MARKET DISCIPLINE, INTEGRITY, INFORMATION DISCLOSURE AND FINANCIAL

MISCONDUCT

By

Qingqiu Li

A DISSERTATION

Michigan State University

in partial fulﬁllment of the requirements

Submitted to

for the degree of

Business Administration - Finance – Doctor of Philosophy

2018

ABSTRACT

MARKET DISCIPLINE, INTEGRITY, INFORMATION DISCLOSURE AND FINANCIAL

MISCONDUCT

By

Qingqiu Li

In my dissertation, I explore the factors which impact the outcome of ﬁnancial misconduct. Financial
misconduct imposes negative externalities on ﬁrm value, inﬂuences investment decisions, and
results in both wealth transfer and destruction. The unique governance structure in the asset
management industry ampliﬁes the role of investors on ﬁrm behavior. I provide novel evidence on
the variation in response towards enforcement actions by investor types; I ﬁnd evidence consistent
with two non-mutually exclusive explanations for this heterogeneity. First, investor sophistication
aﬀects the eﬀective cost of information acquisition and processing, making the fund ﬂow discipline
less prevalent for retail investors. Second, investors are less likely to punish funds when the
costs of moving capital become substantial. Besides investors, other market participants also
signiﬁcantly aﬀect ﬁrm behavior. Internal factors, such as corporate culture, along with external
factors, such as product market competition, have signiﬁcant impact on corporate fraud. For
example, a lack of focus on integrity in corporate culture is associated with unethical corporate
behavior, cultures that neglect integrity are associated with a greater probability of SEC enforcement
actions for accounting misstatements. In addition, ﬁrms with lower product market diﬀerentiation
exhibit signiﬁcantly lower rates of fraud; the relationship is more pronounced for complex ﬁrms
and is robust to controlling for various measures of competition, predictors of fraud, and industry
heterogeneity. Overall, the ﬁndings suggest that lower diﬀerentiation disciplines ﬁrms by facilitating
fraud detection through a benchmarking channel.

Copyright by
QINGQIU LI
2018

This thesis is dedicated to my husband Alexander Ferko, my parents Zhao Hanxiang and Li Gang.

iv

ACKNOWLEDGEMENTS

I am honored to have completed my dissertation under the mentoring of Dr. Andrei Simonov
(dissertation committee chair), Dr. Charles Hadlock (dissertation committee member), Dr. Zoran
Ivkovic (dissertation committee member) and Dr. Miriam Schwarts-Ziv (dissertation committee
member). Their advice and guidance have contributed greatly to my development as a scholar. I
would also like to express my gratitude to Dr. Audra Boone, Dr. Nickolay Gantchev, Dr. William
Grieser, Dr. Stacey Jacobsen, Dr. Chotibhak Jotikasthira, Dr. Mattia Landoni, Dr. Mehrdad
Samadi, Dr. Parth Venkat for their support and insightful comments.

v

TABLE OF CONTENTS

x

.
.

.
.

.
.

.

.

. .

. .

. .

. .

.
.
.

.

.
.

.

.
.

.

.

.
.

.

.
.

.

.
.

.

. .
. .
.
.

CHAPTER 1

Introduction .

1.5.1 Monitoring .
1.5.2

1.4.1 Visibility .
1.4.2
Severity . .
1.4.3 Relevancy .

LIST OF TABLES .
LIST OF FIGURES .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
INVESTOR REACTION AND MUTUAL FUND MISCONDUCT . . . . .
1.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Data and Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
1
7
1.2.1 Data .
7
1.2.2 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Empirical Design and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.1 Empirical Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.2
Fund Flow and Regulatory Disclosure . . . . . . . . . . . . . . . . . . . . 12
1.3.3 Management Fees and Regulatory Disclosures . . . . . . . . . . . . . . . . 14
1.4 Eﬀective Information Costs and Regulatory Disclosure . . . . . . . . . . . . . . . 15
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5 Monitoring, Investor Mobility and Regulatory Disclosure . . . . . . . . . . . . . . 19
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Investor Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.2.1 Capital Gains Overhang . . . . . . . . . . . . . . . . . . . . . . 20
1.5.2.2
Investment Options . . . . . . . . . . . . . . . . . . . . . . . . . 21
Sales Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.2.3
. 22
CHAPTER 2 CORPORATE CULTURE AND CORPORATE FRAUD . . . . . . . . . . . 24
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.1 The AshleyMadison Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.2 Other data .
2.3 Corporate Fraud and Ethics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4
Internal vs. External CEOs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5 Conclusion .
CHAPTER 3 PRODUCT COMPETITION AND CORPORATE FRAUD . . . . . . . . . 39
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1
3.2 Data .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Product Diﬀerentiation and Alternative Measures of Competition . . . . . . . . . . 47
Product Market Diﬀerentiation . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3.1
3.3.2 Alternative Measures of Competition . . . . . . . . . . . . . . . . . . . . 49
. 50

1.6 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Introduction .

2.1
2.2 Data .

. .

.

3.4 Empirical Results .

Introduction .

.
. .

.
. .

.
.

.

.

.
.

.
.

.

.

.
.

.
.

.

.

.
.

.

.

.

. .

.

. .

.

. .

. .

. .

.

.

. .

vi

Product Market Diﬀerentiation and Corporate Fraud . . . . . . . . . . .
. 51
3.4.1
3.4.2
Firm Complexity and Product Diﬀerentiation . . . . . . . . . . . . . . . . 52
3.4.3 Change to Information Environment Associated with Rival IPOs and M&A 54
3.4.4 Alternative Measures of Competition, Product Similarity, and Fraud . . . . 57
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5 Conclusion . .
.
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
INVESTOR REACTION AND MUTUAL FUND MISCONDUCT . 61
CORPORATE CULTURE AND CORPORATE FRAUD . . . . . . 76
PRODUCT COMPETITION AND CORPORATE FRAUD . . . . . 87
.
. 97

APPENDICES .
APPENDIX A
APPENDIX B
APPENDIX C
BIBLIOGRAPHY .

. .
.
.

.
.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.

.

.

.

.
.

.

.

.

.
.

.

vii

LIST OF TABLES

Table 1.1: Variable Description .

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

Table 1.2: Descriptive Summary Statistics

. . . . . . . . . . . . . . . . . . . . . . . . . . 65

Table 1.3: Fund Flow and Regulatory Disclosure . . . . . . . . . . . . . . . . . . . . . . . 66

Table 1.4: Sales Channel and Regulatory Disclosures . . . . . . . . . . . . . . . . . . . . . 67

Table 1.5: Fund Flow and Regulatory Disclosure - Visibility . . . . . . . . . . . . . . . . . 68

Table 1.6: Fund Flow and Regulatory Disclosure – Fraud Severity . . . . . . . . . . . . . . 69

Table 1.7: Fund Flow and Regulatory Disclosure – Information Quality . . . . . . . . . . . 70

Table 1.8: Fund Flow and Regulatory Disclosure - Monitoring Eﬀort

. . . . . . . . . . . . 71

Table 1.9: Fund ﬂow and Regulatory Disclosure – Investor Mobility . . . . . . . . . . . . . 72

Table 1.10: List of Largest Financial Conglomerate in the U.S.

. . . . . . . . . . . . . . . . 73

Table 1.11: Fund ﬂow before and after Regulatory Event

. . . . . . . . . . . . . . . . . . . 74

Table 1.12: Fund Flow and Regulatory Disclosure - Sales Loads

. . . . . . . . . . . . . . . 75

Table 2.1: Description of the Variables

. . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

Table 2.2: Descriptive Statistics .

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

Table 2.3: AAER Misstatements and AshleyMadison Membership . . . . . . . . . . . . . . 80

Table 2.4: KLD Ethics Ratings and AshleyMadison Membership . . . . . . . . . . . . . . 82

Table 2.5: Tax Policy and AshleyMadison Membership . . . . . . . . . . . . . . . . . . . 83

Table 2.6: AshleyMadison Membership and the Choice of Internal vs. External CEO . . . . 85

Table 2.7: Determinants of AshleyMadison Membership . . . . . . . . . . . . . . . . . . . 86

Table 3.1: Variable Description .

Table 3.2: Summary Statistics .

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

viii

Table 3.3: Product Market Diﬀerentiation and Corporate Fraud . . . . . . . . . . . . . . . 90

Table 3.4: Product Diﬀerentiation and Fraud by Complexity Quartiles . . . . . . . . . . .

. 91

Table 3.5:

IPOs and Acquisitions of Rivals as Change to Information Environment

. . . . . 92

Table 3.6: Product Market Diﬀerentiation and Corporate Fraud (Controlling for Alterna-

tive Measures of Competition) . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

Table 3.7: Alternative Measures of Competition and Corporate Fraud . . . . . . . . . . .

. 94

Table 3.8: Product Market Diﬀerentiation and Corporate Fraud âĂŞ Standardized Variables

95

Table 3.9: Product Market Diﬀerentiation and Corporate Fraud Alternate Similarity Scores

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

(Standardized)

. .

ix

LIST OF FIGURES

Figure 1.1: Regulatory Disclosure .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

Figure 1.2: Fund Flow Before and After Regulatory Disclosure . . . . . . . . . . . . . .

. 63

x

CHAPTER 1

INVESTOR REACTION AND MUTUAL FUND MISCONDUCT

1.1 Introduction

Detection of ﬁnancial fraud depends on both internal and external corporate governance control
mechanisms (Dyck et al., 2010; Fang et al., 2016). Internal corporate governance is negatively
associated with ﬁnancial fraud (Dechow et al., 1996; Farber, 2005), whereas external governance
depends on regulators and competitors facilitating information disclosures and the market’s re-
sponses revealed the allocation decisions of investors (Giannetti and Wang, 2016; Gurun et al.,
2018). The extent to which regulatory enforcement and market enforcement complement each
other in protecting stakeholder interests depends on the behavior of market participants (DeMarzo
et al., 1998; Garoupa, 2002). Yet, little is known about how investor heterogeneity aﬀects this rela-
tion. In this paper, I study how shareholder composition aﬀects market discipline by investigating
variation in investors’ responses to regulatory disclosures of mutual fund misconduct.

Why is the mutual fund a good laboratory to examine the role of investor heterogeneity in
external governance? First, nearly half of U.S. households own mutual funds, whose assets
under management exceed $10 trillion by 2016. Such high reliance on investment companies and
investors’ limited attention (Hirshleifer and Teoh, 2003; Sialm et al., 2015) expose investors to a
greater risk of mutual fund fraud.1 Second and more importantly, the unique governance structure
of mutual funds disciplines management through redeemable residual claims (Fama and Jensen,
1983)2, as mutual fund boards have limited oversight over the management and advisers3.
In
addition, unlike banking regulators, the Securities and Exchange Commission (SEC) facilitates the

1See ICI Research Perspective and ICI Fact book (2017).
2Fama and Jensen (1983) posit that redeemable residual claims should serve as a strong disciplining mechanism
to resolve ﬁduciary conﬂicts of interest between shareholders and management, which is very diﬀerent from equity
shares are not redeemable, but rather transferred on secondary market.

3The 1940 Act and its rules set forth speciﬁc duties of mutual fund directors because mutual fund has no
employees, the operations of mutual funds relies on the adviser and other service providers. (See Tufano and Sevick,
1997; Del Guercio et al., 2003; Khorana et al., 2007; Ding and Wermers, 2012)

1

transparency of mutual fund operations, but has no authority to intervene if an investment adviser
is taking excessive risks (Jickling and Murphy, 2010). With weak governance structures in place
and very limited scope for regulatory enforcement, market discipline plays a vital role in mutual
fund governance.

I present novel evidence about the heterogeneity in responses to disciplinary actions for diﬀerent
investor types using regulatory disclosures against investment advisers from Form ADV. First, I
show that investment adviser misconduct results in strong negative ﬂows from institutional investors,
but not from other investors (e.g, direct-investing retail clients). Second, I ﬁnd supportive evidence
for two non-mutually exclusive explanations of such asymmetric reactions. One explanation is that
investor sophistication and information noise aﬀect information transmission. Another explanation
is that due to restricted investor mobility, elevated asset redemption costs reduce investors’ incentive
to reallocate assets. These ﬁndings contribute to a better understanding of how shareholder
characteristics aﬀect the extent to which market disciplines management.

Form ADV is a mandatory disclosure form used by investment advisers to register with both the
SEC and state securities authorities, and must include any disciplinary events involving the ﬁrm
and/or its employees.4 For example, ADV disclosures reveal that Virtus Investment Advisers used
false and misleading advertisements to grow its assets under management from $191 millions in
2009 to $11.5 billion by 2013; Barclays advisory programs overbilled their clients by nearly $50
million; one of the portfolio managers from Morgan Stanley Investment Management unlawfully
conducted prearranged trading that favored certain advisory client accounts over other. These
instances of ﬁnancial fraud not only cost investors millions of dollars but also destroy investors’
trust in the ﬁnancial system. Following disclosures of such regulatory disciplinary events, investors
exert market discipline on ﬁnancial institutions through asset redemption; however, the extent to
which they do so depends on the prevalence of information friction and restrictions on investor
mobility (Bliss and Flannery, 2002).

4Form ADV is used in recent studies regarding investment adviser fraud. Form ADV discloses information
about investment advisers’ aﬃliation, disciplinary events, and other material facts. The information is used to assess
fraudulent risks of investment advisers (Dimmock and Gerken, 2012).

2

Consistent with previous studies, such mutual fund misconduct is signiﬁcantly curtailed by
asset redemptions upon the revelation of disclosed events. I ﬁnd that in the month after regulatory
disclosures, funds on average experience negative net ﬂows of almost 1%. Such negative ﬂows
amounts to a reduction in asset under management of over $24.5 million, on average, assets under
management. During the six months following the regulatory disclosures, the cumulative fund
ﬂows total to negative 3.6%. More importantly, I ﬁnd signiﬁcant variation in fund ﬂow reactions
across investor types. The magnitude of negative net ﬂows is twice as large among institutional
share classes compared to retail share classes. The lack of fund ﬂow response among retail investors
is primarily found in direct-sold funds. This asymmetric market discipline is alarming considering
the increased reliance of individual investor on mutual funds.

Next, I explore two non-mutually exclusive explanations for the heterogeneous fund ﬂow re-
sponse by investor types. First, I ﬁnd that elevated information costs signiﬁcantly reduce the
prevalence of the investors’ response. Even though SEC enforcement information is public infor-
mation, the eﬀective information costs depend on investors sophistication and information quality.
Moreover, the ﬂows from broker-sold funds are sensitive to regulatory disclosures, suggesting that
the costly service provided by full-service brokers (compared to discount brokers who exclusively
carry out trade orders) can reduce such information frictions signiﬁcantly. Second, capital reallo-
cation can be costly for investors, search costs for investment options and tax liabilities can impair
investors’ ability to reallocate assets, therefore aﬀect their responses to mutual fund fraud.

Investor sophistication has direct impact on their responses and market discipline through the
mechanism of elevated eﬀective information costs. There exists information frictions that could
interfere information communication for less sophisticated investors and contribute to the variation
in the eﬃcacy of market discipline. For example, negative fund ﬂows are more pronounced among
highly visible funds, such as funds from larger institutions and star fund families.5 Media coverage
can signiﬁcantly reduce the search costs of such information. These results suggest that retail

5Following Nanda et al. (2004), star funds are the top performing funds ranked by risk-adjusted return for the
past 12 months. “Star" family is an indicator variable equal to one if a fund family has at least one star fund under
management.

3

investors may rely on diﬀerent information sources (see Sirri and Tufano, 1998; Chevalier and
Ellison, 1997), and the heterogeneity in investor sophistication can drive market segmentation
(Guercio et al., 2010).

Indeed, the diﬀerences in fund ﬂow responses between institutional and retail investors dis-
appear when the enforcement events are against repeat oﬀenders. The magnitude of the negative
ﬂow reaction from retail investors more than doubles for repeated oﬀenses (−0.6%) compared to
ﬁrst-time oﬀenses (−0.2%). The negative ﬂow response is also stronger when regulatory actions
are initiated by the SEC, compared to other regulators, including state regulators, other federal
regulators, foreign regulators, and self-regulating organizations such as Financial Industry Reg-
ulatory Authority (FINRA). Fund ﬂow responses from retail investors also increase as the fraud
charges gets more severe,6 and increase with the media coverage of the mutual fund misconduct7.
These results further convey that retail investors have an inferior ability to access information and
therefore face a higher level of eﬀective costs to public information.

There has been a long-standing debate over whether mandatory disclosure requirements improve
transparency and ﬁnancial eﬃciency (i.e. Hermalin and Weisbach, 2012; Agarwal et al., 2014;
Edmans et al., 2016). However, the relation between investor response and information quality
is overlooked, especially when considered in combination with investor sophistication.
I use
irrelevant disciplinary events to study how information quality aﬀects investor response. I ﬁnd that
both enforcement events against unrelated mutual fund products (i.e., derivatives and insurance)
and against funds associated with ﬁnancial institutions (no indication of wrong-doings by funds
themselves) result in negative fund ﬂows by retail share classes. In addition, investment advisers
are also forced to reduce management fees by 5% after disciplinary events, with the exception of
direct-sold retail funds. These results raise further questions about the scope of market discipline
in the presence of investor segmentation.

The second explanation for the variation in intensity of market discipline depends on investor’s

6The severity of fraud is measured by the monetary ﬁne as a result of regulatory enforcement.
7Media coverage is an indicator variable that takes value one if there is mentioning of ﬁnancial fraud in the major
U.S. business news outlet, such as The Wall Street Journals, The Dow Jones Newswire, Yahoo Finance, and The New
York Times etc.

4

incentive to monitor, measuring investors’ monitoring eﬀort remains chalenging. Evans and
Fahlenbrach (2012) argue that retail investors can beneﬁt from the ability and willingness of
institutional investors to exercise market governance; retail investors beneﬁt from monitoring
performed by institutional “twin" funds8 and can mimic the asset allocation decisions of their twin
funds. I ﬁnd that the retail funds show much stronger negative fund ﬂow responses when they have
an institutional twin. The results suggest that investor’s incentive to monitor has signiﬁcant impact
on the eﬀectiveness of market discipline.

Another factor that aﬀect monitoring incentives is investor mobility.9 For instance, tax liabil-
ity aﬀect capital allocation decisions (Sialm and Starks, 2012); capital gains overhang increases
investors’ tax liabilities and costs associated with asset redemptions (Barclay et al., 1998; Ivkovic
et al., 2005). Indeed, I only observe increased outﬂows among funds with low capital gains over-
hang. My results suggest that the lack of investor response is an outcome of the trade-oﬀ between
the costs of asset redemptions and the beneﬁts of switching funds.

In addition to capital gains overhang, outside options and other transaction costs also aﬀect
investor mobility. Fund ﬂow responses are particularly strong during the late-trading scandal
period (2003 to 2005), and became much weaker during the ﬁnancial crisis period (2007 - 2009).
The late-trading scandal is associated with extensive scrutiny by media, which reduces the search
costs for alternative investment options. The trade-oﬀ between transactions costs (direct and
indirect) and monitoring beneﬁts can result in investor inattention and inaction.

To my knowledge, this is the ﬁrst paper about how investor segmentation aﬀects the eﬃcacy of
enforcement in case of mutual fund fraud by showing heterogeneity in market responses, based on
investor types and regime of enforcement. Previous studies show that mandatory disclosures have
signiﬁcant power in predicting investment fraud (Dimmock and Gerken, 2012), and that disclosed
ﬁnancial misconduct aﬀects mutual fund ﬂows (Houge and Wellman, 2005; Brown et al., 2008;
Qian and Tanyeri, 2017). In more recent work, Wu (2017) shows that mutual fund companies tend

8The “twin" funds are the funds with the same manager and similar performance but sold to diﬀerent investors of

a diﬀering ability to select and monitor managers.

9Hubbard et al. (2010) argue that investor mobility signiﬁcantly aﬀects investors’ asset allocation decisions.

5

to reduce contractual incentives, raise marketing expenditures, and relax investment restrictions
after a revelation of misconduct.

Firms cooking the books are disciplined heavily by investors in the market place; for each dollar
of value inﬂation, ﬁrms lose up to four dollars (Karpoﬀ et al., 2008; Dyck et al., 2013; Johnson et al.,
2014). Mixed evidence exists on whether capital market participants eﬀectively discipline banks’
risk taking behavior through deposit withdrawals and requests for higher interest rates (Billett
et al., 1998; Martinez Peria and Schmukler, 2001; Bliss and Flannery, 2002). Asset management
companies diﬀer from both typical corporations and banks. As opposed to providing products that
require active decisions, the service-for-fee model naturally entails inertia on the consumer’s part.
In addition, unlike highly-regulated banks, the main role of the regulator in the asset management
industry is to maintain fair and orderly markets with no authority to limit risks taken by investment
advisers. Thus, the mutual fund industry relies to a large extent on market discipline in governing
fund managers, with the regulator’s responsibility limited to information disclosure (Jickling and
Murphy, 2010). This paper sheds light on how information frictions aﬀect the outcomes of market
discipline, in the presence of less-informed investors.

