Measuring securities litigation risk

Transcription

Measuring securities litigation risk
Journal of Accounting and Economics 53 (2012) 290–310
Contents lists available at SciVerse ScienceDirect
Journal of Accounting and Economics
journal homepage: www.elsevier.com/locate/jae
Measuring securities litigation risk$
Irene Kim a, Douglas J. Skinner b,n
a
b
George Washington University, School of Business, United States
University of Chicago, Booth School of Business, United States
a r t i c l e in f o
abstract
Article history:
Received 25 August 2010
Received in revised form
6 September 2011
Accepted 21 September 2011
Available online 6 November 2011
Extant research commonly uses indicator variables for industry membership to proxy
for securities litigation risk. We provide evidence on the construct validity of this
measure by reporting on the predictive ability of alternative models of litigation risk.
While the industry measure alone does a relatively poor job of predicting litigation,
supplementing this variable with measures of firm characteristics (such as size, growth,
and stock volatility) considerably improves predictive ability. Additional variables such
as those that proxy for corporate governance quality and managerial opportunism do
not add much to predictive ability and so do not meet the cost–benefit test for
inclusion.
& 2011 Elsevier B.V. All rights reserved.
JEL classification:
K22
K41
M41
Keywords:
Litigation risk
Securities litigation
Corporate disclosure
1. Introduction
A large body of research in accounting and finance investigates whether litigation risk (the risk of securities class action
lawsuits) affects corporate decisions. While much research investigates the effect of litigation risk on managers’ disclosure
choices, authors also investigate how litigation affects a large array of managerial decisions.1
Much of this research measures litigation risk using an industry-based proxy, either alone or in conjunction with other
variables. A common proxy is based on membership in the biotechnology, computers, electronics, and retail industries.
This proxy originates from Francis, Philbrick and Schipper (1994a, 1994b; hereafter FPS), who sample firms drawn from
these industries to study the relation between litigation and disclosure because those industries were subject to ‘‘a high
incidence of litigation during 1988–1992’’ (1994a, p. 144). These authors do not advocate the use of industry membership
generally, or these industries in particular, as a universal proxy for litigation risk. However, the use of this industry proxy
(hereafter, the FPS measure) has become pervasive in the literature.
$
We appreciate the comments of Bill Mayew, Karen Nelson, Adam Pritchard, Jonathan Rogers, Jerry Zimmerman (editor), Kin Lo (referee), and seminar
participants at the University of Chicago, Duke University, Northwestern University, and the DC Area Accounting Symposium. Kim and Skinner appreciate
financial support from the George Washington University School of Business and the University of Chicago Booth School of Business, respectively.
n
Corresponding author. Tel.: þ 1 773 702 7137.
E-mail addresses: [email protected] (I. Kim), [email protected] (D.J. Skinner).
1
Papers that investigate the relation between managers’ financial reporting and disclosure decisions and litigation risk include Skinner (1994, 1997),
Francis et al. (1994a, 1994b), Johnson et al. (2000, 2001), Baginski et al. (2002), Frankel et al. (2002), Matsumoto (2002), Field et al. (2005), Lennox and
Park (2006), Rogers and Van Buskirk (2009), and Donelson et al. (2010), among others. Research also examines how litigation risk affects cash holdings
(Arena and Julio, 2010), equity-based compensation (Dai et al., 2008; Jayaraman and Milbourn, 2009), conservatism in debt contracting (Beatty et al.,
2008), IPO underpricing (Lowry and Shu, 2002; Weiss Hanley and Hoberg, 2010), institutional monitoring and board discipline (Cheng et al., 2010; Laux,
2010), MD&A disclosures (Brown and Tucker, 2011), audit fees (Seetharaman et al., 2002), and auditors’ resignation decisions (Shu, 2000).
0165-4101/$ - see front matter & 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.jacceco.2011.09.005
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
291
It is reasonable to expect that litigation is associated with industry membership. Stock volatility and stock turnover
directly affect litigation risk because both are directly related to measures of stockholder damages that drive plaintiff
lawyers’ decisions to file lawsuits (e.g., Alexander, 1991; Jones and Weingram, 1996a). Both of these variables are likely to
be associated with industry; for example, high technology stocks are by their nature inherently more uncertain with more
variable earnings, and hence are more volatile.
The use of industry to proxy for litigation risk results from a cost–benefit tradeoff by researchers. While this proxy is
simple and readily available, it likely captures industry characteristics that are unrelated to litigation risk but that affect
managers’ decisions, creating a potential correlated omitted variables problem. The fact that this proxy is ubiquitous in the
literature seems to indicate that it passes the cost–benefit test. However, there is little evidence (of which we are aware)
on the construct validity of this proxy or whether other proxies are available that might represent a better cost–benefit
tradeoff. Further, beyond reporting pseudo-R squareds, there is little systematic evidence on the ability of extant measures
to actually predict litigation. We report on some relatively simple and low cost models that significantly outperform the
industry-based proxies in terms of predictive and discriminatory ability.
The use of industry membership to capture litigation risk makes it difficult to ensure that industry captures litigation
risk as opposed to different underlying factors that affect managers’ disclosure decisions. Consider a study that investigates
whether litigation risk affects managers’ disclosure choices and uses industry to proxy for litigation risk. If managers’
disclosure decisions depend on their firms’ information environments (Einhorn and Ziv, 2008) and information
environment varies systematically across industry, disclosure will be associated with industry for reasons that have little
to do with litigation risk.2 A similar problem arises if firms in high technology industries have higher proprietary costs than
firms in more mature industries and proprietary costs systematically affect disclosure.
The existence of a well-developed theory of litigation would allow us to identify all of the economic determinants of
litigation, in which case the FPS measure would presumably no longer be useful in explaining litigation risk. Although we
do not have such a theory (we discuss previous literature in Section 2), one goal of our research is to investigate
systematically whether the inclusion of an extensive set of firm-specific characteristics reduces the usefulness of the FPS
variable in predicting litigation, as would be expected if these characteristics directly capture litigation risk.
We provide two sets of empirical analyses to evaluate how well industry membership proxies for securities litigation
risk. We first provide evidence on how litigation rates vary across industries and through time. This evidence shows that
while litigation tends to cluster in certain industries, the set of industries varies over time. Nevertheless, the FPS industries
generally have consistently higher litigation rates than other industries, although this result is weaker when we focus the
analysis on large firms generally subject to higher rates of litigation.
Second, we provide evidence on the predictive ability of alternative models of litigation risk. We show that while the
relationship between the FPS industry measure and litigation is robust in a statistical sense, using industry membership
alone does a relatively poor job of predicting litigation. However, when we supplement this variable with measures of firm
characteristics that include size, growth, and stock performance and volatility, predictive ability improves considerably.
These variables are readily available to researchers in a broad variety of settings. Further, including additional variables,
such as proxies for corporate governance quality, issuance of securities, insider trading, and so forth, adds relatively little
to predictive ability. Given the cost of obtaining these variables (which includes possible sample selection biases), more
sophisticated models that include these variables are unlikely to be cost beneficial.
Conventional measures of goodness of fit (such as pseudo-R-squareds) do not perform well in assessing the fit and
predictive ability of these models. We use a number of alternative approaches suggested in the statistics literature (e.g.,
Hosmer and Lemeshow, 2000; Long and Freese, 2006) to evaluate model fit and predictive ability, most notably the area
under the receiver operating characteristic (ROC) curve, or AUC.3 These techniques confirm that models that supplement
the FPS measure with a small set of variables that are readily available from CRSP/Compustat provide significant
improvements in predictive ability relative to a model that includes the FPS measure alone.
By securities litigation risk, we are referring specifically to the risk of securities class action lawsuits, as opposed to the
risk of legal action brought by government agencies such as the U.S. Securities and Exchange Commission (SEC), the U.S.
Department of Justice, or state attorney generals, which we view as a related but distinct form of litigation risk. SEC
Accounting and Auditing Enforcement Releases (AAERs) have been extensively studied in the accounting literature (see
Feroz et al., 1991; Beneish, 1999; Dechow et al., 1996, 2011; Schrand and Zechman, 2011, among others).4 As noted in
those studies, SEC enforcement actions typically result from cases of serious accounting irregularities, including fraud.
While such cases are likely to lead to securities class actions, many securities class actions involve less serious allegations,
2
For example, financial statements are relatively less useful for firms in high technology industries with significant intellectual property and other
intangibles (e.g., Lev and Zarowin, 1999; Tasker, 1998) so that managers of firms in these industries have stronger incentives to provide voluntary
disclosures as a substitute for mandated disclosure.
3
To this point, these methods have been used infrequently in the accounting literature but, as argued below, are useful for evaluating and comparing
the predictive ability of different models. Recent accounting papers that use this measure include Batta and Wongsunwai (2011), Correia et al. (2011),
Demers and Joos (2007), Hobson et al. (2011), and Larcker and Zakolyukina (2011).
4
Actions brought by other government agencies, such as the U.S. Department of Justice and state attorney generals usually involve criminal
allegations, and are a subset of SEC enforcement actions (i.e., not all SEC enforcement actions relate to allegations serious enough to warrant allegations
of criminal wrongdoing).
292
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
such as failure to disclose in a timely manner, and so do not result in SEC enforcement actions. Consistent with this, we
provide evidence that less than 10% of securities class actions are associated with SEC enforcement actions.
Our goal is to measure ex ante litigation risk. It is well known that certain factors, most notably large and sudden
declines in stock price at the time of an information release, increase the risk of litigation considerably ex post
(Alexander, 1991; Jones and Weingram, 1996a). Our goal is not to examine whether outcomes such as stock price
declines or accounting frauds result in litigation—the evidence confirms that this is the case (e.g., Hennes et al., 2008).
Instead, our goal is to capture factors that make firms more vulnerable to litigation before such ‘‘triggering events’’ occur,
which is the construct likely to be of most interest to researchers investigating, say, how litigation risk affects firms’
ongoing disclosure practices.
Our paper contributes to the literature in several ways. First, we provide comprehensive evidence on the usefulness of
the FPS industry variable as a measure and predictor of litigation risk, an important task given the ubiquity of this measure
in the literature. Second, we provide evidence that allows us to better understand what makes particular firms and
industries vulnerable to litigation. Third, we provide more precise measures of the predictive ability and goodness of fit of
models of litigation risk than those typically used in prior literature.
Section 2 reviews previous research. Section 3 details the sample and provides evidence on how litigation rates vary
over time and across industries and sectors. Section 4 provides evidence on the determinants of litigation risk, comparing
the predictive ability of the conventional FPS industry proxy to models that supplement this proxy with additional drivers
of litigation risk. Section 5 concludes.
2. Previous research and empirical predictions
2.1. Previous research on litigation risk
A considerable body of research in accounting and finance investigates how private securities litigation affects various
corporate policies and managerial decisions. Much of this research uses some variant of the FPS industry proxy for
litigation risk. For example, many papers use some form of dummy variable for membership in the FPS industries to
measure litigation risk (e.g. Matsumoto, 2002; Ajinkya et al., 2005; Beatty et al., 2008; Jayaraman and Milbourn, 2009;
Bhojraj et al., 2010; Brown and Tucker, 2011; Donelson et al., 2010; Hribar et al., 2010).
Other authors who examine the determinants and effects of litigation risk limit their samples to firms in the FPS
industries.5 Ali and Kallapur (2001), Johnson et al. (2000, 2001, 2007), and Choi (2006), examine various hypothesized
effects of the Private Securities Litigation Reform Act of 1995 (PSLRA).6 These authors restrict attention to samples of firms
drawn from the three high technology industries identified by FPS (Ali and Kallapur show that their results are robust to
using a broader sample of firms). Choi (2006) uses a high technology industry dummy as an explanatory variable for
litigation pre- and post-PSLRA. Chandra et al. (2004) limit their sample to firms in high technology industries because high
litigation risk is hypothesized to explain heightened income conservatism in these industries.
A number of papers use predicted probabilities from models of litigation risk to measure litigation risk. These models
typically include firm characteristics such as market capitalization, stock volatility, and stock turnover as well as, in some
cases, industry dummies based on FPS.7 Johnson et al. (2000) estimate a probit model that explains lawsuit filings as a
function of market capitalization, stock beta, cumulative stock return, minimum stock return, return skewness, stock
turnover, CEO power, management monitoring, external financings, and insider trading. The market capitalization and
stock return variables, including turnover, come from previous research (Alexander, 1991; Jones and Weingram, 1996a;
Skinner, 1997) based on the idea that damages in Rule 10b-5 litigation depend on the size of the price decline, the number
of shares traded during the period of the alleged fraud, and the stock price.8 (Larger potential damages amounts make
firms more attractive to plaintiffs’ attorneys, ceteris paribus, which explains why market capitalization is strongly
associated with litigation risk.)
The inclusion of the CEO power and monitoring variables is motivated by the fact that CEOs who have more power
and/or are less closely monitored are more likely to engage in aggressive financial reporting and other types of
5
Field et al. (2005) also use industry membership to measure litigation risk but develop their own measure by looking at industry litigation rates
during the period before the sample test period (1988–1994) and sorting industries according to whether they had litigation rates above or below the
median.
