You are on page 1of 51

Measuring Accounting Fraud and Irregularities Using Public and Private Enforcement

Dain C. Donelson
University of Iowa
dain-donelson@uiowa.edu

Antonis Kartapanis
Texas A&M University
akartapanis@mays.tamu.edu

John McInnis
The University of Texas at Austin
John.Mcinnis@mccombs.utexas.edu

Christopher G. Yust
Texas A&M University
cyust@mays.tamu.edu

Abstract: Most accounting studies use only public enforcement actions (SEC cases) to measure
accounting fraud. However, private cases (securities class actions) also play an important
enforcement role. We discuss the legal standards and processes for both public and private
enforcement regimes, emphasize the importance of screening cases for credible fraud allegations,
and show both yield credible fraud measures. Further, we demonstrate these research design
choices affect inferences from prior research and a hypothetical research setting. Finally, we
show common measures of accounting irregularities using Audit Analytics to proxy for fraud
result in significant false positives and negatives and develop a fraud prediction model for use in
future research. We recommend using both public and private enforcement with appropriate
screening when examining accounting fraud to reduce Type I and II errors, or reporting the
sensitivity of findings across regimes. This is particularly important given the reduction in
accounting-related enforcement after 2005.

Keywords: Financial Reporting Fraud; Securities Litigation; AAER; SEC Enforcement;


Restatements; Irregularities; Fraud Prediction Models

JEL Codes: G38; K22; K41; K42; M41; M42; M48

We thank Christopher Armstrong (the editor), two anonymous referees, Matt Ege, Jen Glenn, Loren Jacobson,
Sudarshan Jayaraman, Jordan Neyland, Sarah Noor, Charles Wasley, Chris Wolfe, Joanna Wu, and workshop
participants at The University of Rochester, Texas A&M University, and the University of Texas at Austin for
helpful comments. We thank Rui Silva and Laura Kettell for research assistance. We gratefully acknowledge
research support provided by the Red McCombs School of Business and Mays School of Business. All errors are our
own.

Electronic copy available at: https://ssrn.com/abstract=3744392


I. INTRODUCTION

Due to its significant impact on the economy and accounting profession, numerous

studies examine the determinants and consequences of corporate accounting fraud (see Amiram

et al. 2018). 1 By accounting fraud, we mean the intentional, material misstatement of financial

statements that causes damages to investors. This definition is consistent with the construct most

fraud studies have in mind: senior management manipulating financial statements to inflate share

prices and deceive investors. However, there is no accepted measure of “true” accounting fraud

as admissions by companies and trials proving fraud are rare (Dyck, Morse, and Zingales 2010).

Thus, researchers studying fraud must rely on proxies for the underlying construct. The

dominant proxy is public enforcement through the Securities and Exchange Commission (SEC),

while relatively few studies use or include private enforcement via securities class actions

(SCAs). Because SEC cases and SCAs typically do not involve an admission of fraud, they can

probably best be characterized as “credible fraud allegations.” 2 Still, these proxies are much

closer to the fraud construct than other measures that do not require external evidence of a

misstatement or an assessment of fraudulent intent (e.g., abnormal accruals).

In this study, we examine the use of both public and private enforcement with appropriate

screening to measure accounting fraud and how this affects research inferences. Understanding

the dual public-private U.S. enforcement regime is crucial since we show that using only public

or private enforcement excludes fraud observations, leading to biased regression estimates and

reduced power. A key limitation of either regime alone is that neither pursues all credible cases.

The SEC has never had a sufficient budget to do so (Grundfest 1994), while private litigants lack

1
For example, the SEC notes that evidence on fraud determinants is “enormously important to us as regulators, as
we have a profound interest in minimizing the instances of fraud and harm to investors” (Richards 2008).
2
Despite this, we hereafter largely refer to these enforcement proxies, once screened, as “fraud” cases for simplicity.

Electronic copy available at: https://ssrn.com/abstract=3744392


incentives to pursue cases where costs exceed expected recovery (Cox, Thomas, and Kiku 2003).

We organize the study into three main parts. In the first part, we show that SEC cases are

the dominant proxy for accounting fraud. This likely began when many SCAs were (reasonably)

viewed as frivolous before the Private Securities Litigation Reform Act of 1995 (PSLRA) (Bohn

and Choi 1996). However, this view is now outdated. We review the relevant legal standards and

processes for fraud enforcement and make two contentions. The first is that, from a legal and

procedural perspective, settled SCAs are also a valid proxy for fraud. The second contention is

that researchers interested in accounting fraud should screen all cases to ensure construct validity

and reduce measurement error. Cases in both regimes often do not involve credible allegations of

accounting fraud because: a) both types of cases often involve non-accounting issues, b) the SEC

often brings cases that do not allege intent (as do some SCAs), and c) SCAs are often dismissed.

In the second part, we use fraud data from both regimes to support these two contentions.

We construct a sample of publicly available settled SEC cases and SCAs that allege accounting

fraud from 1998 to 2014. Interestingly, there is a significant decrease in both types of fraud

enforcement over time. This paradigm shift is critical and shapes much of our analysis due to the

lower power of smaller sample sizes. Related to our first contention, we show SCAs are rare and

do not inevitably occur after large stock price drops. Further, settled SCAs exhibit higher merit

proxies than dismissed cases, consistent with a now robust screening process for less meritorious

cases. We highlight several large, high profile SCAs, such as the Lehman Brothers “Repo-105”

case, which have credible fraud allegations but would be left out of an SEC-only sample. We

also find that settled SCAs and SEC cases that allege fraud target similar financial reporting

behavior: abnormal accruals and fraud propensity scores (i.e., F-scores from Dechow, Ge,

Larson, and Sloan 2011 [DGLS]). Additionally, consistent with Choi and Pritchard (2016), both

Electronic copy available at: https://ssrn.com/abstract=3744392


types are linked to higher executive turnover and lower institutional ownership after fraud

revelation. Thus, despite the low overlap between SCAs and SEC cases, both “look” like fraud.

Supporting our second contention above, settled SEC cases alleging intentional

misstatements have higher fraud propensity scores and restatements rates compared to SEC cases

that do not allege fraud, and they are associated with more adverse consequences such as

executive turnover. Thus, it is critical to also screen SEC cases for allegations of intentional

misstatements to increase the likelihood that researchers are capturing accounting fraud.

In the third part, we consider how these issues affect research inferences. Most prior

studies focus on enforcement using either regime in isolation (Schantl and Wagenhofer 2020),

resulting in many “false negatives.” Likewise, failing to screen out dismissed SCAs or SEC cases

that do not allege intent creates “false positives.” For binary variables, such as fraud indicators,

random misclassifications introduce measurement error and bias coefficient estimates toward

zero (Hausman, Abrevaya, and Scott-Morton 1998), increasing Type II errors. However, non-

random misclassifications (i.e., those correlated with the variable of interest) bias coefficient

estimates toward or away from zero, based on the correlation (Meyer and Mittag 2017),

increasing Type I errors. 3 Combining both enforcement measures mitigates both types of biases.

To illustrate these issues, we examine two research settings. The first is the effect of Big

N auditors on reducing accounting fraud. Similar to Lennox and Pittman (2010), we find a

negative relation between Big N auditors and SEC enforcement, consistent with Big N auditors

reducing fraud. However, Holzman, Marshall, and Schmidt (2019) show that firms with Big N

auditors receive less SEC investigation scrutiny. As a result, using only SEC cases could bias

inferences because the negative relation may be due to case selection, rather than fraud

3
For example, if the probability that the SEC does not pursue a case when fraud exists is correlated with the variable
of interest, this induces a statistical relation due to case selection, as opposed to firms’ fraud commission.

Electronic copy available at: https://ssrn.com/abstract=3744392


commission. When we use only SCAs or cases from both regimes to proxy for fraud, we find a

weakened or insignificant relation between Big N auditors and fraud, depending on the sample

period used. Thus, using an SEC-only sample may lead to Type I error.

For the second setting, we show that a viable, hypothetical research project – developing

an abnormal inventory model for misreporting, similar to Stubben (2010) – might be pursued but

abandoned due to a lack of statistically significant results. While we find positive and significant

results using SEC cases in earlier periods, we find insignificant results in recent periods, due to a

significant decline in SEC enforcement. However, our combined proxy for fraud yields positive

and significant estimates consistently over time. Thus, using a combined fraud measure can

reduce the risk of Type II error, particularly in the current enforcement environment.

We conduct two additional analyses to complement these findings. First, we examine

restatements, which have been used in their entirety or in subsets (e.g., “irregularities” in

Hennes, Leone, and Miller 2008 [HLM]) to proxy for fraud or related constructs (e.g., Ham,

Lang, Seybert, and Wang 2017; Hobson, Mayew, and Venkatachalam 2012). Many recent

studies rely on “fraudulent” restatements as identified in Audit Analytics (e.g., Hobson et al.

2012; Hobson, Mayew, Peecher, and Venkatachalam 2017; Brown, Crowley, Elliot 2020). We

find that many of these restatements lack credible fraud allegations from public or private

enforcement and appear to be false positives (i.e., they have lower merit proxies, garner less

market reaction, and often involve actions of lower level employees). Further, many enforcement

cases have no related restatement and are thus excluded from restatement samples (i.e., false

negatives). Thus, studies using restatements to proxy for fraud could benefit by following our

recommended approach to reduce false positives and negatives.

Second, in the spirit of DGLS, we create a fraud prediction model using our combined

Electronic copy available at: https://ssrn.com/abstract=3744392


measure of fraud and both the accounting-based SEC fraud prediction variables from DGLS and

the primarily market-based SCA litigation risk variables from Kim and Skinner (2012) [KS].

Consistent with public and private enforcement both capturing fraud, we find cross-regime

predictive ability: DGLS variables derived from SEC cases predict fraud measured with SCAs,

and vice versa. We also show this expanded model increases explanatory power while reducing

both false positives and false negatives. We provide this model primarily to assist researchers

who wish to control for fraud probability, similar to the way many researchers currently use the

F-score (e.g., Chan, Chen, and Chen 2013; Bradley, Gokkaya, Liu, and Xie 2017).

This study makes five contributions. First, we provide a framework to consider when

evaluating fraud proxies for construct validity. Many proxies need to be broadened (to include

private enforcement) or narrowed (to exclude cases that do not allege fraud). For example,

Karpoff, Koester, Lee, and Martin (2017) [KKLM] use a large sample of SEC and Department of

Justice (DOJ) cases as a benchmark to compare other “financial misconduct” databases. This

sample is thus both under-inclusive, as it excludes SCAs, and over-inclusive, as it starts with

Section 13(b) cases that do not require material or intentional misstatements. An open question is

“To what extent does our current system in both public and private enforcement systematically

commit more Type I than Type II errors?” (Amiram et al. 2018, 737). Our framework shows

both regimes commit these errors and how researchers can minimize their effect.

Second, we provide comparative empirical evidence on public and private enforcement

regimes. Our analysis suggests that many of the criticisms of SCAs appear outdated because,

over our sample period, both private and public enforcement cases, with proper screening,

exhibit similar merit proxies. Thus, there is no clear reason to focus only on public enforcement,

even though this is the most common design choice in the accounting literature. In fact, the use

Electronic copy available at: https://ssrn.com/abstract=3744392


of public enforcement only as a proxy for fraud is increasingly problematic due to a dramatic

decrease in the SEC’s focus on accounting fraud in recent years.

Third, we show that a fraud proxy based on both public and private enforcement can

affect inferences in common research settings. Our main takeaway is not that one enforcement

type is “better,” but that each is incomplete in isolation. This takeaway complements the recent

analytical study by Schantl and Wagenhofer (2020, 6), who argue that it is important to jointly

consider public and private enforcement to “focus on overall deterrence of misreporting.” We

also encourage researchers to assess and report the sensitivity of results to each enforcement type

in isolation since results obtained from only one regime may indicate a selection bias. 4

Fourth, we show that the overlap between restatements and fraud is low and demonstrate

that filtering restatements – even “fraudulent” restatements – from databases such as Audit

Analytics significantly improves their merits as fraud proxies. Moreover, the use of such

restatements as a proxy for fraud results in significant false negatives. These findings should be

helpful for future research interested in using restatements to proxy for fraud. Finally, we

develop a fraud prediction model that can be used by future research and show cross-regime

predictive ability of factors that have been shown to predict only public or private enforcement.

Our study has some similarities to KKLM, but there are critical differences. Most

importantly, the conclusions differ. KKLM question the usefulness of combining databases (p.

146) and state that researchers should use the individual database that matches their research

question (p. 159), with a tacit endorsement of their database as the preferred “financial

misconduct” measure. We agree that matching data and research question is important. However,

for studies on accounting fraud in general, we strongly caution against using any single database

4
This is also important as firms may take actions to minimize the risk of a certain type of enforcement. For example,
firms can minimize scrutiny from plaintiffs’ lawyers by bundling disclosures (Bliss, Partnoy, and Furchtgott 2018).

Electronic copy available at: https://ssrn.com/abstract=3744392


that focuses on one enforcement regime and/or does not screen for credible fraud allegations.

II. PRIOR LITERATURE AND LEGAL BACKGROUND

Fraud Proxies in Recent Research

We start by reviewing all articles on accounting fraud from 2006-17 in: The Accounting

Review, Contemporary Accounting Research, Journal of Accounting and Economics, Journal of

Accounting Research, and Review of Accounting Studies. 5 Online Appendix A lists these 46

studies and the fraud proxy used. 54 percent use SEC cases but no SCAs, 9 percent use SCAs but

no SEC cases, 20 percent use a combination of SEC cases and SCAs, and 17 percent use only

restatements. 6 Overall, SEC cases alone are by far the most common proxy in accounting.

SEC enforcement is the dominant fraud proxy based on the historical view that SCAs are

low quality (e.g., Bohn and Choi 1996; DGLS). For example, KS (293) state that most SCAs

relate to allegations “that are not sufficiently serious to warrant SEC enforcement actions.”

While this concern may have been valid before the PSLRA, as we discuss below, it is outdated. 7

Most legal scholars concur the PSLRA strengthened SCA merits (e.g., Johnson, Nelson, and

Pritchard 2007). Another potential reason for the prior focus on SEC enforcement was the SEC’s

previous commitment to targeting accounting fraud in earlier periods (see Levitt 1998).

However, as discussed in more detail in Section V, that commitment has changed in recent years.