This paper also relates to the literature on information quality, investor mobility and asset
allocation decisions. Previous research documents that retail investors use less sophisticated
measures to evaluate funds, and their behavior incentivizes fund managers to alter risks (Chevalier
and Ellison, 1997; Sirri and Tufano, 1998); in contrast, sophisticated institutional investors show
more rational behavior (Del Guercio and Tkac, 2002). In addition, Evans and Fahlenbrach (2012)
ﬁnd greater monitoring and lower transaction costs when an institutional twin fund exists. This
reduces agency problems and enhances the retail fund’s performance. My paper complements this
strand of the literature by showing that lack of investor sophistication together with information
frictions can inﬂate the eﬀective information costs, and signiﬁcantly impair investors’ ability to
discipline mutual fund management. This casts doubt on whether market-based governance alone
can safeguard investors.

The rest of the paper is organized as follows. Section 2 describes the data sources used in the

6

paper and presents summary statistics. Section 3 introduces the empirical design and empirical
results. How information costs inﬂuence the eﬀectiveness of market discipline is examined empir-
ically in section 4. Section 5 investigates how monitoring incentives and investor mobility aﬀects
market discipline. Section 6 brieﬂy discusses the implications of the paper and concludes.

1.2 Data and Summary Statistics

1.2.1 Data

Four main data sources are used in this study. First, fund ﬂows are from two separate data sources.
Fund ﬂows at the share-class level are directly obtained from the Survivor-ship bias-free mutual
fund database provided by the Center for Research in Security Prices (CRSP) from year 2000 to
2016. As in other mutual fund studies (see Chevalier and Ellison, 1997; Barber et al., 2016), I
exclude funds which have data of less than 24 months, and total assets under management below
$10 million at year end. Closed-end funds and variable annuity funds are also excluded. In the ﬁnal
sample, only equity funds, which contain 5438 fund share classes and 1,654 investment advisers,
are included.

The second data source used to obtain fund ﬂows are N-SAR ﬁlings from years 2000 to 2016.
N-SAR is a mandatory semi-annual ﬁling for investment companies. The main advantage of N-SAR
data is that monthly gross inﬂows and outﬂows are reported separately. N-SAR ﬁlings also report
other fund characteristics such as fees, expenses, investment objectives, etc. N-SAR ﬁlings are
collected from the SEC’s Electronic Data Gathering, Analysis, and Retrieval system (EDGAR). In
my sample, there are 13,931 funds from 1,627 fund families and 1,232 unique investment advisers.10
The third main data source regarding regulatory disclosures are Form ADV obtained through
a Freedom of Information Act request. All investment advisers are required to ﬁle Form ADV
under the Investment Advisor Act of 1940. U.S mutual funds are subject to comprehensive

10I use CRSP data to diﬀerentiate institutional share classes from retail share classes. And in N-SAR, fund inﬂows
and outﬂows are reported separately. However, studies have found that the matching rate between CRSP mutual fund
database and N-SAR ﬁlings are lower than 40%. To avoid selection bias, I use both sources of data independently
throughout this study.

7

requirements under the Investment Company Act of 1940. These regulations provide investors
with transparency, assure daily liquidity, and ensure the trustworthiness of funds’ stated returns.
On the one hand, disclosure requirements aim to provide adequate and prompt information to help
investors make better decisions. On the other hand, regulatory enforcement actions aim to seek
injunctions, prohibit future violations and disgorge illegal proﬁts. Disclosure and enforcement
together can protect investors through reduced information costs and ensure the proper functioning
of the market.

Form ADV contains each adviser’s SEC number, which is used as the primary identiﬁer to
merge with its N-SAR ﬁlings. For the remaining advisers not matched using and SEC number, as
well as merging Form ADV with CRSP data, I match the investment advisers by their legal company
names. Regulatory disclosures obtained from Form ADV contain all regulatory events initiated
and amended by all regulators, including the SEC, the Commodity Futures Trading Commission,
self-regulatory organizations such as FINRA, state regulators and foreign regulators. In this paper,
I focus on the initial regulatory events to assess investors’ initial response to ﬁnancial misconduct,
which is the ﬁrst time public become aware of a particular investigation on investment adviser’s
misconduct.

Form ADV also discloses the principal products that are related to each regulatory event, such as
“Debt," “Equity," “Insurance" and “Mutual Fund(s)." These disclosures include all types violations,
e.g. speciﬁc trading regulation, insurance laws, failure to disclosure requirement, or failure to meet
the registration requirement, and most of these regulatory disclosures have no direct impact on the
value of assets in the funds. The main variable of interest in this study is an indicator variable MF
Fraud SEC, which takes the value of 1 if there is a new regulatory disclosure about the investment
adviser that is initiated by the SEC, with the principal product being “Mutual Fund(s)." I also deﬁne
MF Fraud All as an indicator variable, using the value of 1 when there is a regulatory disclosure
related to “Mutual Fund(s)" initialed by any regulators. Additionally, an indicator variable Other
Regulatory SEC utilizes the value of 1 if there are regulatory events initiated by the SEC against
investment advisers and related to products that are unrelated to mutual funds.

8

Additionally, SEC enforcement actions data against ﬁnancial institutions are obtained from
the SEC’s website.11 The names of ﬁnancial institutions are matched to SEC enforcement data;
these names are also matched with mutual fund names to determine whether a fund and ﬁnancial
institution are associated. The association is identiﬁed through one of the following channels.
One channel is through name association whereby mutual funds and ﬁnancial conglomerates share
common parts in their names. (For example, Goldman Sachs Trust: Goldman Sachs Asia Equity
Fund Class A Shares are associated with Goldman Sachs.) The other channel is by business
association whereby the mutual fund management company is a registered subsidiary of a ﬁnancial
conglomerate. Using 10-K ﬁlings, I manually verify whether the management company is listed
as a subsidy of the ﬁnancial institution. The ﬁnancial institutions included in the sample consist of
the top 40 ﬁnancial conglomerates ranked by total assets.12

To capture the severity of fraud and the media coverage.

I manually check the amount of
monetary ﬁne as a result of each regulatory enforcement from form ADV. For all the investment
advisers with at least one regulatory disclosure, I manually check the media coverage for the
misconduct. Media coverage data is from Factiva. By searching key words “fraud, misconduct,
scandal, enforcement", the indicator variable “media" takes value one if the mutual fund misconduct
has media mentioning in the headlines. I restrict the news source to the major business media in
the U.S. (i.e. Yahoo Finance, The Wall Street Journal, and The Dow Jones Newswire etc.)

In the study, the main dependent variable of interest is fund ﬂows as a proxy for residual claim
redemption. I deﬁned fund ﬂow following Sirri and Tufano (1998). For each fund class each month,
fund ﬂow is deﬁned as follows:

Flow f ,t =

T N A f ,t − T N A f ,t−1 × (1 + R f ,t)

T N A f ,t−1

(1.1)

where T N A f ,t is a fund’s total net asset at month t, and R f ,t is the fund’s return over the prior
month.

11All SEC enforcement actions can be found at he SEC website: Litigation and Administrative proceedings. The
format of judicial and administrative proceedings is variable; however there is no rigid formula dictating the choice of
forum. Each year, the SEC brings hundreds of civil enforcement actions against individuals and entities that violate
federal securities laws.

12The list of ﬁnancial conglomerates can be found in table 1.10 in Appendix.

9

From N-SAR ﬁlings, monthly fund gross inﬂows and gross outﬂows are directly reported in
Item 28. N-SAR ﬁlings also reports average fund TNA over the report period in Item 75. The
variable In f lowi,t (Out f lowi,t) is calculated by dividing the grossInﬂowi,t (grossOutﬂowi,t) by
T N Ai,t , and fund net ﬂows are calculated as follows:

Net f lowi,t = In f lowi,t − Out f lowi,t =

grossInﬂowi,t − grossOutﬂowi,t

T N Ai,t

(1.2)

Following the mutual fund literature, the following control variables are included in the regres-
sion analyses (see Chevalier and Ellison, 1997; Nanda et al., 2004; Cooper et al., 2005; Barber et al.,
2016). Size f ,t is the natural logarithm of TNA of each fund share class at month t. E xpenseRatio f ,t
is the funds’ expense ratio, which is the ratio of total investment that shareholders pay for the fund’s
operating expenses, which include 12b-1 fees. Age f ,t measures how long the funds have been
publicly traded, and I use the natural logarithm of fund age in the regression. Return f ,t−12,t−1
is the accumulated return of the fund in the previous year. Fund returns from N-SAR ﬁlings are
calculated using net asset value and distributions reported in N-SAR (Item 73 and Item 74) as
follows:

(1.3)

Returni,t =

N AVi,t + Payouti,t − N AVi,t−1

N AVi,t−1

To account for the common shocks to economic cyclicality, I include βt to capture time
ﬁxed eﬀects.
In accounting for potential preference over certain investment styles, I include
AvgSt yleFlowt as a control, where funds are grouped based on funds investment objectives. All
the continuous variables are winsorized at the 1% and 99% level.

1.2.2 Descriptive Statistics

Table 1.2 reports summary statistics of the characteristics for all equity funds from 2000 to 2016.
Panel A reports the summary statistics for the CRSP sample and Panel B reports the summary
statistics for the N-SAR Sample. Table 1.2 also reports control variables including past returns,
TNA, fund ﬂows, expense ratio, management fees, and fund age each month.

10

In the N-SAR sample, on average, funds experience monthly inﬂows of 5.1% and average fund
outﬂows is 4.2%. At some point in time, 9.4% of funds have regulatory actions taken against them.
In the CRSP sample, about 31.4% of observations are institutional share classes and 67.6% are
retail share classes, and on average fund ﬂows are 0.3%. At some point in time, 15.4% of funds
have regulatory actions taken against them, and 12.3% in the sample are associated with ﬁnancial
conglomerates, including both name association and business association.

Institutional and retail share classes have comparable sizes in terms of dollar value and both
have a highly skewed distribution. The mean TNA is around $1,060 million, while the median
TNA is $171 million. The annualized fund return is about 6.3% for the overall sample, with a
slightly higher return of 7.0% observed within the institutional share classes versus 5.9% for the
retail share classes. Expense ratios for retail share classes is much higher, equaling 1.3%, while
institutional share classes have an expense ratio of around 0.7%.

Figure 1.1 plots the overall number of regulatory disclosures from Form ADV as well as the
overall number of regulatory disclosures for the mutual fund principal product. The seasonality in
the ﬁgures is consistent with ﬁnancial crisis-related fraud, as well as the late-trading scandal that
relates to mutual fund regulatory events. Mutual fund-related regulatory actions are heightened
from year 2003 to 2005; other regulatory actions (such as asset-backed securities related allegations)
are clustered around the recent ﬁnancial crisis.

1.3 Empirical Design and Results

1.3.1 Empirical Design

The baseline speciﬁcation to test mutual fund investors’ response to regulatory disclosure is as
follows.

Flow f ,t+s = β1MF Fraud SEC f ,t + β2Past Fraud f ,t + γ f ,tControls f ,t + βt +  f ,t

(1.4)

11

Where Flow f ,j,t+s is the fund ﬂow deﬁned in (equation 1.1 and equation 1.2) of fund f (fund
share class f ) from t to t + s, where s equals 1 and 3 windows13. Using N-SAR ﬁling data, I
examine the eﬀects of In f low f ,j,t+s and Out f low f ,j,t+s separately. PastFraud f ,t is an indicator
variable accounting for regulatory events in the recent past one (ﬁve) year(s) against an investment
management company. Controls include fund size, expense ratio, age, past return and average fund
ﬂows within the same investment objectives deﬁned in the data section. Time ﬁxed eﬀects are
included in all regressions. Standard errors are clustered at the mutual fund family level to account
for serial correlation within a fund family.14

Using both inﬂows and outﬂows, separately, from N-SAR ﬁlings, allows for the examination
of the diﬀerent aspects of investors’ reactions. At the same time, using the share class-level data
from CRSP facilitates the comparison of the two distinct investor groups, institutional investors
and retail investors. Existing evidence suggests a high likelihood that these two groups of investors
face diﬀerent information and monitoring costs. To examine the market discipline outcome among
diﬀerent investor types, the ﬁrst variations comes from separating retail investors from institutional
investors. In additional to controlling for investors’ ﬁnancial literacy, the eﬀects between separate
sales channels are further explored.

1.3.2 Fund Flow and Regulatory Disclosure

Figure 1.2 presents the investment style-adjusted fund ﬂows (inﬂows and outﬂows) six months
before and after regulatory disclosures about mutual fund fraud. A signiﬁcant drop in net ﬂows
and inﬂows is observed in the month after regulatory disclosures, whereas outﬂows increase in the
month of disclosures and is even anticipated one month before. Table 1.11 presents consistent t-test
results that investment style-adjusted net ﬂows and inﬂows are signiﬁcantly 0.67% (0.89%) lower
after regulatory disclosures.

To control for other identiﬁed factors that aﬀect fund ﬂows, I use speciﬁcation (1.4) to test
the change in fund ﬂows following regulatory disclosures about investment advisers’ misconduct.

13In unreported analyses, most results are robust when extending the window to 6 months and 12 months.
14see Gaspar et al. (2006) and Nanda et al. (2004).

12

Panel A of Table 1.3 uses N-SAR data to show the fund ﬂows (inﬂows, outﬂows) from month t to
t+1 and t+3, respectively. On average, one and three months after regulatory disclosure, fund ﬂows
decrease by 1% and 1.9%, respectively. In other words, the month following a regulatory disclosure,
10.8% of a standard deviation decrease is observed in fund ﬂows. Moreover, outﬂows are more
pronounced immediately after regulatory events, and reverse overtime (captured by the negative
coeﬃcient of variable Past Fraud in columns 5 and 6). Alternatively, negative future inﬂows have
a long-lasting eﬀect. One drawback using monthly ﬂow data is the diﬃculty in pinpointing the
exact time of response when regulatory disclosures occur in the middle of the month. The eﬀect
could be underestimated using speciﬁcation 1.4. Nevertheless, together these results suggest that
investors show an immediate response to negative regulatory signals.

Panel B of Table 1.3 reports very similar results using the CRSP sample. Columns (1) - (2)
in Panel B of Table 1.3 report all equity fund share classes. Columns (3) to (4) report the fund
ﬂows for institutional shares classes, and columns (5) to (6) report those of the retail counterparts.
Importantly, the results show that the eﬀect of regulatory disclosures on fund ﬂows is twice as strong
among institutional share classes. For example, in the month following regulatory disclosure,
institutional share classes experience on average a 0.6% negative ﬂow, which is equivalent to
13.7% of a standard deviation decrease. On the contrary, retail share classes only experience a
0.3% negative ﬂow in the month following regulatory disclosure, equivalent to 8.2% of a standard
deviation decrease. Also, there is a signiﬁcant delay in retail investors’ response, indicated by the
signiﬁcant negative coeﬃcient of the variable Past Fraud (5-year).

Heterogeneity between retail investors and institutional investors has been documented in pre-
vious studies. Investors’ asset allocation decisions are largely aﬀected by factors beyond rational
expectations. Individual investors are net buyers of attention grabbing stocks (Barber and Odean,
2008), and salience aﬀects their investment decisions signiﬁcantly (Cooper et al., 2001; Rashes,
2001; Cooper et al., 2005; Sialm and Tham, 2015). On the contrary, professional investors respond
to proper performance signals and act (Del Guercio and Tkac, 2002). My results further conﬁrm
that institutional investors are better informed about regulatory disclosures and market discipline is

13

a much more eﬀective force levied at the hands of better informed investors.

Investor sophistication contributes to the discrepancy in reactions between institutional and
retail investors. There are two channels through which individuals can buy and sell funds, either
through a full-service broker or discount broker. With discount brokers, investors can place buy
and sell orders at a reduced commission. Full-service brokers provide retail investors professional
investment advices that help mitigate investment sophistication discrepancy, which are not available
through discount brokers. Following ICI criteria, the direct-sold funds are deﬁned as funds without
front or rear loads, and the 12b-1 fee being less than 25 basis points.15

In Panel A of Table 1.4, I ﬁnd that negative fund ﬂows are only signiﬁcant within broker-sold
funds, and direct-sold funds show lack of ﬂow discipline. Previous research suggests that individual
investors do not always make asset allocation decisions rationally, and their trading activity can
negatively impact their wealth (i.e. Barber and Odean, 2000; Cooper et al., 2005). My results
suggest that with professional advice, the sub-optimal behavior of retail investors can be mitigated,
at least partially. These results support the ﬁndings in Bergstresser et al. (2009), that brokers
can provide intangible beneﬁts to investors even if there is virtually no diﬀerence in fund returns
between broker-sold and direct-sold retail funds.

1.3.3 Management Fees and Regulatory Disclosures

Investors could also discipline fund managers by negotiating lower fees.

Investment advisers compete on both the size of assets under management as well as management
fees.
In addition to
fund ﬂows, it is also important to understand the impact of regulatory disclosure on mutual fund
management fees. Similar to fund ﬂow analyses, the speciﬁcation below is used to analyze the
eﬀect of regulatory disclosures on future management fees paid by investors.

∆Mgmt Feet,j,t+s = β1MF Fraud SEC f ,j,t + γ f ,j,tControls f ,j,t + βt + βj +  f ,j,t

(1.5)

In Panel B of Table 1.4, I ﬁnd a signiﬁcant reduction in fund management fees of 2% to 5%
within two years after a regulatory disclosure, with the only exception being direct-sold retail

15Also see Sun (2014).

14

funds. Among most investor classes, mutual fund misconduct is quite costly. Over the following
six months after a regulatory action, funds experience cumulative -3.4% net ﬂows. Based on
the average fund size of $2,362 million, mutual funds lose up to $80 million of assets under
management. Moreover, fund advisers have to lower management fees to prevent further negative
ﬂows; on average, each regulatory disclosure can reduce the investment advisers’ proﬁt by about
3.8%. Under necessary conditions, market discipline is rather eﬀective. However, the eﬀectiveness
of market discipline varies among investor bases.

The weaker reaction from retail investors, especially among direct-sold funds, is noteworthy.
Theory suggests that market discipline prevails only if (a) information is publicly available with no
(very low) cost and (b) incentives to monitor are suﬃciently high and cost of actions are suﬃciently
low.
In the next section, I explore the possible mechanisms that contributes to the asymmetric
market discipline outcome.

1.4 Eﬀective Information Costs and Regulatory Disclosure

Heterogeneity in information costs beneﬁts informed traders at the cost of the uninformed
(Grossman, 1976). Regulatory disclosures are public signals, however, investor sophistication
can aﬀect their ability to eﬃciently process information (Barber and Odean, 2000). As a result,
the eﬀective cost of information increases which ultimately inﬂuence their ability to exert market
discipline. In this section, I show that investor sophistication together within information quality
can hinder the eﬃcacy of market discipline.

1.4.1 Visibility

Previous studies also showed that fund ﬂows are aﬀected by fund family size, media attention
and past performance (see Chevalier and Ellison, 1997; Sirri and Tufano, 1998; Berk and Green,
2004). A nudge, in behavior ﬁnance, inﬂuences investors’ decision making, making it more likely
that investor will make a particular choice with the desired outcome (Thaler and Sunstein, 2008).
Investors’ inertia can, therefore, be oﬀset by increasing visibility, serving as information nudge.

15

To further examine its impact on market discipline, I separate the funds into groups with diﬀerent
search costs based on fund family visibility.

First, the funds are ranked based on the size of the families to which the funds belong, where
fund family size is measured by the total net assets under management within the family. Large-
fund groups represent funds with a family size that is above the median. Large-fund families are
expected to have better marketing and distribution outcomes, as well as generating media attention,
which results in higher investor familiarity. Information costs are much lower among large funds
in comparison to that of small funds.

Table 1.5 provides results to support this hypothesis. Panel A of Table 1.5 ﬁnds a signiﬁcant
negative ﬂow reaction to regulatory disclosures among the funds belonging to large-fund families.
Column (2) and (4) in Panel A shows that large-fund families experience 0.5% and 1.1% negative
fund ﬂows, respectively, one and three months following the regulatory disclosures. Column (1) and
(3) shows virtually no ﬂow discipline among funds from small families during the same duration.
In unreported analysis, the institutional funds are separated from retail funds. I ﬁnd that the large
discrepancy in fund ﬂows reaction come mainly from retail share classes. Both large and small
institution funds experience negative fund ﬂows following regulatory disclosures (columns 1 to 4);
whereas the negative fund ﬂows are only observed among large retail shares (columns 5 to 8).

On the other hand, funds with better past performance exhibit the so-called “star" phenomenon.
These “winner" funds are often subject to better exposure, which generates spillover eﬀects to
other funds within the same fund family. To identify a star family, the procedure in Nanda et al.
(2004) is followed. First, the Fama-French three-factor adjusted return is calculated for each fund
each month; then the risk-adjusted returns are ranked into percentile, and the funds are ﬂagged that
belong to the top 5th percentile as a star fund. For each month, I name a fund family as “star
family" if at month t any fund belonging to the fund family is identiﬁed as a star fund.

