularity results
The results for testing whether complexity is associated with errors or irregularities
are found in Table 4. The table presents results using only RRC SCORE, but results
are consistent when using the other two proxies. The coefficient on RRC SCORE is
positive, but not significant at the 10 percent level. Since revenue complexity is not
a significant predictor of the restatement being an irregularity, these results
123
87
0.135
-0.63
Z-stat
1.03
-1.55
MISSFCST
0.198
0.46
RESTLEN
0.042**
2.05
AUDITOR
0.509
1.33
MULTIPLE
0.250
0.90
LOGMVE
0.290***
3.77
CHREV
-0.305
-0.29
CHNI
-0.036
-0.47
333
Pseudo R2
0.126
This table contains coefficient estimates of a logistic regression of IRREG (whether the firms restatement
was an irregularity as defined in Hennes et al. 2008) on revenue recognition complexity and control
variables. MISS FCST is an indicator equal to one if the restatement caused the firm to miss the sales
forecast for the first period of the restatement and zero otherwise. RESTLEN is the number of firm
quarters the firm restated. AUDITOR is an indicator equal to one if the auditor identified the restatement
and zero otherwise. MULTIPLE is an indicator equal to one if the firms restatement included additional
areas of restatement besides revenue and zero otherwise. CHREV (CHNI) is the percentage change in
revenue (net income) over all periods of the restatement due to the restatement. All other variables are
defined in prior tables. Z-statistics are listed below each coefficient, using Huber/White Robust standard
errors with firm-level clustering. *, **, and *** Indicate significance at 10, 5, and 1%
combined with the results from H1 suggest complex firms are likely to engage in
both intentional and unintentional misreporting. However, these results are
inconsistent with managers of complex firms pervasively exploiting complexity to
manipulate financial reporting.
5.3 Consequences of misreporting tests
I also examine the consequences of misreporting to determine whether stakeholders
response to misreporting is affected by complexity. If stakeholders are aware of
complexity when they observe misreporting, it is possible they temper their
reactions to restatements for complex firms. I examine three reactions to
misreporting that provide evidence of intent: SEC Accounting and Auditing
Enforcement Releases (AAERs), restatement announcement returns, and CEO
turnover.
First, the issuance of an AAER represents a greater likelihood of intentional
actions.14 I test the following logistic regression model, where the dependent
14
Erickson et al. (2006) correctly argue that SEC actions do not necessarily imply fraud or gross
negligence. In these cases, the action ends with a settlement and an AAER, the firm admits to no
wrongdoing but agrees to avoid future securities violations. However, Karpoff et al. (2008) find that 79
percent of enforcement actions in their sample from 1978 through 2006 include charges of fraud.
123
88
K. Peterson
variable is one if the firm has an AAER associated with revenue or receivables
within 3 years of the restatement announcement and zero otherwise:
PAAER f b0 b1 Complexity b2 MULTIPLE b3 AUDITOR b4 IRREG
b5 MISSFCST b6 RESTLEN b7 CHREV b8 CHNI b9 LOGMVE
b10 BIGN b1119 INDUSTRY
Generally, studies on AAERs (Dechow et al. 1996, 2007; Beneish 1999) have
compared AAER firms with either a large sample of public firms or to small
matched-samples but have not modeled the probability of SEC involvement for a
specific misreporting event. I conjecture that restatement characteristics are
important in determining if the SEC issues an AAER in this setting. These
characteristics include managers intent to manipulate revenue, the magnitude of the
misstatement, and SEC exposure from issuing the AAER. I include the three
variables to identify intent as used in the irregularity regression (MULTIPLE,
AUDITOR, and MISS FCST) plus IRREG as previously defined. I also include three
measures of the magnitude of the restatement (RESTLEN, CHREV, and CHNI) as
defined previously. Finally, the SEC may target large firms (LOGMVE) and firms
audited by large accounting firms (BIGN) because it benefits from enforcement of
those firms relative to smaller firms.