Legal Standards Potentially Relevant to Accounting Fraud

In this subsection, we discuss the legal framework relevant to accounting fraud. As a

5
We include related concepts such as irregularities but not generic “earnings management.” We search articles using
keywords such as “fraud,” “securities class”, and “AAER.” We read the papers and exclude cases where the authors
do not examine the construct of fraud. In some cases, we use judgment. For example, Desai, Hogan, and Wilkins
(2006) examine how earnings restatements affect management turnover. However, they refer to fraud several times
and state that the study is motivated by “the lack of evidence…about adverse consequences to the manager of firms
that have restated earnings (GAAP violations) or committed other types of corporate fraud” (108, emphasis added).
6
Restatements alone are not indicative of fraud as discussed in more detail in Section VI.
7
The low regard for private enforcement in accounting is in stark contrast to finance, where it is seen as critical in
financial market development (e.g., La Porta, Lopez-de-Silanes, and Shleifer 2006; Jackson and Roe 2009).

Electronic copy available at: https://ssrn.com/abstract=3744392


practical matter, only credible allegations of violations of Section 10(b) or Rule 10b-5 are related

to the construct of accounting fraud. Common law fraud requires: an (1) intentional, (2) material

(3) misstatement that (4) causes (5) damage (Prentice 2009). For accounting fraud, the critical

elements are misstatement, intent, and materiality because causation and damage are less

commonly contested. Online Appendix B summarizes potentially relevant legal standards.

Section 10(b) of the Securities Exchange Act of 1934 (’34 Act) and related Rule 10b-5,

which was issued by the SEC under its rule-making authority to implement Section 10(b), are

broad anti-fraud remedies. The intent standard for these provisions is scienter, which is a

knowing misstatement or recklessness. These provisions specifically require materiality. Thus,

credible allegations of violations of Section 10(b) or Rule 10b-5 are inherently related to fraud.

Section 17(a) of the Securities Act of 1933 (’33 Act) is similar to Section 10(b) and,

unlike many ’33 Act provisions, applies to after-market transactions. However, there are two

crucial differences. First, Section 17(a) is enforceable only by the government. Second, only

subsection (1) requires scienter, while negligence satisfies the other two subsections (Block

1981). Thus, alleged violations of Section 17(a) must be screened for allegations of intentional

misconduct. As a practical matter, SEC cases rarely cite Subsection (1) of 17(a), particularly

without also citing Section 10(b) and/or Rule 10b-5. 8

Similarly, only the government may enforce the internal control and books and records

provisions of the FCPA, which amended the ’34 Act (Section 13(b)). These sections require

firms to keep “accurate” books and records and maintain controls to prevent misappropriation

and misstatements. Two parts of the FCPA do not require scienter, the third part prohibits

knowing control avoidance and misstatements, and none requires materiality (Prentice 2009).

8
In fact, in our sample (described in Section III), we do not exclude a single case by screening only on Section
10(b)/Rule 10b-5 that would have been included had we also screened on subsection (1) of Section 17(a).

Electronic copy available at: https://ssrn.com/abstract=3744392


Thus, violations of Section 13(b) alone are insufficient to proxy for accounting fraud. 9

Public and Private Enforcement

Securities laws are enforced by public and private actors. The SEC is responsible for

public civil enforcement, while criminal enforcement is handled by the DOJ. Private

enforcement is available to any market participant meeting the statutory qualifications, which is

generally a transaction in the relevant securities combined with a loss caused by the defendant.

Public Enforcement

The SEC’s enforcement process begins with informal investigations, which are converted

to formal investigations if warranted (Karpoff, Lee, and Martin 2008). The SEC has subpoena

power, giving it access to private information. The SEC also has discretion regarding defendants’

rights because it can pursue enforcement either in federal court or administrative proceedings.

Administrative proceedings are tried before an administrative law judge (ALJ), an SEC

employee. While appeals of ALJ decisions are allowed, they rarely succeed because defendants

must show the decision is outside the “range of reasonable interpretations” (Rakoff 2014, 10). 10

Like SCAs, defendants in SEC cases can move for case dismissal. However, the standard

for SEC cases to avoid dismissal is not as high (see Federal Rules of Civil Procedure 12(b)(6)

and 9(b)). Thus, while some cases are dismissed, most SEC cases are settled when announced.

Private Enforcement

Private enforcement of securities laws has long existed alongside public enforcement.

Due to concerns over frivolous lawsuits, Congress passed the PSLRA, which includes procedural

9
These provisions have been criticized for giving the SEC excessive leverage, as nearly any firm could be liable for
accounting errors or control failures (Karmel 2005). Alleged Section 13(b) violations form the main KKLM sample,
but are poor proxies for fraud without screening due to the lack of materiality and often intent requirements. Amiram
et al. (2018, 734) agree that 13(b) violations are “financial misrepresentation” and “fraud would be an inaccurate
characterization.” However, many researchers do not appear to recognize this distinction.
10
Recent Supreme Court decisions ruled the SEC’s appointment process for ALJs unconstitutional and constrained
the SEC’s disgorgement powers, limiting its overall enforcement powers (Hackbrath and Stedman 2018).

Electronic copy available at: https://ssrn.com/abstract=3744392


obstacles beneficial to defendants. First, there is a stay on discovery until cases survive a motion

to dismiss. Thus, plaintiffs’ attorneys largely build cases from public information or confidential

witnesses (see Choi 2007). Also, a motion to dismiss invokes a strict pleading standard requiring

detailed factual allegations that provide a “strong inference” of fraudulent intent (i.e., scienter).

In an accounting case, this means the plaintiff must allege how GAAP was violated with specific

allegations to convince the judge that the misstatement was intentional, which is difficult without

discovery. Thus, about 40 percent of SCAs are dismissed (Donelson, McInnis, and Mergenthaler

2012), compared to 4 percent of SEC cases in federal court and 1 percent of SEC cases in

administrative courts (Zheng 2016). Like SEC cases, SCAs frequently do not allege GAAP

misstatements, so they should be screened for researchers interested in accounting fraud.

Comparing Public and Private Enforcement

There are several similarities between public and private enforcement, such as: 1) most

cases settle, 2) settlements almost never result in fraud admissions, 3) neither attempt to target all

cases of fraud (i.e., both have significant false negatives), and 4) insurance is involved in

settlements. 11 However, there are several important differences, particularly the power of the

SEC compared to private litigants, that may result in settlements in SEC cases where a

comparable SCA would fail. The SEC: 1) can pursue technical, immaterial violations through the

Foreign Corrupt Practices Act (FCPA), 2) has historically brought cases before its own ALJs

instead of in federal court, and 3) faces a lower pleading standard and, thus, has far lower

dismissal rates. The SEC also has subpoena power to compel discovery in ways that private

11
Directors’ and officers’ (D&O) insurance covers litigation costs, such as attorney’s fees, for both types of cases as
long as there is no fraud admission. SEC settlements classified as fines, penalties or disgorgement of ill-gotten gains
are not covered, while insurance covers nearly all SCA settlements (Donelson, Hopkins, and Yust 2015). Also, both
types of enforcement have been criticized for agency problems. While commonly discussed for SCAs (Coffee
1986), the SEC has “no real clients who must pay the marginal costs of pursuing weak or nonexistent claims,” and,
due to their resources, “even the most innocent client will find it in its best interests to settle” (Macey 2010, 659).

10

Electronic copy available at: https://ssrn.com/abstract=3744392


litigants cannot (Choi and Pritchard 2016). Further, the SEC can gain leverage by coordinating

with other parts of the federal government, unlike private litigants (see Macey 2010). 12

From the above discussion, we make two key contentions. First, there is no clear reason

that SCAs that pass a motion to dismiss and settle are a poor proxy for fraud, relative to SEC

cases. Thus, the construct of fraud should be broadened to include private enforcement. Second,

researchers interested in accounting fraud should narrow the scope of both SEC cases and SCAs

to ensure they allege intentional GAAP violations. Relatedly, researchers should screen out

dismissed cases due to a lack of credible fraud allegations. In the next section, we detail sample

construction using both public and private enforcement and provide descriptive statistics. In

Section IV, we offer empirical evidence to support these two contentions.

III. SAMPLE AND DESCRIPTIVE STATISTICS


Sample Selection

For our sample, we collect SEC cases and SCAs with fraud periods ending between 1998

and 2014. Starting in 1998 ensures a consistent legal environment since it is the year all SCAs

alleging fraud were required to be filed in Federal courts under the Securities Litigation Uniform

Standards Act. Ending in 2014 allows several years for SEC cases to be filed and SCAs to settle

or be dismissed. 13 We require both SEC cases and SCAs to be settled, involve Rule 10b-5 (fraud)

allegations, and allege GAAP violations. We obtain SCAs from the Stanford Securities Class

Action Clearinghouse (SCAC) and SEC cases primarily using CFRM from DGLS. 14 We then

12
Further, the SEC routinely uses “no admit/no deny” settlement terms, which its officials claim enhances efficiency
(see Ceresney 2013). However, as discussed, the SEC has the power to force settlement in nearly any case. The
potential to force settlements in non-meritorious cases was the reason for the PSLRA in 1995. However, while SEC
enforcement actions can yield false positives for fraud, it can also yield false negatives (e.g., the Madoff scandal).
13
As reported in Cox, Thomas, and Kiku (2005), most SEC and SCA cases require two to three years to be resolved,
although some take considerably longer (see Solomon and Soltes 2019).
14
If there are multiple SEC cases or SCA complaints related to the same underlying misconduct, we ensure that the
group collectively contains at least one instance of GAAP violations and Rule 10b-5 allegations and one settles.
Because the last AAER in CFRM was filed on September 30, 2016, we supplement it with SEC cases and details

11

Electronic copy available at: https://ssrn.com/abstract=3744392


identify Rule 10b-5 allegations through keyword searches and then manually identify GAAP

violations for SCAs, as discussed in Online Appendix C. We report the construction of our main

public and private enforcement samples in Table 1. We use financial statement data from

Compustat; stock market data from CRSP; restatement data from Audit Analytics, supplemented

with the GAO data relating to irregularities as defined in HLM; executive turnover data from

BoardEx; analyst data from IBES; and institutional ownership data from Thomson Reuters.

Panel C depicts changes in enforcement over time by reporting the number of annual

cases using the final year alleged in each case. 15 There is a dramatic decrease in enforcement in

the latter half of our sample, more so for SEC cases. While the average number of annual fraud

firm-years subject to SCAs fell significantly in this period (64 percent, untabulated), the number

of annual fraud firm-years subject to SEC enforcement actions nearly disappeared, resulting in

only 28 firm-years (a 79 percent decrease, untabulated). As discussed in Section V, the decrease

in SEC cases appears to be largely due to shifting priorities over time. However, while not the

focus of this study, it is also possible that underlying fraud commission has also decreased over

time, which would explain at least part of the decrease in enforcement across both regimes. 16

Descriptive Statistics

Table 2 provides descriptive statistics for our main samples of screened SCAs without

SEC enforcement (Panel A), SEC cases without a SCA (Panel B), cases with both types of

from Advisen and the SEC website to identify all fraud years in our sample. Advisen also allows us to supplement
missing information, such as CIK and fraud period, from CFRM, but the additional cases in Advisen are also
available on the SEC’s website. Our sample of SEC cases differs significantly from KKLM’s (unreleased) primary
sample of cases alleging Section 13(b) violations because not all of their cases allege GAAP misstatements and/or
fraud (i.e., Rule 10b-5 violations). Our later tests using all SEC cases (Table 7) indicate about 34 percent of
accounting-related SEC cases do not allege fraud, similar to KKLM in their Table 6.
15
We report them by the final year in the class period, rather than the filing year, to hold the scrutinized enforcement
years relatively constant as SEC cases are filed years after both the fraud ends and the SCA is filed (see KKLM).
16
The decrease in SCAs may also be partly attributable to the decrease in SEC enforcement since the presence of
SEC investigations or enforcement can help plaintiffs develop cases (Johnson et al. 2007; Donelson et al. 2015).

12

Electronic copy available at: https://ssrn.com/abstract=3744392


enforcement (Panel C), and cases that involve either type of enforcement (Panel D). Figure 1

presents a Venn diagram to represent the observations across these four panels. Variables are

defined in detail in Appendix A. For benchmarking purposes, we include descriptive statistics for

the Compustat population for the same time frame (Panel E).

We highlight three things from Table 2 and Figure 1. First, the overlap in settled

accounting SEC and SCAs is small (236 cases or about 21 percent). Second, using only SEC

cases with Rule 10b-5 violations (408 observations in Panels B and C) cuts the sample by nearly

65 percent, compared to a proxy that captures either regime. This drastic reduction in fraud

observations has the ability to significantly reduce power, as we discuss further in Section V.

Third, there are key differences between firms targeted in SEC cases and those targeted

in SCAs. Firms facing only SCAs: 1) are larger, 2) more profitable (ROA), and 3) have higher

maximum damages (p < 0.01). Firms facing only SEC cases: 1) are more financially distressed

(p < 0.01) and 2) have higher book-to-market ratios (p < 0.05). Thus, SCAs appear relatively less

likely to target smaller distressed firms, likely because such cases are not profitable (Coffee

2006; Baker and Griffith 2010), while the SEC is more likely to target these firms (Cox et al.

2003). Cox et al. (2003) speculate the SEC focuses on these firms because of its concern that

investors have a high likelihood of being harmed and the visibility of such harm. 17

We further illustrate this differential selective enforcement by examining two recent,

high-profile accounting fraud scandals: stock option backdating and Chinese Reverse Mergers

(CRM). As Table 3 shows, the SEC handled 31 settled backdating cases versus only 28 settled

SCAs. The limited number of SCAs was the result of relatively small investor losses in many

backdating cases, which meant that many were not worth pursuing as SCAs. While the SEC and

17
Cox et al. (2003) also note that these results are consistent with the SEC pursuing weak opponents.

13

Electronic copy available at: https://ssrn.com/abstract=3744392


SCAs handled a similar number of backdating cases, this enforcement ratio differs substantially

from the total enforcement ratio in our sample, where SCAs outnumber SEC enforcement actions

by a ratio of approximately 2.4 to 1 (970 to 408; see Figure 1). Said more directly, the SEC was

relatively more important at pursuing accountability for backdating misconduct.

In contrast, the SEC did not vigorously pursue CRM cases. Instead, activist short-sellers

primarily uncovered CRM frauds, which resulted in large investor losses (Chen 2016), so SCAs

were profitable to pursue. Thus, there were 35 settled SCAs but only eight settled SEC actions

related to CRMs, a ratio of roughly 4.4 to 1. Thus, SCAs were relatively more important in

pursuing CRM enforcement. As a result, researchers using only SEC cases (SCAs) would have

missed a large proportion of CRM (backdating) fraud. The fact that enforcement is non-random

raises important concerns about Type I error, which we discuss in Section V.