In Panel B of Table 1.5, the negative fund ﬂows are found to be more signiﬁcant among the funds
from a star family, both statistically and economically. Column (4) shows that funds from a star
family experience 1.2% negative ﬂows after regulatory disclosures, whereas funds from non-star

16

families experience only one-third of the eﬀect (0.4%).

1.4.2 Severity

Investors’ inattention aﬀects investments performance (DellaVigna and Pollet, 2009; Dahlquist and
Martinez, 2015; Sialm et al., 2015). Investors’ inertia can be oﬀset by external mechanisms that
reduce eﬀective search costs. For example, severe events are more attention grabbing; investors and
update their information set through learning from the past. In addition, previous results suggest
that media, compare to public disclosure platform, a direct way to attract investors’ attention is
through media coverage of mutual fund misconduct events.

In Table 1.6, I examine the possible explanation for heterogeneous information friction. First,
If retail investors face an
I examine whether investors learn about misconduct from the past.
information disadvantage, multiple oﬀenses should oﬀer learning opportunity for investors; thus
the search costs for the repeated oﬀense should be substantially lower. The results in Table 1.6
are consistent with this hypothesis. For retail investors, the response to the repeated oﬀense is
about -0.6% (0.2% + 0.4%), which is comparable to the fund ﬂow responses from their institutional
counterpart. These results suggest that learning is one way to mitigate information friction. And
once the eﬀective information costs are reduced, retail investors behave very similarly to institutional
investors in disciplining ﬁnancial misconduct.

Investor’s attention are also likely to be correlated with the severity of an event. Retail investors,
in particular, have limited time and resources to obtain ﬁnancial information. Severe events are
more likely to attract their attention. For example, regulatory actions can be initiated by the SEC
and other regulators. The median monetary ﬁne for SEC enforcement is $8 million dollars, whereas
the median monetary ﬁnd for enforcement by other regulators is about $400,000. In Table 1.6, I
ﬁnd that retail investors are less likely to respond to less severe fraud disclosures, measured both
by regulator and monetary ﬁne. For example, negative fund ﬂow responses are about 0.6% from
retail investors if the monetary ﬁne is above median. Also, more severe or high proﬁle mutual
fund fraud are likely to be covered by major media outlets, further reducing information friction.

17

Indeed, the marginal eﬀect on fund ﬂows from media coverage is about -0.3%, both economically
and statistically signiﬁcant.

1.4.3 Relevancy

Less sophisticated investors also have inferior ability to process information. Therefore information
quality also aﬀects investment decision signiﬁcantly, especially for the investors with lower ﬁnancial
sophistication. The ability to eﬀectively extract relevant information from noisy signals is crucial
for investment decision making. Market discipline breaks down if investors under-react to relevant
information or over-react to irrelevant information. Two types of ﬁnancial misconduct are consid-
ered to be irrelevant to mutual funds. First, the group of funds whose investment management
companies are aﬃliated with large ﬁnancial institutions is analyzed.16 The aﬃliation is identiﬁed
through two non-mutually exclusive channels. The funds can be associated with a ﬁnancial con-
glomerate when they share common elements in their names; funds and ﬁnancial institutions can
be aﬃliated if mutual fund management companies are registered subsidiaries of large ﬁnancial
institutions. When the SEC initiates regulatory actions against the ﬁnancial conglomerates that
mutual funds are associated with, rational investors should show no response if such misconduct
has no impact on the value of mutual fund assets.

The results in Table 1.7 indicate divergent responses. In Panel A, results show that institutional
investors are able to detect the fallacy; we observe no fund ﬂow reaction following regulatory
disclosure against the ﬁnancial institution. On the contrary, retail investors react as if such regulatory
actions are indicative of investment adviser fraud and punish the management companies. More
interestingly, the magnitude, 0.4% negative ﬂows, is even stronger than the actual mutual fund
misconduct. One explanation for this result is that the enforcement actions against large ﬁnancial
institutions are much more visible via media coverage to the general population. The results convey
that individual investors cannot eﬀectively distinguish noise from signals, whereas institutional
investors are less aﬀected by information noise.

16Table 1.10 provide the list of large ﬁnancial institutions based on their assets, I only include the top 40 largest

ﬁnancial institution in this study.

18

Second, the group of funds with misconduct not related to the mutual fund itself is analyzed.
Form ADV provides information about the speciﬁc product to which the regulatory charges are
related. Mutual fund investors should be concerned about misconduct related to “Mutual Fund(s)."
In Panel B of Table 1.7, fund ﬂows following all other types of regulatory disclosures initiated by
the SEC are examined. Institutional investors show no response to such disclosures; whereas we
observe moderate levels, or 0.4% negative ﬂows over three months from retail investors. It suggests
irrational behavior by retail investors in a noisy environment. These results are noteworthy as we
have the tendency to equate information quantity to information quality. Transparency matters. The
manner in which information is communicated is also important. Improving information quality
so that disclosures help safeguard investors and reduce information noise deserves more attention.

1.5 Monitoring, Investor Mobility and Regulatory Disclosure

Whether market discipline is eﬀective also depends on investor’s incentive to monitor. Monitor-
ing costs are high for retail investors, which would reduce their incentive to exert eﬀort. However,
retail investors can beneﬁt from the ability and willingness of institutional investors to exercise
market governance through reduced agency problems from greater institutional monitoring (Evans
and Fahlenbrach, 2012). Also, some of the retail investors are active traders (Barber and Odean,
2000), compared to others such as retirement funds investors. Among funds where investors closely
monitor fund performance, we should observe a strong disciplining eﬀect.

1.5.1 Monitoring

First, I split the retail funds into two groups to proxy for investor monitoring, retail funds with an
institutional fund twin and the rest. Retail investors can beneﬁt from monitoring by institutional
twin fund, and follow the investment strategy by their twin. In Panel A of Table 1.8, I ﬁnd that
the retail funds show much stronger negative fund ﬂow responses when they have an institutional
twin fund. The negative fund ﬂows are 0.5% for the month following the regulatory disclosures
for the retail funds with an institutional twin, compared to 0.2% without. The results suggest that

19

investor’s incentive to monitor has signiﬁcantly impact on the eﬀectiveness of market discipline.
In Panel B of Table 1.8, I use volatility of abnormal outﬂows as a proxy for the intensity of investor
monitoring. When funds are under high scrutiny, regulatory disclosure results in much stronger
negative fund ﬂows (1.92%). The fund ﬂow responses are insigniﬁcant (-0.26%) among the funds
with low level of monitoring intensity.

These ﬁndings suggest that professional investors monitor investments more closely than retail
investors and react more strongly to negative signals. However, when retail investors who actively
monitor funds, or have reduced cost to monitor through institutional twin funds, the fund ﬂow
responses become much more sensitive to regulatory disclosures.

1.5.2

Investor Mobility

Hubbard et al. (2010) argue that the mutual fund industry is price competitive only if investors can
easily “ﬁre" their investment advisers. Investor mobility facilitates competition in the mutual fund
industry. Factors such as capital gains taxes, sales loads and outside investment options can aﬀect
investors’ ability to switch funds. Market discipline is eﬀective only when capital can be moved
freely into and out of funds (Dangl et al., 2008). In this section of the paper, I further link how
changes in investor mobility aﬀect the eﬃcacy of market discipline.

1.5.2.1 Capital Gains Overhang

Barclay et al. (1998) show that capital gains overhang increases expected future taxable distributions,
and hence the present value of new investors’ tax liabilities. Ivkovic et al. (2005) also ﬁnd supportive
evidence that capital gains overhang aﬀects investors’ trading behavior. Unrealized capital gains
aﬀect investor mobility; particularly for taxable investments, high levels of unrealized returns
increase the investor’s tax liability when redeeming assets. Next, I explore how the level of capital
gains overhang can aﬀect investors’ ability to discipline funds. Following Barclay et al. (1998),
capital gains overhang is calculated for each fund each month as in 1.6. Then, each month funds
are ranked into terciles based on the level of capital gains overhang to examine the eﬀect of ﬂows

20

following regulatory disclosure.

Overhangt = Overhangt−1 + (N AVt − N AVt−1) × Sharest−1

+ (Sharest − Sharest−1) × (N AVt − Acg Price Paid for New Shares)

(1.6)

The results are presented in Panel A of Table 1.9. Outﬂows after regulatory disclosures increase
signiﬁcantly only for the funds with low level of capital gains overhang. High levels of overhang
deter future inﬂows, that the reduced inﬂow is more pronounced within the group of funds with
higher level of capital gains overhang. The results suggest that the level of unrealized capital
gains indeed aﬀect investors’ disciplinary power, however, through diﬀerent channels. Disciplining
capacity of existing investors and future investors are signiﬁcantly aﬀected by the tax liabilities they
face. Whether there exist conﬂict of interest between fund management companies, existing and
future investors, and the related potential wealth transfer implications needs to be further explored.

1.5.2.2

Investment Options

Capital reallocation is associated with a considerable amount of transaction costs, both direct
(loads) and indirect (outside investment options). How indirect costs (opportunity costs associated
with the search for alternative investments) aﬀect the eﬃcacy of market discipline are examined
in this section. When outside investment options are limited, investors have lower incentives to
react. Therefore, the sample is split into diﬀerent time periods based on outside investment options’
availability.

More speciﬁcally, the late-trading scandal period is examined, as well as during the ﬁnancial
crisis. The ﬁnancial crisis has had a major impact on investors’ perceptions about the market
environment as well as conﬁdence in the asset management industry. Overall, a loss of conﬁdence
in the asset management industry results in increased search costs and reduced investor mobility.
On the other hand, the late-trading scandal provides a strong signal for investors to ﬁlter investment
opportunities and results in lower monitoring costs.

In Panel B Table 1.9, I ﬁnd that fund ﬂows are statistically insigniﬁcant from zero following
In contrast, during the late-trading scandal

regulatory disclosures during ﬁnancial crisis era.

21

period, it required much less eﬀort to distinguish good funds from bad funds. I ﬁnd that negative
fund ﬂows are 50% stronger (1.5% decreasing in ﬂows compared for 0.5% of the average eﬀect)
during late-trading periods. More interestingly, the increased outﬂow is only signiﬁcantly when
outside options are cheap. The results suggest that elevated level of search costs for alternative
investment options can signiﬁcantly aﬀect market discipline negatively.

1.5.2.3 Sales Loads

Sales loads increases transaction costs of asset redemption as well. In Panel A of Table 1.4, I show
that broker-sold funds are more sensitive to regulatory disclosure, and the beneﬁt from professional
investment advice seems to outweigh the sales costs. In Table 1.12, I further investigate inﬂows
and outﬂows separately to understand the role brokers play.

The results in Table 1.12 indicate that the net ﬂow of broker-sold funds is predominantly
derived from increasing outﬂows. Compared to direct-sold funds, broker-sold funds with either
front sales loads or rear sales loads both experience high levels of outﬂows immediately following
regulatory disclosures; decreasing inﬂows have a smaller magnitude. These results suggest that
the beneﬁts investors gain from professional investment advice (active investing) can outweigh the
direct transaction costs. This can be one of the important intangible beneﬁts the broker can provide
to retail investors, mitigating the high cost of inattention.

Overall, evidence convey that restricted investor mobility hinders the eﬃcacy of market disci-
pline. More speciﬁcally, tax liabilities and search costs of alternative investment options are the
primary factors aﬀect investors’ ability of asset reallocation. Transaction costs paid by the investors
are somewhat oﬀset by the reduced information friction through the investment advice provided by
brokers.

1.6 Discussion and Conclusion

Enforcement mechanisms are used to deter future violations to protect the public interest. The
eﬃcacy of enforcement is determined by the costs and beneﬁts from such actions. In particular,

22

mutual fund enforcement deserves special attention due to the unique governance structure and lack
of authoritarian role of regulators in the mutual fund market. Regulatory enforcement in mutual
funds is limited to facilitating transparency, whereas market enforcement complements to a large
extent through asset redemption.

By using public regulatory disclosures against mutual fund investment advisers, I ﬁnd that as
theory predicted, when information costs are suﬃciently low and investor mobility remains unhin-
dered, market discipline serves as an eﬀective governance mechanism in mutual funds. However,
heterogeneity among investors exists in their ability to respond to enforcement. Among possi-
ble channels, ﬁnancial sophistication, information costs and investor mobility have a considerable
impact on investors’ disciplinary power. Both search costs and information quality contribute
signiﬁcantly to investors’ ability to discipline investment advisers, in the presence of heteroge-
neous investor sophistication. Also debatable is whether mandatory information disclosure itself
is suﬃcient to guarantee low information costs. Ultimately, regulatory disclosures should aim at
improving information quality that better communicates with less informed investors. I also ﬁnd
that limited investor mobility can prevent the necessary disciplinary action from investors. When
outside options are expensive, or when faced with high levels of penalties to transfer assets (e.g.,
tax liabilities), the market’s disciplining mechanism is less eﬀective.

This paper highlights the importance of this unintended outcome of segmentation in market
discipline. The asymmetric investor reactions could aﬀect the managers’ incentives, ex ante,
which could results in wealth transfer at the cost of certain types of investors. Moreover, investor
heterogeneity exists beyond the scope of mutual funds, that shareholder characteristics could aﬀect
the corporate governance similarly. My results suggest that future research could be focus on how
shareholder heterogeneity and composition could aﬀect external corporate governance outcomes.

23

CHAPTER 2

CORPORATE CULTURE AND CORPORATE FRAUD

2.1 Introduction

Managers often assert that having an appropriate culture is critical to a ﬁrm’s success. For
example, in a recent survey of CEOs and CFOs by Graham et al. (2015), 91% of respondents
said that they thought that culture was “Important” or “Very Important,” and 78% think that it is
a top 5 value driver for their ﬁrm. One particular dimension of culture that has received much
attention from regulators, auditors, and academics, is the degree of integrity in a ﬁrm’s culture and
its relationship with corporate fraud.1 Yet, there is relatively little quantitative research that tests
whether an unethical culture actually predicts corporate fraud. This is perhaps because culture, by
its very nature, is diﬃcult to deﬁne and measure. Most research relies on surveys and interviews
to measure culture, which provide important insights into some facets of culture. However, self-
reporting biases may be particularly severe for surveys of integrity: for example, employees of ﬁrms
with questionable ethics are perhaps more likely to lie or embellish their responses to surveys.

In this paper, we move beyond self-reported measures to study a measure of integrity based on
individual employee actions: the decision to register for, and use, AshleyMadison.com (“AM”), a
website that facilitates extramarital aﬀairs.2 We assign AM users to ﬁrms based on the domain
name taken from their email accounts, resulting in a sample of approximately 47,000 individuals
who used their corporate email account to register and actively use an AM account over the 2002-

1For example, in a speech to members of the ﬁnancial services industry on October 20, 2014, William Dudley,
President and Chief Executive of the Federal Reserve Bank of New York says: “Supervisors simply do not have
suﬃcient “boots on the ground” to ferret out all forms of bad behavior within a giant, global, ﬁnancial institution.
Moreover, regardless what supervisors want to do, a good culture cannot simply be mandated by regulation or imposed
by supervision...It is up to you to address this cultural and ethical challenge.

2We use anonymized data on individual users and do not conduct any analysis at the user level. Furthermore, we
do not disclose in any way the names of corporations with employee email accounts in the database. We have received
exemption from Institutional Review Board and approval by the universities with which we are associated because
of the anonymization process, public availability of the data, and the aggregate nature of the measures that enter our
analysis.

24

2014 period. Our key variable of interest is the number of active users at any point in time in a
given ﬁrm, where active means the user has not only registered, but also exhibited some activity in
the account (e.g. purchased credits to send a message).

We hypothesize that AM membership reﬂects a ﬁrm’s emphasis on integrity. Because a ﬁrm
is more likely to attract, select, and retain employees who match its culture (Schneider, 1987), we
expect that individual employee traits provide information about corporate priorities. Firms that do
not emphasize integrity in their cultures are more likely to employ individuals who display a lack
of integrity. Overall, AM membership reﬂects both a focus on integrity at an individual employee
level, and the focus on integrity in the systems and policies in place at ﬁrms.

We test whether AM membership predicts future corporate fraud. In particular, we test whether
ﬁrms with greater AM membership are more likely to be subject to SEC enforcement actions
due to accounting misstatements. Dechow et al. (2011) ﬁnd that a host of ﬁnancial variables
predict SEC enforcement actions. We ﬁnd that after controlling for all these variables, greater AM
membership predicts a greater likelihood of future enforcement actions. Moreover, these results
are economically signiﬁcant: a one standard deviation increase in AM membership is associated
with a 0.104 percentage point increase in the rate of detected fraud, which is more than double the
unconditional mean.

Despite its advantages, AM membership is by no means a perfect measure of corporate ethics.
One potential concern is that we can only observe the fraction of employees who use their oﬃcial
email account to register for AM. Besides perceptions of systems in place, there could be other
reasons to use an oﬃcial email account that are likely to make AM membership a noisy measure
of corporate ethics.3 Another related concern is that the number of AM users that we observe
constitutes a small fraction of the ﬁrm’s workforce. Around 50% of ﬁrms have zero AM membership
and the mean AM membership is 5.4 employees for the ﬁrms that have at least one AM member.
While the small fraction of AM membership likely adds noise to the measure, it should also bias
against ﬁnding results.

3For example, employees could be unfamiliar with the perils of electronic communications, or may believe that

their spouse does not have access to oﬃcial email identiﬁcation.

25

Nevertheless, we run a battery of tests to ensure that the relation between AM membership and
corporate fraud is robust. In addition to our variable on the number of AM employees, which could
be a noisy measure of culture, we also construct a dummy variable equal to one if a ﬁrm has at
least one AM account for at least two consecutive years. We ﬁnd that this dummy variable also
predicts future misstatements after controlling for standard determinants. Another concern is that
there may be heterogeneity across industries or geographies in AM usage. All our results hold after
including industry and geography ﬁxed eﬀects. Additionally, there may be non-linearities in fraud
with respect to ﬁrm size. We therefore match ﬁrms on size (number of employees) and ﬁnd that
AM ﬁrms are 75% more likely to have accounting misstatements than size-matched non-AM ﬁrms.
Finally, it is possible that these results may be sensitive to the choice of accounting misstatements
as a measure of corporate misconduct. We therefore test whether AM membership predicts an
alternative measure of corporate ethics: ratings of ﬁrms on ethical issues by external analysts
at KLD. We ﬁnd that a one standard deviation increase in AM membership is associated with a
2.65 percentage point increase in analysts perception of signiﬁcant concerns regarding bribery and
corruption. This eﬀect is economically quite large, since the unconditional average is only 4.7%,
and the average for ﬁrms with no membership is 3.39%. We also ﬁnd that AM ﬁrms are more likely
to be involved in tax avoidance via use of tax havens, pay lower taxes than similar companies, and
are more likely to be rated as having tax-related concerns by KLD analysts.

Our results provide insight that, at a minimum, AM membership captures an important source
of unobserved heterogeneity across ﬁrms, which predicts substantive ﬁrm-level outcomes. After
controlling for standard variables, AM membership has incremental predictive power for future
accounting misstatements and external analyst perceptions of unethical behavior. These results are
consistent with the hypothesis that ﬁrm culture and ethical behavior are closely linked.

Our paper is speciﬁcally related to research that attempts to quantify corporate culture. Kim
et al. (2012) use analyst ratings to examine whether socially responsible ﬁrms are also responsible
along various dimensions of ﬁnancial reporting. Popadak (2013) measures culture based on a
textual analysis of employee reviews of ﬁrms from career intelligence websites, and ﬁnds that

26

stronger shareholder governance causes ﬁrms to focus on observables and neglect intangibles such
as collaboration and integrity. Guiso et al. (2015) and Garrett et al. (2014) measure integrity using
surveys that ask employees whether they believe that senior managers in their ﬁrms are ethical. We
also focus on integrity, but our measure is akin to a revealed preference. Rather than surveying
employees, we infer the importance of integrity in a ﬁrm’s culture using the actions of a subset of
the ﬁrm’s employees.

Moreover, our results are related to prior research that examines the eﬀect of CEO personality
on ﬁrm outcomes (e.g., Jia et al. (2014), Schrand and Zechman (2012), and Gormley et al. (2013)).
In particular, recent work by Mironov (2015), Cline et al. (2016), and Griﬃn et al. (2016), shows
that a CEO’s personal indiscretions and corrupt behavior are associated with ﬁrm level corruption,
ethical violations, and class action lawsuits. While we also document a strong association between
personal and professional ethics, our analysis is broader in the sense that it includes employees
from all levels of a ﬁrm, not only upper management. This is consistent with anecdotal evidence
that suggests that “rank and ﬁle” employees, rather than top management, were responsible for
unethical corporate behavior in a number of recent corporate scandals.4 Moreover, the choice of
CEO is endogenous with respect to ﬁrm culture; we ﬁnd that ﬁrms with lax cultures are more likely
to choose internal CEOs relative to ﬁrms with more ethical cultures, thereby perpetuating their
current culture.