6
Choi et al. (2009) use a sample of firms from the high-technology sector because for other sectors ‘‘the incidence of litigation has fluctuated over
time for reasons unrelated to the passage of the PSLRA’’ (page 46). They also improve the generalizability of their sample by including other randomlyselected non-financial industries.
7
Jones and Weingram (1996b) use a regression model to explain why technology and financial services firms experienced a high level of securities
litigation in the period from 1989 to 1992. They include several stock market explanatory variables in a logit regression, along with a technology and
financial services indicator. They find that the technology dummy is statistically insignificant after controlling for stock return variables, but the financial
services dummy remains significant. They attribute this finding to an industry effect of the savings and loan crisis.
8
There is little in the way of theory to guide the choice of variable selection in these models. Most research in the law and economics literature
discusses the incentives of plaintiffs’ attorneys to file suit (e.g., Alexander, 1991; Coffee, 1985, 1986; Cox and Thomas, 2006; Dunbar et al., 1995; Johnson
et al., 2000; Jones and Weingram, 1996a, 1996b; Romano, 1991). These incentives are directly related to potential damages, which explains the inclusion
of variables such as firm size, stock volatility, stock turnover, and the likelihood of price drop. This literature is predicated on the idea that most suits are
‘‘nuisance’’ suits that are unrelated to the extent of managerial wrongdoing (e.g., Alexander, 1991; Romano, 1991).
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
293
opportunistic behavior that increases exposure to securities litigation (Dechow et al., 1996). This is supported by evidence
that underwriters of D&O liability insurance focus on corporate governance quality to assess liability risk (Baker and
Griffith, 2007) and that consulting groups use governance quality to predict litigation risk (The Corporate Library, 2009).9
These studies also commonly include measures of the extent of insider trading which arguably increases the likelihood
of stockholder litigation, especially after PSLRA, as well as the extent to which firms have recently issued debt and/or
equity. Insider trading activity and external financing provide opportunities for managers to exploit high market
valuations; if the valuations are achieved using what can be alleged to be false or misleading information, these activities
increase the probability of a lawsuit filing.10
Other papers that model litigation risk include Brown et al. (2005) and Rogers and Stocken (2005), both of which use
largely the same set of variables as in Johnson et al. (2000) but supplement them with FPS industry dummy variables.
While corporate governance and insider trading variables are plausible measures of managerial opportunism that
increase firms’ exposure to litigation, there are two issues when including these types of variables in litigation risk models.
First, it is not clear a priori that most securities litigation results from opportunism by managers as opposed to being driven
by adverse outcomes. While it is clear that extreme forms of opportunism such as accounting frauds lead to litigation,
these suits form a relatively small part of the population of securities class action suits (Section 3). Second, corporate
governance and insider trading data are less widely available than data on firm characteristics such as size and volatility,
constraining sample sizes and perhaps also biasing sample selection.
Although there is some agreement in the literature in terms of the set of covariates usually included in litigation risk
models, apart from reporting pseudo-R-squareds, these studies rarely report or discuss in much detail the goodness of fit
or predictive ability of these models. Consequently, there is little evidence on the construct validity of the litigation risk
measures used in previous research. This is a primary motivation for our research.
In the introduction we draw a distinction between the risk of securities class action lawsuits (which we refer to as
litigation risk) and the risk of SEC enforcement actions and other types of government legal actions against firms and
managers. While securities class actions can allege a range of management improprieties, these are for the most part less
serious than those alleged in SEC enforcement actions (AAERs), many of which relate to accounting fraud. Moreover,
securities class actions often result from bad outcomes as opposed to malevolent management actions. These differences
imply that variables used to predict SEC enforcement actions will be somewhat different to those used to predict securities
class actions. Consistent with this, Dechow et al. (2011), who use prior literature to develop a comprehensive prediction
model for AAERs, include a number of accruals quality variables (which are likely linked to deliberate earnings
overstatements), as well as other variables that capture managerial incentives to overstate earnings. While we include
some of these variables in our analysis, the main focus of our prediction model is on variables that make firms and
industries vulnerable to lawsuit filings.
2.2. Assessing the validity of the FPS litigation risk proxy
The ability of a fixed industry proxy to reliably capture litigation risk is reduced if litigation rates in particular industries
vary over time. If economy-wide events cause time-series variation in the fortunes of different industries, litigation risk is
not likely to be specific to particular industries or firms. Instead, it is more likely that economic shocks cause losses in
value that vary across industries and through time, and that these losses trigger litigation.11 This means that it will be hard
to identify particular industries that are always subject to higher litigation risk. FPS identify their four industries based on
observed litigation rates during the 1988–1992 period, which includes the recession of the early 1990s.
On the other hand, it is possible that there are firm and industry characteristics that make particular firms and
industries generally more susceptible to litigation. For example, firms that operate in more volatile operating environments have greater levels of stock volatility, which makes them more susceptible to litigation.
This discussion motivates two principal types of empirical tests. First, we provide evidence on the extent to which
litigation rates vary across economic sectors and industries over time. Second, we compare the predictive ability of the FPS
industry dummy to that of models that also include other firm characteristics likely to drive litigation risk. While including
additional variables will usually increase predictive ability, our purpose is to gauge empirically the magnitude of these
improvements to provide researchers with information they can use to make the tradeoffs necessary in choosing a proxy
for litigation risk. Further, to the extent that underlying firm characteristics (rather than industry membership per se) drive
litigation risk, we expect the predictive power of the FPS measure to decline as these more direct measures are included in
predictive models.
9
Daines et al. (2010) find that corporate governance and transparency ratings, such as those produced by Risk Metrics/ISS, Governance Metrics
International and the Corporate Library, do not have predictive power for identifying lawsuit filings.
10
Johnson et al. (2007) find that abnormal insider selling is more strongly associated with litigation after PSLRA, consistent with their prediction that
the Act’s more stringent pleading requirements encourages lawyers to focus on more objective evidence of managerial malfeasance. See also Pritchard
and Sale (2005), who discuss the role of insider trading and securities issuances in litigation after PSLRA.
11
Consider two economy-wide events that occur during our sample period—the large shock to prices of technology stocks in 2001 and the financial
crisis of 2007–2008. The damage, measured in terms of stockholders’ value losses, of both shocks was concentrated in particular industries. If the effect of
shocks varies across firms within a given industry (Albuquerque, 2009), industry will be even less useful in predicting litigation risk.
294
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
We investigate a number of variables in addition to those conventionally employed in the literature to predict litigation
risk.12 We add a measure of economic performance (ROA) because as performance deteriorates and scrutiny on managers
increases, the risk of aggressive actions that potentially lead to litigation also increases. Similarly, we include Altman’s
(1968) Z-score to proxy for the likelihood of financial distress, which should also increase litigation risk. We include
several proxies for the nature of the firms’ investment opportunities (working capital, market-to-book, R&D intensity, ratio
of PP&E to total assets) because these variables affect corporate policies such as executive compensation, capital structure,
and payout policy that could affect litigation risk (Smith and Watts, 1992). We include the ratio of goodwill to assets to
measure the extent of the firms’ M&A activity, which likely increases litigation risk. We also include exchange listing (NYSE
dummy) to see if listing on the NYSE (as opposed to the NASDAQ) increases the risk of litigation. Finally, we include the
percentage of institutional ownership, and debt and equity issuance proceeds.
3. Sample and evidence on the relation between industry and litigation risk
We obtain data on filings of securities class action lawsuits from the Stanford Law School Securities Class Action Clearinghouse.
These data begin in 1996 and continue through the current time. We include lawsuits filed against public companies (listed on
the NYSE, ASE, or NASDAQ) and exclude the IPO allocation, mutual fund, and analyst lawsuits common around 2001.13
The Stanford database also has data on lawsuit outcomes. We restrict attention to lawsuit filings because we believe that
managers’ main goal is to avoid filings (and the associated legal, reputational, and time costs).14 Once suits are filed, there are
effectively two outcomes—dismissal or settlement. Since 1996, 43.0% of cases were dismissed while 56.7% were settled and 0.3%
went to trial (Cornerstone, 2011a, p. 14). Defendants typically wait to see whether their initial motion to dismiss is successful; if
it is not, they eventually settle the case.15 Variation in settlements is relatively small: of those cases settled between 1996 and
2010, 57% settled for less than $20 million while 80% settled for less than $25 million (Cornerstone, 2011b, p. 3).
We do not take a position on whether ‘‘bad managers’’ get sued; that is, whether managers took some type of
opportunistic or unlawful action that resulted in litigation. Instead, we are agnostic about managers’ guilt or innocence
and simply address whether, as an empirical matter, certain firm characteristics and/or industry membership increase the
likelihood of litigation. Clearly, managers who engage in malfeasance (illegal insider trading, accounting fraud) expose
their firms to greater overall litigation risk. However, it is also the case that firms and managers can be sued in situations
where there is no obvious wrongdoing. Whether the ‘‘merits matter,’’ in the sense that the legal process correctly identifies
and punishes managers who violate the securities laws (as opposed to being unlucky and getting a bad outcome), is the
subject of considerable discussion in the law and economics literature (e.g., Alexander, 1991; Romano, 1991).
Table 1, Panel A shows the number of lawsuits by year and sector (as defined by Bloomberg); the data are also shown in
Fig. 1. There are 2,497 filings from 1996 to 2009, for an average of approximately 178 per year. The number of lawsuits
each year varies considerably; for example, there were 220 lawsuits in 2004 but only 112 in 2006. It is not clear what
explains this variation. Cornerstone (2011a, p. 4) plots a measure of equity market volatility against filing activities. This
variable seems correlated with filings, perhaps because higher volatility increases the likelihood of the sharp declines in
firms’ stock prices that can trigger litigation.16
Panel A of Table 1 also reports the fraction of lawsuits that allege a violation of SEC Rule 10b-5 (a misstatement or omission
of material information), which are the lawsuits of most interest to researchers interested in financial reporting and disclosure
issues. The fraction of 10b-5 cases is high, averaging 89% for the overall period. The most common allegations (untabulated) are
material misrepresentations regarding the business, failure to warn, and accounting or internal control problems.
Panel B of Table 1 reports the percentage of lawsuits in each sector by year. It is clear that lawsuits cluster by sector. The most
lawsuits overall are in the technology (29%), services (21%), healthcare (15%), and financial (15%) sectors, consistent with
concentration by industry but somewhat inconsistent with the FPS measure, which is mainly focused on technology firms. At the
other end of the spectrum, the basic materials, capital goods, conglomerate, energy, and transportation sectors generally account
for a small fraction of litigation (in the 1% to 5% range). The rate of litigation also varies through time for a given sector. For
example, the share of litigation attributable to firms in the technology sector is high in 2000 and 2001 (38% and 42%) when stock
prices of technology firms fell dramatically but falls to 15% and 11% in 2008 and 2009. Conversely, financial firms had a relatively
12
We acknowledge that this choice of variables, while based on the law and economics literature discussed above, is rather ad hoc. However, part of
the motivation for this research is to see whether we can identify new factors that predict litigation risk, and in particular to see whether these factors
supplant industry as a predictor of litigation risk.
13
The Stanford database is commonly used as a source of lawsuit filings. To provide some assurance as to the completeness of these data, we search
the 10-K filings of all S&P 500 companies from 2007 through the present for references to 10b-5 litigation. Of the 500 companies, 46 had 10-K disclosures
indicating involvement in a 10b-5 securities class action. We found all of these cases in the Stanford database, providing assurance that it is reasonably
complete.
14
The law and economics literature suggests that the merits of the case do not greatly affect resolutions (e.g. Alexander, 1991; Romano, 1991;
Baker and Griffith, 2007; Choi, 2006; Johnson et al., 2007). This reinforces managers’ incentives to avoid litigation.
15
The tendency for both sides of securities class action lawsuits to have strong incentives to settle is well known; see Alexander (1991) and Romano
(1991) for further discussion.
16
Alexander (1991) points out that a lawsuit filing following a large stock price decline supports ‘‘an award of attorneys’ fees that would make it
worthwhile to bring a case’’ (p. 513). As an empirical matter it is well-established that sharp, significant stock price declines are associated with lawsuit
filings (Francis et al., 1994b; Jones and Weingram, 1996a).
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
295
Table 1
Number and percentage of lawsuits by year and sector.
In this table, Panels A and B provide the number and percentage of lawsuits by year and sector, respectively. The sector definitions are provided by
Bloomberg. The lawsuits in this table are lawsuits filed against publicly-held firms and exclude IPO allocation, mutual fund, and analyst lawsuits. Panel A
shows that the number of lawsuits can vary considerably across years. Panel A also shows the percentage of lawsuits alleging a Rule 10b-5 violation
(these data are also shown in Fig. 1). The percentages in Panel B are calculated by dividing the number of lawsuits for the sector-year by the total lawsuits
for all sectors in the relevant year; in other words, the percentages are the breakdown of sector litigation for the year.