I also test whether the market reaction to revenue restatement announcements
differs based on revenue recognition complexity using an OLS regression where the
dependent variable is cumulative abnormal market adjusted daily returns over a
5-day window (CAR).
CAR b0 b1 Complexity b2 MULTIPLE b3 AUDITOR b4 IRREG b5 CHREV
b6 CHNI b7 LOGMVE b8 PREPET b917 INDUSTRY e
123
89
Consistent with these prior studies, I include control variables that are associated
with CEO turnover following restatements. I include IRREG and MULTIPLE as
previously defined as partial controls for managerial culpability. I control for firm
size by including LOGMVE as previously defined. I also include both CHREV and
CHNI to capture the magnitude of the restatement. Prior firm performance is also
associated with CEO turnover decisions (Engel et al. 2003). Therefore, I include the
cumulative stock returns for the year prior to (PRERET) and the year following
(POSTRET) the restatement announcement to control for market-based performance, and return on assets (ROA) prior to the restatement to control for operatingbased performance. I include the restatement announcement return (CAR) to capture
the markets assessment of the restatement. I also include CEO controls that should
influence the turnover decision including the CEOs age (CEO AGE), the CEOs
tenure (CEO TENURE), and whether the CEO is also the chair of the board
(CHAIR).
5.4 Consequences of misreporting results
Descriptive statistics on the consequences of misreporting (untabulated) show
AAERs were enforced on 20 percent of revenue restatements in the sample and 31
percent of revenue restatement firms have CEO turnover in the 2 years following
the restatement. The mean announcement CAR is -10 percent consistent with the
findings in Palmrose et al. (2004).
Table 5 contains regression estimates for the consequences of misreporting tests.
The results are presented with the RRC SCORE complexity proxy only but are
similar when using WORDS and METHODS as proxy. The results for AAERs show
RRC SCORE is negatively associated with AAERs, indicating restatements
involving complex revenue recognition are less likely to receive an AAER,
consistent with the SEC recognizing the role of complexity in misreporting. The
results also show the SEC targets firms with irregularities (IRREG) and pervasive
restatement issues (MULTIPLE). Finally, CHREV has a significant negative
coefficient, which is expected if the SEC is more concerned with revenue
overstatements.
The results for announcement returns in Table 5 also show that firms with
complex revenue recognition have less negative announcement returns (coeff 0.029,
t-stat 3.05). The economic effect of complexity on returns is also significant. A one
standard deviation increase in RRC SCORE (1.22) increases announcement returns
by 3.5 percent. With an average market capitalization of $1.9 billion prior to the
restatement, the mean change in announcement return dollars is $68 million. The
results also show that understatements of revenue (CHREV) have higher
announcement returns, and restatements that are irregularities have much lower
123
90
K. Peterson
RRC SCORE
BIGN
CAR
Coeff
Z-stat
-0.534***
-3.22
0.869
Coeff
0.029***
CEO TURN
t-stat
3.05
Coeff
Z-stat
-0.305**
-2.15
1.33
MISSFCST
0.092
0.14
RESTLEN
-0.004
-0.11
AUDITOR
-0.313
-0.67
0.044
1.38
IRREG
4.892***
4.67
-0.097***
-4.40
1.012***
3.31
MULTIPLE
1.240***
2.76
-0.032
-1.43
0.469
1.43
0.62
-0.003
-0.51
-0.140
-1.52
3.24
-0.860
-0.82
LOGMVE
CHREV
CHNI
0.062
-4.001***
0.256**
PRERET
-3.02
2.32
0.428***
0.001
-0.038***
0.22
0.090
1.02
-3.55
-0.302
-1.06
POSTRET
-0.520*
-1.82
ROA
-1.257**
-2.26
CAR
-0.543
-0.79
CEO AGE
0.681
CEO TENURE
CHAIR
0.74
-0.057**
-2.18
-0.631**
-2.20
333
333
326
Pseudo R2/R2
0.375
0.187
0.193
This table contains logistic and OLS regression estimates to test if revenue recognition complexity affects
the consequences of restatement. AAER is an indicator equal to one if the firm has an SEC AAER related
to revenue or receivables within 2 years of the restatement announcement. CAR is the 5-day cumulative
abnormal return (market adjusted return) centered on the restatement announcement date. CEO TURN is
an indicator set to one if the CEO resigns or is terminated within 2 years of the restatement but excludes