2.2 Data

2.2.1 The AshleyMadison Data

AshleyMadison.com is a dating website for people who are married or in a committed relationship.
The website was created in 2002 and quickly became the world’s largest online social networking

4For example, AIG’s Joseph Cassano and Drexel Burnham, and Lambert’s Dennis Levine (both employees well
below the level of corporate executive), each played a large role in their ﬁrm’s troubles during the ﬁnancial crises of
2008 and the late 1980’s, respectively. Similarly, it appears that engineers, and not top executives, at Volkswagen
installed software intended to mislead emissions testing. While it is likely that Martin Winterkorn (the CEO) played a
role in determining the culture, it was the ethics of “rank and ﬁle” employees that led to scandal, and ultimately a large
loss in shareholder wealth.

27

community for people who wish to engage in extramarital aﬀairs.5 While signing up on Ashely-
Madison is free, users must purchase credits to send custom messages, initiate chat sessions, send
priority messages, or send virtual gifts. By late August 2015, information for the majority of AM
accounts became available on a variety of websites and received a great deal of media attention.6
Many of the accounts on AM were registered using corporate email addresses. We link these
email accounts to ﬁrms using corporate domains identiﬁed from the WebURL from Compustat
and LexisNexis corporate aﬃliations. We merge the matched list to the Compustat database using
ticker symbol and company name. We then hand-check each domain-company link to verify its
validity. We exclude certain domains that are likely being used by people who are not employed
at the ﬁrm to which the domain belongs. For example, we exclude domains such as “yahoo.com,"
“facebook.com," “aol.com," “google.com," and “verizon.com." After applying these ﬁlters, our
ﬁnal sample includes 12,687 company domains in the Compustat database from 2002-2014. Using
these domains, we are able to match 46,649 employees to companies who used the corporate domain
name with which they created an AM account from 3,469 diﬀerent companies. We do not in any
way disclose the names of individuals or corporations that have accounts in our dataset.

For each account we observe the date that the account was created, the age of the user, the
gender of the user, the city (zip-code) in which the account was created, the ﬁrst date that an email
or message was sent, the last date that an email or message was sent, and whether the account
user purchased any credits. For the majority of our analysis we restrict our focus to accounts that
exhibited some level of activity (e.g., a custom message was sent, a chat session was initiated, or
credits were purchased for the account). This excludes “phantom" accounts that were created by
mistake, as a practical joke, or by someone who immediately appears to have had second thoughts
about their actions.7 Furthermore, since we can only observe the dates for the ﬁrst and last email,

5http://www.prnewswire.com/news-releases/hollywood-courts-toronto-based-ashley-madison-75587257.html

"Hollywood Courts Toronto-based AshleyMadison." www.prnewswire.com. Retrieved 2015-10-24.

6For example, on August 19, 2015 the Washington Post published that thousands of accounts were linked to the U.S.
military and the U.S. government. Inside Higher Ed reported that more than 74,000 accounts at AshleyMadison.com
were from universities and colleges with ‘.edu’ email accounts.

7In unreported results, we relax this restriction to include possible “phantom" accounts and the results are largely

unchanged.

28

or message, we assume that an account is active in the intermediate time between its inception and
the last observed activity. We deﬁne the variable activeaccountj,t as a binary variable equal to
unity for the years in which an account is active according to our deﬁnition, and zero otherwise.

We create our primary variable, active AM accountsi,t, by summing the number of accounts
with a corporate domain name that belongs to ﬁrm i and that have exhibited some level of activity
on or before time t:

1[domain(activeaccountj,t) = corpdomaini] + 1)

activeAMaccountsi,t = ln( t

N

τ=0

j=1

We use the natural log of the number of active AM accounts as our main variable, and not the
ratio of AM accounts to the total number of employees at a ﬁrm, because the Compustat item, emp
(i.e., the number of employees), is only a noisy approximation.8 However, we control for the (log)
number of employees and (log) market capitalization in all our speciﬁcations.

Table 2.2, Panel A, reports basic descriptive statistics. The average number of AM accounts
per ﬁrm is 2, and conditional on having at least one account, the average number of accounts is 5.4.
Given the relatively small number of AM accounts per ﬁrm, it is possible that the number of AM
accounts is a noisy measure. In additional robustness tests, we use a dummy variable equal to one
if the number of active AM accounts is greater than zero for at least two consecutive years, and zero
otherwise as an alternative explanatory variable. The average age of an AM user is 39 year old. The
ratio of males to females is roughly two to one.9 In Table ??, we report industry and geographic
statistics for our sample. As Table ?? documents, AM membership services are used by high-tech
industries, while low-tech and defense industries are less frequented. In later analyses, we control
for heterogeneity across industries through geography and industry ﬁxed eﬀects transformations.

8The number of employees at a ﬁrm is not an audited number and ﬁrms strategically misreport employment
numbers (e.g., Beatty and Liao, 2012). As a result, there is not a standard way for ﬁrms to report this number (e.g.,
some ﬁrms report the average number of employees and some report the number at year-end). In addition, the emp item
typically includes part-time, seasonal, and foreign employees. Scaling by a number that includes foreign employees
could potentially bias our results, since our AM membership measure is composed of only domestic employees. Finally,
there are only a few AM accounts per ﬁrm, relative to the total number of employees at the ﬁrm. Taking the ratio would
result in a denominator that is several orders of magnitude larger than the numerator and that exhibits a large degree of
measurement error.

9In unreported results, we used only males or females, or the number of AM users along with the ratio of

males/females as an additional control. The results are qualitatively similar to those reported in the paper.

29

We report regression results for the determinants of AM membership in Table 2.7. The larger
(as measured by market capitalization or number of employees) and older ﬁrms have a higher rate
of AM users. Firms with higher AM membership rates also tend to be headquartered in areas with
large populations and relatively low incomes. For these reasons it is important that our results are
robust to controlling for various measures of ﬁrm size and regional heterogeneity in AM rates.

2.2.2 Other data

For data on corporate social responsibility, we use the MSCI KLD STATS from 2002-2014. KLD
data are detailed annual statistics of performance indicators developed by MSCI analysts who pro-
vide research for institutional investors. To create these performance indicators, MSCI analysts use
government databases, company disclosures, and macroeconomic data to assess company perfor-
mance with respect to meeting stakeholder needs regarding environmental, social, and governance
factors. Mattingly and Berman (2006) and Kim et al. (2012) suggest that the KLD data is well
suited for studying corporate social responsibility. Note that Kim et al. (2012) document a strong
association between KLD ratings and ﬁnancial reporting standards, which is reassuring for our
analysis since we use both variables as proxies for corporate ethics. For the purpose of our study,
we focus on the particular indicators we consider to be closely related to integrity, which is the
dimension of corporate culture we intend to study. The KLD indicators are broken down into
strength and weakness categories.

Our ﬁrst variable, Bribery and Fraud is a binary variable equal to unity if a ﬁrm has experienced
severe controversies related to bribery, tax evasion, insider trading, and accounting irregularities
in a given year, and zero otherwise. Similarly, Tax Disputes indicates whether a ﬁrm has had
major tax disputes within a given year. The variable Product Quality assesses how companies
manage their risk of facing major product recalls or losing customer trust through major product
quality concerns. Companies that score higher are those that proactively manage product quality by
achieving certiﬁcation to widely acceptable standards, undertaking extensive product testing, and
building processes to track raw materials or components. The variable Human Rights measures

30

the severity of controversies related to a history of involvement in human rights-related legal
cases; widespread or egregious complicity in killings, physical abuse, or violation of other rights;
resistance to improved practices; and criticism by NGOs or other third-party observers. Firms
that are guilty of worse human rights violations have negative scores. Lastly, proﬁt sharing
indicates whether a company has a cash proﬁt-sharing program through which they have recently
made distributions to a signiﬁcant proportion of their workforce. Note that the ﬁrst two variables
(Bribery and Fraud and Tax Disputes) are binary, and the other KLD variables are the sum of
binary sub-components and hence can take on values other than 0 or 1. All variables are deﬁned in
detail in Table 2.1.

Data on misstatements from 2002-2014 come from the AAER data set discussed in Dechow
et al. (2011). This dataset provides detailed information regarding SEC investigations of public
corporations for ﬁnancial misstatements and has been commonly used in accounting research
to study misreporting. Feng et al. (2011) study the AAER database and provide evidence that
pressure from CEO’s cause CFO’s to become involved in material accounting misstatements. In a
closely related study to ours, Garrett et al. (2014) use the AAER database to show that trust in top
management, measured at various employee ranks, strongly predicts ﬁnancial reporting quality.

Our tax havens data come from Dyreng and Lindsey (2009), who download every 10-K through
the SEC’s Edgar database between 1994 and 2014 and search every 10-K ﬁling (Exhibit 21) for
country names. Countries are identiﬁed as tax havens if they are deﬁned as such by by three of the
four following sources: (1) Organization for Economic Cooperation and Development (OECD),
(2) the U.S. Stop Tax Havens Abuse Act, (3) The International Monetary Fund (IMF), and (4) the
Tax Research Organization.

Firm accounting and ﬁnancial information come from Compustat from 2001-2014. We also
use stock price and return data from CRSP to calculate volatility measures and portfolio returns. A
full description of all variable deﬁnitions is provided in Table 2.1.

31

2.3 Corporate Fraud and Ethics

In this section we examine whether greater AM membership among the employees in a ﬁrm is
related to unethical behavior by the ﬁrm. We consider three sets of measures of unethical behavior:
SEC enforcement actions for accounting misstatements, KLD ratings on ethics-related variables,
and measures of tax avoidance. The ﬁrst two are clearly related to corporate ethics. Tax avoidance
is more nuanced. Managers have a duty to act in the interest of shareholders, and minimizing a
ﬁrm’s tax bill using legal strategies is consistent with this duty. However, if managers use extremely
aggressive measures such as tax havens, this suggests that they are willing to walk close to the line
of legality. We therefore include tax avoidance as an additional measure, but with the qualiﬁcation
that it is not as clear a measure of ethics as the two other measures we consider.

First, we follow related work by Schrand and Zechman (2012), Garrett et al. (2014), and Jia
et al. (2014) and use SEC enforcement actions due to misstatements as a measure of corporate
ethics. We follow Dechow et al. (2011) and examine three models. Model 1 (Speciﬁcations
1 and 2) includes AM-variables and ﬁnancial statement variables. Model 2 (Speciﬁcations 3
and 4) adds non-ﬁnancial statement and oﬀ-balance-sheet variables, and Model 3 (Speciﬁcations
5 and 6) incorporates market-based measures. As in Dechow et. al.
(2011), we use a logit
speciﬁcation. We augment the models in Dechow et al. (2011) by adding industry and year ﬁxed
eﬀects transformations. We report our results in Panel A of Table 2.3.

AM membership strongly predicts the probability of accounting misstatements, after controlling
for other potential determinants studied by Dechow et al. (2011). The unconditional probability of
misstatements in our sample is 0.67%. Increasing Active AM Accounts by one standard deviation
results in an increase in misstatement probability by 0.36-0.53 percentage points (54-79% of the
unconditional mean). The eﬀect is even more pronounced in speciﬁcations in which we use a
dummy variable (Dummy(AM > 0)) as our explanatory variable. Companies with at least one
AM account have a 0.84-1.06% higher probability of accounting misstatements (126-159% of the
unconditional mean). Note that these eﬀects hold after controlling for all the other determinants of
fraud studied by Dechow et al. (2011). The magnitudes on the control variables are similar to the

32

results in Dechow et. al. (2011).10

Speciﬁcations 5 and 6 also include the log number of employees as an additional control
besides the variables from Dechow et al. (2011). The AM variables retain signiﬁcance after
controlling for the number of employees. It is possible that non-linearities in size drive these results.
This explanation requires that AM membership is non-linear in ﬁrm size and the probability of
misstatements is non-linear in size as well (perhaps due to greater monitoring of the largest ﬁrms).11
To mitigate concerns that our results on the predictive ability of AM membership for accounting
misstatements are due to non-linearities in size, we use a matching approach. The ‘treatment’
sample consists of ﬁrms with at least one AM account. For controls, we match ﬁrms in the same
year, and same Fama-French 48 industry with the closest number of employees. We impose the
restriction that the number of employees must be within 10% of the ﬁrm with AM membership
to be included in our sample.12 It is important to note that the matching criteria is quite stringent
and results in a relatively high loss of observations. In the matched sample, ﬁrms with a positive
number of AM accounts are 1.19% likely to misstate ﬁnancials, whereas matched ﬁrms without AM
membership are only 0.68% likely to misstate ﬁnancials. The diﬀerence of 0.51% is economically
large and statistically signiﬁcant at the 10% level.

We then turn to KLD ratings as a measure of ethics in Table 2.4. We examine four categories
that are related to corporate ethics: Bribery and Fraud, Tax Disputes, Human Rights, and Product
Quality. The Bribery and Fraud variable is related to accounting misstatements, but is wider in
scope. Bribery and Fraud (reported in Speciﬁcations 1 and 2) is based on expert opinion on a
wider range of compromised ethics, not just misstated ﬁnancials. A one standard deviation higher
AM Membership is associated with a 56 percentage point increase in ethical concerns by analysts.
When we use the dummy variable (Dummy(AM > 0)) as our explanatory variable, ﬁrms with
positive AM membership are 51% more likely to have ethical concerns by analysts when compared

10We follow Dechow et al. (2011) and estimate logit speciﬁcations. Our results are qualitatively similar using probit
and linear probability model speciﬁcations. We include year dummies and cluster our standard errors at the year level
because the rate of enforcement actions exhibit strong heterogeneity through time.

11An earlier version of the paper reported results when cubic polynomials in log employees and log assets were

used as controls. These results are available from the authors.

12In unreported analyses, we also match on market capitalization and value of assets and obtain similar results.

33

to ﬁrms without AM membership. Among other variables, Tobin’s Q and tangibility are negatively
related to ethical concerns regarding Bribery and Fraud, while size (log(employees)) is positively
related to analysts concerns.

We observe similar results for Tax Disputes (Speciﬁcations 3 and 4). A one standard deviation
increase in Active AM Accounts (positive AM membership) is associated with a 1.9 percentage point
(1.66 percentage point) increase in the likelihood of tax disputes (72% (79%) of the unconditional
mean). The results for human rights violations are consistent, albeit weaker (Speciﬁcations 5 and
6). Companies with higher AM Membership are more likely to be involved with regimes that
violate human rights (have conﬂicts with indigenous people, or have labor disputes), and less likely
to have proactive policies that prevent such involvement.13 A one standard deviation increase in AM
membership is associated with a 1.5percentage point increase in ethical concerns regarding human
rights violations (39% of the unconditional mean). We do not ﬁnd a statistically signiﬁcant eﬀect at
conventional levels when we use a dummy variable as the explanatory variable (Dummy(AM > 0)).
In Columns 7 and 8, we report results for product quality concerns. Higher AM Membership
is related to more concerns (negative values), and fewer strengths (positive values). A one standard
deviation increase in the number of AM Accounts (Dummy(AM > 0)) is associated with a 63 (67)
percentage point increase in concerns regarding product quality.14

Panel B of Table 2.4 reports results for the matched sample discussed previously. The results
from the matched sample corroborate our initial evidence. Out of ﬁve KLD variables, only Product
Quality is insigniﬁcant in the matched sample. The coeﬃcient for Bribery and Fraud is 57% larger
for ﬁrms with positive AM membership than for matched ﬁrms with zero-AM membership. This
is very close to the economic eﬀect reported in Panel A. Similarly, Tax Disputes are three times
more likely to occur for a ﬁrm with positive AM membership than a comparable ﬁrm in the same

13This variable has a negative value if the company is involved with the regimes that violate human rights, violate
the rights of indigenous people, or are involved in labor disputes and anti-union policies. The variable has a positive
value if the company has proactive policies to prevent such involvement and has pro-indigenous and pro-labor-union
policies. Unreported results showed that what we observe in Speciﬁcation 5 is driven mostly by violations rather than
strong anti-human rights violations policies.

14In unreported results we ﬁnd that the eﬀects in speciﬁcations 7 and 8 are driven mostly by product quality concerns

rather than product quality strengths.

34

industry and with roughly the same size, but without AM membership. Additionally, AM-ﬁrms are
69% more likely to be ﬂagged by analysts regarding Human Rights concerns. Finally, AM-ﬁrms
are 66% more likely to be involved in proﬁt sharing programs.

Finally, we examine the relation between AM membership and the use of tax havens/avoidance.
Building on the work of Dyreng and Lindsey (2009), we use the proportion of tax havens among
the countries the ﬁrm does business with. We assume that it is unlikely that a Compustat-listed
ﬁrm would have a legitimate reason for the majority of its international business dealings to be
tax havens. We use three cutoﬀs, 50% (i.e., more than half of the countries the ﬁrm is doing
business in are tax havens), 75%, and 90%. Table 2.5 (cont’d), Panel A, reports the results for
use of tax havens. Even in our weakest test (50% cutoﬀ, Columns 1 and 2), the eﬀect of AM
membership is both economically and statistically signiﬁcant. A one standard deviation increase in
active AM Accounts leads to a 1.05 percentage point increase in abnormal tax haven use. Similarly,
Dummy(AM>0) is associated with a 2.6 percentage point increase in the use of tax havens, relative
to the unconditional mean. At the 90% cutoﬀ (Columns 5 and 6), the eﬀect is considerably larger.
For Active AM Accounts (Dummy(AM>0)) a one standard deviation increase is associated with a 4.0
(5.3) percentage point increase in the use of tax havens.15 The presence of institutional investors,
greater competition, a higher Tobin’s Q, and larger market capitalization are all associated with
greater use of tax havens. Family ﬁrms use tax havens less extensively.

In Table 2.5 (cont’d) Panel B, we report results on the relationship between AM membership
and eﬀective tax rates (deﬁned as tax expense divided by pretax income). Even after controlling
for the use of tax havens, the eﬀective tax rate is about 0.16 percentage points lower for high AM
companies. A one standard deviation increase in AM membership is associated with a decrease
in the total taxes paid by approximately 1.1% of an average ﬁrm’s total tax bill. The results using
a dummy variable as our explanatory variable Dummy(AM>0) and reported in speciﬁcations 2, 4,
and 6, are not statistically signiﬁcant at the 10% level. However, the results are similar in magnitude
to the estimates reported for the continuous measure of Active AM Accounts.

15For consistency with earlier tables, we report results from logit speciﬁcations in Panel A of Table 2.5 (cont’d).

We obtain qualitatively similar results for probit and linear probability speciﬁcations.

35

2.4 Internal vs. External CEOs

In this section, we ask the question: do ﬁrms make an attempt to transform a culture with low
integrity? Prior literature suggests that culture is one the most diﬃcult organizational attributes
to change; it outlasts organizational products, services, founders, leadership, and the physical
attributes of an organization (Schein, 1992). However, as the ﬁrm’s business environment changes,
its former culture may no longer be appropriate. “When basic survival is threatened in terms
of an organization’s ultimate mission, there is a very strong external impetus to make a radical
change in culture” (Flanagan, 1995). Research in management science has suggested that such a
transformation often begins when an organization has a new, strong, leader who understands the
need for a major change (Kotter, 1995). This literature also recommends that such ﬁrms should
hire CEOs from outside the ﬁrm—or even outside the industry—if changing the existing culture is
a primary goal (Bailey and Helfat, 2003).16

Thus, the literature suggests that if a ﬁrm wants to change its culture, an eﬀective way to
do so is to hire an external CEO. In our context, we ask whether ﬁrms with high levels of AM
membership attempt to change their culture in this manner. This would be the case if there were
no tradeoﬀs to consider in the attempt to enforce stricter standards of integrity. We exploit CEO
changes to examine whether ﬁrms with high AM membership are more likely to hire external CEOs,
suggesting shareholders and directors want to purge “miscreant" cultures. We acknowledge that a
ﬁrm’s culture is likely diﬃcult to change, and therefore we do not attempt to measure the success or
failure of a regime shift. However, a ﬁrm’s board or shareholders trying to institute a deep cultural
shift are more likely to do so by appointing an external CEO rather than by hiring someone who
has been a part of the very culture the ﬁrm is trying to change.

We use Boardex data from 2003-2013 to identify internal versus external CEO hires. We deﬁne
internal CEOs as CEOs who were employed at the hiring ﬁrm for at least two years before their

16Lou Gerstner, the former IBM CEO is an example of an outsider who was brought in to change the corporate
culture (and succeeded). Many attempts to replicate this story have failed. For example, Hewlett-Packard’s Carly
Fiorina, and Procter & Gamble’s Durk Jager, are cited as examples of CEOs that tried to change too much, too soon.
Research has documented that many outside CEOs have not made meaningful changes at all (Karaevli and Zajac,
2013).

36

appointment. Table 2.6 presents the results from our analysis. The unconditional probability of
hiring internal CEOs in our sample is 0.378. After controlling for time eﬀects, as well as industry
and geography ﬁxed eﬀects, the probability of choosing an internal CEO is signiﬁcantly higher for
ﬁrms with higher AM membership. Speciﬁcally, a one standard deviation increase in the number of
AM accounts leads to a 6.5-18.1% increase in the probability that a new CEO appointment comes
from within the ﬁrm, or between 17.0-48.0% of the unconditional probability.