Panel A: Number of lawsuits by year and sector
Sector
Basic materials
Capital goods
Conglomerates
Consumer cyclical
Consumer non-cyclical
Energy
Financial
Healthcare
Services
Technology
Transportation
Utilities
No sector provided
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Total
2
4
0
3
6
2
10
11
20
32
2
0
2
7
4
1
11
4
1
18
17
37
58
3
2
0
4
3
0
11
5
2
22
36
52
81
4
4
1
5
5
3
19
11
4
22
22
51
59
2
0
0
6
4
2
10
8
2
27
19
47
78
3
6
0
6
4
1
10
5
1
13
17
38
76
4
4
0
3
5
3
4
4
6
28
31
52
53
2
21
1
7
5
3
9
7
2
26
38
38
45
1
5
0
5
11
1
8
5
6
34
39
46
55
6
4
0
7
0
1
13
6
2
26
31
31
53
1
1
1
2
4
0
5
4
2
12
19
23
39
2
0
0
2
6
0
6
4
4
40
26
44
33
1
2
0
7
6
2
3
9
6
89
23
19
27
2
3
0
7
3
3
4
3
2
67
22
21
14
0
1
6
70
64
20
116
81
42
434
351
519
703
33
53
11
Total filings by year
94
163
225
203
212
179
213
186
220
173
112
168
196
153
2,497
% of Filings alleging Rule 10b-5 violation
89
90
91
93
91
88
93
96
94
95
90
82
79
75
89
Panel B: Percentage of total lawsuits by year and sector
Sector
Basic materials
Capital goods
Conglomerates
Consumer cyclical
Consumer non-cyclical
Energy
Financial
Healthcare
Services
Technology
Transportation
Utilities
No sector provided
Total
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 All Yrs
2
5
0
4
6
1
11
11
21
35
2
0
2
5
3
1
7
3
1
11
10
19
37
2
1
0
2
1
0
5
2
0
10
17
24
35
2
1
1
2
3
1
9
6
2
11
12
24
29
1
0
0
2
2
1
5
4
1
12
9
21
38
2
3
0
4
2
1
6
3
1
6
8
21
42
3
3
0
2
3
2
2
2
3
12
15
24
24
1
10
0
3
3
2
5
4
1
14
21
19
24
1
3
0
2
5
1
4
2
2
13
18
22
26
3
2
0
4
0
1
7
4
1
14
19
16
31
1
1
1
2
4
0
3
4
2
11
16
22
34
2
0
0
0
4
0
3
3
2
24
19
25
18
1
1
0
4
3
1
2
5
3
40
14
12
15
0
1
0
5
2
3
4
2
1
34
17
16
11
0
1
4
3
3
1
5
3
2
15
15
21
29
1
2
0
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
small share of litigation in 2000 and 2001 (12% and 6%) but a much larger share in 2008 and 2009 (40% and 34%) as a result of
the financial crisis. This suggests that industry-based proxies may not be consistently reliable measures of litigation risk.
In Section 2.1 we argue that the majority of securities class action filings do not result in SEC enforcement actions
(AAERs) because most filings relate to allegations (such as timeliness of disclosure) that are not sufficiently serious to
warrant SEC enforcement actions. To provide evidence on this, we examine how many of the 2,497 lawsuit filings (Table 1)
involve accounting allegations that could potentially results in enforcement actions. We find that 774 filings (31%) involve
accounting allegations. We then obtained data on AAERs, which are available through August of 2006, rather than through
the end of our sample period.17 Of the 631 filings that involve an accounting allegation for the period over which the two
samples overlap, 171 (27%) are associated with SEC enforcement actions. This implies that less than 10% of all securities
filings (0.27 0.31) are associated with SEC enforcement actions. Alternatively, if we take the 1,948 filings for the period
over which the samples overlap, 171 filings (9%) are associated with enforcement actions. Thus, it seems clear that the
large majority (over 90%) of class action lawsuit filings do not involve the types of allegations that typically result in SEC
enforcement actions, consistent with our argument that the two types of risk are related but distinct.
Table 2, Panel A reports the number of firms sued in each industry along with industry litigation rates, where industry
is measured using two-digit SIC codes.18 To derive this sample we merge the set of lawsuits from Table 1 with Compustat.
17
See Karpoff et al. (2008) for details of the AAER sample.
Francis et al. (1994a) sometimes define their high litigation set more finely than two-digit SIC codes. Specifically, they define this group as biotech
firms (SIC codes 2833–2836 and 8731–8734), computer firms (3570–3577 and 7370–7374), electronics firms (3600–3674), and retail firms (5200–5961);
subsequent research typically follows these definitions. Francis et al. limit their sample to four industries because ‘‘yinvestigations of the information
18
296
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
250
Lawsuit Filings Count
200
150
100
50
0
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Year
Rule 10b-5 Filings
Other Class Action Securities Filings
Fig. 1. Number of lawsuit filings by year. This figure shows the number of lawsuit filings by year, which is also shown in Table 1, Panel A. This figure
shows the fraction of lawsuits that allege a violation of SEC Rule 10b-5 (a misstatement or omission of material information). The unshaded portion of the
figure are alleged violations of other class action securities laws, for example violation of Securities Exchange Act of 1933 section 11, 12(a), 15, and
Securities Exchange Act section 20(a).
To keep the table manageable, we only tabulate FPS industries and industries for which litigation rates exceed 5% in at
least one year and eliminate industries with less than ten firms.19
The FPS industries tend to have litigation rates that exceed the overall rate of 1.6%. Chemicals (SIC 28) has a litigation
rate of 2.4%, machinery (SIC 35, which includes computers) has a rate of 2.4%, and business services (SIC 73) has a rate of
2.8%, although these numbers are understated by the inclusion of non-FPS industries. On the other hand, some non-FPS
industries experience relatively high litigation rates. Litigation rates in financial services (SIC 61–64) are at or above those
for the FPS industries, with rates of 2.6% to 5.0%, as are personal services (SIC 72, 4.4%) and health services (SIC 80, 3.2%).
Panels B and C of Table 2 report the number and percentage rate of litigation for firms in the FPS industries compared to
firms in all non-FPS industries. We report litigation rates for each of the four FPS industry groups, for the FPS industries as a
whole, and for the remaining industries as a group.
The data in these panels show that, overall, litigation rates in the FPS industries are higher than those of other
industries. The litigation rate for the four FPS industries considered together is 2.7% versus 1.2% for all non-FPS industries
considered together (w2 ¼13.7, significant at the 1% level). The litigation rate for the FPS industries is higher than for nonFPS industries in all years, with rates that are significantly higher (at the 5% level or better) in eight of 13 years. These
results reinforce the validity of the FPS measure.
There is evidence that litigation risk increases with firm size (Jones and Weingram, 1996a) and that the effect of industry
shocks varies across firm size within a given industry (Albuquerque, 2009). Given this, we examine whether the relation
between industry and litigation rates differs for large firms. In Table 3, we reperform the analysis reported in Table 2 for
firms in the largest 5% of the size distribution (by year) as measured by assets. Litigation rates are significantly higher for
larger firms. The overall litigation rate here is 5.1% (Table 3, Panel B), roughly three times the overall litigation rate of 1.6%
shown in Table 2. In addition, the litigation rate for large firms varies more from one year to the next, from lows of around
0.8% in 1996 and 1998 to highs of 12.2% in 2002 and 12.7% in 2008, than does the rate for the overall sample.20
Panel A of Table 3 shows that large firms in some non-FPS industries have relatively high litigation rates. Firms in
transportation (SIC 37, rate of 6.5%), instruments (SIC 38, rate of 9.1%), utilities (SIC 49, rate of 5.5%), and financial services
(SIC 60–63, rates of 3.0% to 23%) all have litigation rates roughly comparable to or higher than those for the large firms in
FPS industries.
Panel B of Table 3 shows that litigation rates for large firms in the FPS industries are again higher than for the non-FPS
industries but that differences are smaller and less consistent than for the overall sample. The overall litigation rate is 7.8% for the
FPS industries compared to 4.8% for the non-FPS industries, a difference that is not significant at the 5% level (w2 ¼ 2.6). Although
differences in particular years are sometimes economically significant, the difference is statistically significant in only one of the
(footnote continued)
mix defense require substantial familiarity with industry-specific information which may temper or offset alleged misleading statements.’’ (Francis et al.
1994a, p. 144). They choose these four industries because they are subject to a ‘‘high incidence’’ of litigation during 1988–1992.
19
These observations are retained in the overall litigation rates reported at the bottom of the table.
20
Size is likely to have counter-veiling effects on the likelihood of litigation. While larger firms have ‘‘deeper pockets’’ and size is likely to increase
expected stockholder damages, larger firms are also more diversified, and so less likely to suffer sharp declines in stock price than smaller firms.
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
297
Table 2
Number and percentage of unique companies subject to filings by industry.
In this table, Panel A shows the number and percentage of unique companies subject to filings by industry. The lawsuits in this table are lawsuits filed
against publicly-held firms (with shares listed on the NYSE, ASE, or NASDAQ) and exclude IPO allocation, mutual fund, and analyst lawsuits listed by
2-digit SIC code. Included in the table are all FPS industries; for non-FPS industries, to make the table wieldier, we only report those industries where
litigation rates exceed 5% in at least one year and eliminate industries with less than ten firms. The % sued column is calculated by dividing the
number sued column by the number of firms in Compustat for the relevant SIC code
Panel A: Number of unique companies subject to filings by industry
SIC
Industry name
Number sued
10
15
16
23
24
26
28a
31
32
35a
36a
37
39
40
42
44
49
51
52a
53a
54a
55a
56a
57a
58a
59a
61
62
63
64
72
73a
78
80
82
87a
99
Metal Mining
Building Construction
Heavy Construction
Apparel
Lumber and Wood Products
Paper and Allied Products
Chemicals and Allied Products
Leather
Stone, Glass, Clay, Concrete
Industrial and Commercial Machinery
Electronic and Other Electrical Equipment
Transportation Equipment
Misc. Manufacturing Inds.
Rail Road Transportation
Motor Freight Transportation
Water Transportation
Electric, Gas, Sanitary Services
Wholesale Trade, Nondurable Goods
Building Materials, Hardware, Garden Supply
General Merchandise Stores
Food Stores
Automotive Dealers
Apparel and Accessory Stores
Home Furniture, Furnishings, Equip.
Eating and Drinking Places
Miscellaneous Retail
Nondepository Credit Institutions
Security and Commodity Brokers
Insurance Carriers
Insurance Agents, Brokers
Personal Services
Business Services
Motion Pictures
Health Services
Educational Services
Engineering and Management Services
Nonclassifiable Establishments
Total Filings
8
5
3
11
3
3
123
4
2
76
111
19
8
1
4
3
41
18
1
9
6
1
10
4
9
30
30
36
52
9
8
195
6
29
7
15
8
1,177
a
Average industry count
% Sued
421
294
188
423
273
500
5,125
182
286
3,167
4,826
1,118
444
167
444
429
2,158
667
71
360
300
250
556
236
818
1,000
600
818
2,000
346
182
6,964
222
906
219
1,154
154
1.9
1.7
1.6
2.6
1.1
0.6
2.4
2.2
0.7
2.4
2.3
1.7
1.8
0.6
0.9
0.7
1.9
2.7
1.4
2.5
2.0
0.4
1.8
1.7
1.1
3.0
5.0
4.4
2.6
2.6
4.4
2.8
2.7
3.2
3.2
1.3
5.2
1.6
Contains FPS industry.
In this table, Panels B and C show the number and percentage of unique firms subject to securities lawsuits in the FPS industries. Biotech firms are classified as
firms in SIC Codes 2833–2838 and 8731–8734; computer firms are in SIC codes 3570–3577 and 7370–7374; electronics firms are in SIC codes 3600–3674,
and retail firms are in SIC Codes 5200–5961. The lawsuits in this table are lawsuits filed against publicly-held firms (with shares listed on the NYSE, ASE, or
NASDAQ) and exclude IPO allocation, mutual fund, and analyst lawsuits. The FPS firms and non-FPS firms percentage sued are calculated by dividing the
number sued by the number of Compustat firms for the relevant SIC code shown in Panel B. nn and n significant at the 1% and 5% level, respectively.
Panel B: Number of unique firms subject to lawsuit filings in FPS classification
Biotech
Computers
Electronics
Retail
# FPS Firms Sued
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
Total
2
286
7
500
2
286
1
333
0
0
18
545
3
300
4
286
7
292
23
547
7
304
3
300
7
292
23
622
3
300
4
286
6
316
26
650
4
333
5
278
9
310
23
511
12
308
5
263
10
313
16
552
11
314
5
263
15
306
12
522
12
308
5
263
18
333
28
509
8
320
9
265
12
333
16
485
11
324
7
259
8
348
9
450
8
320
3
273
15
357
7
438
13
317
6
250
8
333
8
444
7
304
7
241
117
4,189
216
6,750
101
4,040
64
3,556
12
1,333
25
1,471
40
1,429
37
1,480
41
1,577
49
1,400
42
1,448
44
1,419
63
1,432
46
1,394
28
1,400
41
1,367
30
1,304
498
18,444
298
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
Table 2 (continued )
In this table, Panels B and C show the number and percentage of unique firms subject to securities lawsuits in the FPS industries. Biotech firms are classified as
firms in SIC Codes 2833–2838 and 8731–8734; computer firms are in SIC codes 3570–3577 and 7370–7374; electronics firms are in SIC codes 3600–3674,
and retail firms are in SIC Codes 5200–5961. The lawsuits in this table are lawsuits filed against publicly-held firms (with shares listed on the NYSE, ASE, or
NASDAQ) and exclude IPO allocation, mutual fund, and analyst lawsuits. The FPS firms and non-FPS firms percentage sued are calculated by dividing the
number sued by the number of Compustat firms for the relevant SIC code shown in Panel B. nn and n significant at the 1% and 5% level, respectively.