IV. SETTLED PUBLIC AND PRIVATE CASES AS A PROXY FOR FRAUD

In this section we offer empirical evidence to support the two contentions from the

discussion in Section II: 1) settled SCAs are a valid proxy for fraud and 2) settled SEC cases

should be screened for intent by researchers interested in fraud.

Contention 1: Securities Class Actions are a Valid Fraud Proxy

We first use the full SCA sample to show that they are rare and dismissal motions screen

out weaker cases. Table 4, Panel A reports firms experiencing one-day market-adjusted stock

price drops of at least 10 percent face a SCA only 4 percent of the time, and only 1 percent of

such firms face SCAs that allege accounting fraud and settle. Panel B reports that even when we

examine one-day drops of at least 20 percent and market value declines of at least $100 million,

only 6 percent of firms experience settled SCAs that allege accounting fraud. This is inconsistent

with the claim in DGLS (25) that “lawsuits…are also very common after a stock has experienced

14

Electronic copy available at: https://ssrn.com/abstract=3744392


a precipitous price decline, even when there is no clear evidence supporting the allegation.”

Instead, settled SCAs alleging accounting fraud are now rare following large stock price drops. 18

As shown in Table 5, motions to dismiss filed by defendants screen out less meritorious

SCAs that are unlikely to represent fraud. Compared to dismissed accounting SCAs with Rule

10b-5 allegations, settled SCAs have higher damages and are more likely to have concurrent

SEC enforcement (Panel A). In terms of proxies for aggressive accounting (Panel B), about 39

percent of dismissed cases involve restatements versus 54 percent of settled cases (p < 0.01).

Also, settled cases have higher accruals (p < 0.10) and F-scores (p < 0.01) from DGLS.

Panel C shows that adverse consequences are higher for settled cases: the CEO and CFO

are more likely to leave the firm (p < 0.10); the increase bid-ask spread is higher (p < 0.01); and

the decreases in forecast accuracy (p < 0.01) and institutional ownership are larger (p < 0.05).

Overall, the legal system appears to screen out less meritorious, non-fraudulent cases. Thus, it is

important to exclude dismissed litigation given that many papers that use SCAs in whole or part

to proxy for fraud or related constructs do not exclude dismissed SCAs from their samples (e.g.,

McNichols and Stubben 2008; Chhaochharia, Kumar, and Niesen-Ruenzi 2012; KKLM).

We next show that properly screened settled SCAs and SEC cases target similar behavior.

In Table 6, we compare our sample of SCAs to SEC cases on proxies for aggressive accounting

during the fraud period (Panel A) and adverse consequences when the fraud is revealed (Panel

B). Moving from left to right, we start with the full Compustat population as a benchmark (1st

Column) and next tabulate: 1) only settled SCAs (2nd Column), 2) only settled SEC cases (3rd

18
The number of accounting-based SCAs with Rule 10b-5 allegations reported in each year is far below the number
in KKLM (see their Table 2, Panel B), despite the fact that KKLM state that they similarly examine litigation that
involves accounting fraud allegations. This difference is both due to the fact that KKLM does not exclude dismissed
litigation and appears to be using all SCAs. The clearest year to see the apparent lack of screening is 2001, when
hundreds of IPO laddering cases were filed, which did not allege accounting fraud (see KS). KKLM report a spike in
annual filings, from 219 in 2000, to 504 in 2001 which is possible only if IPO laddering cases are included.

15

Electronic copy available at: https://ssrn.com/abstract=3744392


Column), and 3) both types of cases (4th Column). Overall, if SCAs and SEC cases with credible

fraud allegations are both reasonable proxies for fraud, as we contend, these observations should

look relatively similar to each other and dissimilar to Compustat. We find such evidence.

For example, Panel A indicates that during the fraud period, firms facing only SCAs have

significantly higher abnormal accruals, performance-adjusted abnormal accruals, and F-Scores (p

< 0.01) than the Compustat population. SCAs also have a restatement rate (44 percent) roughly

four times higher than the Compustat average (p < 0.01). Firms facing only SEC actions have

abnormal accruals that are insignificantly different from the Compustat population but

significantly higher F-scores and restatement rates (p < 0.01). Firms facing both types of

enforcement exhibit more aggressive accounting than Compustat firms across all four proxies (p

< 0.10). Most importantly for our study, firms facing only SCAs have aggressive accounting

measures that look much closer to firms facing only SEC actions (Column 2 versus 3) than the

Compustat population (Column 2 versus 1), where the fraud rate is lower. 19

Panel B examines adverse consequences after the fraud ends for SEC cases and SCAs,

similar to Choi and Pritchard (2016), and we reach similar conclusions. They compare the

consequences of firm-disclosed SEC investigations to filed SCAs. They do not require cases

from either regime to involve alleged accounting fraud, the SEC to issue an enforcement action

or cases from either regime to settle. Thus, it is important to examine adverse consequences in

our setting of accounting-related settled SEC cases and settled SCAs that allege fraud.

When fraud ends, firms facing only SCAs have significantly higher CEO and CFO

19
Some of these findings appear to conflict with KKLM, who examine how accruals and other firm characteristics
change following SCAs and AAERs in their Table 1. As discussed in Online Appendix D, it appears that the
differences reported between SCAs and SEC cases in KKLM are not attributable to differences in the fraud proxies
underlying each database but rather to the event date they use. Specifically, KKLM fail to find significant results
because they examine changes around the AAER issuance date, which is often years after the fraud ends.

16

Electronic copy available at: https://ssrn.com/abstract=3744392


turnover, increases in bid-ask spreads, and decreases in analyst forecast accuracy and

institutional ownership (p < 0.01) compared to the Compustat population. We see similar, but

weaker, patterns for firms facing only SEC cases. Except for increases in bid-ask spreads and

decreases in institutional ownership being higher for SCAs (p < 0.05), there are no significant

differences in consequences between SCAs and SEC cases. Thus, like Panel A, SCAs look

similar to SEC cases and dissimilar to the population of firms as a whole.

Nonetheless, given the common claim that SCAs target less meritorious conduct, we also

highlight in Online Appendix E several high profile SCAs that would be excluded from a sample

using only SEC fraud cases, despite credible evidence that the firms engaged in accounting

fraud. These cases received significant media coverage, settled for at least $150 million, and

involved high profile firms, such as Lehman Brothers, AIG, Sears, Bank of America, and others.

Contention 2: Screening SEC Cases for Intent

Many papers examining fraud use all SEC cases (e.g., Caskey and Hanlon 2013; Li

2016), perhaps assuming the SEC would not bring weak cases, so requiring intent to be alleged is

unnecessary (see Beneish 1997). We examine this issue by testing whether settled SEC cases

alleging Rule 10b-5 violations appear more indicative of fraud than those that do not. We use the

same proxies for aggressive accounting and adverse consequences from prior tables. Results are

presented in Table 7. Firms with settled SEC cases alleging Rule 10b-5 violations have higher F-

Scores and are more likely to restate (p < 0.01), have higher CEO and CFO departures (p <

0.05), and have larger spread changes (p < 0.10) and decreases in institutional ownership (p <

0.01). 20 Thus, if focusing on the construct of fraud, it is important to screen SEC cases.

20
We focus on SEC cases in this contention since the vast majority of accounting-related SCAs involve Rule 10b-5
allegations (KS). However, we find similar evidence comparing SCAs with and without Rule 10b-5 violations, and
SCAs with Rule 10b-5 violations are more likely to settle and result in higher expected payoffs (untabulated).

17

Electronic copy available at: https://ssrn.com/abstract=3744392


Consistent with this, we discussed our findings with a former senior SEC lawyer, who confirmed

the SEC chooses whether to allege fraud based on case merits and if it believes fraud occurred.

Overall, both SEC cases and SCAs, when properly screened, are credible proxies for

fraud. However, focusing on only one enforcement regime to measure fraud is likely problematic

and can predictably affect empirical inferences, which we explore next.

V. PROXIES AND RESEARCH INFERENCES

Potential for Type I and Type II Errors

To demonstrate the implications of our findings, we consider how the measurement of

fraud using public and private enforcement affects Type I and II errors (Amiram et al. 2018).

Focusing on one regime classifies valid fraud observations as non-fraud, creating false negatives

(Figure 1). Likewise, failing to screen out cases that do not allege intent and/or are dismissed

creates false positives (Tables 5 and 7). Random misclassifications due to false negatives/

positives in a binary fraud variable cause attenuation bias, whether it is the independent (Aigner

1973) or dependent variable (Hausman et al. 1998), increasing Type II errors, regardless of

whether a linear probability model or a logit/probit is used (Meyer and Mittag 2017). 21

Further, if the misclassification is non-random and is correlated with the covariates,

coefficient estimates are biased in any direction, toward or away from zero, depending on the

direction of the correlations (Meyer and Mittag 2017). This increases the risk of Type I error.

Unfortunately, there is no foolproof way to eliminate either of these biases. One approach is to

try to jointly model the fraud commission and enforcement/selection processes (e.g., Wang

2013). However, this is challenging because it requires proper specification of these processes,

21
This contrasts with the usual effects of measurement error in a continuous dependent variable, where the error
becomes part of the residual, which lowers R2 and increases standard errors but does not bias coefficients (Greene
2003). This attenuation bias for binary dependent variables is because the error is unidirectional (i.e., a true “1” can
only be misclassified downward as a false negative and vice versa).

18

Electronic copy available at: https://ssrn.com/abstract=3744392


which are largely unobservable and change over time (see Cox et al. 2005). Relatedly,

identification hinges on finding plausibly exogenous instruments that affect fraud enforcement,

but not commission, and vice versa.

We advocate using properly screened fraud observations from both regimes to mitigate

bias. For random misclassifications, measuring fraud when either a public or a private case exists

reduces the risk of false negatives and proper screening reduces false positives. This reduction to

random misclassifications mitigates attenuation bias and reduces Type II error risk. For non-

random misclassifications, using observations from both regimes, both on a combined and

separate basis, can mitigate effects of selection bias within one regime or across regimes. 22

Demonstration of Type I and II Errors

To provide evidence on these points, we examine two research settings, one from a

published study and one from a hypothetical study. In both settings, we start with SEC cases as

the only fraud proxy, because this is the dominant choice in the accounting literature as shown in

Online Appendix A. The first setting is the effect of Big N auditors on fraud (Lennox and

Pittman 2010). There is an ex-ante reason to suspect non-random misclassification in this setting,

which raises Type I error concerns. The second setting is a hypothetical study of how an

abnormal inventory measure may signal increased fraud risk. In this setting, random

misclassification is more likely, which raises concerns about low power and Type II error. We

use a hypothetical study because Type II error leading to “null results” likely leads many

22
For example, if private litigants are less likely to target distressed firms, this could induce a relation between
proxies correlated with distress and private enforcement. If the SEC does not have this selection bias or pursues the
opposite strategy of targeting distressed firms, combining observations across regimes can help attenuate a
potentially spurious association between distress and fraud enforcement. However, we would expect similar non-
random and random misclassifications in studies that only use SCAs.
This approach also has limitations. If random false negatives exist even after considering both public and
private enforcement, attenuation bias will still exist. Also, if selection bias related to a variable of interest is similar
in both regimes, combining fraud observations will not help mitigate this bias.

19

Electronic copy available at: https://ssrn.com/abstract=3744392


researchers to abandon projects due to publication bias (see Hubbard and Armstrong 1997;

Franco, Malhotra, and Simonovits 2014). We examine if researchers in our hypothetical study

could have obtained significant results.

In addition to examining these settings over our entire sample period, we split the sample

into two time periods: 1998-2005 and 2006-2014. We perform this split due to the significant

changes in public and private enforcement over time (Table 1, Panel C). The first time period

captures a time when accounting issues were a relatively high priority at the SEC, first under the

leadership of Arthur Levitt (see Levitt 1998; Eichenwald 2005) and then in the aftermath of the

Enron/WorldCom crises. 23 However, SEC priorities shifted to other issues including the

financial crisis and Ponzi schemes after 2005 (see Choi, Wiechman, and Pritchard 2013). This

illustrates another danger of relying only on SEC enforcement to measure fraud as it can be

significantly influenced by the agency’s (largely unobservable) priorities (see Choi et al. 2013). 24

While SCAs also decreased in the post period, the decrease is less dramatic in percentage terms.

More broadly, the significant decrease in both types of enforcement will result in reduced

power due to the increase in fraud false negatives and may increase the risk of spurious results.

Absent the increased power from combining these datasets, many future scholars may abandon

valuable research projects due to insignificant results. We discuss this in more detail below.

Setting 1 – Big N Auditors and Fraud Risk

The first setting we examine is whether having a Big 5 auditor is associated with a lower

probability of committing accounting fraud. Lennox and Pittman (2010) focus on accounting

23
Richard Walker (1999), the Director of the SEC Division of Enforcement, announced “combating financial fraud
was the Division of Enforcement's number one priority” during this period, and the SEC announced 30 enforcement
actions on the anniversary of Levitt’s “Numbers Game” speech to show commitment to reducing accounting fraud.
24
Similarly, as noted by former SEC Chair Mary Jo White (2014), the DOJ had not actively pursued criminal
accounting fraud prosecutions before the earlier period, and Eisinger (2017) finds that such prosecutions essentially
ended after 2005. Thus, the threat of both civil and criminal public enforcement changed substantially over time.

20

Electronic copy available at: https://ssrn.com/abstract=3744392


fraud in part because such fraud was the catalyst for major legislative and regulatory changes

(e.g, the Sarbanes Oxley Act [SOX] and the PCAOB). Using SEC cases that allege fraud from

1981-2001, they find a negative relation between having a Big N auditor and subsequently being

targeted by SEC enforcement, consistent with Big N auditors reducing fraud risk.

However, SEC enforcement selection could lead to non-random misclassification of false

negatives. Holzman et al. (2019) find that firms with Big N auditors receive less SEC

investigative scrutiny. Although their findings relate to investigations, enforcement actions are

the direct result of investigative scrutiny, so a relation with enforcement actions is a logical

extension. As a result, a negative relation between having a Big N auditor and SEC cases may be

due to leniency in the overall enforcement process, not probable fraud. Said more directly, Big N

auditors may be negatively associated with fraud by their clients not because Big N auditors

detect or reduce fraud, but rather because the SEC is less likely to investigate their clients.