These results show that boards and shareholders of ﬁrms with more AM membership do not
exhibit a strong tendency to want change their culture by hiring an external CEO. One possible
explanation follows from Fiordelisi and Ricci (2014), who show that companies with more creative
cultures are more likely to choose an internal CEO in order to continue their creative success.
Furthermore, our evidence suggests that ﬁrms’ directors and shareholders are content with a culture
that supports a relatively high level of AM membership. This is consistent with our hypothesis that
there are inherent tradeoﬀs to engineering a corporate culture. These results are also consistent
with those of Parrino (1997).17

2.5 Conclusion

We ﬁnd that individual decisions by employees provide substantial information about their
employer. Firms that have a greater number of employees registered on AM are not only more
likely to behave more unethically, they are also likely to be more innovative. Overall, our results
suggest that the personality traits of employees vary systematically across ﬁrms. Firm culture is
related to corporate outcomes, and ﬁrms and employees tend to have matching personality types.

17In unreported tests, we examine whether CEO characteristics can explain our results. In particular, we examine
the overconﬁdence measure in Malmendier and Tate (2005). We construct the backward-looking measure, Holder 67,
that describes the exercise decision of a CEO in the ﬁfth year prior to expiration. Five years before expiration is the
earliest point we can consider, since most options in our sample have a ten-year duration and are fully vested only after
year four. Under Malmendier and Tate (2005) assumptions of constant relative risk aversion and diversiﬁcation, the
new exercise threshold in the Hall-Murphy framework is 67%. We set Holder 67 equal to 1 if a CEO fails to exercise
options with ﬁve years remaining duration despite a 67% increase in stock price (or more) since the grant date. We
ﬁnd no correlation between Holder 67 and Active AM Accounts (it is 0.026). We do not see any signiﬁcant changes in
the coeﬃcient on our variable of interest in all regressions in our paper after controlling for CEO overconﬁdence, age,
and gender. For consistency we report results from logit speciﬁcations in Table 2.6. We obtain qualitatively similar
results for probit speciﬁcations.

37

An interesting avenue for further research is understanding whether a causal relation extends
to more general settings. Research argues that culture ﬁts the ﬁrm’s business environment, and
employee personalities are selected to ﬁt the culture. Yet, research also argues that the corporate
culture is persistent. Thus, rapid changes in the ﬁrm’s external environment might lead to a culture
that is no longer suited to the ﬁrm’s environment. Are such ﬁrms the proverbial “dinosaurs” that
cannot adapt to changes in their environment and thus go extinct? Anecdotal evidence suggests
that even CEOs ﬁnd it diﬃcult to change a ﬁrm’s culture.

38

CHAPTER 3

PRODUCT COMPETITION AND CORPORATE FRAUD

3.1 Introduction

Trust is a central factor in the proper functioning of ﬁnancial markets (Greenspan, 2008). In
particular, distrust in the accuracy of ﬁnancial statements stemming from corporate fraud can
erode ﬁrm value (Karpoﬀ et al., 2008; Dyck et al., 2010), impose negative externalities on other
ﬁrms (Guiso et al., 2008; Kedia and Philippon, 2009), and inﬂuence investors’ allocation decisions
(Giannetti and Wang, 2016; Gurun et al., 2018). These consequences have obliged auditors,
regulators, and researchers to improve their understanding of managers’ incentives to commit fraud
and the ability of various parties to detect ﬁnancial reporting manipulations.

A survey of CFOs by Dichev et al. (2013) indicates that comparability between rival ﬁrms is
an important means for identifying ﬁnancial reporting abnormalities. Building on their insight,
we posit that greater product market overlap can enrich ﬁnancial statement comparability, thus
facilitating monitoring and improving fraud detection. This benchmarking eﬀect could discipline
managers’ reporting practices, leading them to commit less fraud. Alternatively, intense product
market competition stemming from less diﬀerentiated products can also pressure managers to
distort perceived relative performance through reporting manipulations (see Shleifer, 2004; Tirole,
2010) (see Shleifer, 2004; Tirole, 2010). Until recently, only coarse industry-level measures of
competition have been available to researchers, which creates challenges in identifying the potential
eﬀect of these opposing forces. As a result, the relationship between corporate fraud and product
market interactions remains underexplored.

In this paper, we move beyond traditional measures of competition to shed light on whether
product market diﬀerentiation imposes discipline or increases pressure, on average, as it pertains
to corporate fraud. To measure corporate fraud, we use the combination of settled Accounting
and Auditing Enforcement Releases (AAER) and settled Securities Class Action Clearinghouse

39

(SCAC) ﬁnancial lawsuits.1 To measure product diﬀerentiation, we use pairwise product market
similarity scores developed by Hoberg and Phillips (2010, 2016) that are based on the product
descriptions in ﬁrms’ annual ﬁnancial reports. Whereas most industry classiﬁcations are binary,
product similarity scores capture the degree of product market overlap between ﬁrms. Furthermore,
these similarity scores improve on the accuracy of other competitor classiﬁcation schemes and
allow for ﬁrm-speciﬁc and time-varying competitor networks. These features enable us to move
beyond identifying associations between industry-level characteristics and fraud by conducting
more powerful tests using ﬁrm-level measures of product market diﬀerentiation.

Our analyses reveal a strong relationship between product market diﬀerentiation and corporate
fraud. We ﬁnd that the incidence of fraud is signiﬁcantly lower for ﬁrms with a less diﬀerentiated
product mix. Speciﬁcally, a one standard deviation higher average product similarity score for
a ﬁrm is associated with a 14.8-23.7% decrease in the rate of SEC enforcement actions, relative
to the unconditional sample average. This result is robust to the inclusion of control variables
that have a documented relation to corporate fraud, such as measures of ﬁrm size, accounting
quality, internal and external corporate governance, the number of distinct markets in which a ﬁrm
operates, industry and industry-year ﬁxed eﬀects, and various levels of clustered standard errors.
Our ﬁndings suggest that the eﬀect of product market diﬀerentiation on the incidence of fraud
has a relatively large economic eﬀect when compared to virtually any predictor of corporate fraud
previously explored in the literature.

To explore whether the apparent disciplining eﬀect of low product diﬀerentiation stems from
a benchmarking channel, we study how ﬁrm complexity impacts the relationship between product
market diﬀerentiation and fraud. On the margin, it is likely more diﬃcult to detect abnormal
behavior for complex ﬁrms operating in many segments than for ﬁrms with a simple organizational
structure (e.g. Cohen and Lou, 2012).2 We proxy for complexity using the number of unique SIC
codes that a ﬁrm’s product mix spans, and split ﬁrms into quartiles based on this proxy. Consistent

1Donelson et al. (2017) suggest the combination of AAER and class action lawsuits provide the most accurate and

complete measure of corporate fraud that is currently available.

2Cohen and Lou (2012) ﬁnd that ﬁnancial markets incorporate information at a slwer pace for complicated ﬁrms.

40

with the benchmarking channel, we ﬁnd that the disciplining eﬀect of product similarity increases
monotonically across complexity quartiles, after controlling for ﬁrm size. Indeed, the coeﬃcient
estimate is more than four times as large for the most complex quartile when compared to the least
complex quartile. This ﬁnding suggests that having less diﬀerentiated rivals generates a larger
marginal impact on the ability to detect corporate fraud for complex ﬁrms.

To further explore the benchmarking channel, we exploit an additional feature of IPO and
M&A activity. Speciﬁcally, we utilize IPOs and M&As of a ﬁrm’s rivals as a shock to the ﬁrm’s
information environment. IPOs increase the publicly available ﬁnancial information of existing
competitors, which in turn enhances the ability to assess, compare, and scrutinize related ﬁrms’
own ﬁnancial statements (e.g. Bauguess et al., 2018). Similarly, M&A activity generates attention
in the merging ﬁrms’ industries due to potential spillover eﬀects on rivals, customers, and suppliers
(e.g. Fee and Thomas, 2004), and due to potential ensuing acquisition activity (Song and Walkling,
2000). We ﬁnd that higher IPO and M&A activity of rival ﬁrms is associated with a higher incidence
of detected fraud. Further, this increase in detection is signiﬁcantly more pronounced for ﬁrms
with less similar rivals, prior to the IPO (i.e., the eﬀect is greater for ex ante less disciplined ﬁrms).
This ﬁnding suggests these events enhance monitors’ eﬀectiveness in detecting fraud. In particular,
IPOs or M&As by rivals increase the saliency of publicly available information in a ﬁrm’s industry
before a ﬁrm has time to fully wind down misreporting. A potential concern is that IPOs and M&A
activity likely impact the competitive nature of the industry by injecting capital into a newly public
ﬁrm or by consolidating market power in existing competitors. While we cannot entirely rule out
the competition eﬀect, we condition on the amount of capital raised in an IPO (size of M&A deal)
to help isolate the eﬀect of information saliency on fraud detection.

Product market diﬀerentiation likely reﬂects aspects of the competitive landscape, in addition
to the amount of comparable information provided by rivals. To isolate the benchmarking aspect
of product market similarity, we control for a variety of measures meant to capture various aspects
of competitive intensity and market power. These measures include industry-wide proﬁt margin,
sales concentration, top-4 ﬁrm market share, the number of competitors a ﬁrm has, and product

41

market ﬂuidity. We ﬁnd that including these measures of competition as control variables does not
signiﬁcantly inﬂuence the coeﬃcient estimates for product market diﬀerentiation. Furthermore,
we fail to ﬁnd a robust statistical relationship between these alternative competition measures and
fraud when examined separately. These results suggest that product market diﬀerentiation plays an
important and unique role in determining fraudulent behavior that is not inﬂuenced by controlling
for other characteristics of competition.

In summary, we ﬁnd a strong and robust link between product market diﬀerentiation and cor-
porate fraud. Indeed, our estimates suggest that product market similarity has an economically
larger eﬀect on fraud than any factor, other than ﬁrm size, previously documented in the literature.
Our initial results suggest that product market similarity imposes a strong disciplining eﬀect on
ﬁnancial reporting misconduct. Further, while none of our follow-up analyses provides incontro-
vertible evidence in isolation, the preponderance of evidence suggests that the disciplining eﬀect
stems through a benchmarking channel. These results indicate that market-based mechanisms play
an important role in both the incentive to commit fraud and the ability of external parties to uncover
fraudulent activities.

Our paper relates to the literature examining the eﬀect of various measures of competition on
managerial discipline. On one hand, competition can diminish conﬂicts of interest by incentivizing
managerial eﬀort (Nickell, 1996) or by reducing resources available for rent extraction (Ades and
Di Tella, 1999; Schmidt, 1997). On the other hand, competition has been argued to pressure
managers to distort the perceived performance relative to rivals (Shleifer, 2004; Tirole, 2010;
Andergassen, 2016). Until recently, only coarse industry-level measures of competition have been
available to researchers, which has introduced challenges in identifying the potential eﬀect of these
opposing forces. We shed light on this relationship by exploiting newly developed, ﬁrm-level
measures of product diﬀerentiation that allow us to conduct more powerful tests. Consistent with
a disciplining channel of competition, we document that product market similarity is strongly
associated with a lower incidence of fraud.

In addition, our work suggests that benchmarking is an important factor to consider in studying

42

competition, as it enhances information, and therefore facilitates monitoring ability. In turn, this
benchmarking channel can inﬂuence managerial behavior. More speciﬁcally, as predicted by
Holmstrom (1982) and Nalebuﬀ and Stiglitz (1983), to the extent that ﬁrms are similarly impacted
by the competitive landscape, more direct competition increases information about the ﬁrms that
could help reduce moral hazard problems.
Indeed, empirical work such as Hsu et al. (2017)
indicates that a ﬁrm’s competitive landscape is an important determinant of analyst coverage and
forecast accuracy. We study whether this eﬀect applies to ﬁnancial fraud. Our results suggest
that having similar rivals can facilitate information acquisition, which is consistent with survey
evidence by Dichev et al. (2013) on factors that help detect ﬁnancial misrepresentation. Thus, our
work suggests an avenue through which external parties, such as short-sellers and analysts, can
obtain information useful in the detection of fraud (Dyck et al., 2010).

Our analysis also complements empirical work by Wang et al. (2010) who show that industry-
level information aﬀects fraud detection. Our results indicate that information contained in ﬁrm-
speciﬁc product markets and unique competitor networks leads to substantial within-industry
heterogeneity in fraud detection, after controlling for the industry-wide measures of information
outlined in Wang et al. (2010). Other work by Balakrishnan and Cohen (2013) investigates
the interplay between traditional measures of competition and the industry-level incidence of
restatements, rather than misreporting and fraud. Our results also suggest that competition matters,
but we focus on a particular dimension of competition (product diﬀerentiation) that facilitates
exploration of the benchmarking channel. We show that the benchmarking ability brought about
by ﬁrms with a similar product mix holds after controlling for various measures of industry
concentration.

3.2 Data

We follow recent empirical work of Donelson et al. (2017) by deﬁning corporate accounting
fraud as, “the intentional, material misstatement of ﬁnancial statements that causes damages to
investors." Donelson et al. (2017) advocate using a combination of public and private enforcement

43

actions through AAER and class action lawsuits to capture ﬁnancial reporting fraud to mitigate
measurement error. While regulatory enforcement is important, other participants, such as the
media, industry regulators, and employees, serve as important actors in this arena (Dyck et al.,
2010).

We obtain AAER data for the sample period 1996-2010. According to the Center for Financial
Reporting and Management, the U.S. Securities and Exchange Commission (SEC) issues AAERs
during, or at the conclusion of, an investigation against a company, an auditor, or an oﬃcer for
alleged accounting or auditing misconduct. The AAER dataset provides information on the nature
of the misconduct, the named individuals, and the entities involved, as well as their eﬀect on the
ﬁnancial statements. The misstatement investigations in our sample occur because of bribery, fraud,
inﬂated assets, ﬁnancial reporting related enforcement actions concerning civil lawsuits brought
by the in federal court, and orders concerning the institution and/or settlement of administrative
proceedings.

We construct our sample of class action lawsuits following the work of Choi et al. (2009),
Griﬃn et al. (2004), Jayaraman and Milbourn (2009), and Thompson and Sale (2003). We start
by downloading all class action lawsuits from the SCAC hosted by Stanford University for 1996
through 2011 and scan each ﬁling to only include 10-b5 class action lawsuits, which eliminates
those lawsuits that occur for non-ﬁnancial reasons.3

We deﬁne each ﬁrm-year as an AAER year, a SCAC year, both, or neither. Our primary
independent variable, fraud, is a binary variable equal to one for all ﬁrm years in which there is an
AAER or SCAC. We exclude ﬁrms in the ﬁnancial and utilities industries and ﬁrms headquartered
outside the United States. Further, we drop ADRs, ﬁrms with assets less than $1M, and ﬁrms with
missing assets or sales items in Compustat. Our ﬁnal sample of corporate fraud events includes
935 ﬁrm-years that are aﬀected by AAER misstatements in at least one quarterly or annual ﬁnancial
statement from 322 unique ﬁrms from 1996 to 2010. In addition, our sample includes 311 class
action lawsuits aﬀecting 299 ﬁrms from 1996 to 2011. In total, our sample contains 498 ﬁrms and

3Karpoﬀ et al. (2017) note the importance of additional checks of the sources to ensure that they contain instances

of fraud.

44

1,217 ﬁrm-years, ﬂagged as years with fraudulent reporting. These ﬁgures are very closely in line
with those of (Dyck et al., 2010). As shown in Table 3.2, the overall incidence of fraud in our
sample is 1.9%.

To construct our set of control variables, we follow work in the ﬁnance and accounting literatures
related to corporate fraud. We include predictors of accounting misstatements from Dechow et al.
(2011). The variable RSST accruals measures the change in non-cash net operating assets, including
both working capital accruals and long-term operating capital. Bergstresser and Philippon (2006)
show that changes in accounts receivable (∆AR) and change in inventory (∆Inventory) are associated
with incentives to improve sales growth and gross proﬁt margin. A ﬁrm’s soft assets as a percentage
of total assets (% soft assets) is associated with more discretion for earnings management. We deﬁne
% soft assets as total assets minus property plant and equipment and cash and cash equivalents, all
scaled by total assets. Change in cash-based sales (∆Cash Sales) excludes accrual-based sales to
measure the portion of sales that are not subject to discretionary accrual management. Change in
ROA (∆ROA) controls for changes in earnings growth. The variable ∆Employees is the percentage
change in employees less the percentage change in total assets. This measure is associated with
labor costs and must be expensed as incurred. Reducing the number of employees can boost a ﬁrm’s
short-term ﬁnancial performance by immediately lowering expenses. Finally, we include a dummy
variable (d_Security Issue) equal to one for ﬁrm years in which a ﬁrm issues debt or equity, which
can increase incentives to manage earnings Rangan (1998). We refer to speciﬁcations including
only the controls from Dechow et al. (2011) as the Dechow set of controls.

We also include speciﬁcations that contain proxies for monitoring mechanisms and corporate
opaqueness, which could potentially inﬂuence the marginal impact of our proposed benchmarking
channel. We include Institutional Ownership, the natural log of the number of analysts covering
a ﬁrm’s stock (Ln Num Analysts), research and development expenses (R&D), and industry stock
return r-squared (Ind R2).4 To construct the industry r-squared, we follow Wang and Winton

4To handle observations with missing R&D, we follow the method outlined in Koh and Reeb (2015) and replace
each missing observation with the industry year average and include a dummy variable for whether the ﬁrm has missing
R&D .

45

(2014) and ﬁrst regress each ﬁrm’s daily stock returns on the weighted-average daily market return
and the weighted-average daily industry return. Then, we take the average r-squared for each
ﬁrm in a given three-digit SIC code. Managers may feel pressured to commit fraud when they
require capital from outside sources (Teoh et al., 1998; Wang and Winton, 2014). Thus, we
include the Whited and Wu (2006) Index for ﬁnancial constraints.5 We include the natural log
of total assets (ln assets) as a measure of ﬁrm size. We also include the variable Book Leverage,
which is deﬁned as long and short-term debt over total assets. Highly levered ﬁrms may have
greater probabilities of ﬁnancial distress, which has been shown to predict fraud (e.g. Healy and
Wahlen, 1999). Alternatively, debt can have a disciplining eﬀect by either mitigating agency issues
between managers and shareholders (Grossman and Hart, 1982), or providing an additional source
of external monitoring vis-a-vis debtholders.

Product diﬀerentiation is likely related to relative performance evaluation (RPE). Firms with less
product market diﬀerentiation might naturally have better benchmarks, and therefore, be more prone
to RPE, which could pressure some managers to cut corners or misstate earnings to outperform
benchmarks (e.g. Cheng, 2011). This eﬀect would work against our hypothesis and ﬁndings. Thus,
to increase the power of our tests, we control for RPE following the work of Wang and Winton
(2014) who construct an indicator variable RPE. First, the authors estimate the following regression
equation:

prob(CEOTurnoveri,t−1) = γ1RP+

i,t + γ2RP−

i,t + i,t

(3.1)

i,t is equal to relative performance when relative performance is above 0, and zero
where RP+
otherwise; and RP−
i,t is equal to relative performance when relative performance is below 0, and 0
otherwise. Relative performance is measured as the diﬀerence in performance between ﬁrm i and
the weighted average of ﬁrm i’s rivals according to its three-digit SIC code. Following Wang and
Winton (2014), we estimate equation (1) separately for each industry (three-digit SIC) and deﬁne
the binary variable RPE equal one for industries where γ2 < 0. We refer to speciﬁcations that
include all our control variables as the full set of controls.

5In unreported analysis, we use an alternative prosy for equity ﬁnance needed, which measure a ﬁrm’s asset growth

rate in excess of the maximum internally ﬁnanceable growth rate. We ﬁnd qualitatively similar results.

46

Table 3.2 provides the number of observations, mean, standard deviation, 10th percentile, and
90th percentile value for our control variables. We estimate all speciﬁcations for both winsorized
and non-winsorized data. Estimates obtained from winsorized data (1% in each tail) are reported.

3.3 Product Diﬀerentiation and Alternative Measures of Competition

3.3.1 Product Market Diﬀerentiation

For our measure of product diﬀerentiation, we use the product similarity score developed by Hoberg
and Hoberg and Phillips (2010, 2016), who use textual analytics to capture the relatedness of a
ﬁrm’s product market with all other ﬁrms in that ﬁle annual reports with the U.S. Securities and
Exchange Commission (SEC). The process involves vectorizing the product market vocabulary
from the business description from each ﬁrm’s annual 10-Ks, according to a dictionary the authors
develop. They then assign pairwise similarity scores based on the cosine similarity between two
ﬁrms’ vectorized product market descriptions. The cosine similarity is higher when the product
market descriptions between the two ﬁrms are more similar. The measure ranges from 0 (no
similarity) to 1 (perfect similarity). We contend that product market overlap measures relatedness
that can be used to assess both ﬁnancial statements and understand the common factors that aﬀect
the performance of related ﬁrms.