Panel B: Number of unique firms subject to lawsuit filings in FPS classification
# non-FPS
Firms Sued
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
Total
8
4,000
27
4,500
29
4,143
57
4,071
48
4,000
31
3,875
73
3,842
59
3,933
70
3,889
66
3,883
37
4,111
53
4,077
83
3,952
641
53,417
Panel C: Percentage of unique firms subject to lawsuit filings in FPS classification
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
Biotech (%)
Computers (%)
Electronics (%)
Retail (%)
0.7
1.4
0.7
0.3
0.0
3.3
1.0
1.4
2.4
4.2
2.3
1.0
2.4
3.7
1.0
1.4
1.9
4.0
1.2
1.8
2.9
4.5
3.9
1.9
3.2
2.9
3.5
1.9
4.9
2.3
3.9
1.9
5.4
5.5
2.5
3.4
3.6
3.3
3.4
2.7
2.3
2.0
2.5
1.1
4.2
1.6
4.1
2.4
2.4
1.8
2.3
2.9
FPS Firms % Sued
Non-FPS Firms % Sued
Chi-Squared for diff.
0.9
0.2
3.83
1.7
0.6
3.90n
2.8
0.7
6.35n
2.5
1.4
2.86nn
2.6
1.2
3.77nn
3.5
0.8
7.10nn
2.9
1.9
2.28
3.1
1.5
3.67nn
4.4
1.8
5.41n
3.3
1.7
3.61
2.0
0.9
3.17
3.0
1.3
3.94n
2.3
2.1
0.25
All Yrs
2.8
3.2
2.5
1.8
2.7
1.2
13.68nn
13 years. In seven of the 13 years, the difference is 2% or less or goes in the ‘wrong’ direction. This means that the validity of the
FPS proxy for litigation risk, at least in terms of industry litigation rates, is more questionable for larger firms.
4. Predicting litigation risk
Our principal goal is to develop and evaluate models that predict litigation risk and to benchmark these models against
the FPS measure widely used in the literature.
Table 4 reports sample formation for these analyses. We begin with the 2,883 lawsuit filings available from the
Securities Class Action Clearinghouse from 1996 to 2008. After eliminating non-Rule 10b-5 cases (120 filings), filings against
firms not listed on a major exchange (584 filings), filings related to IPO allocation, mutual fund, and analyst cases (279
cases), and firms without the requisite Compustat and CRSP data, we are left with 720 lawsuit filings. This set of filings
translates into 1,562 firm-years which include a lawsuit class period, which when added to our sample of 31,344 nonlawsuit firm-years yields a final sample of 32,906 firm-years.21
As a baseline, we first report the results of an approach under which membership in one of the FPS industries is
interpreted as predicting a lawsuit, consistent with how many prior studies operationalize litigation risk. The results of
this analysis are shown in Table 5. This approach predicts lawsuits in 26.3% of firm/years, which reflects FPS industry
membership. The classification table also reports a number of statistics conventionally used to evaluate model success in
predicting lawsuits. In this case the Type I error rate is high, at 93.6%, while the Type II error rate is low, at 4.2%. These rates
are expected because predicting that all firms in particular industries will be subject to litigation naturally results in a
relatively large number of false positives and a relatively small number of false negatives. We also report model sensitivity,
which is 35.3%, and specificity, which is 74.2%.22 If researchers (and managers) are concerned about litigation risk, model
sensitivity (or ‘‘hit rate’’) is important because it tells us how often the model correctly forecasts that a firm will be sued.
Overall model accuracy (the overall fraction of correctly classified observations) is 72.3%.
The advantage of the FPS measure is that it is available in virtually all research settings—other than industry membership,
no data are required. Because we are interested in assessing the usefulness of a parsimonious litigation model for researchers
in a wide array of settings, we first limit our model to a small set of explanatory variables that we believe will deliver
increased predictive ability at relatively modest cost in terms of data requirements. To do this we augment the FPS measure
with explanatory variables that are readily available from Compustat and CRSP. We include firm size (log of total assets), sales
21
We require complete data on lawsuit filing (a dummy variable set to one for firm/years for which there is a lawsuit), lagged accounting variables (assets
and revenue growth rate), the return variables (prior and contemporaneous skewness of returns, volatility of returns, stock turnover, and year t, t 1, t 2, and
t 3 twelve-month cumulative abnormal return), and the FPS industry indicator (a dummy variable that turns on for firm/years in the FPS industries).
22
Sensitivity and specificity are common measures of the performance of prediction models (Hosmer and Lemeshow, 2000, Chapter 5). Sensitivity
reports the fraction of true positives correctly predicted. Specificity reports the fraction of true negatives correctly predicted. Similar to Type I and Type II
errors, there is usually a tradeoff between sensitivity and specificity; however, it is possible to attain 100% sensitivity and specificity. An analogy can be
drawn to metal detectors used in airport security screening. One can set the detector so that sensitivity is high and specificity is low to be sure all true
security risks are detected (with the cost that it will incorrectly identify lots of innocent people as security risks). Conversely, if it is relatively costly to
pull people aside when they are not true security risks, one can lower sensitivity to increase specificity.
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
299
Table 3
Number and percentage of unique companies subject to filings by year and industry for large firms.
In this table, Panel A shows the number and percentage of unique companies subject to securities lawsuits by industry for large firms only. The lawsuits in this
table are lawsuits filed against publicly-held firms (with shares listed on the NYSE, ASE, or NASDAQ) above the 95th percentile in size (total assets), and
exclude IPO allocation, mutual fund, and analyst lawsuits listed. Included in the table are all FPS industries; for non-FPS industries, to be included in this table,
the observation count must be at least 20. Some industries are included in this table but not in Table 2 due to the fact that the minimum % sued in Table 2
criteria is 5% in any year. The % Sued column is calculated by dividing the number sued column by the number of firms in Compustat for the relevant SIC code.
Panel A: Number and percentage of companies subject to filings by industry for large firms
SIC
Industry name
10
20
28a
29
33
35a
36a
37
38
40
48
49
52a
53a
54a
55a
56a
57a
58a
59a
60
61
62
63
73a
87a
99
Metal Mining
Food and Kindred Products
Chemicals and Allied Products
Petroleum Refining and Related Industries
Primary Metal Industries
Industrial and Commercial Machinery
Electronic and Other Electrical Equipment
Transportation Equipment
Measuring, Analyzing, Controlling Instruments
Rail Road Transportation
Communications
Electric, Gas, Sanitary Services
Building Materials, Hardware, Garden Supply
General Merchandise Stores
Food Stores
Automotive Dealers
Apparel and Accessory Stores
Home Furniture, Furnishings, Equip.
Eating and Drinking Places
Miscellaneous Retail
Depository Institutions
Nondepository Credit Institutions
Security and Commodity Brokers
Insurance Carriers
Business Services
Engineering and Management Services
Nonclassifiable Establishments
a
Number sued
% Sued
42
78
170
172
30
90
112
123
33
43
386
254
8
46
2
1
0
0
11
2
733
119
94
426
41
0
61
0.0
2.6
8.2
1.2
0.0
7.8
4.5
6.5
9.1
2.3
3.4
5.5
12.5
4.3
0.0
0.0
0.0
0.0
9.1
0.0
3.0
11.8
23.4
5.6
7.3
0.0
13.1
Contains FPS industry.
In this table, Panel B, shows the percentage of companies subject to securities lawsuits by year for large firms in the FPS industries. Biotech firms are
classified as firms in SIC Codes 2833–2838 and 8731–8734; computer firms are firms in SIC codes 3570–3577 and 7370–7374; electronics firms are
firms in SIC codes 3600–3674, and retail firms are firms in SIC Codes 5200–5961. The lawsuits in this table are lawsuits filed against publicly-held
firms (with shares listed on the NYSE, ASE, or NASDAQ) above the 95th percentile in size, and exclude IPO allocation, mutual fund, and analyst
lawsuits listed by year. The FPS firms % large firms sued and non-FPS % large firms sued rows are calculated by dividing the number sued by the
number of firms in Compustat for the relevant SIC codes. The bottom of the panel compares the percentage sued of large FPS firms and large non-FPS
firms, and a chi-squared for the difference between the two groups.
Panel B: Percentage of companies subject to filings by year for large firms in the FPS classification
1996
Biotech (%)
Computers (%)
Electronics (%)
Retail (%)
All Large Firms % Sued
FPS % Large Firms Sued
Non-FPS % Large Firms Sued
Chi-squared for difference
n
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
All Yrs
0
0
0
0
0
0
0
14
0
13
0
0
0
25
0
0
0
20
0
0
0
11
11
0
30
0
11
20
10
0
0
0
8
14
11
25
10
14
0
0
10
13
0
0
50
0
22
25
9
11
0
0
10
10
4
6
0.8
0.0
0.9
1.9
3.0
1.7
0.8
2.9
0.4
4.9
6.3
4.7
5.4
5.7
5.3
3.3
6.3
2.8
12.2
16.7
11.6
5.8
3.2
6.2
7.8
12.5
7.0
4.1
7.1
3.7
2.1
6.3
1.4
6.6
26.7
3.8
12.7
6.3
13.7
5.1
7.8
4.8
2.58
0.54
1.58
0.37
0.10
1.01
0.79
0.65
1.08
0.87
1.80
4.71n
1.18
2.58
Significant at the 5% level.
growth (the change in sales deflated by total assets), as well as a number of stock-return based measures—abnormal returns,
return volatility, return skewness, and stock turnover—designed to capture potential stockholder damages.
To predict litigation we deliberately avoid variables that directly reflect events that trigger the litigation—as discussed
above, we are interested in ex ante litigation risk. In some previous studies, the independent variables are measured in the
same period as the litigation. By including abnormal stock returns measured during the period during which the lawsuit is
filed, the researcher includes the downward stock price movement that triggers the litigation (as well as any valuation
300
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
Table 4
Sample selection.
This table reports sample formation: the progression from securities lawsuit filings from the Securities Class Action Clearinghouse (http://securities.
stanford.edu/) from 1996 to 2008 to the ultimate sample used for the regression analysis in Table 7. The sample consists of firm-years in which a lawsuit
filing occurred, and firm-years without a lawsuit filing from 1996 to 2008.
Lawsuit filings spanning 1996 2008 from Stanford Law School
Securities Class Action Clearinghouse
2,883
Less:
Less:
Less:
Less:
Less:
(120)
(584)
(279)
(818)
(362)
Non rule 10b-5 violation cases
Filings against private companies or filings against firms not listed on the NYSE, ASE, or NASDAQ
IPO allocation cases, hedge fund, mutual fund, and analyst cases
firm-years for which Compustat or Crsp identifiers are not available
no Compustat or Crsp coverage or certain missing data items in the relevant firm-year
Total useable lawsuit filings
Total useable lawsuit filing firm-years
(a class period occurred during the firm-year)
720
1,562
Firm-years for which no lawsuit filing/class period occurred
31,344
Total useable observations
32,906
Table 5
Prediction model under which membership in the FPS industries is interpreted as predicting a lawsuit.
This table shows a baseline model under which membership in one of the FPS industries is interpreted as predicting a lawsuit, consistent with how
much of the relevant prior literature has operationalized litigation risk. FPS is set to 1 for biotech firms (SIC codes 2833–2836 and 8731–8734), computer
firms (3570–3577 and 7370–7374), electronics firms (3600–3674), and retail firms (5200–5961), and 0 otherwise. The table calculates cases where FPS
predicts a lawsuit (26%), correctly classified cases (72%), sensitivity (35%), specificity (74%), type I error (94%), and type II error (4%).
Lawsuit filing
No lawsuit filing
FPS¼ 1
FPS¼ 0
551
1,011
1,562
8,098
23,246
31,344
FPS Predicts Lawsuit
Correct Classification
Sensitivity
Specificity
Type I Error
Type II Error
Goodman and Kruskal Gamma Statistic
8,649/32,906
(551þ 23,246)/32,906
551/1,562
23,246/31,344
8,098/8,649
1,011/24,257
8,649
24,257
32,906
26.28%
72.32%
35.28%
74.16%
93.62%
4.17%
0.0021
effects of the litigation). To avoid this, we measure the stock return variables in the fiscal year before the lawsuit filing. To
examine the effect of this choice, we present the analyses with the stock return-based variables measured both ways: over
the class period (misrepresentation period) and over the prior fiscal year.