However, we would not expect a similar selection effect between Big N auditors and plaintiffs’

lawyers. 25 We use our sample of settled SEC cases and SCA cases alleging Rule 10b-5 violations

and accounting issues from 1998-2014 to estimate the following logit model: 26

𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝑑𝑑𝑖𝑖,𝑡𝑡 = 𝛼𝛼0 + 𝛼𝛼1 𝐵𝐵𝐵𝐵𝐵𝐵 5𝑖𝑖,𝑡𝑡 + 𝛼𝛼𝑛𝑛 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑠𝑠𝑖𝑖,𝑡𝑡 + 𝑌𝑌𝑌𝑌𝑌𝑌𝑌𝑌 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 + 𝜀𝜀𝑖𝑖,𝑡𝑡

Our sample includes: 1) 1,262 firm-years with settled 10b-5 SEC cases, 2) 2,401 firm-

25
In untabulated analyses, we estimate the relation between BigN and the filing of accounting-related SCAs over
our time period following the approach in KS (their Table 7, Model 2) and in a sample of irregularity restatements
using the Lennox and Pittman (2010) model and find no significant relation. We also discuss the case selection
process with a senior plaintiff’s lawyer who served as the lead counsel in securities litigation of SEC registrants. He
confirmed that, in his experience, the potential defendant’s external auditor is not a determinant of litigation.
26
We use a logit model, as opposed to a probit model like Lennox and Pittman (2010), because Wooldridge (2010)
states that fixed effects in probit models can result in inconsistent estimations. However, inferences are similar if we
use a probit model (untabulated). We recognize that recent research (e.g., Lawrence, Minutti-Meza, and Zhang
2011) emphasizes it is important to match Big N clients to non-Big N clients. In untabulated analyses, we use
propensity score matching (see Lawrence et al. 2011; Shipman, Swanquist, and Whited 2017) and entropy balancing
(see Hainmueller 2012) to match BigN and non-BigN firms on the controls used by Lennox and Pittman (2010) and
those further motivated by Lawrence et al. (2011). We find Big N is statistically insignificantly related to fraud
using any of the proxies for fraud across any of the time periods that we examine.

21

Electronic copy available at: https://ssrn.com/abstract=3744392


years with settled 10b-5 SCAs, and 3) 2,843 firm-years with either settled 10b-5 SEC or SCA

firm-year observations from 1998-2014. We start with the first part of our sample period (1998-

2005) because this is closer to the period in Lennox and Pittman (1980-2001). Results are

presented in Panel A of Table 8. Similar to Lennox and Pittman (2010), we find a negative

relation between Big N auditors and SEC cases (t = -1.87). However, the coefficient on Big N

using SCAs as the dependent variable is insignificant, as is the coefficient when a combined

measure is used. One explanation for this difference, consistent with Holzman et al. (2019), is a

selection effect, whereby the SEC is more lenient against Big N auditors but shareholders are

not. Thus, inferences are sensitive to the proxy used.

To examine enforcement leniency further, in Panel B, we examine the latter half of our

sample (2006-2014), when public and private enforcement both fell. In this period, when SEC

cases are used as the fraud measure, the Big N coefficient nearly doubles in size (-0.315 to

-0.565), and the statistical significance increases (t = -2.23), consistent with the SEC relying on

the Big N auditors as substitute monitors, particularly when regulatory scrutiny decreased in the

post-2005 period. When SCAs are used as the fraud measure, the coefficient on Big N remains

insignificant. However, when a combined measure is used, the coefficient on Big N is again

significant (t = -2.35), driven by the SEC sample. Similar inferences are obtained for the full

period (Panel C). Thus, to the extent that selection bias confounds findings, use of a combined

measure may mitigate, but not necessarily eliminate, this possibility. Nevertheless, the results

demonstrate the importance that results in both time periods are sensitive to the fraud proxy used.

Setting 2 – Abnormal Accruals Related to Inventory

To examine the possibility of Type II error in fraud research, we ask a hypothetical

research question: do abnormal inventory levels increase the risk that fraud is occurring?

22

Electronic copy available at: https://ssrn.com/abstract=3744392


Stubben (2010) develops a model aimed at measuring abnormal revenue accruals and provides

evidence that it accurately predicts SEC enforcement in cases targeting revenue. However,

inventory (and the offsetting income account, cost of goods sold [COGS]) likely runs a close

second as a common source of fraud (e.g., Lee and Fargher 2013; Deloitte 2009). Donelson et al.

(2012) find that inventory accounting standards are some of the most commonly named

standards in accounting-related SCAs. 27

Despite this, to our knowledge, no study examines how abnormal inventory accruals

relate to fraud. We develop the following abnormal inventory model, consistent with the general

intuition of Stubben (2010) but using the relation between inventory and COGS:

ΔInventoryi,t = 𝛼𝛼0 + 𝛼𝛼1 ΔCOGS Qtrs1-3i,t + 𝛼𝛼2 ΔCOGS Qtr4i,t + 𝜀𝜀 i,t

We use the error term as the firm’s abnormal inventory level (Abnormal Inventory). To

examine the association between abnormal inventory accruals, we employ a model that includes

similar control variables to Lennox and Pittman (2010) and also control for abnormal revenue

due to the relation between revenue and inventory frauds. 28 As before, Panel A of Table 9

presents the results using the first half of our sample. We expect the risk of Type II error to be

lower in this period due to a relatively high number of SEC cases and SCAs in this period.

Column 1 (2) [3] presents results using SEC cases (settled SCAs) [either settled SEC case or

SCA] to identify fraudulent firm-years. As expected, Abnormal Inventory is positively associated

with fraud across all measures (t = 4.32, 4.71, and 5.12, respectively).

However, in Panel B, we examine the same model in the latter half of our sample, when

27
Anecdotally, many famous accounting frauds involve inventory, such as the McKesson & Robbins and Crazy
Eddie frauds (see Clikeman 2009; Dollarhide 2019). In fact, the McKesson & Robbins fraud is credited with leading
to the creation of generally accepted auditing standards after the creation of nearly $300 million in fictitious
inventory in current dollars (Shinde, Willems, and Sallehu 2015; Dollarhide 2019).
28
Untabulated results are similar if we exclude Abnormal Revenue or if we also control for total abnormal accruals.

23

Electronic copy available at: https://ssrn.com/abstract=3744392


enforcement decreased. Here, we expect the issue of low power to be more of a concern. When

SEC cases are used, the coefficient on Abnormal Inventory is insignificant, likely due to the low

number of SEC cases. However, when SCAs or a combined proxy is used, the coefficient is

significantly positive, as expected (t = 3.16 and 3.41, respectively). Thus, a researcher wanting to

pursue a similar abnormal inventory model in a post-SOX period may have abandoned such a

project using only SEC cases. However, using a combined measure of fraud increases power and

reduces Type II error risk. Due to larger power over the full period in Panel C, we obtain

significant results across all measures (t = 4.48, 5.70, and 6.12, respectively).

VI. RESTATEMENTS

Restatements are related to public or private enforcement and have been used as a proxy

for fraud or financial “irregularities” (i.e., intentional misreporting) in prior research as shown in

Online Appendix A. Irregularity restatements are closely related to the construct of fraud, and

some researchers use the terms interchangeably (e.g., Masulis, Wang, and Xie 2012), even

though irregularities do not require users rely on financial statements to their detriment (HLM, p.

1488). HLM are the first to provide a systematic approach to distinguish between severe

restatements (which proxy for “irregularities”) and “errors” and show the importance of doing

so. HLM define irregularity restatements as those where the firm admits to a fraud or irregularity

or if an internal or external investigation is announced. While HLM significantly contributed to

the literature with a relatively easy way to split restatements into those more similar to fraud

versus errors, they acknowledge these severe restatements may still be false positives if, for

example, an SEC investigation concludes fraud did not occur. Thus, the key difference between

our fraud proxy and irregularity restatements is that the former requires a credible allegation of

intentional misstatement by plaintiffs’ lawyers or the SEC, while the latter does not.

24

Electronic copy available at: https://ssrn.com/abstract=3744392


Recent research continues to acknowledge the challenge of restatement false positives

when used as a fraud proxy and follows various approaches to identify severe restatements (e.g.,

Hobson et al. 2012, 2017). Building on the importance of reducing false positives (see Beneish

and Vorst 2020) and recent research that roughly 60 percent of SEC investigations do not result

in formal enforcement (Solomon and Soltes 2019), we contend our methodology may allow

future restatement research to reduce measurement error by reducing these false positives. That

is, while severe restatements may be suspected fraud, further screening can isolate those that

have credible fraud allegations and are closer to the fraud construct. 29

We focus on the overlap between severe restatements in Audit Analytics, using screens

from recent research, and our measure of fraud to identify the proportion of likely false positives

and negatives. 30 For example, some fraud studies use all restatements (e.g., Ham et al. 2017),

others exclude those classified by Audit Analytics as clerical mistakes (e.g., Bens, Goodman, and

Neamtiu 2012), and others use the Audit Analytics flag for fraud, SEC investigation, or a

combination of these flags (e.g., Hobson et al. 2012, 2017; Brown et al. 2020). 31

To provide evidence on whether those filters are successful in isolating restatements

consistent with the accounting fraud construct that most researchers are interested in, we first

reconcile our fraud sample to the following Audit Analytics categories: a) all restatements, b) all

29
One limitation of our approach is that researchers cannot identify fraud observations in “real time,” such as at the
time of a restatement. While most research settings would allow for the revelation of whether the fraud allegations
are credible, researchers may not be able to employ this approach in a setting, for example, involving the passage of
regulatory reform in the past year. In such scenarios, it may be reasonable to use an approach such as that of HLM
with information available at the time of the restatement announcement. However, researchers should acknowledge
the additional measurement error of such an approach if the construct of interest that they are focused on is fraud.
30
We focus on Audit Analytics to be most relevant for future research, but we perform a similar analysis using the
GAO restatement data from HLM and other earlier papers and find similar inferences (see Online Appendix F).
31
The Audit Analytics fraud (SEC investigation) flag is “res_fraud” (“res_sec_invest”). Specifically, “res_fraud”
indicates that the restatement identified financial fraud, irregularities, and misrepresentations; “res_sec_invest”
indicates SEC involvement in the restatement process is noted either in the form of an SEC comment letter that
triggered the restatement or in the form of a SEC inquiry into the circumstances surrounding the restatement

25

Electronic copy available at: https://ssrn.com/abstract=3744392


restatements other than clerical errors, and c) restatements explicitly flagged as fraudulent by

Audit Analytics. 32 To preserve sample size, we use all restatements we can match to Compustat.

Results are reported in Table 10, Panel A. One interesting takeaway is how few

restatements are dropped when omitting those flagged as clerical errors. 33 However, given the

apparent false positive rate of over 90 percent when using all restatements or those excluding

clerical mistakes, we focus our discussion on restatements flagged as fraudulent by Audit

Analytics. Of the 202 such restatements in our period, only 87 would be classified as fraud using

our recommended approach for identifying credible fraud allegations. 34 In other words, over half

appear to be false positives. We explore these presumed false positives below, but it is also worth

emphasizing that a fraud sample constructed using only Audit Analytics fraud restatements

versus our approach for identifying fraud would exclude nearly 88 percent of the fraud

observations in this time period since many fraud observations lack restatements (untabulated). 35

Thus, fraud samples using only restatements lack substantial power due to false negatives.

Examining the 115 restatements classified as fraud by Audit Analytics but not using our

methodology, we identify five main categories (see Panel B of Table 10). The most common

type (60 percent) involves misstated financials, but the fraud is committed by lower level

employees seeking personal gain. Since only senior management is ultimately responsible for the

32
We focus only on the res_fraud flag, rather than also including res_sec_invest, to minimize false positives due to
our focus on accounting fraud and that most SEC investigations do not result in formal enforcement (Solomon and
Soltes 2019). We limit restatements to those that have restatement period ends between 1998 and 2014 and require
that those restatements would be included in a Compustat sample in the year prior to the restatement announcement.
33
It is worth noting “res_cler_err” is not mutually exclusive from other categories. If “res_cler_err” restatements are
purely clerical errors, it is counterintuitive to observe flags such as “res_fraud” also turned on in some cases.
34
We link restatements to our fraud sample based on the overlap between the restatement periods and SEC or SCA
class periods as follows: 1) check whether there is such enforcement with a class end period +/- 90 days from
restatement date; 2) check whether there is such enforcement with class period end +/- 90 days from restatement
period end; 3) check whether there is such enforcement filed within +/- 90 days restatement announcement; and 4)
check whether the restatement period covers the whole class period or vice versa.
35
As discussed in more detail in Online Appendix F, we find that the most common reasons firms do not restate
even in the presence of enforcement are the firm: 1) delists and/or goes bankrupt or 2) neither admits to nor denies
the allegations as part of the settlement and therefore admits no GAAP misstatement.

26

Electronic copy available at: https://ssrn.com/abstract=3744392


financial statements and can intend to deceive investors, these restatements appear inconsistent

with the construct of accounting fraud. 36 In our view, such restatements are not consistent with

the accounting fraud construct usually motivated by prior research (e.g., research motivated by

frauds such as those perpetrated by Enron, WorldCom, and others), nor would they map in well

to common fraud research using executive characteristics. The second most common type only

indirectly involves accounting (17 percent). 37 In all these cases, an illegal or unethical action

occurred, but accounting is only indirectly involved, and they do not meet the definition of fraud

that accounting researchers generally have in mind.

From the remaining restatements, the most common category involves subsequent

dismissals of SCAs or SEC investigations or SEC enforcement that does not allege Rule 10b-5

violations. Thus, they may have been classified as fraud due to discussion of a related SEC

investigation or SCA in the restatement announcement, but ultimately they lack credible fraud

allegations. 38 The fourth category involves restatements where indeed the announcement

includes potentially credible accounting fraud allegations, but there is no related SEC or SCA

case. However, there are only 10 such cases. The last category includes six restatements that face

subsequent enforcement that would meet our criteria but are not in our sample. 39 Thus, relatively

few of these restatements appear relevant for researchers interested in accounting fraud.

36
For example, in one restatement, employees overstated proved reserves, but the investigation from an independent
counsel concluded that the senior executive team did not participate. In another instance, the misstatement was
committed by subordinate divisional employees.
37
For example, in one case, a trader made illegal trades, and once the firm found out, they issued a restatement. In
another example, the firm discovered that a non-regulated, wholly-owned subsidiary entered into unauthorized
financial derivative energy trading. In yet another case, the company “identified certain loans extended by a former
employee in violation of The Bank's lending policies and procedures,” which ultimately required a restatement.
38
In private correspondence, Audit Analytics noted that they usually make their fraud assessment on the restatement
announcement and do not revise the flags later on unless “overwhelming evidence” is brought to their attention that
the classification is incorrect, which they cannot recall happening.
39
This is due to either missing data in our primary enforcement sources or a fraud period per our sources which
extends beyond our sample period end.