We also make use of the text-based network industry competitors (TNIC) that Hoberg and
Phillips deﬁne as a byproduct of their product similarity score. The TNIC competitor set includes
all ﬁrms with a similarity score above a given threshold. Importantly, TNIC allows the ﬂexibility
for each ﬁrm to have its own distinct set of competitors. For example, Nike competes with Callaway
in golf, and competes with Head in tennis, but Callaway and Head are not direct competitors with
each other. This intransitive feature better reﬂects economic reality, and it allows us to exploit
granularity in the data that is not possible using measures created from standard transitive industry
classiﬁcations, such as SIC codes.

Furthermore, these industry classiﬁcations are updated annually, which provides more ﬂexibility
and accuracy in empirical design. For example, when Exxon sold its retail gas stations in 2008,

47

this event was reﬂected by the change in its competitor set (TNIC) and average product similarity
score (from 0.035 to 0.012). However, the divestment from Exon was not reﬂected by a change
in its SIC code or other industry classiﬁcations. As a result, the level of competition that Exxon
faced according to SIC code-based Hirschman-Herﬁndahl Index (HHI) measures did not change
in response to its large divestment. These features allow us to conduct more powerful tests than
transitive and time-invariant industry classiﬁcations would allow.

Using the TNIC competitor classiﬁcation and product similarity scores, we create our main
variable of interest; Average Similarity Score, as the average pairwise similarity score of all
competitors within a ﬁrm’s TNIC-3 classiﬁcation in each year. As shown in Table 3.2, the ﬁrms
in our sample have 49 competitors on average, with an Average Similarity Score of 0.03 above the
threshold set by Hoberg and Phillips (2016).

There are potential issues when using the Average Similarity Score based on the TNIC classi-
ﬁcation. First, the TNIC only includes pairs of ﬁrms over a certain threshold of similarity. While
imposing this threshold allows us to focus on closely related rivals, there can be substantial varia-
tion in the number of competitors being averaged across for each ﬁrm. The wide variation in both
the number of competitors each ﬁrm has, and the degree of similarity with each competitor, can
obfuscate the association between fraud and product diﬀerentiation. Two ﬁrms, for example, could
have the same average product market similarity scores for diﬀerent reasons. One ﬁrm could have
several moderately close rivals while another ﬁrm could have some nearly identical rivals and some
that are barely related. While both ﬁrms could have the same average product similarity score,
we would expect the ﬁrm with the near identical rivals to have stronger discipline eﬀect through
benchmarking.

To address such concerns, we implement a series of alternate methods for aggregating product
similarity scores. Rather than averaging across all competitors in a ﬁrm’s TNIC, we average
across the top 5, 10 or 15 closest competitors. This process creates more homogeneity by utilizing
the same number of competitors for each ﬁrm and focuses on each ﬁrm’s closest rivals, which
should provide the greatest benchmarking externalities. As an alternative approach, we count of

48

the number of competitors each ﬁrm has that are in the top percentile (95th, 90th and 75th) of
the overall distribution of similarity scores across all ﬁrms in the sample. This process allows
us to count the number of rivals that each ﬁrm has that are very similar relative to the complete
cross-section of ﬁrms.

In addition to the simple Average Similarity Score and rival counts, we develop a more nuanced
measure that helps account for the degree of similarity between rivals. In particular, rivals provide
signals about similar ﬁrms, with greater similarity between two rivals producing a less noisy signal.
It follows that both the similarity with a given rival, as well as the number of rivals, impact the total
signal provided by a ﬁrm’s product market competitors. If signal noise is normally distributed,
then there is an inverse squared relationship between product market similarity and the quality of
the signal. We deﬁne a measure of precision as:

precisioni,t =(cid:0) 1

Ni

Ni

j=1

(cid:1)0.5

1

(1 − scorei,j,t)2

(3.2)

where Ni is the number of competitors in ﬁrm i’s TNIC, and scorei,j,t is the product similarity
score between ﬁrm i and competitor j in year t.6 Higher precision is indicative of a greater signal
provided by a ﬁrm’s product market rivals.

3.3.2 Alternative Measures of Competition

The similarity measures based on product market diﬀerentiation should capture the extent to which a
ﬁrm’s rivals provide suitable benchmarking of performance, thus facilitating the detection of fraud.
However, the degree of product market diﬀerentiation also reﬂects the notion that competition is an
endogenous outcome of market forces and that ﬁrms choose to diﬀerentiate to the greatest degree
possible (e.g. Tirole, 1988). Therefore, less diﬀerentiation suggests more intense competition, all
else equal. To isolate the benchmarking eﬀect, we control for commonly used measures in the
literature that are designed to capture other aspects of competition, including HHI (Herﬁndahl,
1950), proﬁt margin (Bain, 1951), and the sales concentration ratio of the largest four ﬁrms in an

6we thank Jerry Hoberg for suggesting this measure

49

industry (Heﬂebower, 1957). The HHI based on SIC code is the most extensively used measure of
competition in studies related to product market competition. The HHI for industry j is calculated
as:

Nj

i=1

HHIj =

((MSi)2

(3.3)

Where MSi is the sales-based markets share of ﬁrm i in industry j, and Nj is the number of ﬁrms
in industry j. HHI has a maximum value of 1 that is attained if a single ﬁrm makes up an entire
industry, and a minimum value of 1/Nj if each ﬁrm has equal weight in industry j.

As additional measures of competition, we include the number of competitors according to a
ﬁrm’s TNIC or three-digit SIC codes. Classic models of competition suggest that the more ﬁrms
there are oﬀering similar products, the competition would be more intense (Tirole, 1988). We
also include proﬁt margin (Bain, 1951) and the sales concentration ratio of the top four ﬁrms in an
industry (Heﬂebower, 1957) as an additional measure of market power at the industry level. Lastly,
we include product market ﬂuidity, which captures competitive pressure from potential entrants
that captures each ﬁrm’s ex ante competitive threats (Hoberg et al., 2014). This measure also
uses textual analytics and compares the use of unique words in each ﬁrm’s product descriptions to
changes in the overall use of a given word by other ﬁrms in their product descriptions. This measure
lies between zero and one and is higher the more the words used a ﬁrm’s product description overlap
with the changes in the word changes by competitor ﬁrms. The intent is to capture threats based on
the actions by rival ﬁrms, rather than changes of the ﬁrm itself.

3.4 Empirical Results

In this section, we discuss results from ﬁrm-level regressions that examine the association
between corporate fraud and product market diﬀerentiation. We ﬁrst explore associations in a
standard panel data framework before exploring an instrumental variables approach. We then move
on to discuss empirical tests that are aimed to highlight the particular channel that could explain
the ﬁndings of our exploratory regressions.

50

3.4.1 Product Market Diﬀerentiation and Corporate Fraud

We report OLS estimates for the association between average product similarity score (Average
Similarity Score) and corporate fraud in Table 3.3.7 The ﬁrm-year is the unit of observation in
all reported speciﬁcations in this section. The speciﬁcation in Column 1 only includes year ﬁxed
eﬀects. Column 2 includes the Dechow set of controls (i.e. accruals, change in accounts receivable
(∆AR), change in Inventory (∆Inventory), the percentage of soft assets (% Soft Assets), change
in cash sales (∆Cash Sales), change in ROA (∆ROA), change in employees (∆Employees), and a
dummy for security issuance (d_Security Issue).

In Column 3, we also include the natural log of total assets (Ln Assets), Book Leverage, Industry
Stock Return R-squared, the Whited-Wu Index, a ﬂag for relative performance evaluation (RPE
ﬂag), R&D, an R&D ﬂag, and Institutional Ownership. Including Institutional Ownership results
in a large drop in the number of observations and does not appear to have a meaningful eﬀect on
the detection of fraud. Furthermore, inclusion of Institutional Ownership only seems to intensify
the relationship between fraud and Average Similarity Score. Considering these issues, we drop
Institutional Ownership from the remaining speciﬁcations and use the speciﬁcation from Column 4
as our primary speciﬁcation throughout the remainder of our analysis. Henceforth, we refer to the
speciﬁcation of control variables in Column 4 as our Full set of control variables. All explanatory
variables are lagged by one year.

The granularity of our data enables us to control for unobserved heterogeneity at the industry
and industry-year level. The speciﬁcation in Column 4 includes industry (three-digit SIC code)
and year ﬁxed eﬀects, and the speciﬁcation in Column 5 includes industry-year ﬁxed eﬀects.
The inclusion of ﬁxed eﬀects improves upon existing studies that are typically unable to account
for unobserved industry heterogeneity because the variables they deploy are often constructed at
the industry level. In particular, inclusion of industry or industry-year ﬁxed eﬀects accounts for
pervasive diﬀerences in the propensity to commit fraud across industries and helps to mitigate the
eﬀects of large industry shocks explained by factors not controlled for in our initial speciﬁcations.

7In untabulated analysis, we estimate this relationship with probit and logit speciﬁcations and ﬁnd similar results.

51

The t-statistics are calculated from standard errors clustered by three-digit SIC code.8

Throughout all speciﬁcations, the coeﬃcient estimate for Average Similarity Score exhibits a
very consistent, economically meaningful, and statistically signiﬁcant, mitigating eﬀect on fraud.
A one standard deviation increase in Average Similarity Score (0.023) is associated with a roughly
0.48 percentage point decline in the rate of fraud. That is, a one standard deviation increase in
average product similarity score is associated with a decline in the rate of fraud from 1.8% to 1.3%.
Thus, the results suggest that product similarity has a large economic eﬀect. Indeed, ﬁrm size is the
only predictor of fraud that we have found documented in the literature to have a larger economic
relation to fraud than Average Similarity Score. These results are robust to several diﬀerent sample
periods (i.e., before and after Sarbanes Oxley) and to the inclusion of controls that proxy for external
monitoring, such as the number of analysts and the degree of institutional ownership.9

3.4.2 Firm Complexity and Product Diﬀerentiation

Thus far, we have documented a strong mitigating eﬀect of product market similarity (lack of
diﬀerentiation) on corporate fraud. In this section, we aim to further establish that the disciplining
eﬀect of product market similarity on earnings manipulation acts through a benchmarking channel.
Tirole (2010) claims that relative performance evaluation (benchmarking) plays an important role
in corporate governance because the performance of rival ﬁrms is partly governed by common
shocks to production cost and demand.

To explore the benchmarking channel, we study the disciplining eﬀects of product market
similarity and a measure of ﬁrm complexity. Cohen and Lou (2012) argue “complicated ﬁrms
require more complicated analysis to impound the same piece of information into the price of a
ﬁrm with multiple operating segments." It stands to reason that regulators, media, and employees
can more easily disseminate information for ﬁrms with a simple organizational structure, and are
therefore, more likely to detect “abnormal" performance or ﬁnancial reporting. Thus, for ﬁrms

8Our results are robust to clustering at broader industry classiﬁcations (e.g., two-digit SIC) and at the ﬁrm level.
9We also control for ﬁnancial statement comparability developed by Franco et al. (2011) and we obtain similar

results.

52

with a very simple organizational structure and product mix, the information provided by having
similar rivals (benchmarks) would have a lower marginal eﬀect on outsiders’ monitoring ability. In
contrast, complex ﬁrms can be very diﬃcult to understand and detect “abnormal" behavior without
a clear benchmark. Thus, having close rivals for complex ﬁrms should intuitively provide a larger
marginal eﬀect on the ability to detect earnings manipulations.

All else equal, a ﬁrm that operates in several product markets has greater scope to conceal
ﬁnancial information. Operating across a multitude of product markets reduces substantive analytic
procedures that auditors can perform and will require more subjective and detailed testing. This
notion is reﬂected in the higher audit fees for ﬁrms with many segments (Brinn et al., 1994). For
example, a ﬁrm that competes in pharmaceuticals, manufacturing, and consumer durables, could
hide information by shifting resources across segments or using complex transactions. Furthermore,
monitors would need to understand all three industries to conﬁdently detect reporting abnormalities.
As such, we deﬁne complexity as the unique number of industries (three-digit SIC codes) in
which a ﬁrm operates each year. To calculate this value, we sum the number of distinct industries
spanned by a ﬁrm’s TNIC-based competitor set. For example, if a ﬁrm has three rivals that each
operate in a diﬀerent three-digit SIC code, then we consider that ﬁrm to be operating in three
distinct markets. A higher score on complexity indicates that a ﬁrm operates in an environment
where rivals are from many diﬀerent industries, and thus the ﬁrm is likely more diversiﬁed and
has a complex basket of products that compete across several markets. Our measure of complexity
builds on the intuition provided by Cohen and Lou (2012) who measure complexity as whether a
ﬁrm operates in multiple markets.

We split our sample into quartiles according complexity rankings. Then, we estimate our main
speciﬁcation for the relationship between corporate fraud and product market similarity separately
for each quartile. The results are presented in Table 3.4. In Panel A, we report the average number
of unique SIC codes and the number of competitors in each ﬁrm’s TNIC. Each speciﬁcation is
estimated using our full set of control variables, described in Section II and in our analysis of
Table 3.3. We estimate regressions separately for each complexity quartile without Institutional

53

Ownership (Panel B) and with Institutional Ownership (Panel C).

Consistent with the benchmarking channel, we ﬁnd that the disciplining eﬀect of product
similarity increases monotonically across complexity quartiles for Panel B. The partial eﬀect for
the top quartile is more than four times as large as that for the lowest quartile. To put this
ﬁnding into perspective, a one standard deviation increase in Average Similarity Score for the least
complex ﬁrms leads to a decrease in propensity of fraud from 1.9% to approximately 1.6%, or a 0.3
percentage point decline. By comparison, a one standard deviation change in Average Similarity
Score for the most complex ﬁrms decreases the propensity of fraud from 1.9% to 0.65%, or a
1.25 percentage point decline. This result is robust to several variations of controls for ﬁrm size.
While large ﬁrms are more complex than small ﬁrms, on average, our sorts capture product market
complexity beyond ﬁrm size.

In Panel C, we include a control for Institutional Ownership because institutions can serve as
monitors (Hartzell and Starks, 2003). We split this speciﬁcation into a separate panel because
the number of observations is reduced substantially. The coeﬃcient estimates continue to exhibit
a strong monotonic relationship across complexity quartiles, and therefore, we omit institutional
ownership form other speciﬁcations to maintain the larger sample size. Again, the eﬀect of product
market similarity in the top quartile is almost four times as large as the eﬀect in the bottom quartile.

3.4.3 Change to Information Environment Associated with Rival IPOs and M&A

To further investigate the benchmarking channel, we exploit an additional feature of IPO and
M&A activity by a ﬁrm’s rivals. In particular, both events plausibly lead to a shock to the ﬁrm’s
information environment. We ﬁrst study the event of IPOs by a ﬁrm’s rivals. These events increase
the publicly-available ﬁnancial information of previously existing, private competitors, which in
turn, enhances the ability to assess, compare, and scrutinize a ﬁrm’s own ﬁnancial statements.
Consistent with this view, Bauguess et al. (2018) provide evidence that IPOs lead to in an increase
EDGAR traﬃc for rival ﬁrms that are already publicly traded. Next, we study acquisitions by a
ﬁrm’s rivals. Acquisitions are material events that can draw considerable scrutiny from investors,

54

analysts, regulators and the media, thus increasing the saliency of existing information in the
industry. For instance, acquisitions often occur in waves, suggesting an increase in attention for
other ﬁrms that could potentially be involved in a deal (Song and Walkling, 2000).

For the IPO tests, we take each pairwise observation of competitors, i and j, and ﬂag whether
ﬁrm j underwent an IPO in year t. We then aggregate the data to the ﬁrm-year level for ﬁrm i,
counting the number of rivals that underwent an IPO in year t. For robustness we also deﬁne a
dummy variable (Competitor IPO) equal to 1 if any of a ﬁrm’s rivals underwent an IPO in year t,
and 0 otherwise. There are 3.2 rival IPOs per ﬁrm-year in our sample, with a median of 0 (43%
of ﬁrms have at least one rival IPO). Contingent on having at least one IPO rival, each ﬁrm has an
average 7.6 rivals undertaking IPOs, which is consistent with the documented evidence that IPOs
occur in waves (e.g. Lowry and Schwert, 2002).

A competitor’s IPO is a shock to competition via two channels. First, as discussed, more
information about economic conditions becomes publicly available for the rival, as well as increased
attention, which enhances monitoring abilities for a ﬁrm’s own investors, regulators, and auditors.
Second, the IPO provides a capital injection for a ﬁrm’s competitor, thus enhancing the intensity of
competition with that rival. We control for the amount of funds raised by the competitor during the
IPO to help separate the shock to a ﬁrm’s information environment from the inﬂuence of changes
in competitors’ capital structure and size. While this solution is imperfect, it assists in isolating the
eﬀect of rival IPOs due to information rather than the intensity of competition.

We report OLS estimates for the relationship between fraud and rival ﬁrm IPOs in Panel A of
Table 3.5. More speciﬁcally, we estimate the interaction between the natural log of the number
of rival ﬁrms undergoing IPOs and a ﬁrm’s product market similarity (Average Similarity Score)
prior to the rivals’ IPOs. A ﬁrm’s pre-existing product market similarity is indicative of the level
of market discipline provided by competitors prior to the rival’s IPO. The coeﬃcients on the level
terms suggest that rival IPOs are positively related to fraud detection and Average Similarity Score
continues to exhibit a negative relation to fraud propensity. The positive eﬀect of rival ﬁrm IPOs on
fraud suggests a shock to detection. In particular, IPOs by rivals change the available information

55

for comparison rather abruptly, before a ﬁrm has time to fully wind down ﬁnancial misconduct.

The coeﬃcient estimate for the interaction term is negative (lower for ﬁrms with greater pre-
existing product market similarity). This ﬁnding suggests that the increased detection resulting
from rival IPOs is signiﬁcantly more pronounced for ﬁrms with more product diﬀerentiation prior
to the rival’s IPO (i.e., the eﬀect is more pronounced for ex ante undisciplined ﬁrms).10 We also
split the sample between pre-IPO year high and low Average Similarity Score ﬁrms to verify that
the IPO-detection eﬀect is greater for ﬁrms that had lower ex ante discipline due to fewer related
ﬁrms before the IPO year. Due to a lack of power in the high pre-IPO regression these coeﬃcients
are not signiﬁcantly diﬀerent from each other. We ﬁnd that these eﬀects hold after conditioning
on the amount of capital raised by rivals during the IPOs. Overall, the ﬁndings in this section are
consistent with rival IPOs having a greater detection eﬀect for ﬁrms with lower pre-existing product
market discipline.

Next, for the M&A tests, for each pairwise observation of competitors, i and j, we ﬂag whether
ﬁrm j was acquired in year t. We then aggregate the data back to the ﬁrm-year level for ﬁrm i, and
take the log of the number of acquired ﬁrms competing with ﬁrm i that were acquired in year t.
For the average ﬁrm in our sample, there are 0.059 competitor acquisitions of rival ﬁrms each year.
Conditional on at least one competitor being acquired, the average increases to 1.09 competitor
acquisitions per year.

Panel B of Table 3.5 reports the results for rival ﬁrms being acquired. Much like the IPO results,
we ﬁnd a negative coeﬃcient on average similarity and a positive coeﬃcient for the number of
competitors being acquired. The interaction term is also negative, indicating that the partial eﬀect
of ﬁrms that had higher score is lower when rivals are being acquired. Thus, ﬁrms that are ex ante
less disciplined exhibit the greatest response to the information/interest generated around takeovers.
The split by ﬁrms’ pre-existing similarity scores, highlights that the eﬀect of the acquisitions exists
predominantly in the subsample with less similar product market rivals (i.e. those less disciplined
ex ante).

10We repeat this test excluding the rival undergoing IPO from a ﬁrm’s Average Similarity Score calculation to ensure

that the similarity with the IPO ﬁrm is not driving the results.

56

3.4.4 Alternative Measures of Competition, Product Similarity, and Fraud

In this section, we show that product diﬀerentiation captures a particular dimension of competi-
tion not explained by traditionally used measures. Traditional competition measures include the
Herﬁndahl-Hirschman Index (HHI) developed in Herﬁndahl (1950), proﬁt margin Bain (1951), the
sales concentration ratio of the top four ﬁrms in an industry Heﬂebower (1957), and the number of
competitors. We also use the newer measure, product market ﬂuidity, which captures the intensity
of product market changes for a given ﬁrm each year (see Hoberg et al., 2014). All measures of
competition are discussed in detail in Section III.

In each Column of Table 3.6, we estimate a speciﬁcation that includes our full set of control
variables, described in Section II. In Column 1, we control for the sales based HHI according to a
ﬁrm’s primary three-digit SIC code. This measure of competition is among the most widely used
in academic research. In Column 2, we also control for the number of competitors that each ﬁrm
has according to its primary three-digit SIC code. Classic models of competition, in which more
ﬁrms oﬀering the same product results in more competition, motivate our inclusion of the number
of competitors. Additionally, in Column 3 we include proﬁt margin (Bain, 1951) and the sales
concentration ratio of the top four ﬁrms in an industry (Heﬂebower, 1957) to account for market
power at the industry level. The eﬀect of Average Similarity Score on fraud remains consistent in
both signiﬁcance and magnitude across Columns 1-3 of Table 3.3.