We report descriptive statistics in Table 6. Panel A reports means and medians for the variables used in the regressions
in Table 7 (basic set of variables) and Table 8 (full set of variables). Mean (median) total assets for these firms is $582m
($566m) with annual sales growth of 11.3% (4.6%). As shown in Table 5, 26.3% of these firms are in FPS industries. Mean
(median) monthly stock volatility is 3.0% (2.6%). Mean stock returns are lower in year t (which includes the class period)
than year t 1, consistent with negative returns in the event period triggering litigation. Similarly, turnover is higher in
year t than in year t 1, consistent with high turnover being associated with litigation.
Panel B of Table 6 reports Pearson correlations among the variables. The indicator for litigation is positively correlated
with the FPS dummy (0.046), size (0.168), sales growth (0.064), and turnover (0.246), and negatively correlated with event
year returns ( 0.104). These correlations are all statistically significant at better than 1%. These correlations again suggest
that event year returns and turnover drive litigation (the corresponding correlations with lagged returns and turnover are
0.032 and 0.069, respectively). The FPS dummy is negatively associated with size ( 0.226), positively associated with sales
growth (0.066), positively associated with return volatility (0.327), and positively associated with lagged turnover (0.322),
consistent with the FPS dummy being associated with variables that drive litigation. Size is negatively correlated with
return volatility ( 0.548), as expected if larger firms are more diversified.
Table 7 reports the results of binomial logistic regressions of the lawsuit dummy variable on the FPS dummy and the
other covariates.23 The first model uses only the FPS industry dummy (model 1). The coefficient on this variable is positive
23
We use a binomial logistic regression rather than a hazard model because the traditional Cox proportional hazard model, used for events such as
bankruptcy or death, is designed for situations in which (1) there is a single event for each firm (or other observational unit) after which that firm
disappears, and (2) the likelihood of that event is likely to systematically change as time elapses for a given firm (or unit). Neither of these conditions is
true of class action lawsuits. First, firms are often sued in multiple years. Second, unlike events such as bankruptcy, it is not clear that the likelihood of
lawsuits changes as a function of elapsed time.
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
301
Table 6
Descriptive statistics.
In this table, Panels A and B present descriptive statistics and correlation coefficients of the variables used in the regressions in Tables 7 and 8,
respectively. All variables are winsorized at the 1% and 99% level. Correlation coefficients in Panel B are based on 32,906 observations (from Table 7).
Pearson correlation coefficients are shown above the diagonal, and Spearman below the diagonal. Variable definitions are provided in the appendix.
Panel A: Descriptive statistics
Variable
All firm-years
(Number of observations: 32,906 (Table 7); 11,597 (Table 8)
Mean
Table 7 Variables
FPSt
LNASSETSt 1
SALES GROWTHt 1
RETURNt
RETURNt 1
RETURN SKEWNESSt
RETURN SKEWNESSt 1
RETURN STD DEVt
RETURN STD DEVt 1
TURNOVERt ($million)
TURNOVERt 1 ($million)
Additional Table 8 variables
NYSE t 1
USINCORPt 1
WCt 1
ROAt 1
R&Dt 1
GOODWILLt 1
PP&Et 1
ALTMAN Zt 1
MBt 1
INSTt 1
EQUITY PROCEEDSt 1
DEBT PROCEEDSt 1
INSIDER TRADINGt 1
INSIDER HOLDINGt 1
Std dev
Minimum
Median
Maximum
0.263
6.368
0.113
0.088
0.123
0.310
0.334
0.030
0.031
1.45
1.21
0.440
2.010
0.268
0.441
0.446
1.095
0.982
0.018
0.017
1.52
1.29
0.000
2.170
0.522
1.035
0.951
3.492
3.026
0.009
0.009
0.06
0.05
0.000
6.339
0.046
0.057
0.080
0.274
0.294
0.026
0.026
0.95
0.79
1.000
11.546
1.331
1.570
1.638
4.564
4.049
0.091
0.093
8.44
7.17
0.348
0.991
0.406
0.000
0.095
0.120
0.486
11.255
3.629
0.553
0.078
0.005
0.371
0.020
0.476
0.095
0.350
0.208
0.134
0.165
0.342
16.140
3.934
0.272
0.235
0.026
3.789
0.049
0.000
0.000
0.131
0.923
0.000
0.000
0.035
19.260
4.935
0.010
0.000
0.000
20.065
0.000
0.000
1.000
0.346
0.050
0.048
0.051
0.405
6.748
2.568
0.592
0.000
0.000
0.028
0.003
1.000
1.000
1.924
0.383
0.719
0.817
1.502
92.374
24.416
0.977
1.210
0.170
20.813
0.322
Panel B: Correlations
SUEDt
SUEDt
FPSt
LNASSETSt 1
SALES
GROWTHt 1
RETURNt
RETURNt 1
RETURN
SKEWNESSt 1
RETURN
STD DEVt 1
TURNOVERt
TURNOVERt 1
0.046
o 0.0001
0.123
o 0.0001
0.056
o 0.0001
0.102
o 0.0001
0.033
o 0.0001
0.031
o 0.0001
0.019
0.0005
0.210
o 0.0001
0.092
o 0.0001
FPSt
LNASSETSt 1
0.046
0.168
o 0.0001 o0.0001
0.226
o0.0001
0.237
o 0.0001
0.096
0.069
o 0.0001 o0.0001
0.003
0.097
0.5405 o0.0001
0.017
0.070
0.0018 o0.0001
0.078
0.251
o 0.0001 o0.0001
0.351
0.583
o 0.0001 o0.0001
0.320
0.104
o 0.0001 o0.0001
0.337
0.059
o 0.0001 o0.0001
SALES
GROWTHt 1
RETURNt RETURNt 1
0.064
o 0.0001
0.066
o 0.0001
0.068
o 0.0001
0.104
o0.0001
0.021
0.0002
0.146
o0.0001
0.056
o0.0001
0.048
o 0.0001
0.144
o 0.0001
0.016
0.0049
0.074
o 0.0001
0.252
o 0.0001
0.201
o 0.0001
0.032
o0.0001
0.037
o0.0001
0.113
o0.0001
0.163
o0.0001
0.015
0.0054
0.047
o0.0001
0.051
0.275
o0.0001 o0.0001
0.169
0.134
o0.0001 o0.0001
0.020
0.020
0.0003
0.0003
0.094
o0.0001
0.5443
RETURN
SKEWNESSt 1
0.031
o 0.0001
0.052
o 0.0001
0.217
o 0.0001
0.008
0.1425
0.058
o 0.0001
0.315
o 0.0001
0.241
o 0.0001
0.107
o 0.0001
0.050
o 0.0001
RETURN STD
DEVt 1
0.009
0.1102
0.327
o 0.0001
0.548
o 0.0001
0.053
o 0.0001
0.237
o 0.0001
0.220
o 0.0001
0.259
o 0.0001
0.014
0.0110
0.289
o 0.0001
TURNOVERt TURNOVERt 1
0.246
o 0.0001
0.299
o 0.0001
0.022
o 0.0001
0.200
o 0.0001
0.005
0.3347
0.159
o 0.0001
0.023
o 0.0001
0.230
o 0.0001
0.069
o 0.0001
0.322
o 0.0001
0.015
0.005
0.142
o 0.0001
0.096
o 0.0001
0.014
0.0104
0.034
o 0.0001
0.312
o 0.0001
0.731
o 0.0001
0.836
o 0.0001
and highly significant, with a marginal effect of 0.020, indicating that FPS industry membership increases the probability of
litigation by 2.0%, roughly consistent with the litigation rates reported in Table 2. Although the coefficient on the FPS
variable is significantly associated with litigation in this model, overall goodness of fit and predictive ability are low. The
conventional measure of goodness of fit is pseudo-R-squared. We report the McFadden (1973) pseudo-R-squared, perhaps
302
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
Table 7
Models of litigation risk.
This table presents the FPS only model in comparison to our multivariate models with contemporaneous and lagged variables. Model (1a) is the logit
model with the FPS variable. Model (2) adds lagged assets, lagged sales growth, and contemporaneous stock return variables (market-adjusted return,
return skewness, return standard deviation, and turnover) to the FPS variable. Model (3a) is the same specification as model (2), except it uses lagged
return variables. Model (4) is the same specification as model (3), except adds two-year and three-year lagged returns. Variable definitions are provided
in the appendix. nnn, nn, and n indicate p-values of 1%, 5%, and 10%, respectively. The p-values are based on robust standard errors that control for
heteroskedasticity and serial correlation. Marginal effects of the coefficients are reported below the coefficients.
SUED ¼ b0 þ b1 ðFPSt Þ þ e
ð1aÞ
SUED ¼ b0 þ b1 ðFPSt Þ þ b2 ðLNASSESTSt1 Þ þ b3 ðSALES GROWTHt1 Þ þ b4 ðRETURN t Þ
þ b5 ðRETURN SKEWNESSt Þ þ b6 ðRETURN STD DEV t Þþ b7 ðTURNOVERt Þþ e
ð2Þ
SUED ¼ b0 þ b1 ðFPSt Þ þ b2 ðLNASSESTSt1 Þ þ b3 ðSALES GROWTHt1 Þ þ b4 ðRETURN t1 Þ
þ b5 ðRETURN SKEWNESSt1 Þ þ b6 ðRETURN STD DEV t1 Þþ b7 ðTURNOVERt1 Þ þ e
ð3Þ
SUED ¼ b0 þ b1 ðFPSt Þ þ b2 ðLNASSESTSt1 Þ þ b3 ðSALES GROWTHt1 Þ þ b4 ðRETURN t1 Þ þ b5 ðRETURN t2 Þþ b6 ðRETURNt3 Þ
þ b7 ðRETURN SKEWNESSt1 Þ þ b8 ðRETURN STD DEV t1 Þþ b9 ðTURNOVERt1 Þ þ e
INTERCEPT
FPSt
ð4Þ
Model (1a)
Model (2)
Model (3)
Model (4)
3.135nnn
0.448nnn
0.020
7.718nnn
0.180
0.007
0.463nnn
0.018
0.553nnn
0.021
0.498nnn
0.019
7.883nnn
0.566nnn
0.024
0.518nnn
0.022
0.982nnn
0.044
7.938nnn
0.567nnn
0.024
0.523nnn
0.022
0.896nnn
0.037
0.379nnn
0.016
0.419nnn
0.018
0.201nnn
0.008
0.273nnn
0.011
0.108nnn
0.005
0.101nnn
0.004
25.635nnn
1.076
25.254nnn
1.057
LNASSETSt 1
SALES GROWTHt 1
RETURNt
RETURNt 1
RETURNt 2
RETURNt 3
0.359nnn
0.014
RETURN SKEWNESSt
RETURN SKEWNESSt 1
14.437nnn
0.550
RETURN STD DEVt
RETURN STD DEVt 1
0.0004nnn
0.00002
TURNOVERt
0.00007nn
0.000003
TURNOVERt 1
Pseudo-R2 (McFadden) (%)
Pseudo-R2 (Cox-Snell) (%)
Area under ROC Curve (AUC)
Mean out of sample AUC
Hosmer-Lemeshow p-value
Observation count
0.01
0.20
0.547
0.558
(0.000)
32,906
25.65
8.14
0.842
0.838
(0.746)
32,906
12.47
4.65
0.756
0.750
(0.747)
32,906
0.00004
0.000002
12.72
4.74
0.759
0.754
(0.165)
32,906
the most common measure, and the Cox and Snell pseudo-R-squared.24 This model has a McFadden pseudo-R-squared of
0.01% and a Cox and Snell pseudo-R-squared of 0.20%, indicative of poor fit.
Another way of assessing predictive ability is to use a classification table (as in Table 5). To generate a classification
table, the researcher first specifies a cutoff probability (estimated probability above which we predict that an observation
will experience a lawsuit). It is common in predicting financial distress to use cutoffs such as 0.5 (Ohlson, 1980) or 0.3
(Altman and Sabato, 2007). However, the predicted probabilities from model 1 are uniformly low, at around 5%, reflecting
24
The McFadden pseudo R2 is based on the ratio of log likelihoods and has the desirable feature of varying between 0 and 1. However, unlike a
conventional R square, it cannot be interpreted as the proportion of variation in the dependent variable explained by the regression covariates. See Long
(1997, pp. 104–108) for more discussion of different pseudo R-squareds. Because of the non-linear nature of logit models, there is no universally agreed
upon pseudo R-squared or, more generally, measure of goodness of fit—there are variety of measures, but each has advantages and disadvantages (e.g.,
see Hosmer and Lemeshow, 2000, Chapter 5; Long and Freese, 2006).