CEO turnover where the former CEO retains a Chair or Director position. All other variables are
previously defined in prior tables. The CEO TURN regressions have seven observations with missing
CEO AGE, CEO TENURE, and CHAIR because the data was unavailable in proxy filings. Coefficients
on the intercept and industry indicators are included but not presented. Z-statistics (Logistic) or t-statistics
(OLS) are listed next to the coefficient, using Huber/White Robust standard errors with firm-level
clustering. *, **, and *** Indicate significance at 10, 5, and 1%
123
91
regressions as in Table 5 but includes interactions with the complexity proxies and
IRREG. As in Table 5, the results are only presented for RRC SCORE but are
similar when using WORDS and METHODS. To make it easier to understand the
interaction effects, the regression includes two interactions with complexity: one for
cases where IRREG is equal to zero and one for IRREG equal to one. For AAERs,
complexity reduces the incidence of AAERs for both irregularities and errors,
although the reduction is larger for errors than irregularities. Announcement returns
are also less negative for both irregularities and errors. In contrast, complexity
reduces the probability of CEO turnover only in the case of irregularities, suggesting
that managers can hide behind complexity when there is some indication of intent.
While the coefficient on the interaction with mistakes (IRREG = 0) and complexity
is insignificant, this may be due to the already low probability of CEO turnover for
mistakes in general. Collectively, these results suggest accounting complexity
tempers restatement consequences for both errors and irregularities.
6 Additional analysis
The existence of a restatement includes the sequential events of misreporting and
detection of the misreporting; therefore, modeling these events separately may yield
better parameter estimates relative to traditional logit estimation (Callen et al.
2009). The two-stage partial observability probit model allows such estimation
when only the combined event is observed. Results for tests of H1 when using this
model are consistent with the results presented in the paper (complexity coefficient
Z-statistics of 5.178.34).
Prior to SAB 101, firms had a choice to disclose their revenue recognition policy
if they thought it was a significant policy. Since my proxy for revenue recognition
complexity relies upon these disclosures, a positive association between complexity
Table 6 Consequences of misreporting regression estimates with irregularity interactions
AAER
Coeff
IRREG
CAR
Z-stat
5.117***
4.89
-1.242***
-4.30
RRC_SCORE (IRREG = 1)
-0.513***
-3.07
CEO TURN
Coeff
t-stat
-0.102***
-4.12
Coeff
Z-stat
1.130***
3.53
0.022*
1.88
-0.093
-0.37
0.033***
2.68
-0.407**
-2.45
Controls included
Yes
Yes
Yes
333
333
326
Pseudo R2/R2
0.378
0.188
0.197
This table contains logistic and OLS regression estimates to test if revenue recognition complexity affects
the consequences of restatement differently if the misstatement is intentional or unintentional. The
models are the same as those presented in Table 5, except I include interactions between the revenue
recognition complexity proxy (RRC SCORE) and IRREG. Coefficients on the intercept and other control
variables are included in the model but are not presented. Z-statistics (Logistic) or t-statistics (OLS) are
listed next to the coefficient, using Huber/White Robust standard errors with firm-level clustering. *, **,
and *** Indicate significance at 10, 5, and 1%
123
92
K. Peterson
and misreporting may be due to a regulation change. I conduct all the prior tests
after splitting the sample into pre- and post-SAB 101 restatements (fiscal years
2001). All results are consistent with the results presented in the paper except
coefficients on complexity are insignificant for the consequence regressions in the
pre-SAB 101 period; results remain consistent in the post-SAB 101 period. The
difference in results pre- and post-SAB 101 may imply that lack of disclosure
guidance in the pre-SAB 101 period caused firm disclosures to be less reliable
measures of the firms real revenue recognition polices, increasing noise in my
measures of complexity in the pre-period.