27

Electronic copy available at: https://ssrn.com/abstract=3744392


To further examine the merits of these restatements, we perform an analysis similar to

that in Table 6. We examine whether restatements flagged as fraudulent by Audit Analytics but

excluded from our fraud sample are similar to: a) all fraud observations (including restatements)

in our sample and b) restatements flagged as fraudulent by Audit Analytics and included in our

fraud sample. Results are shown in Panel C. 40 Despite the small sample of restatements flagged

as fraudulent but excluded from our sample, we find that they have less aggressive accounting

and face lower adverse consequences relative to all cases in our fraud sample or the restatements

flagged as fraudulent that are included in our sample. 41 Collectively, these results should provide

caution to researchers relying on the “fraud” coding by Audit Analytics as it appears broader

than the classification in HLM and results in many false positives for the fraud construct that

most accounting researchers examine.

Additionally, many restatements we would classify as fraudulent are not identified as

such by Audit Analytics, possibly because the company did not discuss related SCAs or SEC

investigations in the restatement announcement. Specifically, we identify a sample of fraudulent

restatements nearly five times larger than Audit Analytics in our sample by using our method to

identify fraud and matching the observations to restatements. 42 Overall, even with screening used

in recent research, many restatements do not result in credible fraud allegations and vice versa.

Thus, if researchers are interested in the construct of accounting fraud, the presence or absence

of a restatement is not dispositive, and Audit Analytics coding alone should not be relied on.

40
We omit executive turnover from the table because we subsequently examine it in Online Appendix F, but we
supplement it by also examining the restatement announcement returns and the restatement size.
41
In untabulated analyses, we also compare them to the remainder of the Compustat population without credible
fraud allegations and find that they appear relatively similar to those observations. We further validate the
importance of the false positives in these restatements in Online Appendix F by examining subsequent executive
turnover, similar to HLM, and find significant turnover only for restatements that we also classify as fraud.
42
Validating the importance of these false negatives, we examine executive turnover using all restatements we
identify as fraud and find significant results with t-statistics around two to three times larger (Online Appendix F).

28

Electronic copy available at: https://ssrn.com/abstract=3744392


VII. PREDICTIVE MODEL

While we focus on research directly interested in proxying for accounting fraud, either as

the dependent variable or an independent variable of interest, we recognize that many researchers

simply want to control for the probability that firms have committed fraud which will ultimately

face enforcement. For example, the F-Score coefficients from DGLS, which was modeled using

SEC enforcement, have been used extensively for this purpose (e.g., Chan et al. 2013; Bradley et

al. 2017). Similarly, many researchers use the coefficients from the KS model to control for the

threat of securities class actions (e.g., Chen, Li, and Zou 2016; Cunningham, Li, Stein, and

Wright 2019; Kim, Wang, and Zhang 2019). Given the usefulness of these models and the fact

that each looked at public or private enforcement in isolation, we create a predictive model using

their explanatory variables and our combined measure of public and private fraud enforcement.

We first validate the models in our sample as shown in Table 11, Panel A. Interestingly,

we find evidence of cross-regime predictability. F-Score determinants, derived using SEC

enforcement, predict settled SCAs (Column 4), consistent with SCAs also representing fraud.

Similarly, but likely more surprisingly, the largely market-related variables from KS, calibrated

to predict SCA filings, predict SEC cases (Column 2). We conjecture these market variables,

such as return volatility and skewness and share turnover, capture firm characteristics that give

managers the incentive and/or ability to commit accounting fraud. For example, for firms with

high valuation uncertainty (i.e., volatile prices), investors do not have the same information on

fundamental value as insiders, so it may be easier for managers to “fool” investors with

fraudulent accounting. Additionally, for firms with market values more sensitive to bad news

(i.e., negative skewness), managers have stronger incentives to issue positively biased financial

29

Electronic copy available at: https://ssrn.com/abstract=3744392


statements. 43 Ultimately, the fact that variables from public enforcement models predict private

enforcement, and vice versa, reinforces our main results that both are valid fraud proxies. 44

As a result, when predicting our combined fraud measure, adding both sets of variables

unsurprisingly improves predictive ability and performs better than either individual model

variables (Panel B). More importantly, the expanded model also has higher precision and

sensitivity and a lower false discovery and false positive rate than either set of variables in

isolation, particularly when compared to the DGLS model (see Online Appendix G for more

details). The lower false positives in particular are important due to the costs associated with

incorrectly flagging non-fraud firms (see Beneish and Vorst 2020). For example, the false

discovery rate (i.e., the number of false positives divided by the total observations positive) is

approximately 11 percent lower using the expanded model than the DGLS variables only.

Overall, this predictive model may be useful for researchers and others interested in predicting or

controlling for misreporting that will result in future public or private enforcement.

VIII. CONCLUSION

We propose measuring accounting fraud with both public and private enforcement after

filtering for credible fraud allegations. Many studies rely solely on public enforcement via SEC

cases. However, based on a review of the legal standards and processes for public and private

enforcement and descriptive evidence on SEC cases and SCAs, we argue that both public and

private enforcement efforts can yield credible allegations of accounting fraud.

We explore two research settings to illustrate these issues. In the first, where a selection

43
Moreover, the SEC also considers information such as the size and speed of market capitalization loss when
deciding whether or not to bring cases (SEC 2017).
44
We note that RSST accruals is somewhat strangely negatively associated with both settled public and private
enforcement when combined with the KS variables. In untabulated analyses, we identify that this result is due to the
inclusion of the Shares Turnover variable.

30

Electronic copy available at: https://ssrn.com/abstract=3744392


effect for SEC cases may exist, the combined fraud proxy weakens results. In the second, where

limited power likely exists in recent periods due to a significant decline in SEC enforcement, we

find insignificant results using only SEC cases but positive and significant results using our

combined fraud proxy. In additional analysis, we show that common proxies for fraud using

restatements in recent research include significant false positives and false negatives. We also

report a fraud prediction model using our combined fraud proxy for use in future research.

Nonetheless, focusing on public or private enforcement alone may sometimes be proper.

For example, this is the case if the study (a) does not require a control sample (and thus does not

run the risk of the control sample containing false negatives) or (b) concerns only enforcement in

one regime. Examples of such research are Files (2012), who investigates the SEC enforcement

process, and Bliss et al. (2018), who examine securities litigation risk. However, for researchers

interested in accounting fraud, we suggest using both public and private enforcement with

appropriate screening and reporting the sensitivity of their findings to such alternative measures.

31

Electronic copy available at: https://ssrn.com/abstract=3744392


REFERENCES
Aigner, D. J. 1973. Regression with a binary independent variable subject to errors of observation. Journal of
Econometrics 1 (1): 49–59.
Altman, E. I. 1968. Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of
Finance 23 (4): 589–609.
Amiram, D., Z. Bozanic, J. D. Cox, Q. Dupont, J. M. Karpoff, and R. G. Sloan. 2018. Financial reporting fraud and
other forms of misconduct: A multidisciplinary review of the literature. Review of Accounting Studies 23 (2):
732–783.
Baker, T., and S. J. Griffith. 2010. Ensuring Corporate Misconduct: How Liability Insurance Undermines
Shareholder Litigation. Chicago, IL: University of Chicago Press.
Begley, J., J. Ming, and S. Watts. 1996. Bankruptcy classification errors in the 1980s: An empirical analysis of
Altman’s and Ohlson’s models. Review of Accounting Studies 1 (4): 267–284.
Beneish, M. D. 1997. Detecting GAAP violation: Implications for assessing earnings management among firms with
extreme financial performance. Journal of Accounting and Public Policy 16 (3): 271–309.
Beneish, M. D., and P. Vorst. 2020. The cost of fraud prediction errors. Working Paper. Available at:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3529662.
Bens, D. A., T. Goodman, and M. Neamtiu. 2012. Does investment-related pressure lead to misreporting? An
analysis of reporting following M&A transactions. The Accounting Review 87 (3): 839–865.
Bliss, B. A., F. Partnoy, M. Furchtgott. 2018. Information bundling and securities litigation. Journal of Accounting
and Economics 65 (1): 61–84.
Block, D. J. 1981. An overview: Responsibilities of attorneys under the federal securities laws. The Business Lawyer
36 (4): 1781–1790.
Bohn, J., and S. Choi. 1996. Fraud in the new–issues market: Empirical evidence on securities class actions.
University of Pennsylvania Law Review 144: 903–982.
Bradley, D., S. Gokkaya, X. Liu, and F. Xie. 2017. Are all analysts created equal? Industry expertise and monitoring
effectiveness of financial analysts. Journal of Accounting and Economics 63 (2-3): 179–206.
Brown, N. C., R. M. Crowley, and W. Brooke Elliot. 2020. What are you saying? Using topic to detect financial
misreporting. Journal of Accounting Research 58 (1): 237–291.
Caskey, J., and M. Hanlon. 2013. Dividend policy at firms accused of accounting fraud. Contemporary Accounting
Research 30 (2): 818–850.
Ceresney, A. 2013. Financial reporting and accounting fraud. 19 September. Available at:
https://www.sec.gov/news/speech/spch091913ac.
Chan, L. H., K. C. W. Chen, and T.-Y. Chen. 2013. The effects of firm-initiated clawback provisions on bank loan
contracting. Journal of Financial Economics 110 (3): 659–679.
Chen, L. 2016. The informational role of short seller: The evidence from short sellers’ reports on US-listed Chinese
firms. Journal of Business Finance and Accounting 43 (9): 1444–1482.
Chen, Z., O. Z. Li, and H. Zou. 2016. Directors’ and officers’ liability insurance and the cost of equity. Journal of
Accounting and Economics 61: 100–120.
Chhaochharia, V., A. Kumar, and A. Niesen-Ruenzi. 2012. Local investors and corporate governance. Journal of
Accounting and Economics 54 (1): 42–67.
Choi, S. J. 2007. Do the merits matter less after the Private Securities Litigation Reform Act? Journal of Law,
Economics, and Organization 23 (3): 598–626.
Choi, S. J., A. C. Wiechman, and A. C. Pritchard. 2013. Scandal enforcement at the SEC: The arc of the option
backdating investigations. American Law and Economics Review 15 (2): 542–577.
Choi, S. J., and A. C. Pritchard. 2016. SEC investigations and securities class actions: An empirical
comparison. Journal of Empirical Legal Studies 13 (1): 27–49.
Clikeman, P. M. 2009. Called to Account: Fourteen Financial Frauds that Shaped the American Accounting
Profession. New York, NY: Routhledge.
Coffee, J. C. 1986. Understanding the plaintiff’s attorney: The implications of economic theory for private
enforcement of law through class and derivative actions. Columbia Law Review 86 (4): 669–727.
Coffee, J. C. 2006. Reforming the securities class action: An essay on deterrence and its implementation. Columbia
Law Review 106 (7): 1534–1586.
Corwin, S. A., and P. Schultz. 2012. A simple way to estimate bid–ask spreads from daily high and low prices.
Journal of Finance 67 (2): 719–760.
Cox, J. D., R. S. Thomas, and D. Kiku. 2003. SEC enforcement heuristics: An empirical inquiry. Duke Law Journal

32

Electronic copy available at: https://ssrn.com/abstract=3744392


53: 737–779.
Cox, J. D., R. S. Thomas, and D. Kiku. 2005. Public and private enforcement of the securities laws: Have things
changed since Enron? Notre Dame Law Review 80: 893–908.
Cunningham, L. M., C. Li, S. E. Stein, and N. S. Wright. 2019. What’s in a name? Initial evidence of U.S. audit
partner identification using difference-in-differences. The Accounting Review 94 (5): 139–163.
Dechow, P. M., W. Ge, C. R. Larson, and R. G. Sloan. 2011. Predicting material accounting
misstatements. Contemporary Accounting Research 28 (1): 17–82.
Deloitte. 2009. Sample listing of fraud schemes. Centre for Corporate Governance. Available at:
https://www2.deloitte.com/content/dam/Deloitte/in/Documents/risk/Corporate%20Governance/Audit%20Com
mittee/in-gc-fraud-schemes-questions-to-consider-noexp.pdf.
Desai, H., C. E. Hogan, and M. S. Wilkins. 2006. The reputational penalty for aggressive accounting: Earnings
restatements and management turnover. The Accounting Review 81 (1): 83–112.
Dollarhide, M. E. 2019. 4 famous inventory frauds you’ve never heard of. Investopedia. 19 February. Available at:
https://www.investopedia.com/articles/economics/12/four-unknown-massive-frauds.asp.
Donelson, D. C., J. J. Hopkins, and C. G. Yust. 2015. The role of directors’ and officers’ insurance in securities
fraud class action settlements. Journal of Law and Economics 58 (4): 747–778.
Donelson, D. C., J. M. McInnis, and R. D. Mergenthaler. 2012. Rules-based accounting standards and litigation. The
Accounting Review 87 (4): 1247–1279.
Dyck, A., A. Morse, and L. Zingales. 2010. Who blows the whistle on corporate fraud? Journal of Finance 65 (6):
2213–2253.
Eichenwald, K. 2005. Conspiracy of Fools: A True Story. New York, NY: Broadway Books.
Eisinger, J. 2017. The Chickenshit Club: Why the Justice Department Fails to Prosecute Executives. New York, NY:
Simon & Schuster.
Files, R. 2012. SEC enforcement: Does forthright disclosure and cooperation really matter? Journal of Accounting
and Economics 53 (1-2): 353–374.
Francis, J., D. Philbrick, and K. Schipper. 1994. Shareholder litigation and corporate disclosures. Journal of
Accounting Research 32 (2): 137–164.
Franco, A., N. Malhotra, and G. Simonovits. 2014. Publication bias in the social sciences: Unlocking the file drawer.
Science 345 (6203): 1502–1505.
Greene, W. H. 2003. Econometric Analysis. Upper Saddle River, NJ: Prentice Hall.
Grundfest, J.A. 1994. Disimplying private rights of action under the federal securities laws: The Commission’s
Authority. Harvard Law Review 107 (5): 961–1024.
Hackbrath, J.W. and E. Stedman. 2018. Kokesh, Lucia, and their impact on SEC enforcement. Quarles & Brady.
Available at: https://www.quarles.com/publications/kokesh-lucia-and-their-impact-on-sec-enforcement/.
Hainmueller, J. 2012. Entropy balancing for causal effects: A multivariate reweighting method to produce balanced
samples in observational studies. Political Analysis 20 (1): 25–46.
Ham, C., M. Lang, N. Seybert, and S. Wang. 2017. CFO narcissism and financial reporting quality. Journal of
Accounting Research 55 (5): 1089–1135.
Hausman, J. A., J. Abrevaya, and F. M. Scott-Morton. 1998. Misclassification of the dependent variable in a
discrete-response setting. Journal of Econometrics 87 (2): 239–269.
Hennes, K. M., A. J. Leone, and B. P. Miller. 2008. The importance of distinguishing errors from irregularities in
restatement research: The case of restatements and CEO/CFO turnover. The Accounting Review 83 (6): 1487–
1519.
Hobson, J. L., W. J. Mayew, and M. Venkatachalam. 2012. Analyzing speech to detect financial misreporting.
Journal of Accounting Research 50 (2): 349–392.
Hobson, J. L., W. J. Mayew, M. E. Peecher, and M. Venkatachalam. 2017. Improving experienced auditors’
detection of deception in CEO narratives. Journal of Accounting Research 55 (5): 1137–1166.
Holzman, E., N. T. Marshall, and B. Schmidt. 2019. Who’s on the hot seat for an SEC investigation? Working
Paper. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3223815.
Hubbard, R., and J. S. Armstrong. 1997. Publication bias against null results. Psychological Reports 80 (1): 337–338.
Jackson, H. E., and M. J. Roe. 2009. Public and private enforcement of securities laws: Resource-based evidence.
Journal of Financial Economics 93 (2): 207–238.
Johnson, M. F., K. K. Nelson, and A. C. Pritchard. 2007. Do the merits matter more? The impact of the Private
Securities Litigation Reform Act. Journal of Law, Economics, and Organization 23 (3): 627–652.
Karmel, R. S. 2005. Realizing the dream of William O. Douglas—The Securities and Exchange Commission takes
charge of corporate America. Delaware Journal of Corporate Law 30: 79–144.