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
303
both the low unconditional incidence of litigation and the modest effect of the FPS industry variable. Moreover,
classification tables cannot be compared for different samples.25
A better way of comparing the predictive ability of different models is to use the Receiver Operating Characteristic, or
ROC curve (e.g., Hosmer and Lemeshow, 2000, Chapter 5). This curve ‘‘plots the probability of detecting a true signal
(sensitivity) and false signal (1—specificity) for the entire range of possible cutpoints’’ (p. 160, our emphasis). The area under
the ROC curve (denoted AUC) provides a measure of the model’s ability to discriminate. A value of 0.5 indicates no ability
to discriminate (might as well toss a coin) while a value of 1 indicates perfect ability to discriminate, so the effective range
of AUC is from 0.5 to 1.0.26
AUC for model 1, at 0.547, is larger than 0.5 but well below the level normally seen as indicating acceptable
discriminatory ability.27 (We also report a corresponding ‘‘out of sample’’ AUC, which is similar, at 0.558.28) This is more
clearly seen in Fig. 2, which we use to plot ROC curves for different models. Here we can see that the ROC curve for model 1
is only modestly above (to the north-west) of the 45-degree line that represents no discriminatory ability. For reference,
we also plot a single point that represents the discriminatory ability of the approach discussed above (Table 5) of simply
using FPS membership to predict litigation risk.29 This point is just above the ROC curve for model 1, also showing
relatively low discriminatory ability. Overall, these results suggest that the FPS variable, considered alone, is not a good
predictor of litigation risk.
We also report the Hosmer-Lemeshow chi-square (Hosmer-Lemeshow, 2000, Chapter 5; Long and Freese, 2006) to
measure discriminatory ability. This measure sorts the sample into (usually) ten groups based on predicted probabilities.
Within these groups, it then compares the observed frequency of the outcome to the expected frequency of that outcome
where the latter is based on the predicted probabilities for the observations within the group. Under the null hypothesis
that the model fits well, the observed and expected frequencies are similar within groups. Conversely, rejection of this null
indicates that the model fits poorly. The Hosmer-Lemeshow chi-squared statistic for this model is highly significant
(p-value of 0.000), which strongly rejects the null that the model fits well, and is again indicative of poor fit.
We next report models (denoted 2–4) that include the size, sales growth, and stock market variables along with the FPS
industry dummy. The first model (model 2) includes stock market variables measured during the event year. As discussed
above, this likely increases predictive ability but is not a realistic approach to measuring litigation risk ex ante. To avoid
this problem (and reduce endogeneity concerns), the second model (model 3) measures stock market variables in the year
before lawsuit filing. In the third specification (model 4) we add two additional lags of annual stock returns to see whether
longer run stock performance helps predict litigation. Size and sales growth are measured in the fiscal year before the
lawsuit filing in all models. The general idea behind these variables is that firms tend to get sued after a period of unusually
strong growth and/or stock price run up that subsequently reverses (a sharp ‘‘reversal of fortune’’), and that the likelihood
of litigation is higher for larger firms and firms with more volatile stock returns.
These specifications have significantly higher predictive ability than model 1. For model 2, the McFadden pseudo-Rsquare is 25.7% while the Cox-Snell measure is 8.1%, both substantially larger than for model 1. AUC increases to 0.842,
substantially higher than model 1, and indicative of ‘‘excellent’’ discriminatory ability (Hosmer-Lemeshow, 2000,
Chapter 5). (The corresponding cross-validation AUC is 0.838.)30 Finally, the Hosmer-Lemeshow chi-squared statistic for
model 2 is insignificant (p-value of 0.746), also indicative of good fit.
The model 2 coefficients have signs that are largely consistent with expectations. The FPS variable is no longer
statistically significant (p-value of 0.12), with a marginal effect of only 0.007, indicating that the inclusion of the other
25
According to Hosmer and Lemeshow (2000, p. 160), ‘‘one cannot compare models on the basis of measures derived from 2 2 classification tables
since these measures are completely confounded by the distribution of probabilities in the samples upon which they are based. The same model,
evaluated in two populations, could give very different impressions of performancey’’.
26
Intuitively, the area can be thought of as follows. Assume we have n1 firm/years subject to litigation and n2 firm/years that are not. We can thus
create n1 n2 pairs. Of the set of all possible pairs, AUC tells us the fraction for which the observation subject to litigation had a higher predicted
probability than its pair. Under the null that the model has no discriminatory ability, this fraction is 0.5.
27
Hosmer-Lemeshow (2000, p. 162) indicate that AUC of 0.5 indicates no discrimination, AUC of between 0.7 and 0.8 indicates acceptable
discrimination, AUC of between 0.8 and 0.9 indicates excellent discrimination, and AUC greater than 0.9 is considered outstanding discrimination.
28
The first AUC number is based on estimating the model in-sample, using all observations. However, in assessing predictive ability, it is useful to be
able to assess how well the model will perform out of sample to avoid an over-fitting problem. One way to do this is to use cross-validation, and in
particular we use the ‘‘K-fold’’ cross validation procedure described by Efron and Tibshirani (1993, Chapter 17). The procedure proceeds as follows (using
ten folds). First, randomly choose 10% of the full set of observations. This is one fold. Second, take the remaining 90% of observations and randomly
choose a second group of equal size (again equal to 10% of the total). This is the second fold. Continue to do this until all observations are classified into
ten folds. Third, estimate the model using nine of the ten folds, and apply the model to the fold that was held out of the estimation. This yields an ‘‘out of
sample’’ estimate of AUC. Repeat this step using nine different sets of nine folds. This yields ten ‘‘out of sample’’ AUC estimates. We report the average of
these ten estimates.
29
This approach plots as a single point because it is not a discriminatory model that generates estimated probabilities. Rather, it is a deterministic
rule that simply says that litigation will occur if firms fall in one of the FPS industries and no litigation will occur otherwise.
30
Although AUC has advantages, a disadvantage is that it naturally increases as covariates are added to the model, in a manner somewhat analogous
to (unadjusted) R-squares. There is no well-accepted way of adjusting for this problem. However, we note that models (2), (3), and (4) all have the same
number of covariates but that model (2) has notably larger AUC which we expect based on our predictions. Conversely, when we estimate a model that
includes only a single variable (for comparison to model (1)), either lagged assets or the contemporaneous abnormal return, we obtain AUCs of 0.667 and
0.639, respectively. Thus, it does not seem to be the case that variation in the AUC across the models in Table 7 is simply due to variation in the number of
covariates.
304
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
Table 8
Expanded model of litigation risk.
This table augments the models in Table 7 with the following additional potential drivers of litigation risk (all lagged): dummy for U.S. incorporation,
dummy for NYSE listing, working capital, ROA, R&D intensity, goodwill intensity, PP&E intensity, Altman’s Z, market-to-book, institutional holdings, debt
and equity issuances in recent past, and insider trading and holdings. Variable definitions are provided in the appendix. Models 1b and 3b include the
same variables as their companion in Table 7, except with the reduced observation count of 11,597. nnn, nn, and n indicate p-values of 1%, 5%, and 10%,
respectively. The p-values are based on robust standard errors that control for heteroskedasticity and serial correlation. Marginal effects of the
coefficients are reported below the coefficients
SUED ¼ b0 þ b1 ðFPSt Þ þ e
ð1bÞ
SUED ¼ b0 þ b1 ðFPSt Þ þ b2 ðLNASSESTSt1 Þ þ b3 ðSALES GROWTHt1 Þ þ b4 ðRETURN t1 Þ þ b5 ðRETURN SKEWNESSt1 Þ
þ b6 ðRETURN STD DEV t1 Þ þ b7 ðTURNOVERt1 Þ þ e
ð3bÞ
SUED ¼ b0 þ b1 ðFPSt Þ þ b2 ðNYSEt1 Þ þ b3 ðUSINCORP t1 Þ þ b4 ðLNASSESTSt1 Þ þ b5 ðWC t1 Þ þ b6 ðROAt1 Þ þ b7 ðSALES GROWTHt1 Þþ b8 ðR&Dt1 Þ
þ b9 ðGOODWILLt1 Þþ b10 ðPP&Et1 Þ þ b11 ðALTMAN Z t1 Þþ b12 ðMBt1 Þ þ b13 ðRETURNt1 Þ þ b14 ðRETURN SKEWNESSt1 Þ
þ b15 ðRETURN STD DEV t1 Þ þ b16 ðTURNOVERt1 Þþ b17 ðINST t1 Þþ b18 ðEQ UITY PROCEEDSt1 Þ
þ b19 ðDEBT PROCEEDSt1 Þ þ b20 ðINSIDER TRADINGt1 Þ þ b21 ðINSIDER HOLDINGt1 Þ þ e
INTERCEPT
FPSt
Model (1b)
Model (3b)
Model (5)
3.451nnn
0.692nnn
0.030
6.184nnn
0.599nnn
0.025
5.555nnn
0.473nnn
0.019
0.146
0.006
0.672
0.027
0.257nnn
0.010
0.124
0.005
0.158
0.006
0.835nnn
0.034
0.584
0.024
0.450
0.018
1.017nnn
0.041
0.002
0.000
0.052nnn
0.002
0.264nn
0.011
0.142nnn
0.006
23.730nnn
0.969
0.00000004
0.000
0.742nn
0.030
0.521nn
0.021
2.974n
0.121
0.050nnn
0.002
0.340
0.014
NYSEt 1
USINCORPt 1
0.278nnn
0.012
LNASSETSt 1
WCt 1
ROAt 1
0.850nnn
0.035
SALES GROWTHt 1
R&Dt 1
GOODWILLt 1
PP&Et 1
ALTMAN Zt 1
MBt 1
0.449nnn
0.019
0.177nnn
0.007
22.597nnn
0.942
0.00005
0.000
RETURNt 1
RETURN SKEWNESSt 1
RETURN STD DEV
t1
TURNOVERt 1
INSTt 1
EQUITY PROCEEDSt 1
DEBT PROCEEDSt 1
INSIDER TRADINGt 1
INSIDER HOLDINGt 1
Pseudo-R2 (McFadden) (%)
Pseudo-R2 (Cox-Snell) (%)
Area under ROC Curve
Mean out of sample AUC
Hosmer-Lemeshow p-value
Observation count
ð5Þ
1.34
0.49
0.5839
0.5825
(0.000)
11,597
6.35
2.31
0.7053
0.7290
(0.5107)
11,597
9.46
3.41
0.7439
0.7831
(0.4048)
11,597
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
305
100%
90%
45°
80%
Sensitivity
70%
60%
(0.37, 0.53)
50%
40%
30%
20%
10%
0%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
1 - Specificity
Model 1, with FPS only
Model 2, (AUC=0.842)
Model 3 (AUC=0.756) and 4 (0.759)
Fig. 2. Receiver operating characteristic (ROC) curve for models in Table 7. This figure shows the ROC curves for regression model 1a, 2, and 3a of Table 7.
The area under the ROC curve (denoted AUC) provides a measure of the model’s ability to discriminate. A value of 0.5 (451 line) indicates no ability to
discriminate (might as well toss a coin) while a value of 1 indicates perfect ability to discriminate, so the effective range of AUC is from 0.5 to 1.0. For
reference, we also plot a single point that represents the discriminatory ability of simply using FPS membership to predict litigation risk (Table 5). This
point is just above the ROC curve for model 1, also showing relatively low discriminatory ability. Sensitivity and (1—) specificity are on the y- and x-axis,
respectively. Sensitivity and specificity are common measures of the performance of prediction models (Hosmer and Lemeshow, 2000, Chapter 5).
Sensitivity reports the fraction of true positives correctly predicted. Specificity reports the fraction of true negatives correctly predicted.
variables reduces its role. Coefficients on size and sales growth are positive and strongly significant, with marginal effects
of 0.018 and 0.021 respectively. The coefficient on event year abnormal stock returns is negative and highly significant,
with a marginal effect of 0.019.31 Because these returns are measured over the class period, this is consistent with the
idea that stock price declines cause litigation (and so are associated with litigation ex post). The coefficient on return
skewness is negative and highly significant while those on return volatility and turnover are both positive and highly
significant, with material marginal effects. Similar to the stock return variable, these variables reflect returns and trading
that directly lead to litigation, as opposed to being predictors of litigation. For these reasons, model 2 likely overstates the
predictive ability of these variables and for the same reason likely understates the predictive ability of the FPS variable.
In model 3 the stock return variables are measured in the period before the filing year. This changes the results in
several respects. First, goodness of fit declines, with the McFadden pseudo-R-square falling to 12.5% and the Cox-Snell
pseudo-R-square to 4.7%, although these are still substantially higher than for model 1. AUC falls to 0.756, which is
indicative of ‘‘acceptable’’ discrimination and still well above that for model 1. (The corresponding cross-validation AUC is
0.750.) The p-value on the Hosmer-Lemeshow chi-squared statistic is 0.747, also consistent with good fit.
Second, lagging the stock return variables changes the coefficients on these variables and causes the FPS variable to
return to significance. The coefficient on the FPS variable in this specification is 0.566, which is highly significant and larger
than for model 1. Its marginal effect is 0.024, also larger than for model 1. The coefficients on size and sales growth remain
positive and highly significant, and are larger here than in model 2. To be clearer about economic magnitude, we again
report the effect of discrete changes in these variables. Holding other variables at their means, increasing size from $583m
(its mean) to $1,000m leads to an increase in the estimated probability from 0.027 to 0.035 when FPS ¼0, and from 0.046
to 0.061 when FPS ¼1. If sales growth increases from its mean of 11.3% to 21.3%, the estimated probability increases from
0.027 to 0.029 when FPS¼0 and from 0.046 to 0.051 when FPS ¼1, a more modest effect. Thus, the effects of size and
growth are more pronounced for firms in the FPS industries.