7 Conclusions
I investigate the effect of accounting complexity on misreporting using a setting of
revenue recognition complexity and revenue restatements. The results suggest that
in the case of revenue recognition, accounting complexity is a key factor in the
occurrence of misreporting. However, firm stakeholders temper the negative
consequences for misreporting when revenue recognition is complex. Given the
FASBs interest in faithfully representing complex transactions, these results help
inform the FASB on stakeholders reactions to misreporting resulting from
complexity. Future research could examine other effects of accounting complexity
besides misreporting.
Acknowledgments This paper is based on my dissertation at the University of Michigan. I appreciate
the guidance and advice of my dissertation committee members, Russell Lundholm and Ilia Dichev, and
especially my chair, Michelle Hanlon. Author also thankful to the following for helpful comments: David
Guenther, Angela Davis, Judson Caskey, Lian Fen Lee, K. Ramesh, Jeff Wilks, Cathy Shakespeare, Chad
Larson, Peter Demerjian, anonymous reviewers, and workshop participants at the University of Michigan,
Washington University (St. Louis), University of Oregon, and Northwestern University.
Appendix
Example revenue recognition disclosures
A.C. Moore Arts & Crafts, 2005 10-K [WORDS: 8; METHODS: 1; RRC SCORE: -1.19]
Revenue is recognized at point of retail sale.
UStel, Inc., 1997 10-K [WORDS: 9; METHODS: 1; RRC SCORE: -1.45]
Revenue is recognized upon completion of the telephone call.
Regal Entertainment Group 2004 10-K [WORDS: 161; METHODS: 4; RRC
SCORE: 0.14]
Revenues are generated principally through admissions and concessions sales with
proceeds received in cash at the point of sale. Other operating revenues consist
primarily of product advertising (including vendor marketing programs) and other
123
93
123
94
K. Peterson
Bergstresser, D., Desai, M., & Rauh, J. (2006). Earnings manipulation, pension assumptions, and
managerial investment decisions. Quarterly Journal of Economics, 121(1), 157195.
Burns, N., & Kedia, S. (2006). The impact of performance-based compensation on misreporting. Journal
of Financial Economics, 79, 3567.
Bushman, R., Chen, Q., Engel, E., & Smith, A. (2004). Financial accounting information, organizational
complexity and corporate governance systems. Journal of Accounting and Economics, 37(2),
139201.
Callen, J., Robb, S., & Segal, D. (2009). Revenue manipulation and restatements by loss firms. Auditing:
A Journal of Practice and Theory, 27(2), 129.
Collins, D., Kothari, S. P., & Rayburn, J. (1987). Firm size and the information content of prices with
respect to earnings. Journal of Accounting and Economics, 9, 111138.
Core, J., & Guay, W. (2002). Estimating the value of employee stock option portfolios and their
sensitivities to price and volatility. Journal of Accounting Research, 40(3), 613630.
Cox, C. (2005). Remarks before the 2005 AICPA national conference on current SEC and PCAOB
developments. Washington, DC, http://www.sec.gov/news/speech/spch120505cc.htm.
Dechow, P., Ge, W., Larson, C., & Sloan. R. (2007). Predicting material accounting misstatements.
Working Paper. University of Michigan and University of Washington.
Dechow, P., Sloan, R., & Hutton, A. (1996). Causes and consequences of earnings manipulation: An
analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting Research,
13, 136.
DeFond, M. L., & Jiambalvo, J. (1991). Incidence and circumstances of accounting errors. The
Accounting Review, 66, 643655.
Desai, H., Hogan, C., & Wilkins, M. (2006). The reputational penalty for aggressive accounting: Earnings
restatements and management turnover. The Accounting Review, 81(1), 83112.
Engel, E., Hayes, R., & Wang, X. (2003). CEO-turnover and properties of accounting information.
Journal of Accounting and Economics, 36, 197226.