In Columns 1-3, sales-based HHI using three-digit SIC codes does not appear to have a
meaningful relationship with Fraud. In Column 4, we include a sales-based Herﬁndahl-Hirschman
Index (HHI) calculated from a ﬁrm’s TNIC, rather than primary SIC code. This speciﬁcation
allows us to explore whether the apparent lack of power exhibited by the HHI in relation to
corporate fraud is driven by the use of SIC codes to deﬁne competitor networks, or by the lack of a
strong relationship between market concentration and fraud. In Column 5, we also control for the
natural log of the number of competitors each ﬁrm has according to its TNIC. Finally, in Column
6 we include product market ﬂuidity and sum similarity from Hoberg et al. (2014). Again, the
relationship between Average Similarity Score on fraud remains consistent in both signiﬁcance and

57

magnitude in Columns 4-6.The key takeaway from this analysis is that the alternative measures do
not appear to aﬀect the association between fraud and product diﬀerentiation. We explore several
combinations of control variables and diﬀerent sample periods and ﬁnd that these results are not
sensitive to model speciﬁcation.

In Table 3.7, we report estimates for the relationship between corporate fraud and traditional
measures of competition, excluding Average Similarity Score. We perform this exercise to ascertain
whether the traditional measures have an association in the absence of product diﬀerentiation,
which could be capturing some of the variation of these traditional measures. For each measure
of competition, we include our full set of control variables from Table 3.3 (described in section
II).11 We estimate the speciﬁcations in Table 3.7 without the inclusion of industry ﬁxed eﬀects to
provide the best chance at highlighting a statistical relationship. The number of rivals in a ﬁrm’s
primary three-digit SIC industry (Ln NCOMP SIC3) is the only competition variable that exhibits a
statistically signiﬁcant relation to fraud (10% level) in Table 3.7. Interestingly, Top 4 Concentration
no longer exhibits a relation to fraud when Average Similarity Score is not included in the same
regression. In untabulated results, we ﬁnd that including industry (SIC3) ﬁxed eﬀects attenuates
the point estimates in Table 3.7 even further. This result highlights one potential reason for a lack
of strong evidence between product market characteristics and corporate fraud documented in the
literature.

At a minimum, the results reported in Tables 3.6 and 3.7 suggest that product similarity captures
a dimension of competition unrelated to these alternative empirical measures. As these alternative
measures are all designed to capture the degree of competition in an industry in various ways,
our results suggest that there is something unique and particularly important about the relationship
between fraud and the degree of similarity with rivals. While we cannot perfectly rule out that
product similarity is merely capturing competition more accurately than these alternative measures,
these results add conﬁdence to the benchmarking channel highlighted throughout our analyses.

11We also estimate speciﬁcations without control variables and report results. The results are substantively very
similar. Note, we include ﬁrm size as a control in all speciﬁcations since size is strongly related to measures of
competition and is a strong predictor of fraud (see Buzby, 1975; Reynolds and Francis, 2000; Graham et al., 2005)

58

3.5 Conclusion

Our paper empirically examines the relationship between product market diﬀerentiation and
the incidence of corporate fraud. Having rivals with signiﬁcant product market overlap can have
two potential eﬀects on ﬁrms’ incentives to commit fraud. On one hand, less product market
diﬀerentiation could facilitate the ability to evaluate common shocks faced by ﬁrms, enhancing
monitoring by external parties such as regulators, auditors, and investors.
In turn, enhanced
monitoring should increase the likelihood that committed fraud would be detected. Whether this
eﬀect would result in more or fewer cases of fraud being observed depends on the extent to which
managers respond to enhanced detection rates by committing less fraud. On the other hand, less
diﬀerentiation could foster more competition, leading ﬁrms to commit more fraud to boost their
own perceived relative performance.

We also show that events that could aﬀect the information environment of ﬁrms are associated
with a greater detection eﬀect for ﬁrms with lower pre-existing market discipline. We contend that
these results are largely due to the ability to benchmark ﬁrm performance when there are more
similar rivals with publicly available information. These results suggest that aspects of competition
faced by a ﬁrm have a disciplining eﬀect on the incentive to commit fraud.

Collectively, our paper provides new insight on how a particular aspect of competition, product
market diﬀerentiation, inﬂuences the incentives to commit fraud via the ability to benchmark a
ﬁrm against similar peers. Thus, our paper highlights the role of one market-based mechanism that
can aﬀect commission and detection of corporate fraud. Our results suggest that external parties
could focus eﬀorts on examining ﬁrms with fewer comparable rivals when looking for fraudulent
reporting.

59

APPENDICES

60

APPENDIX A

INVESTOR REACTION AND MUTUAL FUND MISCONDUCT

61

Figure 1.1: Regulatory Disclosure

This ﬁgure reports time-series of the total number of regulatory disclosures, and total number of regulatory disclosures
due to misconduct related to “Mutual Funds". Regulatory disclosures are from mandatory ADV ﬁling by mutual
fund management companies. The regulatory events only include the cases that are initially ﬁled against investment
advisers. The “Mutual Funds" related misconduct is identiﬁed using the item “Principal Product" reported in Form
ADV.

62

400600800100012001400Total Regulatory Event20012003200520072009201120132015YearTime-series of Total Regulatory Events050100150Total Mutual Funds Fraud Event20012003200520072009201120132015YearTime-series of Mutual Fund Fraud EventsFigure 1.2: Fund Flow Before and After Regulatory Disclosure

This ﬁgure reports, at the fund level, the level of investment adjusted net ﬂow (inﬂow and outﬂow) six months before
and six months after regulatory disclosure. Fund ﬂow data is calculated based on semi-annual N-SAR ﬁlings.

63

-.015-.01-.0050.005Style Adjusted Flow-505Regulatory DisclosureFlow-.015-.01-.0050.005Style Adjusted Inflow-505Regulatory DisclosureInflow-.008-.006-.004-.0020Style Adjusted Outflow-505Regulatory DisclosureOutflowSource: N-SAR formFlow before and after regulatory disclosureTable 1.1: Variable Description

This table provides a detailed description of all variables used in my analysis. The main variable of interest is MF
Fraud SEC, deﬁned as an indicator variable that takes value one if there is an SEC-initiated regulatory action taken
against mutual fund investment adviser company.

Variable
MF Fraud SEC

MF Fraud All

Other SEC Regulatory

Institutional Funds
Retail Funds
Enforcement Bank

Fund Flow

Expense Ratio

TNA

Bank Association

Direct-sold Fund

Broker Fund

Age
Management Fee

Past return
Capital Gains Overhang

Media

Fraud Severity

T N A f ,t−1

, where T N A f ,t is a fund’s total

T N A f ,t−T N A f ,t−1×(1+R f ,t)

Deﬁnitions
Indicator variable equals 1 if there is SEC initiated regulatory action taken
against mutual fund investment advisor company.
Indicator variable equals 1 if there is regulatory action (by any regulator)
taken against mutual fund investment advisor company.
Indicator variable equals 1 if there is SEC initiated regulatory action taken
against investment advisor company related to other products other than
“Mutual Funds".
Indicator variable equals 1 if it’s institutional share class
Indicator variable equals 1 if it’s retail share class.
Indicator variable equals 1 if there is enforcement action taken against fund
associated ﬁnancial conglomerate at quarter t.
Flow f ,t =
net asset at quarter t, and R f ,t is the fund’s return over the prior month.
Expense Ratio as of the most recently completed ﬁscal year. Represented
in decimal format. Ratio of total investment that shareholders pay for the
fund’s operating expenses, which include 12b-1 fees exp_ratio may include
waivers and reimbursements, causing it to appear to be less than the fund
management fee.
Quarterly TNA is equal to total assets minus total liabilities as of quarter-
end. Reported in millions of dollars.
Indicator variable equals 1 if there is fund is associated ﬁnancial conglom-
erate at quarter t. The association can be established through either of the
following ways, (1) fund and ﬁnancial conglomerate share common name;
(2) fund is a registered subsidiary of ﬁnancial conglomerate.
Indicator variable equals 1 if fund share class has no front or rear load and
the 12b-1 fee is less than 25 basis point.
Indicator variable equals 1 if fund share class has either front or rear load
or the 12b-1 fee is less than 25 basis point.
The number of years since fund was available to investors.
Management fee ($)/ Average Net Assets ($). The fee is calculated using
ratios based on the line items reported in the Statement of Operations. The
management fee can be oﬀset by fee waivers and/or reimbursements which
will make this value diﬀer from the contractual fees found in the prospectus.
Reimbursements can lead to negative Management Fees.
The cumulative return for each fund from t-12 to t-1.
Overhang of unrealized gains at the beginning of each month t, Overhangt =
Overhangt−1+(N AVt−N AVt−1)×Sharest−1+(Sharest−Sharest−1)×(N AVt−
Acg Price Paid for New Shares)
Indicator variable equals 1 if the regulatory disclosure is covered in major
business media outlet.
Factor variable equals 0 if monetary ﬁne is $ 0; equals 1 if monetary ﬁne is
above $ 0 and below $ 8 million dollars; equals 2 if monetary ﬁne is above
$8 million dollars.

64

Table 1.2: Descriptive Summary Statistics

This table reports descriptive summary statistics for primary variables of interest and other controls. These variables
include fund ﬂow, regulatory disclosures against investment management companies, regulatory disclosures related to
Mutual Fund, regulatory disclosures initiated by SEC, regulatory disclosures against ﬁnancial conglomerates, as well
as the other fund level control variables used in our analysis. Panel A is based on the monthly level data from CRSP
Survivorship-bias free database. Panel B presents the summary statistics of data from N-SAR ﬁlings. Sample period
is from year 2000 to 2016. The number of observations, means, standard deviations, and the 10th and 90th percentiles
for each variable are reported in this table. All deﬁnitions are provided in detail in Table 1.1.

Panel A: CRSP Sample

Broker Fund
Expense Ratio
ADV Regulatory
ADV MF Fraud
MF Fraud (SEC)
Insitutional Fund
Ln TNA
Past Return
Age
Management Fee
Flow
Enforement (Bank)

Panel B: NSAR Sample
MF Fraud (SEC)
Flow
Inﬂow
Outﬂow
Overhang
Past Return
Expense Ratio
Ln TNA

Num Obs Mean

Std.Dev.

10th Percentile

90th Percentile

581,266
558,074
581,266
581,266
581,266
581,266
579,651
544,183
581,266
562,895
569,354
581,266

0.582
0.011
0.028
0.004
0.002
0.314
5.315
0.063
11.431
0.588
0.003
0.001

0.493
0.006
0.164
0.059
0.041
0.464
1.614
0.180
9.969
0.333
0.040
0.037

0.000
0.004
0.000
0.000
0.000
0.000
3.314
-0.185
2.170
0.082
-0.029
0.000

1.000
0.019
0.000
0.000
0.000
1.000
7.519
0.267
21.923
0.989
0.038
0.000

Num Obs Mean

Std.Dev.

10th Percentile

90th Percentile

608,792
608,792
608,792
608,792
590,999
603,789
608,792
608,792

0.002
0.009
0.051
0.042
0.016
0.062
0.008
12.861

0.041
0.085
0.135
0.173
0.472
0.127
0.006
1.953

0.000
-0.027
0.000
0.001
-0.180
0.000
0.003
10.323

0.000
0.047
0.112
0.087
0.165
0.191
0.015
15.164

65

Table 1.3: Fund Flow and Regulatory Disclosure

In Panel A of this table, we report OLS estimates for fund ﬂows from month t to month t+1 and t+3 after regulatory
disclosure of investment adviser misconduct (initiated by SEC), which happens at month t, respectively. Fund inﬂows
and outﬂows in Panel A are directly obtained from N-SAR ﬁlings. Speciﬁcation (1) and (2)examines the eﬀect on fund
net ﬂows, speciﬁcation (3) and (4) examines fund inﬂows and speciﬁcation (5) and (6) examines fund outﬂows. Our
main variable of interest is the MF Fraud (SEC) at month t. In panel B, we report OLS estimates for fund ﬂows from
month t to month t+1 and t+3 after regulatory disclosure of investment management misconduct at time t. Panel B uses
CRSP survivorship-bias free mutual fund dataset. The results are reported separately for institutional share classes
and retail share classes. Speciﬁcation (1) and (2) includes all equity funds, speciﬁcation (3) and (4) includes only
institutional share classes and speciﬁcation (5) and (6) includes only retail share classes. All speciﬁcations include time
ﬁxed eﬀects. All other variables are deﬁned in Table 1.1. The t-statistics, calculated from standard errors clustered at
the fund management company level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance
(two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***, respectively.

Panel A

(1)

(2)

(3)

(4)

(5)

(6)

Flow

(t, t+1)

(t, t+3)

(t, t+1)

Inﬂow

(t, t+3)

(t, t+1)

Outﬂow

(t, t+3)

MF Fraud (SEC)

-0.010***
(-4.258)

-0.019***
(-3.174)

Past Fraud (1-year)

Past Fraud (5-year)

-0.003**
(-1.974)

-0.001
(-0.609)

-0.008
(-1.605)

0.001
(0.156)

-0.004*
(-1.794)

-0.001
(-1.218)

-0.001
(-0.747)

-0.012*
(-1.659)

-0.009*
(-1.878)

-0.001
(-0.132)

0.006**
(2.139)

-0.000
(-0.301)

-0.001
(-0.751)

Past Return

Ln TNA

-0.000***
(-3.735)

-0.001***
(-4.550)

-0.000**
(-2.502)

-0.001**
(-2.369)

-0.000**
(-2.011)

-0.001***
(-5.299)

-0.005***
(-6.413)

-0.003***
(-12.117)

-0.010***
(-10.062)

-0.003***
(-11.524)

Expense Ratio

-0.150**
(-2.241)

-0.541***
(-2.942)

-0.316***
(-3.543)

-1.742***
(-5.058)

-0.171**
(-2.134)

Age

Avg Style Flow

Past Flow
Year Month FE
Observations

Panel B

-0.006***
(-6.606)

-0.013***
(-5.260)

0.077
(0.964)

Yes
Yes

574168

0.177
(0.794)

Yes
Yes

523282

0.001
(0.788)

0.070
(0.951)

Yes
Yes

628194

0.005
(1.228)

0.267
(0.782)

Yes
Yes

587411

0.003***
(2.728)

0.030
(0.456)

Yes
Yes

628194

0.009
(1.060)

-0.006*
(-1.724)

-0.001
(-0.421)

-0.001**
(-2.264)

-0.008***
(-10.474)

-0.993***
(-4.299)

0.006**
(1.964)

0.254
(1.014)

Yes
Yes

587411

Yes
Yes

All Equity Funds

Institutional Shares

Retail Shares

Flow(t,t+1) Flow(t,t+3) Flow(t,t+1) Flow(t,t+3) Flow(t,t+1)

Flow(t,t+3)

MF Fraud (SEC)

-0.004***
(-2.672)

-0.007**
(-2.534)

-0.006*
(-1.977)

-0.013**
(-2.009)

-0.003**
(-2.316)

-0.006**
(-2.262)

Controls
Year Month FE
Observations
*Diﬀerence between ﬂow from institutional shares and retail shares are statistically signiﬁcant, p-value = 0.075

487614

345270

516873

155694

142344

361179

Yes
Yes

Yes
Yes

Yes
Yes

Yes
Yes

Yes
Yes

66

Table 1.4: Sales Channel and Regulatory Disclosures

In this table we report OLS estimates of ﬂows from t to t+1 and t+3, and change in management fee from month t to
month t+12 and t+24 after regulatory disclosure of investment management misconduct at time t, for retail share classes.
The dependent variable, ∆MgmtFee(t +12) (∆MgmtFee(t +24)), is calculated using MgmtFee(t +12)− MgmtFee(t)
(MgmtFee(t + 24) − MgmtFee(t)). Panel A reports the fund ﬂow following the regulatory disclosures, separately for
broker and direct sold funds. Panel B reports the change of management fees following the regulatory disclosures,
separately for broker and direct sold funds. All speciﬁcations include time ﬁxed eﬀect and investment style ﬁxed
eﬀects. All other variables are deﬁned in Table 1.1. The t-statistics, calculated from standard errors clustered at
the fund management company level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance
(two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***, respectively.

Panel A

MF Fraud (SEC)

Controls
Year Month FE
Observations

Panel B

MF Fraud (SEC)

Controls
Year Month FE
Observations

(1)

(2)

(3)

(4)

Broker

Flow(t,t+1)

Flow(t,t+3)

Flow(t,t+1)

Direct

Flow(t,t+3)

-0.003**
(-2.272)

Yes
Yes

280864

-0.006**
(-2.167)

Yes
Yes

268899

-0.002
(-0.334)

Yes
Yes
80315

-0.009
(-0.921)

Yes
Yes
76371

∆ Mgmt Fee (t,t+12) ∆ Mgmt Fee (t,t+24) ∆ Mgmt Fee (t,t+12) ∆ Mgmt Fee (t,t+24)

Broker

Direct

-0.008**
(-2.372)

Yes
Yes

219498

-0.008*
(-1.877)

Yes
Yes

193198

0.004
(0.547)

Yes
Yes
58107

0.014
(0.979)

Yes
Yes
51735

67

Table 1.5: Fund Flow and Regulatory Disclosure - Visibility

In this table we report OLS estimates for fund ﬂow from month t to month t+1 and t+3 after regulatory disclosure of
investment management misconduct at time t. Panel A reports separately the funds depend on the size of the fund
complex which fund belongs to. Size of fund complex is calculated based on total net assets of all fund share classes
within each fund family. Panel B reports separately the ﬂows of fund depend on the past performance of the fund
complex which fund belongs to. Star family is calculated following Nanda et al. (2004). All speciﬁcations include time
ﬁxed eﬀects. All other variables are deﬁned in Table 1.1. The t-statistics, calculated from standard errors clustered at
the fund management company level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance
(two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***, respectively.

Panel A:
Family Size

MF Fraud (SEC)

Controls
Year Month FE
Observations

Panel B:
Family Performance

(1)

(2)

(3)

(4)

Small

Large

Flow(t,t+1) Flow(t,t+1) Flow(t,t+3) Flow(t,t+3)

Small

Large

0.000
(0.024)

Yes
Yes

177,326

-0.004**
(-2.488)

Yes
Yes

183,485

0.005
(1.056)

Yes
Yes

168,033

-0.011***
(-3.164)

Yes
Yes

176,895

Non-star
Flow(t,t+1) Flow(t,t+1) Flow(t,t+3) Flow(t,t+3)

Non-star

Star

Star

MF Fraud (SEC)

Controls
Year Month FE
Observations

-0.002
(-1.609)

Yes
Yes

202,131

-0.004*
(-1.801)

Yes
Yes

158,882

-0.003
(-1.225)

Yes
Yes

194,006

-0.011**
(-2.143)

Yes
Yes

151,117

68

Table 1.6: Fund Flow and Regulatory Disclosure – Fraud Severity

In this table, we report OLS estimates for fund ﬂow from month t to month t+1 after regulatory disclosure of investment
management misconduct at time t. Speciﬁcation (1) and (2) report the results for the regulatory disclosure from
SROs and other state and federal regulators other than the SEC. Speciﬁcation (3) reports the fund ﬂows response for
repeated oﬀenses. Speciﬁcation (4) reports the fund ﬂows response for severe fraud charges. Speciﬁcation (5) reports
the fund ﬂows response for mutual fund fraud covered by the major ﬁnancial media. All speciﬁcations include time
ﬁxed eﬀects. All other variables are deﬁned in Table 1.1. The t-statistics, calculated from standard errors clustered at
the fund management company level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance
(two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***, respectively.

(1)

(2)

(3)

Flow(t,t+1) Flow(t,t+1) Flow(t,t+1) Flow(t,t+1) Flow(t,t+1)

(4)

(5)

0.001
(0.512)

-0.000
(-0.080)

-0.006*
(-1.744)

-0.003*
(-1.653)

Yes
Yes

351,516

0.000
(0.014)

-0.003
(-1.584)

-0.002*
(-1.697)
-0.004*
(-1.679)

MF Fraud SRO

MF Fraud State

MF Fraud

Repeated Oﬀense

Severe Fraud

Media Coverage

Controls
Year Month FE
Observations

Yes
Yes

361,176

Yes
Yes

361,176

Yes
Yes

361,176

Yes
Yes

361,176

69

-0.004**
(-2.530)

-0.010***
(-2.777)

Yes
Yes

106692

Yes
Yes
97651

-0.000
(-0.005)

Yes
Yes

255141

-0.010*
(-1.748)

-0.000
(-0.115)

Yes
Yes

243570

Bank Aﬃliate

Controls
Year Month FE
Observations

Panel B:

Table 1.7: Fund Flow and Regulatory Disclosure – Information Quality

In Panel A of this table we report OLS estimates for fund ﬂow from month t to month t+1, and t+3 after SEC
enforcement action against the ﬁnancial conglomerate at time t, separately for institutional share class and retail share
class. Our regressor of interest is Enforcement Bank at t, and we identify the fund-ﬁnancial institution association
through common name or subsidiary relationship. In Panel B of this table we report OLS estimates for fund ﬂow
from month t to t+1, and t+3 after regulatory disclosure at time t that is unrelated mutual fund product, separately for
institutional share class and retail share class. Our regressor of interest is Other SEC Regulatory at t. All speciﬁcations
include time ﬁxed eﬀects. All other variables are deﬁned in Table 1.1. The t-statistics, calculated from standard errors
clustered at the fund management company level, are reported in parentheses below coeﬃcient estimates. Statistical
signiﬁcance (two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***, respectively.