The abnormal return variable remains highly significant when lagged but reverses sign, from significantly negative in
model 2 (coefficient of 0.498) to significantly positive when lagged in this model (coefficient of 0.379). We interpret this
reversal as saying that strong prior period stock performance increases the likelihood of a reversal of fortune, which then
triggers litigation. Comparing results for these two specifications implies that including contemporaneous stock returns is
what led to the insignificance of the FPS variable. The coefficient on return skewness remains significantly negative but is
smaller in magnitude here, consistent with its significance in model 2 being attributable to contemporaneous
31
To give a better sense for economic magnitude, we looked at the change in probability for a discrete change in these covariates with other
variables held at their means. If we increase firm size from its mean ($583m) to $1,000m, the estimated probability increases from 2.08% to 2.66% when
FPS ¼0 and from 2.49% to 3.17% when FPS ¼1. When sales growth increases from the mean of 0.113 to 0.213 (annual growth rate increases from 11% to
21%), the estimated probability increases from 2.08% to 2.20% when FPS¼ 0 and from 2.49% to 2.62% when FPS ¼ 1. Finally, when the annual stock return
decreases from its mean of 8.8% to 20.0% (which might be typical for a firm that is sued), the estimated probability increases from 2.08% to 2.40% for
FPS ¼0 and from 2.49% to 2.86% for FPS ¼1.
306
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
Table 7 Model 3
80
70
70
Predicted Litigation Risk (%)
Predicted Litigation Risk (%)
Table 7 Model 1, with FPS Only
80
60
50
40
30
20
60
50
40
30
20
10
10
0
0
0
10
20
30
40 50 60
Percentile (%)
70
80
90
100
0
10
20
30
40 50 60
Percentile (%)
70
80
90
100
Fig. 3. Predicted litigation risk. This figure compares the distributions of predicted probabilities for Table 7, Models 1a and 3a. For model 1, the
distribution is simple, reflecting the fact that the model assigns probabilities according to whether the firm is in an FPS industry. For this model, the
predicted probability is 0.064 for firms in an FPS industry and 0.042 for firms that are not. In contrast, model 3 produces a continuous distribution of
predicted probabilities, ranging from just over 0 to over 70%.
measurement. Stock volatility becomes more significant in this specification, with the marginal effect almost doubling,
which is expected because a more volatile stock price makes large stock price declines more likely ex ante.32 The effect of
trading volume (turnover) declines in this specification, perhaps because turnover is no longer measured in the event year.
Overall, model 3 yields a relatively large improvement in predictive ability over the basic FPS approach (model 1) at
relatively low cost. Explanatory power is substantially higher under all measures, with AUC increasing from 0.547 to 0.756.
Another way of comparing the usefulness of models 1 and 3 in predicting litigation risk is to compare the distributions of
predicted probabilities for these models, as shown in Fig. 3. For model 1, the distribution is simple, reflecting the fact that the
model assigns probabilities according to whether the firm is in an FPS industry. For this model, the predicted probability is
0.064 for firms in an FPS industry and 0.042 for firms that are not. In contrast, model 3 produces a continuous distribution of
predicted probabilities, ranging from just over 0 to over 70%. Around 10% of observations have predicted probabilities that
exceed 0.10, and a small number have substantially higher predicted probabilities. To give a sense for what types of firms have
these relatively high predicted probabilities, if we move all variables to one standard deviation above their means, the predicted
probability of litigation is 0.151 when FPS¼0 and 0.238 when FPS¼1.33 If we move firm size, sales growth, and return volatility
to one standard deviation above their means (holding the other variables at their means), the respective probabilities become
0.136 and 0.218, which indicates that these three variables are relatively more important than the other variables. These results
show that the probabilities increase more than proportionately when several variables increase together. In particular, the FPS
variable has a larger effect (increasing probability by about 8%) when size, growth, and volatility are all above their means, and
vice versa, so the level of all of these variables is important in determining litigation risk.
We report a final specification (model 4) that augments model 3 with two additional lags of annual abnormal stock
returns. Similar to the single lag of abnormal returns, the coefficients on both variables are positive and significant, with
marginal effects of 0.008 and 0.011, respectively. The addition of these variables adds relatively little explanatory power/
predictive ability, with little or no improvement in pseudo-R squared or AUC. The McFadden pseudo-R square is 12.7%
while the Cox-Snell pseudo-R square is 4.7%, both very similar to those for model 3. AUC is 0.759 (cross validation AUC of
0.754), also similar to that for model 3.
The regressions in Table 7 use a panel of around 33,000 observations. In Table 8, we augment these models with the
following additional potential drivers of litigation risk discussed in Section 2 (all lagged): dummy for U.S. incorporation,
dummy for NYSE listing, working capital, ROA, R&D intensity, goodwill intensity, asset tangibility (PP&E intensity),
Altman’s Z, market-to-book, institutional holdings, recent debt and equity issuances, and insider trading and holdings.
Because of the additional data required to compute these variables, sample size is reduced substantially, to around 11,600
observations. We view this as a serious cost for researchers, especially if these data requirements induce survival or other
biases in sample composition, so the benefits in terms of improved model predictive ability would have to be substantial to
justify the inclusion of these additional variables.
32
If we move stock volatility from its mean of 0.031 to 0.046, the estimated probability of litigation increases from 0.027 to 0.039 when FPS ¼ 0 and
from 0.047 to 0.067 when FPS ¼ 1. These are large changes compared to those for the other covariates included in this model. And here again, the effect is
larger for firms in the FPS industries.
33
For comparison, if all variables are at their means, the predicted probabilities are 0.027 for FPS¼ 0 and 0.047 for FPS¼ 1.
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
307
The results in Table 8 show this not to be the case. We again compare a simple model containing only the FPS dummy
to a model that includes the full set of covariates; we refer to these as models 1 and 5, respectively (model 3 from Table 7 is
also reported for comparison). Similar to the result in Table 7, in model 1 the FPS variable is highly significant, with a
material marginal effect (0.030) but relatively low predictive ability. The McFadden and Cox-Snell pseudo-R squareds are
1.34% and 0.49%, respectively, and AUC is 0.584.
The inclusion of the full set of covariates increases predictive ability, as expected, over model 1, but this model does not
dominate models 3 and 4 in Table 7. The McFadden and Cox-Snell pseudo-R squareds are 9.5% and 3.4%, respectively, for
this model, compared to 12.5% and 4.7%, respectively, for model 3 in Table 7. AUC is 0.744, slightly below the
corresponding numbers for models 3 and 4 in Table 7.34 Overall, given the large cost in terms of data requirements, the
payoff to including these additional explanatory variables is relatively small, which reinforces our conclusion that models
along the lines of model 3 in Table 7 are probably the most cost effective solution for researchers interested in measuring
litigation risk.
It is clear from the results in Tables 7 and 8 that the FPS variable generally remains significant (economically and
statistically) even when a large set of firm characteristics are included in the regressions.35 Further, the effects of size,
growth, and volatility are all larger for firms in the FPS industries. This implies that there is something about industry
membership that is important in explaining litigation that is not captured by the set of other variables (i.e., that is not
explained by size, volatility, growth, asset tangibility, etc.) and that, in fact, complements those variables. One possibility is
that this has something to do with how the information environment and managerial disclosure vary by industry. Another
possibility is that plaintiffs’ attorneys specialize by industry, perhaps because there are economies of scale in bringing suit
in particular sectors. Whatever the explanation, it seems important to continue to include the FPS variable in models of
litigation risk, although it is also clear that using this variable alone does a poor job of capturing litigation risk.
5. Conclusion
We provide evidence on the validity of the industry-based litigation risk proxy commonly used in previous research.
We define litigation risk as the risk of private securities class action lawsuits, as opposed to more serious legal actions such
as SEC enforcement actions. We provide two principal empirical findings. First, we show that although litigation rates vary
significantly across sectors and industries over time, litigation rates in the four FPS industries (biotechnology, computers,
electronics, and retail) are generally consistently higher than those in other industries. While the overall litigation rate
across all firm/years in our sample is 1.6%, the rate for firms in the FPS industries is 2.7%, a difference that is statistically
significant. Differences in litigation rates between the FPS industries as a group and other industries are statistically
significant in 8 of 13 sample years. For the largest firms in the economy (those in the top 5% of the size distribution), the
litigation rate is 5.1% across all firm/years, with the rate for firms in the FPS industries at 7.8% (this rate is not significantly
higher than that for non-FPS industries).
Second, we estimate and compare a number of models of litigation risk. While the FPS industry measure is simple,
readily available, and associated with higher litigation rates, it is nevertheless unclear how well this variable performs as a
predictor of litigation risk. We evaluate predictive ability using a number of measures in addition to the pseudo-R squareds
usually reported in extant research. While the FPS variable is clearly associated with litigation risk—the coefficients on this
variable are both economically and statistically significant—the ability of this variable to predict litigation is modest.
Pseudo-R squareds from models that only include the FPS variable are around 1%. This conclusion is supported by
alternative measures of predictive ability, such as AUC and the Hosmer-Lemeshow chi-squared, which the statistical
literature suggests as better measures of predictive ability.
When the FPS variable is augmented with measures of firm size, sales growth, and return characteristics, predictive
ability increases markedly, suggesting that the inclusion of a few widely available variables can result in significant
improvements in model performance. AUC is around 0.55 for models that include the FPS variable alone, only marginally
higher than 0.5, the benchmark for no predictive ability. When we augment the FPS variable with size, growth, and return
volatility, AUC increases substantially, to 0.76, which indicates good predictive ability. This improvement in predictive
ability is achieved at relatively low cost because these additional variables are readily available to researchers.
Interestingly, the FPS variable complements these other variables, which have larger effects on estimated litigation risk
for FPS firms than for non-FPS firms. Augmenting the model with additional covariates, such as those that measure the
quality of firms’ corporate governance, insider trades, the extent to which firms are raising capital, etc., does little to
further increase predictive ability, and so fails the cost–benefit tradeoff given the costs usually associated with obtaining
these variables.
Our suggested model of litigation risk generates predicted probabilities that have desirable properties. This model
generates a relatively continuous distribution of predicted probabilities, ranging from close to zero to over 70%. While
most observations have predicted probabilities of less than 10%, some firm/years have probabilities well in excess of this
34
When we estimate model 3 using the smaller sample in Table 8, the numbers for this model drop somewhat, with pseudo-R squareds of 6.4% and
2.3%, respectively, which are not as high as those for model 5. AUC for model 3 in Table 8 is 0.705, slightly below that for model 5 and also below that for
model 3 in Table 7.
35
This continues to be true for various sample subperiods.
308
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
level. We show that firms in FPS industries that are relatively large, with high volatility and sales growth, have litigation
rates that can be substantially higher than 10%, and that the joint distribution of these variables is important in
determining litigation risk (firms that are both relatively large and volatile face substantially higher litigation rates than
firms that are relatively large or volatile).
Appendix A
See Table A1.
Table A1
Variable definitions.
This table provides definitions of the variables used in the tables in alphabetical order. For all non-stock return and turnover variables, the year t
represents the filing year for sued firms, and the firm-year for non sued firms. Accumulation period for stock return and turnover variables is provided in
the data definition.
Variable
Definition
ALTMAN Zt 1
DEBT PROCEEDSt 1
Altman (1968) Z score at the end of year t 1
Dollar amount of public debt proceeds issued by the firm during year t 1 and year t 2 scaled by beginning of year
t 1 total assets
Dollar amount of equity proceeds issued by the firm during year t 1 and year t 2 scaled by beginning of year t 1
total assets
Equals 1 if the firm is in the biotech (SIC codes 2833–2836 and 8731–8734), computer (3570–3577 and 7370–7374),
electronics (3600–3674), or retail (5200–5961) industry, and 0 otherwise
End of year t 1 goodwill scaled by beginning of year t 1 total assets
Average of all insider shares held in year t 1 scaled by beginning of year t 1 total shares outstanding
Average of year t 1 and t 2 insider sales net of acquisitions scaled by year t 1 revenue
Percentage of market value held by institutional investors at the end of year t 1
Natural log of total assets at the end of year t 1
Market value of equity scaled by book value of equity at the end of year t 1
Equals 1 if the firm is listed on the New York Stock Exchange, and 0 otherwise
Property, plant and equipment at the end of year t 1 scaled by beginning of year t 1 total assets
Research and development expenses in year t 1 scaled by beginning of year t 1 total assets
Market-adjusted 12-month stock return. For sued firms, the accumulation period ends with the lawsuit class period
end month. For non sued firms, the accumulation period ends with the fiscal year-end month
Market-adjusted 12-month stock return. For sued firms, the accumulation period ends with the fiscal year-end before
the filing year. For non sued firms, the accumulation period ends with year t 1 fiscal year-end month
For sued firms, the market-adjusted 12-month stock return for year t 2 before the filing year. For non sued firms, the
market-adjusted 12-month stock return for year t 2
For sued firms, the market-adjusted 12-month stock return for year t 3 before the filing year. For non sued firms, the
market-adjusted 12-month stock return for year t 3
Skewness of the firm’s 12-month return
Skewness of the firm’s 12-month return for year t 1
Standard deviation of the firm’s 12-month returns
Standard deviation of the firm’s 12-month returns for year t 1
Return on assets, defined as year t 1 net income scaled by beginning of year t 1 total assets
Year t 1 sales less year t 2 sales scaled by beginning of year t 1 total assets
Equals 1 if a class period of a lawsuit filing occurred during the year, and 0 otherwise
Trading volume accumulated over the 12-month period ending with the lawsuit class period end month (for sued
firms), and the fiscal year-end month (for nonsued firms) scaled by beginning of the year shares outstanding. Note that
the coefficient on TURNOVER is multiplied by 1000 for expositional convenience
Trading volume accumulated over the 12-month period ending with the fiscal year-end before lawsuit filing (for sued
firms), and year t 1 fiscal year-end month (for non sued firms) scaled by beginning of year t 1 shares outstanding.