33

Electronic copy available at: https://ssrn.com/abstract=3744392


Karpoff, J. M., A. Koester, D. S. Lee, and G. S. Martin. 2017. Proxies and databases in financial misconduct
research. The Accounting Review 92 (6): 129–163.
Karpoff, J. M., D. S. Lee, and G. S. Martin. 2008. The cost to firms of cooking the books. Journal of Financial and
Quantitative Analysis 43 (3): 581–611.
Kim, C., K. Wang, and L. Zhang. 2019. Readability of 10-K reports and stock price crash risk. Contemporary
Accounting Research 36 (2): 1184–1216.
Kim, I., D. J. Skinner. 2012. Measuring securities litigation risk. Journal of Accounting and Economics 53: 290–
310.
La Porta, R., F. Lopez-de-Silanes, A. Shleifer. 2006. What works in securities laws? Journal of Finance 61 (1): 1–
32.
Lawrence, A., M. Minutti-Meza, and P. Zhang. 2011. Can big 4 versus non-big 4 differences in audit-quality proxies
be attributed to client characteristics? The Accounting Review 86 (1): 259–286.
Lee, G., and N. Fargher. 2013. Companies’ use of whistle-blowing to detect fraud: An examination of corporate
whistle-blowing policies. Journal of Business Ethics 114: 283–295.
Lennox, C., and J. A. Pittman. 2010. Big five audits and accounting fraud. Contemporary Accounting Research 27
(1): 209–247.
Levitt Jr., A. 1998. The numbers game. The CPA Journal 68 (12): 14.
Li, V. 2016. Do false financial statements distort peer firms’ decisions? The Accounting Review 91 (1): 251–278.
Macey, J. R. 2010. The distorting incentives facing the U.S. Securities and Exchange Commission. Harvard Journal
of Law and Public Policy 33 (2): 639–670.
Masulis, R.W., C. Wang, and F. Xie. 2012. Globalizing the boardroom – The effects of foreign director on corporate
governance and firm performance. Journal of Accounting and Economics 53 (3): 527–554.
McNichols M. F., and S. R. Stubben. 2008. Does earnings management affect firms’ investment decisions? The
Accounting Review 83 (6): 1571–1603.
Meyer, B., and N. Mittag. 2017. Misclassification in binary choice models. Journal of Econometrics 200 (2): 295–
311.
Prentice, R. A. 2009. The Ethical and Legal Environment of Accounting. Mason, OH: Cengage Learning.
Rakoff, J. S. 2014. Is the SEC becoming a law unto itself? PLI Securities Regulation Institute Keynote Address. 5
November. Available at: https://securitiesdiary.files.wordpress.com/2014/11/rakoff-pli-speech.pdf.
Richards, L. 2008. Why does fraud occur and what can deter or prevent it? Southwest Securities Enforcement
Conference Address. 9 September. Available at: https://www.sec.gov/news/speech/2008/spch090908lar.htm.
Schantl, S. F., and A. Wagenhofer. 2020. Deterrence of financial misreporting when public and private enforcement
strategically interact. Journal of Accounting and Economics, forthcoming.
Securities and Exchange Commission (SEC). 2017. Enforcement Manual. Division of Enforcement. 28 November.
Available at: https://www.sec.gov/divisions/enforce/enforcementmanual.pdf.
Shinde, J. S., J. Willems, M. M. Sallehu, and M. P. Merkle. 2015. Establishment of GAAS: Impact of an auditing
fraud. Journal of Accounting & Finance 15: 40–54
Shipman, J. E., Q. T. Swanquist, and R. L. Whited. 2017. Propensity Score Matching in Accounting Research. The
Accounting Review 92 (1): 213–244.
Solomon, D. H., and E. F. Soltes. 2019. Is ‘not guilty’ the same as ‘innocent’? Evidence from SEC financial fraud
investigations. Working Paper. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3402780.
Stubben, S.R. 2010. Discretionary revenues as a measure of earnings management. The Accounting Review 85 (2):
695-717.
Walker, R. H. 1999. Behind the number of the of the SEC’s recent financial fraud cases. 7 December. Available at:
https://www.sec.gov/news/speech/speecharchive/1999/spch334.htm.
Wang, T. Y. 2013. Corporate securities fraud: Insights from a new empirical framework. Journal of Law,
Economics, and Organization 29 (3): 535–568.
White, M. J. 2014. All-encompassing enforcement: The robust use of civil and criminal actions to police the
markets. 31 March. Available at: https://www.sec.gov/news/speech/2014-spch033114mjw.
Wooldridge, J. M. 2010. Econometric analysis of cross section and panel data. MIT Press. Second Edition.
Zheng, X. 2016. A tale of two courts: Determinants and consequences of the SEC’s choice of enforcement venue
after the Dodd–Frank Act. Working Paper.

34

Electronic copy available at: https://ssrn.com/abstract=3744392


APPENDIX A
Variable Definitions

Variable Description
Abn. Accruals Abnormal accruals calculated using the modified Jones model and data from
Compustat. We require 20 observations per industry-year.
Abnormal Inventory The residual from regressing change in inventory on the change in cost of goods
sold for the first three quarters of the year (relative to prior year’s first three
quarters) and the change in cost of goods sold for the fourth quarter. The model
is estimated by industry-year using data from Compustat. We exclude firms in
the financial services, insurance, and utilities industries.
Abnormal Revenue Abnormal revenue as calculated by Stubben (2010). More specifically, it is the
residual from regressing change in accounts receivable on the change in
revenue for the first three quarters of the year (relative to prior year’s first three
quarters) and the change in revenue for the fourth quarter. The model is
estimated by industry-year using data from Compustat and excludes firms in the
financial services, insurance, and utilities industries.
Actual Issuance An indicator variable set to one if the firm issued securities during the year and
zero otherwise. We define issuance taking place if the sum of Sale of Common
and Preferred Stock and Long-Term Debt Issuance exceed 0.1 on Compustat.
Altman Z – rank Decile rank of the firm’s yearly Altman Z score, which ranges from 0 to 9. A
lower rank implies higher bankruptcy risk. Altman's (1968) Z-score is
calculated using the updated coefficients from Begley, Ming, and Watts (1996)
and data from Compustat.
Audit Tenure Log of the number of years the company has had the same auditor from
Compustat.
Big 5 An indicator variable set to one if the firm is being audited by a Big 5 auditor or
one of their predecessors on Compustat and zero otherwise.
Book to Market Ratio of book value of common equity to market value of equity calculated
using data from Compustat.
CAR[-1, 1] Three-day cumulative abnormal returns centered on the restatement
announcement date from CRSP.
CEO left / CFO left Indicator variables set to one if the executive (CEO or CFO, respectively) left
his/her position at the company by the end of the first fiscal year following the
fraud period end as per BoardEx and zero otherwise. We do not consider
changes of interim CEOs/CFOs or promotions from CFO to CEO as turnover.
Cumul. Abn. Return Cumulative abnormal return over the fiscal year from CRSP.
Debt & Equity Iss. An indicator variable set to one if newly issued long-term debt and equity
exceeds 20 percent of total assets on Compustat and zero otherwise.
F-score Refers to F-Score 1 calculated using Dechow et al. (2011) methodology and
data from Compustat.
Firm Age Logged years the company is listed on Compustat.
FPS An indicator variable set to one for firms operating in high litigation industries
(Francis, Philbrick, and Schipper 1994) using data from CRSP and zero
otherwise.
Leverage Leverage ratio, measured as total long-term and short-term debt scaled by the
sum of total long-term and short-term debt and market value of equity on
Compustat. Missing values of long-term and short-term debt have been replaced
with zero.
Log Max Damages The log value of the difference between the maximum market cap during the
class period less the market cap the day following class period end from CRSP.
Log MVE The log value of the market value of equity in millions on Compustat.

35

Electronic copy available at: https://ssrn.com/abstract=3744392


Log Net Income Effect The log value of the absolute value of the net income effect for a given
restatement per Audit Analytics. We only include restatements for which the
net income effect was negative.
Log Size The log value of total assets in millions on Compustat.
Loss An indicator variable set to one if net income is negative from Compustat and
zero otherwise.
M&A An indicator variable set to one if the firm had an acquisition that contributed to
sales during the fiscal year from Compustat and zero otherwise.
Neg. Equity An indicator variable set to one if the company's total liabilities exceed its total
assets on Compustat and zero otherwise.
Performance-Adj. Abnormal accruals calculated using the modified Jones model and data on
Abn. Accruals Compustat, controlling for ROA as a covariate in the estimation of expected
accruals. We require 20 firms per industry-year.
Restatement An indicator variable set to one if Audit Analytics indicates a restatement for
the year and zero otherwise. We supplement Audit Analytics with GAO data
relating to irregularities as defined by Hennes et al. (2008). We exclude
restatements due to clerical errors as identified by Audit Analytics. We include
restatements with restatement periods ending 1998 to 2014.
Return on Assets Net income scaled by total assets on Compustat.
Return Skewness Skewness of monthly return over the fiscal year on CRSP.
Return Volatility Standard deviation of monthly return over the fiscal year on CRSP.
RSST accruals Change in working capital accruals plus change in long-term operating assets
plus change long-term operating liabilities and then scaled by average total
assets on Compustat as per Dechow et al. (2011).
Sales Growth Change in sales scaled by prior year’s total assets on Compustat.
SEC w/ 10b-5 An indicator variable set to one for firms facing both a SCA and an SEC case
with a Rule 10b-5 allegation and zero otherwise.
Shares Turnover Trading volume over the fiscal year scaled by beginning of the year shares
outstanding from CRSP. Similar to Kim and Skinner (2012), we multiply the
variable by 1,000 for presentation purposes.
Δ Cash Sales Percentage change in cash sales on Compustat. Cash sales is defined as sales
less change in accounts receivable.
Δ Forc. Accuracy Change in forecast accuracy between the fiscal year of fraud end and the prior
calculated using IBES unadjusted file. Accuracy is calculated as the absolute
value of actual less median forecast divided by prior year’s end price,
multiplied by negative one so lower values imply worse forecast accuracy.
Δ Inst. Own. Change in average yearly institutional ownership between the fiscal year of
fraud end and the prior year as per Thomson Reuters.
Δ Inventory Change in inventory scaled by average total assets on Compustat.
Δ Receivables Change in accounts receivable scaled by average total assets on Compustat.
Δ ROA Change in ROA. ROA is defined as income before extraordinary items scaled
by average total assets on Compustat.
Δ Spread Change in spread between the fiscal year of fraud end and the prior year.
Calculated using Corwin and Schultz (2012) methodology and multiplied by
100 for presentation purposes.
% of Soft Assets Total assets excluding PP&E and cash and cash equivalent scaled by total assets
on Compustat.

36

Electronic copy available at: https://ssrn.com/abstract=3744392


FIGURE 1
Enforcement Regime Overlap

This figure provides a visual representation of the observations listed in Table 1 and used in Table 2.

37

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 1
Enforcement Database Sample Selection
Panel A: SCAs Cases
SCAs with class end periods between 1998 and 2014 from SCAC database 3,572
Less: Pending resolution (29)
No Rule 10b-5 allegations (519)
No alleged GAAP violations (1,512)
Missing Compustat identifiers (25)
Total Accounting 10b-5 SCAs 1,487
Less: Dismissed cases (495)
Settled Accounting 10b-5 SCAs (Sample used in Section V replications) 992
Less: Missing total assets or income during the final year in class period (22)
Settled Accounting 10b-5 SCAs (Sample used in Figure 1) 970
Panel B: SEC cases
GAAP SEC cases with misconduct periods ending after 1998 from CFRM database 541
Less: Missing CIK information in CFRM or Compustat identifiers (62)
No Rule 10b-5 allegations (142)
Total Settled Accounting 10b-5 SEC cases from CFRM 337
Supplementing with Settled GAAP SEC cases from Advisen
Cases with class periods ending in 2013 or 2014 14
Cases with missing information in CFRM 57
Additional cases identified in Advisen, but not included in CFRM 37
Settled Accounting 10b-5 SEC cases (Sample used in Section V replications) 445
Less: Missing total assets or income during the final year in fraud period (37)
Settled Accounting 10b-5 SEC cases (Sample used in Figure 1) 408
Panel C: Fraud cases by year using the final year of the fraud period
Settled Accounting Settled Accounting Total Non-Unique Settled Accounting
Year 10b-5 SCAs 10b-5 SEC cases 10b-5 SEC or SCA cases
(1) (2) (1) + (2)
1998 89 17 106
1999 74 26 100
2000 95 45 140
2001 74 41 115
2002 110 58 168
2003 71 40 111
2004 67 37 104
2005 63 39 102
2006 59 32 91
2007 46 25 71
2008 65 18 83
2009 27 17 44
2010 27 13 40
2011 49 11 60
2012 23 9 32
2013 26 11 37
2014 27 6 33
Total 992 445 1,437
This table presents a reconciliation from all observations in each enforcement database with fraud periods ending
between 1998 and 2014 and the final observations used in the paper. For Panel A, we note “No Rule 10b-5
allegations” also includes SCAs for which the SCAC database does not provide the relevant complaints and Rule
10b-5 is not mentioned in the summary description provided by SCAC. For Panel B, we note we manually adjust
some observations for the CFRM sample as some links between the ‘detail’ tab and the ‘qtr’ tab were broken or
fraud years needed to be updated. We ensure SEC cases from CFRM with available information in Compustat settle
or are related to other SEC cases that settle. For Panel C, we tabulate cases using the last year of the fraud period. It
should be noted that the last column which aggregates SCA and SEC cases overstates the total enforcement years
available to researchers because some SEC and SCA target the same frauds as shown in Figure 1 and Table 2.