Erickson, M., Hanlon, M., & Maydew, E. (2006). Is there a link between executive equity incentives and
accounting fraud? Journal of Accounting Research, 44(1), 113143.
Ertimur, Y., Livnat, J., & Martikainen, M. (2003). Differential market reactions to revenue and expense
surprises. Review of Accounting Studies, 8(23), 185211.
Ertimur, Y., & Stubben, S. (2005). Analysts incentives to issue revenue and cash flow forecasts, Working
Paper. Stanford University.
Financial Accounting Standards Board (FASB). (2008). Conceptual framework for financial reporting:
The objective of financial reporting and qualitative characteristics and constraints of decisionuseful financial reporting information. FASB exposure draft.
Freeman, R. (1987). The association between accounting earnings and security returns for large and small
firms. Journal of Accounting and Economics, 9, 195228.
Government Accountability Office (GAO). (2002). Financial statement restatements, trends, market
impacts, regulatory responses, and remaining challenges. GAO-03-138.
Government Accountability Office (GAO). (2006). Financial restatements: Update of public company
trends, market impacts, and regulatory enforcement activities. GAO-06-678.
Greene, W. (1997). Econometric analysis (3rd ed., pp. 339350). Upper Saddle River, NJ: Prentice Hall.
Hayn, C. (1995). The information content of losses. Journal of Accounting and Economics, 7, 85107.
Hennes, K., Leone, A., & Miller, B. (2008). The importance of distinguishing errors from irregularities in
restatement research: The case of restatements and CEO/CFO turnover. The Accounting Review,
83(6), 14871519.
Herz, R. (2005). Remarks before the 2005 AICPA national conference on current SEC and PCAOB
developments. Speech given by the FASB Chairman, Washington DC.
Herz, R. (2007). Specialized industry accounting and reporting. Too much of a good thing? Speech given
by the FASB Chairman, 26th Annual SEC and Financial Reporting Institute Conference.
Johnson, S., Ryan, H., Jr, & Tian, Y. (2009). Managerial incentives and corporate fraud: The sources of
incentives matter. The Review of Finance, 13(1), 115145.
Karpoff, J., Lee, D., & Martin, G. (2008). The Consequences to managers for financial misrepresentation.
Journal of Financial Economics, 88(2), 193215.
Kinney, W., & McDaniel, L. (1989). Characteristics of firms correcting previously reported quarterly
earnings. Journal of Accounting & Economics, 11(February), 7193.
Palmrose, Z., Richardson, V., & Scholz, S. (2004). Determinants of market reactions to restatement
announcements. Journal of Accounting & Economics, 37, 5989.
123
95
Picconi, M. (2006). The perils of pensions: Does pension accounting lead investors and analysts astray?
The Accounting Review, 81(4), 925955.
Plumlee, M. (2003). The effect of information complexity on analysts use of that information. The
Accounting Review, 78(1), 275296.
Plumlee, M., & Yohn, T. (2009). An analysis of the underlying causes attributed to restatements.
University of Utah and Indiana University Working Paper.
Schick, A., Gorden, L., & Haka, S. (1990). Information overload: A temporal approach. Accounting,
Organizations and Society, 15, 199220.
Securities and Exchange Commission (SEC). (2008). Final Report of the advisory committee on
improvements to financial reporting to the United States securities and exchange commissions.
August, 2008, link: http://www.sec.gov/about/offices/oca/acifr/acifr-finalreport.pdf.
Sondhi, A., & Taub, S. (2006). Miller revenue recognition guide. CCH Incorporated.
Turner, L. (2001). The times, they are a-changing. Speech by SEC chief accountant, 33rd rocky mountain
securities conference, Denver, Colorado. www.sec.gov/news/speech/spch493.htm.
Tversky, A., & Kahneman, D. (1974). Judgement under uncertainty: Heuristics and biases. Science, 185,
11241131.
Zhang, Y. (2006). An empirical analysis of revenue manipulation. Working Paper, Columbia University.
123
Copyright of Review of Accounting Studies is the property of Springer Science & Business Media B.V. and its
content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for individual use.