Note that the coefficient on TURNOVERt 1 is multiplied by 1000 for expositional convenience
Equals 1 if the firm is incorporated in the United States, and 0 otherwise
Working capital accruals (current assets current liabilities) at the end of year t 1 scaled by beginning of year t 1
total assets
EQUITY PROCEEDSt 1
FPSt
GOODWILLt 1
INSIDER HOLDINGt 1
INSIDER TRADINGt 1
INSTt 1
LNASSETSt 1
MBt 1
NYSEt 1
PP&Et 1
R&Dt 1
RETURNt
RETURNt 1
RETURNt 2
RETURNt 3
RETURN SKEWNESSt
RETURN SKEWNESSt 1
RETURN STD DEVt
RETURN STD DEVt 1
ROAt 1
SALES GROWTHt 1
SUEDt
TURNOVERt
TURNOVERt 1
USINCORPt 1
WCt 1
References
Ajinkya, B., Bhojraj, S., Sengupta., P., 2005. The association between outside directors, institutional investors, and the properties of management earnings
forecasts. Journal of Accounting Research 43, 343–376.
Albuquerque, A., 2009. Peer firms in relative performance evaluation. Journal of Accounting & Economics 48, 69–89.
Alexander, J., 1991. Do the merits matter? A study of settlements in securities class actions. Stanford Law Review 43, 497–598.
Ali, A., Kallapur, S., 2001. Securities price consequences of the Private Securities Litigation Reform Act of 1995 and related events. The Accounting Review
76, 431–460.
Altman, E., 1968. Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance 23, 589–609.
Altman, E., Sabato, G., 2007. Modelling credit risk for SMEs: evidence from the U.S. market. Abacus 43, 332–357.
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
309
Arena, M., Julio, B., 2010. Litigation Risk and Corporate Cash Holdings. Working Paper. Marquette University.
Baginski, S., Hassell, J., Kimbrough, M., 2002. The effect of legal environment on voluntary disclosure: evidence from management earnings forecasts
issued in U.S. and Canadian markets. The Accounting Review 77, 25–50.
Baker, T., Griffith, S., 2007. Predicting corporate governance risk: evidence from the directors’ and officers’ liability insurance market. The University of
Chicago Law Review 74, 487–544.
Batta, G.E., Wongsunwai, W., 2011. Equity Misvaluation and Corporate Bankruptcy Prediction. February 9, 2011. Available at SSRN: /http://ssrn.com/
abstract=1671685S.
Beatty, A., Weber, J., Yu, J., 2008. Conservatism and debt. Journal of Accounting & Economics 45, 154–174.
Beneish, M., 1999. Incentives and penalties related to earnings overstatements that violate GAAP. The Accounting Review 74, 425–457.
Bhojraj, S., Libby, R., Yang, H., 2010. Analyzing guidance at the firm level: the association between guidance frequency, guidance properties, and market
reactions. Unpublished Paper, Available at /http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1545864S.
Brown, S., Hillegeist, S., Lo, K., 2005. Management Forecasts and Litigation Risk. Working Paper. Emory University.
Brown, S., Tucker, J., 2011. Large-sample evidence on firms’ year-over-year MD&A modifications. Journal of Accounting Research 49, 309–346.
Chandra, U., Wasley, C., Waymire, G., 2004. Income Conservatism in the U.S. Technology Sector. Working Paper. University of Kentucky.
Cheng, C.S., Huang, H., Li, Y., Lobo, G., 2010. Institutional monitoring through shareholder litigation. Journal of Accounting & Economics 95, 356–383.
Choi, S., 2006. Do the merits matter less after the Private Securities Litigation Reform Act? The Journal of Law, Economics, & Organization 23, 598–626.
Choi, S., Nelson, K., Pritchard, A., 2009. The screening effect of the Private Securities Litigation Reform Act. Journal of Empirical Legal Studies 6, 36–68.
Coffee, J., 1985. The unfaithful champion: the plaintiff as monitor in shareholder litigation. Law and Contemporary Problems 48, 5–81.
Coffee, J., 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, 669–727.
Cornerstone Research, 2011a. Securities Class Action Filings 2010: A Year in Review. Available at /http://www.cornerstone.comS.
Cornerstone Research, 2011b. Securities Class Action Filings Settlements: 2010 Review and Analysis. Available at /www.cornerstone.comS.
The Corporate Library, 2009. Securities Litigation Risk Analyst.
Correia, M., Richardson, S., Tuna, I., 2011. Value Investing in Credit Markets. Working Paper. London Business School.
Cox, J., Thomas, R., 2006. Does the plaintiff matter? An empirical analysis of lead plaintiffs in securities class actions. Columbia Law Review 106,
1587–1640.
Dai, Z., Jin, L., Zhang, W., 2008. Litigation Risk and Executive Compensation. Working Paper. University of Texas Dallas.
Daines, R., Gow, I., Larcker, D., 2010. Rating the Ratings: How Good are Commercial Governance Ratings? Working Paper. Stanford University.
Dechow, P., Sloan, R., Sweeney, A., 1996. Causes and consequences of earnings manipulations: an analysis of firms subject to enforcement actions by the
SEC. Contemporary Accounting Research 13, 1–36.
Dechow, P., Ge, W., Larson, C., Sloan, R., 2011. Predicting material accounting misstatements. Contemporary Accounting Research 28, 17–82.
Demers, E., Joos, P., 2007. IPO failure risk. Journal of Accounting Research 45, 333–371.
Donelson, D.C., McInnis, J.M., Mergenthaler, R., Yu, Y., 2010. The timeliness of earnings news and litigation risk. Unpublished Paper, Available at
/http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1641342S.
Dunbar, F., Juneja, V., Martin, D., 1995. Shareholder Litigation: Deterrent Value, Merit and Litigants’ Options. Working Paper. NERA Economic Consulting.
Efron, B., Tibshirani, R.J., 1993. An Introduction to the Bootstrap. Chapman & Hall/CRC.
Einhorn, E., Ziv, A., 2008. Intertemporal dynamics of corporate disclosures. Journal of Accounting Research 46, 567–589.
Feroz, E., Park, K., Pastena, V., 1991. The financial and market effects of the SEC’s Accounting and Auditing Enforcement Releases. Journal of Accounting
Research 29 (Suppl.), 107–142.
Field, L., Lowry, M., Shu, S., 2005. Does disclosure deter or trigger litigation? Journal of Accounting & Economics 39, 487–507.
Francis, J., Philbrick, D., Schipper, K., 1994a. Shareholder litigation and corporate disclosures. Journal of Accounting Research 32, 137–164.
Francis, J., Philbrick, D., Schipper, K., 1994b. Determinants and Outcomes in Class Action Securities Litigation. Working Paper. University of Chicago.
Frankel, R., Joos, P., Weber, J., 2002. An Empirical Investigation of Pre-earnings Announcement Quiet Periods. Working Paper. Washington University.
Hennes, K., Leone, A., Miller, B., 2008. The importance of distinguishing errors from irregularities in restatement research: the case of restatements and
CEO/CFO turnover. The Accounting Review 83, 1487–1519.
Hobson, J.L., Mayew, W.J., Venkatachalam, M., 2011. Analyzing speech to detect financial misreporting. Available at SSRN: /http://ssrn.com/
abstract=1531871S.
Hosmer, D., Lemeshow, S., 2000. Applied Logistic Regression, second ed. John Wiley & Sons.
Hribar, P., Kravet, T., Wilson, R., 2010. A New Measure of Accounting Quality. Working Paper. The University of Iowa.
Jayaraman, S., Milbourn, T., 2009. Does Equity-based CEO Compensation Really Increase Litigation Risk? Working Paper. Washington University.
Johnson, M., Kasznik, R., Nelson, K., 2000. Shareholder wealth effects of the Private Securities Litigation Reform Act of 1995. Review of Accounting Studies
5, 217–233.
Johnson, M., Kasznik, R., Nelson, K., 2001. The impact of securities litigation reform on the disclosure of forward-looking information by high technology
firms. Journal of Accounting Research 39, 297–327.
Johnson, M., Nelson, K., Pritchard, A., 2007. Do the merits matter? The impact of the Private Securities Litigation Reform Act. The Journal of Law,
Economics, & Organization 23, 627–652.
Jones, C., Weingram, S., 1996a. The Determinants of 10b-5 Litigation Risk. Working Paper. George Washington University.
Jones, C., Weingram, S., 1996b. Why 10b-5 Litigation Risk is Higher for Technology and Financial Services Firms. Working Paper. George Washington
University.
Karpoff, J., Lee, D., Martin, G., 2008. The consequences to managers for financial misrepresentation. Journal of Financial Economics 88, 193–215.
Larcker, D., Zakolyukina, A., 2011. Detecting deceptive discussions in conference calls. Unpublished Paper, Available at /http://papers.ssrn.com/sol3/
papers.cfm?abstract_id=1572705S.
Laux, V., 2010. Effects of litigation risk on board oversight and CEO incentive pay. Management Science 56, 938–948.
Lennox, C., Park, C., 2006. The informativeness of earnings and management’s issuance of earnings forecasts. Journal of Accounting & Economics
42, 439–458.
Lev, B., Zarowin, P., 1999. The boundaries of financial reporting and how to extend them. Journal of Accounting Research 36, 353–385.
Long, J.S, 1997. Regression Models for Categorical and Limited Dependent Variables. SAGE Publications.
Long, J.S., Freese, J., 2006. Regression Models for Categorical Dependent Variables Using Stata. Stata Press.
Lowry, M., Shu, S., 2002. Litigation risk and IPO underpricing. Journal of Financial Economics 65, 309–335.
Matsumoto, D., 2002. Management’s incentives to avoid negative earnings surprises. The Accounting Review 77, 483–514.
McFadden, D., 1973. Conditional logit analysis of qualitative choice behavior. In: Frontiers in Econometrics, Academic Press, pp. 105–142.
Ohlson, J., 1980. Financial ratios and the probabilistic prediction of bankruptcy. Journal of Accounting Research 18, 109–131.
Pritchard, A., Sale, H., 2005. What counts as fraud? An empirical study of motions to dismiss under the Private Securities Litigation Reform Act. Journal of
Empirical Legal Studies 2, 125–149.
Rogers, J., Stocken, P., 2005. Credibility of management forecasts. The Accounting Review 80, 1233–1260.
Rogers, J., Van Buskirk, A., 2009. Shareholder litigation and changes in disclosure behavior. Journal of Accounting & Economics 47, 136–156.
Romano, R., 1991. The Shareholder Suit: litigation without Foundation. The Journal of Law, Economics, and Organization 7, 55–87.
Schrand, C., Zechman, S., 2011. Executive overconfidence and the slippery slope to financial misreporting. Unpublished Paper, Available at /http://papers.
ssrn.com/sol3/papers.cfm?abstract_id=1265631S.
310
I. Kim, D.J. Skinner / Journal of Accounting and Economics 53 (2012) 290–310
Seetharaman, A., Gul, F., Lynn, S., 2002. Litigation risk and audit fees: evidence from UK firms cross-listed on US markets. Journal of Accounting &
Economics 33, 1–115.
Shu, S., 2000. Auditor resignations: clientele effects and legal liability. Journal of Accounting & Economics 29, 173–205.
Skinner, D.J., 1994. Why firms voluntarily disclose bad news. Journal of Accounting Research 32, 38–60.
Skinner, D.J., 1997. Earnings disclosures and stockholder lawsuits. Journal of Accounting & Economics 23, 249–282.
Smith, C.W., Watts, R.L., 1992. The investment opportunity set and corporate financing, dividend, and compensation policies. Journal of Financial
Economics 32, 263–292.
Tasker, S., 1998. Bridging the information gap: quarterly conference calls as a medium for voluntary disclosure. Review of Accounting Studies 3, 137–167.
Weiss Hanley, K., Hoberg, G., 2010. Are Strategic Disclosure and Underpricing Decisions Influenced by Liability Risk? Working Paper, Federal Reserve
Board.