38

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 2
Descriptive Statistics
Panel A: Settled Accounting 10b-5 SCAs and no Settled Accounting 10b-5 SEC cases
N Mean Median P10 P90 Std
Log Size 734 6.67 6.36 3.95 10.07 2.36
Log MVE 718 6.59 6.36 4.31 9.32 1.94
Book to Market 718 0.57 0.40 0.09 1.28 0.65
Return on Assets 734 -0.08 0.01 -0.32 0.11 0.34
Leverage 718 0.27 0.18 0.00 0.76 0.28
Altman Z – rank 577 5.26 5.00 2.00 9.00 2.61
Log Max. Damages 694 6.57 6.45 4.45 8.94 1.78
Panel B: Settled Accounting 10b-5 SEC cases and no Settled Accounting 10b-5 SCAs
N Mean Median P10 P90 Std
Log Size 172 5.39 4.98 1.74 8.96 2.74
Log MVE 166 5.28 4.82 2.45 8.65 2.43
Book to Market 166 0.70 0.45 0.03 1.61 0.84
Return on Assets 172 -0.26 0.00 -1.13 0.12 0.65
Leverage 166 0.26 0.16 0.00 0.75 0.29
Altman Z – rank 138 4.48 5.00 0.00 9.00 2.95
Log Max. Damages 111 5.22 5.27 2.03 8.27 2.24
Panel C: Overlap of Settled Accounting 10b-5 SCAs and SEC cases
N Mean Median P10 P90 Std
Log Size 236 6.73 6.46 3.76 10.16 2.42
Log MVE 232 6.60 6.32 4.21 9.38 2.04
Book to Market 232 0.56 0.41 0.07 1.22 0.65
Return on Assets 236 -0.13 0.01 -0.50 0.08 0.40
Leverage 232 0.26 0.18 0.00 0.79 0.29
Altman Z – rank 199 5.16 5.00 2.00 9.00 2.61
Log Max. Damages 197 7.11 6.66 4.35 10.25 2.13
Panel D: Settled Accounting 10b-5 SCAs or SEC cases
N Mean Median P10 P90 Std
Log Size 1,142 6.49 6.23 3.63 9.98 2.48
Log MVE 1,116 6.40 6.21 3.87 9.30 2.09
Book to Market 1,116 0.59 0.41 0.08 1.29 0.68
Return on Assets 1,142 -0.12 0.01 -0.42 0.11 0.42
Leverage 1,116 0.27 0.17 0.00 0.77 0.28
Altman Z – rank 914 5.12 5.00 2.00 9.00 2.67
Log Max. Damages 1,002 6.53 6.38 4.20 9.21 1.97
Panel E: All Compustat Firms with fiscal years between 1998 and 2014
N Mean Median P10 P90 Std
Log Size 140,222 5.40 5.64 1.62 8.97 2.93
Log MVE 123,552 5.13 5.13 1.74 8.56 2.63
Book to Market 123,305 0.37 0.48 -0.03 1.41 2.47
Return on Assets 140,219 -0.46 0.01 -0.71 0.11 2.37
Leverage 123,563 0.26 0.18 0.00 0.69 0.27
Altman Z – rank 98,349 4.50 5.00 1.00 8.00 2.87
This table presents descriptive statistics for: 1) only settled accounting 10b-5 SCAs (Panel A), 2) only settled
accounting 10b-5 SEC cases (Panel B), 3) overlap settled accounting 10b-5 SEC and SCA cases (Panel C), 4) either
settled accounting 10b-5 SCA or SEC cases (Panel D), and 5) all firms listed on Compustat with fiscal years
between 1998 and 2014 (Panel E). Thus, Panel D can be obtained by combining Panels A through C. All continuous
variables are winsorized at the 1st and 99th percentile. The fiscal year included is the last fiscal year affected by
fraud or, if only quarters were affected, the prior fiscal year. For a case to be included, it needs to have available data
for total assets and income before extraordinary items in the last fiscal year prior to fraud period end. Bolded mean
values in Panel A indicate significant mean differences between only settled accounting 10b-5 SCAs (Panel A) and
only accounting 10b-5 SEC (Panel B) cases at 10 percent, two-tailed. Cases that initially had both an accounting
10b-5 SEC enforcement and a SCA but eventually one of the two was dismissed are classified as either accounting
10b-5 SEC only or SCA only based on the settled case.

39

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 3
Systematic Differences Across Enforcement Regimes
Settled Acct Settled Acct
10b-5 SEC Only 10b-5 SCA Only Both
Chinese Reverse Mergers 4 31 4
10.3% 79.5% 10.3%

Option Backdating 18 15 13
39.1% 32.6% 28.3%

Overall (See Figure 1) 174 735 236


15.2% 64.2% 20.6%
This table presents descriptive statistics regarding enforcement actions surrounding specific events. We tabulate
results regarding 1) Chinese firms that became cross listed in the US through a reverse merger (CRM), 2) stock
option backdating, and 3) overall. To identify CRM, we read 8-Ks, including Item 5.06 (i.e., Change in Shell
Company Status), to identify whether it relates to a non-U.S. reverse merger. Then, we supplement this list with
Chinese reverse mergers identified by Bloomberg. After identifying these firms, we check our sample to identify the
number of accounting 10b-5 cases were brought against these firms. We rely on Advisen to identify enforcement
actions related to stock option backdating.

40

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 4
Stock Price Drops and Securities Class Actions
Panel A: One day stock price drops of 10% or more
(1) (2) (3) (4)
Year Drop Firms SCAs Accounting 10b-5 SCAs Settled Accounting 10b-5 SCAs
1998 6,131 198 96 76
1999 5,780 169 98 72
2000 6,360 174 113 85
2001 5,435 412 89 69
2002 4,486 187 131 94
2003 3,165 134 92 61
2004 2,603 148 94 59
2005 2,398 131 76 50
2006 2,269 84 57 38
2007 2,628 111 61 42
2008 5,516 143 74 48
2009 4,001 88 45 23
2010 2,225 88 35 22
2011 2,424 118 59 35
2012 2,069 92 31 18
2013 1,807 106 40 22
2014 2,102 121 31 21
Total 61,399 2,504 1,222 835
% of Drops 100.00% 4.08% 1.99% 1.36%
Overall SCAs 2,915 1,291 869
% of Total 85.90% 94.66% 96.09%
Panel B: One day stock price drops of 20% and $100 million or more
(1) (2) (3) (4)
Year Drop Firms SCAs Accounting 10b-5 SCAs Settled Accounting 10b-5 SCAs
1998 311 70 38 30
1999 430 68 36 29
2000 909 90 57 46
2001 460 115 34 27
2002 383 86 70 49
2003 155 38 21 16
2004 199 64 41 29
2005 172 43 20 16
2006 190 32 23 18
2007 233 50 27 18
2008 577 76 45 30
2009 232 32 17 9
2010 110 24 8 6
2011 198 39 15 11
2012 172 39 8 3
2013 156 42 14 10
2014 208 47 6 5
Total 5,095 955 480 352
% of Drops 100.00% 18.74% 9.42% 6.19%
Overall SCAs 2,915 1,291 869
% of Total 32.76% 37.18% 40.51%
This table presents summary statistics related to SCAs and stock price drops from all observations on CRSP.
Panel A tabulates firms that experienced at least a 10 percent negative market-adjusted return in a single day, and
Panel B tabulates firms that experienced at least a 20 percent negative market-adjusted return and a $100M
decline in market cap. Column 2 presents the number of those firms that faced a SCA with a class period ending
the same calendar year as the price drop. Similarly, Column 3 includes only accounting-related 10b-5 SCAs, and
Column 4 includes settled accounting-related 10b-5 SCAs. Overall SCAs includes SCAs that can be linked to
CRSP.

41

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 5
Settled versus Dismissed Securities Class Actions
Dismissed
Settled Acct
Acct 10b-5 Mean Diff.
10b-5 SCAs
SCAs
N Mean N Mean
Panel A: Case merits
SEC w/ 10b-5 487 0.06 970 0.24 -0.19
Log Max Damages 458 6.50 891 6.69 -0.19

Panel B: Accounting aggressiveness during fraud period


Abn. Accruals 371 0.02 766 0.04 -0.02
Performance-Adj. Abn. Accruals 371 0.01 766 0.03 -0.03
F-score 369 1.38 747 1.60 -0.22
Restatement 487 0.39 970 0.54 -0.15

Panel C: Adverse consequences after fraud


CEO left 303 0.19 464 0.27 -0.07
CFO left 289 0.22 404 0.29 -0.07
Δ Spread 413 0.20 709 0.38 -0.18
Δ Forc. Accuracy 354 -0.01 562 -0.02 0.01
Δ Inst. Own. 440 0.31 799 -1.79 2.10
This table presents summary statistics for settled accounting 10b-5 SCAs versus dismissed accounting 10b-5
SCAs and presents mean differences. Panel A presents common proxies for case merits, Panel B presents
financial statement characteristics during the fraud period, and Panel C presents adverse consequences between
the fiscal year of fraud end and the prior year. For a SCA to be included, it needs to have available data for total
assets and income before extraordinary items in the last fiscal year prior to fraud period end. Differences are
bolded if they are statistically significant at 10 percent levels, two-tailed. Variables are defined in Appendix A.
All continuous variables are winsorized at the 1st and 99th percentile.

42

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 6
Do Settled Accounting Securities Class Actions and SEC Cases that Allege Rule 10b-5 Violations “Look Like” Fraud Observations?
Compustat Settled Acct
Settled Acct 10b-5 Both
Population 10b-5 SEC Mean Difference Tests
SCA Only (2) (4)
(1) Only (3)
N Mean N Mean N Mean N Mean 1 vs. 2 1 vs. 3 1 vs. 4 2 vs. 3 2 vs. 4 3 vs. 4
Panel A: Accounting aggressiveness during fraud period
Abn. Accruals 99,713 0.00 566 0.04 133 0.00 200 0.05 -0.04 0.00 -0.04 0.04 -0.01 -0.04
Performance-Adj.
Abn. Accruals 99,462 0.00 566 0.03 133 0.01 200 0.04 -0.03 -0.01 -0.03 0.02 0.00 -0.03
F-score 95,084 0.95 553 1.53 125 1.52 194 1.79 -0.58 -0.57 -0.84 0.01 -0.26 -0.27
Restatement 140,927 0.12 734 0.44 172 0.56 236 0.87 -0.32 -0.44 -0.76 -0.12 -0.44 -0.32

Panel B: Adverse consequences after fraud


CEO left 75,918 0.08 358 0.23 73 0.27 106 0.37 -0.15 -0.19 -0.28 -0.04 -0.13 -0.09
CFO left 66,523 0.10 318 0.26 59 0.20 86 0.40 -0.15 -0.10 -0.29 0.06 -0.14 -0.19
Δ Spread 90,486 -0.01 558 0.40 104 0.10 151 0.30 -0.42 -0.11 -0.31 0.30 0.11 -0.19
Δ Forc. Accuracy 61,179 -0.01 446 -0.02 72 -0.01 116 -0.02 0.02 0.01 0.02 -0.01 0.00 0.01
Δ Inst. Own. 133,266 1.98 616 -1.17 125 2.34 183 -3.91 3.15 -0.36 5.89 -3.51 2.74 6.25
This table presents financial statement characteristics during the fraud period (Panel A) and adverse consequences between the fiscal year of fraud end and the prior year
(Panel B). Column 1 presents results for the Compustat population, Column 2 for firms facing settled accounting 10b-5 SCAs only, Column 3 for firms facing settled
accounting 10b-5 SEC enforcement only, and Column 4 for firms facing both settled accounting 10b-5 SCA and SEC enforcement. The table also presents mean difference t-
tests across regimes. For a case to be included, it needs to have available data for total assets and income before extraordinary items in the last fiscal year prior to fraud period
end. Differences are bolded if they are statistically significant at 10 percent levels, two-tailed. Variables are defined in Appendix A. All continuous variables are winsorized at
the 1st and 99th percentile.

43

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 7
SEC Cases that Allege Rule 10b-5 Violations Versus Those that Do Not
Settled Acct Settled Acct Mean
Non-10b-5 SEC Cases 10b-5 SEC Cases Diff.
N Mean N Mean
Panel A: Accounting aggressiveness during fraud period
Abn. Accruals 171 0.04 333 0.03 0.01
Performance-Adj. Abn. Accruals 171 0.02 333 0.03 0.00
F-score 169 1.29 319 1.68 -0.39
Restatement 226 0.54 408 0.74 -0.20

Panel B: Adverse consequences after fraud


CEO left 135 0.19 179 0.33 -0.14
CFO left 128 0.2 145 0.32 -0.12
Δ Spread 188 0.09 255 0.22 -0.12
Δ Forc. Accuracy 154 -0.01 188 -0.02 0.01
Δ Inst. Own. 208 2.02 308 -1.37 3.39
This table presents summary statistics for settled accounting 10b-5 SEC cases versus settled accounting non-10b-5
SEC cases and presents results of mean differences. Panel A presents financial statement characteristics during the
fraud period, and Panel B adverse consequences between the fiscal year of fraud end and the prior year. For a case
to be included, it needs to have available data for total assets and income before extraordinary items in the last
fiscal year prior to fraud period end. Differences are bolded if they are statistically significant at 10 percent levels,
two-tailed. Variables are defined in Appendix A. All continuous variables are winsorized at the 1st and 99th
percentile.

44

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 8
Big 5 Audits and Accounting Fraud
Panel A: Fiscal Years 1998-2005
Settled Acct Either Settled Acct
Settled Acct 10b-5 SCA
10b-5 SEC 10b-5 SEC or SCA
(1) (2) (3)
Big 5t -0.315* 0.023 -0.140
(-1.87) (0.18) (-1.22)
Log Sizet 0.280*** 0.340*** 0.308***
(9.58) (17.14) (16.50)
Firm Aget -0.063 -0.074 -0.089*
(-0.81) (-1.40) (-1.75)
Audit Tenuret 0.014 -0.134*** -0.076*
(0.20) (-2.93) (-1.76)
Neg. Equityt 0.052 0.010 -0.001
(0.29) (0.08) (-0.01)
M&At 0.739*** 0.583*** 0.598***
(7.22) (7.48) (8.13)
Debt & Equity Iss.t 0.170 0.206*** 0.225***
(1.56) (2.76) (3.25)
Losst 0.212** 0.614*** 0.522***
(2.08) (8.13) (7.32)
Fraud firm-years 1,019 1,708 2,031
Year FE Yes Yes Yes
Pseudo R-squared 0.05 0.08 0.06
Observations 59,476 59,476 59,476
Panel B: Fiscal Years 2006-2014
Settled Acct Either Settled Acct
Settled Acct 10b-5 SCA
10b-5 SEC 10b-5 SEC or SCA
Big 5t -0.565** -0.142 -0.338**
(-2.23) (-0.92) (-2.35)
Controls Included? Yes Yes Yes
Fraud firm-years 243 693 812
Year FE Yes Yes Yes
Pseudo R-squared 0.04 0.06 0.05
Observations 57,986 57,986 57,986
Panel C: Fiscal Years 1998-2014
Settled Acct Either Settled Acct
Settled Acct 10b-5 SCA
10b-5 SEC 10b-5 SEC or SCA
Big 5t -0.430*** -0.084 -0.250***
(-3.01) (-0.82) (-2.73)
Controls Included? Yes Yes Yes
Fraud firm-years 1,262 2,401 2,843
Year FE Yes Yes Yes
Pseudo R-squared 0.08 0.09 0.08
Observations 117,462 117,462 117,462
This table reports a logit model to examine the effect of having a Big N auditor on the probability of fraud, similar to
Lennox and Pittman (2010), and how inferences change based upon the use of public and private enforcement to
proxy for fraud. Panel A [B] (C) uses fiscal years 1998 to 2005 [2006 to 2014] (1998 to 2014). Column 1 uses
settled accounting 10b-5 SEC cases as indicators for fraud, Column 2 uses settled accounting 10b-5 SCA cases, and
Column 3 uses settled accounting 10b-5 SEC or SCA cases. Following Lennox and Pittman (2010), the dependent
variable takes the value of one if the company engaged in accounting fraud at any point during the fiscal year.
Variables are defined in Appendix A. All continuous variables are winsorized at the 1st and 99th percentile. Standard
errors are adjusted for clustering at the company level, and year fixed effects are included. *, **, and *** denote
statistical significance levels of 0.10, 0.05, and 0.01, respectively, two-tailed.

45

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 9
Effect of Accounting Fraud on Abnormal Inventory Turnover
Panel A: Fiscal Years 1998-2005
Either Settled Acct
Settled Acct 10b-5 SEC Settled Acct 10b-5 SCA
10b-5 SEC or SCA
(1) (2) (3)
Abnormal Inventoryt 5.060*** 4.451*** 4.249***
(4.32) (4.71) (5.12)
Abnormal Revenuet 2.060* 3.836*** 2.963***
(1.89) (4.49) (3.87)
Log Sizet 0.273*** 0.358*** 0.313***
(8.85) (16.82) (15.38)
Firm Aget -0.113 -0.215*** -0.197***
(-1.24) (-3.24) (-3.13)
Audit Tenuret -0.032 -0.129** -0.080
(-0.41) (-2.36) (-1.53)
Neg. Equityt 0.027 0.015 0.016
(0.14) (0.10) (0.12)
M&At 0.675*** 0.564*** 0.576***
(6.16) (6.55) (7.16)
Debt & Equity Iss.t -0.009 0.073 0.110
(-0.08) (0.92) (1.48)
Losst 0.204* 0.616*** 0.505***
(1.84) (7.15) (6.25)
Year FE Yes Yes Yes
Pseudo R-squared 0.05 0.08 0.06
Observations 42,786 42,786 42,786
Panel B: Fiscal Years 2006-2014
Either Settled Acct
Settled Acct 10b-5 SEC Settled Acct 10b-5 SCA
10b-5 SEC or SCA
Abnormal Inventoryt 2.023 4.937*** 4.581***
(1.15) (3.16) (3.41)
Abnormal Revenuet 5.630*** 4.650*** 4.678***
(3.79) (3.49) (4.03)
Controls Included? Yes Yes Yes
Year FE Yes Yes Yes
Pseudo R-squared 0.05 0.05 0.04
Observations 36,916 36,916 36,916
Panel C: Fiscal Years 1998-2014
Either Settled Acct
Settled Acct 10b-5 SEC Settled Acct 10b-5 SCA
10b-5 SEC or SCA
Abnormal Inventoryt 4.499*** 4.555*** 4.306***
(4.48) (5.70) (6.12)
Abnormal Revenuet 2.790*** 4.032*** 3.389***
(2.96) (5.71) (5.32)
Controls Included? Yes Yes Yes
Year FE Yes Yes Yes
Pseudo R-squared 0.08 0.08 0.07
Observations 79,702 79,702 79,702
This table reports a logit model to examine abnormal inventory and the probability of fraud and how inferences
change based upon the proxy for fraud. Panel A [B] (C) uses fiscal years 1998 to 2005 [2006 to 2014] (1998 to
2014). Column 1 uses settled accounting 10b-5 SEC cases as indicators for fraud, Column 2 uses settled accounting
10b-5 SCA cases, and Column 3 uses settled accounting 10b-5 SEC or SCA cases. Following Lennox and Pittman
(2010), the dependent variable takes the value of one if the company engaged in accounting fraud at any point
during the fiscal year. Variables are defined in Appendix A. All continuous variables are winsorized at the 1st and
99th percentile. Standard errors are adjusted for clustering at the company level, and year fixed effects are included.
*, **, and *** denote statistical significance levels of 0.10, 0.05, and 0.01, respectively, two-tailed.

46

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 10
Distinguishing between Restatements and Fraud using Audit Analytics Data

Panel A: Reconciliation of restatements in Audit Analytics to accounting fraud


Total Included in our Excluded from our
Restatement Categories Restatements Fraud Sample Fraud Sample
All restatements 10,166 748 9,418
All non-clerical restatements 9,722 729 8,993
All restatements flagged as
fraudulent by Audit Analytics 202 87 115

Panel B: Breakdown of restatements flagged as fraudulent by Audit Analytics and which are not in our
Fraud Sample
N
Fraud committed by lower level employees 69
Indirectly related to accounting 19
Dismissed accounting 10b-5 SCA, SEC investigation, or settled non-10b-5 SEC cases 11
Potentially credible fraud allegations with no enforcement 10
Restatements with related enforcement but not in our sample due to missing data in our
6
primary data sources or with class periods extending past our sample period end
Total 115

47

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 10, Continued
Panel C: Do Audit Analytics fraudulent observations that are not in our sample look similar to fraudulent cases in our sample?

Restatements classified as Restatements classified as fraudulent


fraudulent by Audit Analytics All fraud observations in our by Audit Analytics that are in our
that are not in our sample sample sample
(1) (2) (3)
Mean Median Mean Diff. Median Diff. Mean Diff. Median Diff.
Panel C1: Accounting aggressiveness during fraud period
Abn. Accruals 0.00 0.01 0.04 0.04 0.02 0.01 0.02 0.02 0.03 0.02
Performance-Adj. Abn.
Accruals 0.01 0.00 0.03 0.02 0.01 0.01 0.01 0.00 0.02 0.02
F-Score 1.15 0.92 1.58 0.43 1.26 0.34 1.63 0.47 1.22 0.31

Panel C2: Adverse consequences after fraud


Δ Spread -0.02 -0.03 0.34 0.37 0.17 0.20 -0.07 -0.04 -0.04 0.00
Δ Forc. Accuracy -0.01 0.00 -0.02 -0.02 0.00 0.00 -0.03 -0.03 0.00 0.00
Δ Inst. Own. 2.49 0.00 -1.23 -3.73 -0.05 -0.05 1.62 -0.88 0.16 0.16
CAR[-1, 1] -0.09 -0.03 -0.27 -0.18 -0.05 -0.02
Log Net Income Effect 15.38 14.99 17.23 1.85 16.88 1.89
Panel A reconciles various populations of Audit Analytics restatements to our sample of fraud observations. The sample consists of restatement
announcements from Audit Analytics with restatement end period between 1998 and 2014 that we are able to match to Compustat. Panel B further examines
restatements classified by Audit Analytics as fraudulent but which are not in our fraud sample. Categories are mutually exclusive. Thus, restatements are
only classified as “Dismissed SCA or SEC investigation or settled non-10b-5 SEC cases”, for example, if it meets both these criteria and is not already
classified into one of the first two categories. Restatements are classified as “Potentially credible fraud allegations” if the fraud is not already classified in
one of the first three categories and the description in the restatement announcement presents severe accounting allegations of accounting fraud committed
by executives. Panel C presents financial statement characteristics (Panel C1) and adverse consequences (Panel C2) between restatements classified by
Audit Analytics as fraudulent but which are not in our sample (Column 1), observations in our sample with either a settled accounting 10b-5 SEC or SCA
case (Column 2), and restatements classified by Audit Analytics as fraudulent and which are in our sample (Column 3). The observations in Column 1 (2)
[3] range from 56 to 110 (634 to 924) (45 to 87) due to data availability. Panel C also presents mean difference t-tests across regimes, as well as Mann-
Whitney median tests between the restatements in Column 1 versus our entire fraud sample in Column 2 or restatements classified as fraudulent and are in
our sample in Column 3. For a case to be included in Panel C, it needs to have available data for total assets and income before extraordinary items in the
last fiscal year prior to restatement period end. Differences are bolded if they are statistically significant at 10 percent levels, two-tailed. Variables are
defined in Appendix A. All continuous variables are winsorized at 1st and 99th percentile.

48

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 11
Prediction Model
Panel A: Cross-Regime Predictability
Predicting Settled Acct 10b-5 SEC Predicting Settled Acct 10b-5 SCA
(1) (2) (3) (4) (5) (6)
Intercept -6.04*** -6.68*** -8.09*** -5.14*** -6.46*** -7.61***
(-24.92) (-20.95) (-21.54) (-34.95) (-34.21) (-32.66)
RSST accrualst -0.12 -0.44*** 0.05 -0.17*
(-0.81) (-3.20) (0.49) (-1.75)
Δ Receivablest 0.96* -0.18 0.61 0.22
(1.67) (-0.28) (1.43) (0.45)
Δ Inventoryt 0.30 -0.73 0.38 -0.01
(0.40) (-0.92) (0.66) (-0.02)
% of Soft Assetst 1.78*** 1.97*** 1.43*** 1.60***
(6.52) (6.82) (8.74) (9.15)
Δ Cash Salest 0.20*** 0.00 0.21*** 0.01
(6.26) (0.01) (9.53) (0.43)
ΔROAt -0.69*** -0.39** -1.26*** -0.51***
(-3.65) (-2.13) (-9.32) (-4.03)
Actual Issuancet 0.73*** 0.46** 0.68*** 0.38***
(4.01) (2.39) (6.29) (3.47)
FPSt 0.69*** 0.70*** 0.33*** 0.33***
(4.61) (4.76) (3.52) (3.54)
Sales Growtht 0.68*** 0.70*** 0.61*** 0.51***
(7.28) (4.86) (9.10) (5.24)
Log Sizet 0.22*** 0.19*** 0.24*** 0.23***
(5.65) (4.90) (10.82) (9.93)
Return Volatilityt 3.26*** 3.08*** 5.29*** 5.21***
(6.06) (5.61) (15.48) (14.77)
Cumul. Abn. Returnt -0.21*** -0.13** -0.55*** -0.49***
(-4.03) (-2.49) (-14.62) (-12.67)
Return Skewnesst -0.17*** -0.16*** -0.31*** -0.30***
(-3.64) (-3.57) (-9.32) (-9.14)
Shares Turnovert 0.07*** 0.09*** 0.09*** 0.11***
(3.18) (4.15) (7.68) (8.72)
AUROC 0.65 0.68 0.73 0.64 0.73 0.75
Pseudo R-squared 0.03 0.05 0.07 0.03 0.07 0.09
Observations 71,756 71,756 71,756 71,756 71,756 71,756

49

Electronic copy available at: https://ssrn.com/abstract=3744392


TABLE 11, Continued
Panel B: Fraud Prediction Model – Predicting Either Settled Acct 10b-5 SEC or SCA
(1) (2) (3)

Intercept -4.92*** -6.01*** -7.13***


(-34.80) (-32.89) (-32.44)
RSST accrualst 0.02 -0.19**
(0.23) (-2.17)
Δ Receivablest 0.63 0.10
(1.58) (0.21)
Δ Inventoryt 0.52 0.03
(0.99) (0.05)
% of Soft Assetst 1.38*** 1.53***
(8.67) (9.06)
Δ Cash Salest 0.20*** 0.01
(9.48) (0.34)
Δ ROAt -1.17*** -0.53***
(-9.14) (-4.48)
Actual Issuancet 0.66*** 0.39***
(6.15) (3.50)
FPSt 0.42*** 0.42***
(4.59) (4.63)
Sales Growtht 0.61*** 0.53***
(9.79) (5.75)
Log Sizet 0.22*** 0.20***
(9.84) (8.79)
Return Volatilityt 4.74*** 4.64***
(14.46) (13.79)
Cumul. Abn. Returnt -0.48*** -0.42***
(-13.36) (-11.35)
Return Skewnesst -0.27*** -0.27***
(-8.86) (-8.70)
Shares Turnovert 0.08*** 0.10***
(7.00) (8.04)

AUROC 0.64 0.71 0.73


Pseudo R-squared 0.02 0.06 0.08
Observations 71,756 71,756 71,756
This table presents results examining predictors of fraudulent firm-years using a logit model. Panel A
examines fraudulent firm-years as identified using settled accounting 10b-5 SEC cases (settled accounting
10b-5 SCA cases) in Columns 1-3 (4-6). Columns 1 and 4 use predictors from DGLS, Columns 2 and 5 use
predictors from KS, and Columns 3 and 6 combine both. Panel B examines fraudulent firm-years as identified
using settled accounting 10b-5 SEC or SCA cases. Following Lennox and Pittman (2010), the dependent
variable takes the value of one if the company engaged in accounting fraud at any point during the fiscal year.
Variables are defined in Appendix A. All continuous variables are winsorized at the 1st and 99th percentile.
Standard errors are adjusted for clustering at the company level. *, **, and *** denote statistical significance
levels of 0.10, 0.05, and 0.01, respectively, two-tailed.

50

Electronic copy available at: https://ssrn.com/abstract=3744392

You might also like