Panel A:

(1)

(2)

(3)

(4)

Institutional

Retail

Flow(t,t+1) Flow(t,t+3) Flow(t,t+1) Flow(t,t+3)

Enforcement Bank

0.007
(1.040)

0.017
(1.393)

-0.004**
(-2.250)

Institutional

Retail

Flow(t,t+1) Flow(t,t+3) Flow(t,t+1) Flow(t,t+3)

Other SEC Regulatory

0.003
(1.335)

-0.002
(-0.489)

-0.001
(-0.841)

-0.004**
(-2.231)

Controls
Year Month FE
Observations

Yes
Yes

155694

Yes
Yes

142344

Yes
Yes

361179

Yes
Yes

345270

70

Table 1.8: Fund Flow and Regulatory Disclosure - Monitoring Eﬀort

In this table we report OLS estimates for fund ﬂow from month t to month t+1, and t+3 after after regulatory disclosure
of investment management misconduct at time t. Panel A reports the fund ﬂows reposes of retail share classes separately
for institutional twins funds and others. Institution Twin funds are deﬁned based on the proportion of institutional
assets within a fund. Retail funds are identiﬁed as having an institutional twin if the proportion of institution assets
is above median. Panel B reports the fund ﬂow responses separately for funds depends on the monitoring intensity.
Monitoring intensity if calculated by the standard deviation of fund outﬂows. Panel B restrict the funds from N-SAR
sample of funds with very limited institutional clients. All speciﬁcations include time ﬁxed eﬀects. All other variables
are deﬁned in Table 1.1. The t-statistics, calculated from standard errors clustered at the fund management company
level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance (two-sided) at the 10% 5%, and
1% level is denoted by *, **, and ***, respectively.

Panel A:

MF Fraud SEC

Controls
Year Month FE
Observations

Panel B:
Monitoring

(1)

(2)

(3)

(4)

Other

Twin

Other

Twin

Flow(t,t+1) Flow(t,t+1) Flow(t,t+3) Flow(t,t+3)

-0.002*
(-1.889)

Yes
Yes

252,338

-0.005*
(-1.715)

Yes
Yes

108,838

-0.005*
(-1.850)

Yes
Yes

241,400

-0.010*
(-1.724)

Yes
Yes

103,869

Low

Flow(t,t+1) Flow(t,t+1) Flow(t,t+3) Flow(t,t+3)

High

Low

High

MF Fraud SEC

-0.0026
(-1.3337)

-0.0192***
(-2.9437)

-0.0011
(-0.2466)

-0.0540***
(-3.1444)

Controls
Year Month FE
Observations

Yes
Yes
83,885

Yes
Yes
90,360

Yes
Yes
76,541

Yes
Yes
78,527

71

Table 1.9: Fund ﬂow and Regulatory Disclosure – Investor Mobility

In this table we report OLS estimates for fund ﬂows (inﬂows, and outﬂows) from month t to t+1, after regulatory
disclosure of investment management misconduct at time t. Panel A reports separately for funds grouped by the capital
gains overhang. Fund ﬂow data are directly obtained from N-SAR ﬁlings. For each fund each month, capital gains
overhang is calculated following the procedure described in Barclay et al. (1998). At the end of each month, we split
funds into terciles based on its level of capital gains overhang. Panel B reports separately for sub-sample of late-trading
scandal period, year 2003 - 2005, and ﬁnancial crisis period, year 2007 - 2009. In unreported analysis, similar results
are found for fund ﬂows from t to t+3. All speciﬁcations include time ﬁxed eﬀects, and controls used in Table 1.3. All
variables are deﬁned in Table 1.1. The t-statistics, calculated from standard errors clustered at the fund management
company level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance (two-sided) at the 10%
5%, and 1% level is denoted by *, **, and ***, respectively.

Panel A:

Overhang

(1)

Low

(2)

Flow(t,t+1)
Mid

(3)

High

MF Fraud (SEC)

-0.009***
(-2.917)

-0.012***
(-2.896)

Controls
Year Month FE
Observations

Yes
Yes

211042

Yes
Yes

148502

-0.012**
(-2.412)

Yes
Yes

206174

Panel B:

All

Flow(t,t+1)

Late-Trading Financial Crisis

MF Fraud (SEC)

-0.010***
(-4.234)

-0.015***
(-4.199)

Controls
Year Month FE
Observations

Yes
Yes

574,168

Yes
Yes
99,102

-0.003
(-0.704)

Yes
Yes

106,812

(4)

Low

-0.002
(-0.715)

Yes
Yes

232151

All

-0.004*
(-1.791)

Yes
Yes

628,194

(5)

Inﬂow(t,t+1)
Mid

(6)

High

(7)

Low

-0.002
(-0.784)

Yes
Yes

159151

-0.010***
(-3.273)

0.008***
(3.018)

Yes
Yes

227170

Yes
Yes

232151

0.007*
(1.786)

Yes
Yes

159151

(8)

(9)

Outﬂow(t,t+1)

Mid

High

0.006
(0.783)

Yes
Yes

227170

Inﬂow(t,t+1)

Late-Trading Financial Crisis

-0.002
(-0.659)

Yes
Yes

109,275

-0.011**
(-2.084)

Yes
Yes

115,943

All

0.006**
(2.124)

Yes
Yes

628,194

Outﬂow(t,t+1)

Late-Trading Financial Crisis

0.014***
(3.099)

Yes
Yes

109,275

-0.006
(-1.081)

Yes
Yes

115,943

72

Table 1.10: List of Largest Financial Conglomerate in the U.S.

This table reports the list of ﬁnancial conglomerates used in the analyses. The list includes the top 40 ﬁnancial
institutions in the U.S. based on total assets.

Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

Name
JPMorgan Chase
Bank of America
Wells Fargo
Citigroup
Goldman Sachs
Morgan Stanley
U.S. Bancorp
PNC Financial Services
Capital One
TD Bank, N.A.
The Bank of New York Mellon
Barclays
HSBC Bank USA
State Street Corporation
Charles Schwab Corporation
BB&T
Credit Suisse
SunTrust Banks
Deutsche Bank
Ally Financial
American Express
Citizens Financial Group
MUFG Union Bank
USAA
Fifth Third Bank
Royal Bank of Canada
UBS
Santander Bank
KeyCorp
BNP Paribas
BMO Harris Bank
Regions Financial Corporation
Northern Trust Corporation
M&T Bank
Huntington Bancshares
Discover Financial
Synchrony Financial
BBVA Compass
First Republic Bank
Comerica

73

Table 1.11: Fund ﬂow before and after Regulatory Event

This table presents t-test of style adjusted fund ﬂows (inﬂows, outﬂows), six months before and after regulatory event
disclosure. Fund ﬂow data are from N-SAR ﬁlings for year 2000 to 2016. The sample only includes equity funds. Net
fund ﬂows are calculate by subtracting outﬂows from inﬂows. All ﬂow measures are scaled by funds’ total net assets
from last period. Investment style of a fund is reported directly in N-SAR formS. For each month, style adjusted ﬂows
are calculated using the diﬀerence between fund ﬂows and average ﬂows of all funds within the same investment style.

Style Adjusted Flow

Before
-0.0012

After
Diﬀ
-0.0079 0.0067

t-stat

6.225***

Style Adjusted Inﬂow

-0.0040 -0.0129

0.0089

5.796***

Style Adjusted Outﬂow -0.0033

-0.0048 0.0015

1.508

74

Table 1.12: Fund Flow and Regulatory Disclosure - Sales Loads

In this table, we report OLS estimates for fund inﬂow and out from t to t+1 after regulatory disclosure about investment
management misconduct at time t, separately for broker-sold funds with front load, broker-sold funds with rear load,
and direct-sold funds using NSAR data. Our regressor of interest is the MF Fraud SEC at month t. Speciﬁcation (1) and
(4) includes only broker-sold funds with positive front load, speciﬁcation (2) and (5) includes only broker-sold funds
with positive rear load, and speciﬁcation (3) and (6) includes only direct-sold funds .. All speciﬁcations include time
ﬁxed eﬀects. All other variables are deﬁned in Table 1.1. The t-statistics, calculated from standard errors clustered at
the fund management company level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance
(two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***, respectively.

(1)

(2)

(3)

(4)

(5)

(6)

Outﬂow (t,t+1)

Inﬂow (t,t+1)

Frontload Rearload Direct

Frontload Rearload Direct

MF Fraud SEC

-0.002
(-0.634)

-0.001
(-0.292)

-0.003
(-1.437)

0.013**
(2.311)

0.013**
(2.065)

0.004
(1.500)

Controls
Past Flow
Year Month FE
Observations

Yes
Yes
Yes

124080

Yes
Yes
Yes
84281

Yes
Yes
Yes

437428

Yes
Yes
Yes

124080

Yes
Yes
Yes
84281

Yes
Yes
Yes

437428

75

APPENDIX B

CORPORATE CULTURE AND CORPORATE FRAUD

76

Table 2.1: Description of the Variables

This table provides a detailed description of all variables used in our analysis. Our main variable of interest is the
AshleyMadison Active Accounts (AM Active Accounts) deﬁned as the natural log of the number of active accounts at
the end of year t (plus one).

Panel A - AshleyMadison Variables
AM Accountsi,t

The total number of AM accounts for ﬁrm i in year t. An account does
not have to have sent a message or purchased credits to be included in this
calculation for a given year. That is, the account does not have to be active.
ActiveAM Accountsi,t The total number of active AM accounts for ﬁrm i in year t. An account is
required to have sent a message or purchased credits to be included in this
measure. If an account is deactivated, then it is excluded from the calculation
in a given year, but still included up until the year of its deactivation. This is
our main variable of interest throughout the text.

Total debt divided by book value of assets. [(dltt+dlc)/at]

ROAi,t
Tangibilit yi,t
#o f Employeei,t
FirmAgei,t
Cash/Asseti,t
LogMarketCapi,t
HHI(sic4)i,t
∆OROAi,t

Panel B - Firm Financial Information
BookLeveragei,t
Debt/MarketEquit yi,t Total debt divided by market value of equity. [(dltt + dlc) / (prcc_f*csho)]
R&D/Sales
R&D expenditures divided by sales. [xrd/sale]
Tobin(cid:48)sQi,t
Total asset minus book value of equity plus the market value of equity divided
by total assets [(at - ceq + me)/at ]
MarkettoBookratioi,t Market value of ﬁrms’ equity divided by the book value of equity, following
Fama-French calculation of book equity [prcc_f * csho / teq - prefered +
txditc]
Return on Asset. [oibdp/l.at]
Net Property, Plant and Equipment divided by total assets [ppent/at]
The natural log of the total number of employee [log(emp)]
Firm age reported in Compustat or the number of years ﬁrm is observed in
Compustat
Cash and short-term investment divided by Assets [(ch + ivst)/at]
Natural log of market cap [log(csho*prcc_f)]
Herﬁndahl index based on sales within 4-digit SIC industries in year t
Diﬀerence between the average operating income scaled by total assets 3
years before and after New CEO was appointed
Diﬀerence between the average operating income scaled by sales 3 years
before and after New CEO was appointed
((∆Current Assets - ∆Cash and Short-term Investments)- (∆Current Liabil-
ities - ∆Debt in Current Liabilities - ∆Taxes Payable) / Average total assets
(∆WC + ∆NCO + ∆FIN) /Average total assets, where WC = (Current Assets -
Cash and Short-term Investments ) - (Current Liabilities - Debt in Current Li-
abilities ); NCO = (Total Assets - Current Assets - Investments and Advances)
- (Total Liabilities - Current Liabilities -Long-term Debt); FIN=(Short-term
Investments +Long-term Investments) - (Long-term Debt+Debt in Current
Liabilities+Preferred Stock)

∆OROSi,t
WC Accruals

RSST Accruals

∆Inventory / Average total assets
(Total Assets - PP&E - Cash and Cash Equivalent) /Total Assets
Percentage change in cash sales (Sales - ∆Accounts Receivable)

Change in receivables ∆Accounts Receivable / Average total assets
Change in inventory
% Soft assets
Change in cash sales
Change in cash margin Percentage change in cash margin, where cash margin is measured as: 1 -
((Cost of Good Sold - ∆Inventory + ∆Accounts Payable) / (Sales - ∆ Accounts
Receivable))
(Earnings t
assets(t−1))
A dummy variable takes value to be 1 if the ﬁrm issued securities during
year t (i.e., A dummy variable takes value to be 1 if ”Sale of Common and
Preferred Stock" > 0 or “Long-Term Debt - Issuance" > 0)

/ Average total assets t ) - (Earnings(t−1)

/ Average total

Actual issuance

Change in ROA

AshleyMadison

AshleyMadison

Compustat
Compustat
Compustat
Compustat

Compustat

Compustat
Compustat
Compustat
Compustat

Compustat
Compustat
Compustat
Compustat

Compustat

Compustat

Compustat

Compustat
Compustat
Compustat
Compustat
Compustat

Compustat

Compustat

77

Table 2.1 (cont’d): Description of the Variables

Panel C - Misstatement Variables
Misstatement

Panel D - Ethics Variables
Briber y and Fraud

Tax Disputes
Cash/StockSharing

Human Rights

Product Qualit y

A dummy variable takes value to be 1 if ﬁrm is during or at the conclusion
of an investigation against a company, an auditor, or an oﬃcer for alleged
accounting and/or auditing misconduct

A discrete variable that indicates the severity of controversies related to a
ﬁrm’s business ethics practices, including bribery, and fraud.
A discrete variable that indicates whether companies have recently been
involved in major tax disputes involving Federal, state, local or
A discrete variable that indicates whether companies have a cash proﬁt-
sharing program through which it has recently made distributions to a sig-
niﬁcant proportion of its workforce. This variable also indicates whether
companies encourage worker involvement via generous employee stock own-
ership plans (ESOPs) or employee stock purchase plans (ESPPs)
A discrete variable that is the net measure of positive features and negative
features regarding human rights for a corporation. Positive features include
quality labor rights, a strong relationship with indigenous peoples in foreign
operations, and other human rights strengths. Negative features include
human rights violations, including freedom of expression and censorship
concerns, indigenous peoples relations concerns, labor rights concerns, op-
erations in Sudan, Mexico , Burma, Norther Ireland and South Africa, and
other human rights concerns.
A discrete variable that is the net measure of positive features and nega-
tives features regarding product category. Positive features include insuring
health and demographic risk, responsible investment, strong privacy and
data security, ﬁnancial product safety, chemical safety, opportunities in nu-
trition and health, access to communications, access to capital, beneﬁts to
economically disadvantaged, R&D innovation, and other product strengths.
Negative features include customer relations concerns, antitrust concerns,
marketing-contracting concerns, product safety concerns, and other product
concerns.

AAER

KLD

KLD

KLD

KLD

KLD

78

This table presents summary statistics for AshleyMadison (AM) variables (panel A), as well as the other variables used in our analysis. The AM data cover the
sample period 2002-2014. We report the number of observations, means, standard deviations, and the 10th and 90th percentiles for each variable. All deﬁnitions
are provided in detail in the Internet Appendix.

Table 2.2: Descriptive Statistics

Variable

Mean

σ

10pct

90pct

N

Panel A: AshleyMadison
Active AM Accounts
all ﬁrms
ﬁrms ≥ 1 account

2.052
5.391
0.712
39.238
0.169

Average Years of Activity
Average Age of AM User
Average Credits
Panel B: Firm and Industry Characteristics
Book Leverage
Debt/Market Equity
Log Market Cap
Tobin’s Q
Market to Book Ratio
ROA
Tangibility
# of Employee
Firm Age
Log Sales
Cash/Asset
Volatility-3 Factor adjusted
HHI(sic4)
Stock Return
Skewness
CDS Spread

0.227
0.645
5.490
2.575
2.719
-0.042
0.233
8.564
17.306
5.152
0.227
0.027
0.122
0.126
0.368
0.026

12.130
19.200
1.105
7.659
6.562

0.254
1.923
2.312
3.131
6.465
0.593
0.228
38.860
12.266
2.626
0.235
0.012
0.176
0.533
0.742
0.076

Variable

Panel D: Governance
Bribery and Fraud
Human Rights Violations
Tax Disputes
Cash/Stock Sharing
Product Quality

Mean

σ

10pct

90pct

N

0.047
-0.032
0.021
0.144
-0.018

0.212
0.262
0.143
0.403
0.294

0.000
0.000
0.000
0.000
0.000

0.000
0.000
0.000
1.000
0.000

3428
18090
11168
14095
18096

Panel G: AAER data
Misstatement

Panel F: Patent data
Ln(R&D)
Patent Cites
Patents
Pat/R&D
Top 10
Pdiv
Cdiv
ACdiv

0.007

0.083

0.000

0.000

40712

1.400
0.079
0.051
0.006
0.023
0.043
0.043
0.043

1.798
0.307
0.259
0.137
1.455
0.182
0.183
0.183

0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000

4.108
0.000
0.000
0.000
0.000
0.000
0.000
0.000

40712
32666
32666
14121
32666
32666
32666
32666

Panel G: Tax Avoidance
MostlyTxh50
MostlyTxh75
MostlyTxh90

0.296
0.279
0.278

0.456
0.449
0.448

0.000
0.000
0.000

1.000
1.000
1.000

10779
10779
10779

0.000
0.000
0.000
30.000
0.000

4.000
34961
10.000 13306
2.000
13303
49.000 13231
0.000
11560

0.000
0.000
2.420
0.953
0.202
-0.385
0.025
0.035
5.000
1.694
0.014
0.013
0.001
-0.486
-0.372
0.003

40712
0.577
40712
1.281
40712
8.416
40712
4.631
40712
6.665
40712
0.259
0.604
40712
17.000 40712
37.000 39056
40712
8.333
40712
0.604
34247
0.044
0.337
40712
34252
0.753
33982
1.119
0.050
4262

79

Table 2.3: AAER Misstatements and AshleyMadison Membership

In Panel A, we report coeﬃcient estimates and marginal eﬀects for logistic regressions of accounting misstatements
on the number of active AshleyMadison (AM) accounts. Data on misstatements and speciﬁcations from 2002-2014
come from the AAER dataset discussed in Dechow, Ge, Larson, and Sloan (2011). This dataset provides detailed
information regarding misstatement investigations for public corporations. Speciﬁcation 1, 3, and 5 reports estimates
for the natural logarithm of the number of AM accounts, while Speciﬁcations 2,4, and 6 use a dummy variable equal
to one if number of AM accounts is greater than or equal to one, and zero otherwise. We include four accruals-related
measures. WC accruals, focuses on working capital accruals and is described in Allen, Larson, and Sloan (2009).
RSST accruals are deﬁned in Richardson et. al. (2005) and Dechow et. al. (2011) and extends the deﬁnition of WC
accruals to include changes in long term operating assets and liabilities. Change in receivables (Change in inventory) is
deﬁned in Dechow et. al. (2011) as the change in accounts receivables (inventory) normalized by average Total Assets.
Soft Assets is deﬁned in Barton and Simko (2002) as Total Assets minus PP&E minus Cash and Cash Equivalent, all
normalized by Total Assets. Performance variables include Change in cash sales (deﬁned as the percentage change
in sales minus the change in Accounts receivables), Change in cash margin (percentage change), and Change in ROA
(deﬁned as ROA(t) minus ROA(t-1)). Actual Issuance is a dummy equal to one if the ﬁrm issued securities during year
t, and zero otherwise. Speciﬁcations (3) and (4) include the Abnormal change in employees (deﬁned as the percentage
change in the number of employees minus the percentage change in assets), and Dummy Lease (deﬁned as one if future
operating lease obligations are greater than zero, and zero otherwise). Speciﬁcations (5) and (6) also include current
and lagged market-adjusted stock return and logarithm of number of employees. The (Lagged) market-adjusted stock
return is the (previous year) annual buy-and-hold return minus the buy-and-hold CRSP value-weighted index return.
All speciﬁcations include year and industry ﬁxed eﬀects, and all explanatory variables are lagged by one year. The
t-statistics, calculated from standard errors clustered at the year level, are reported in parentheses below coeﬃcient
estimates. Statistical signiﬁcance (two-sided) at the 1% 5%, and 10% level is denoted by *, **, and ***, respectively.
Marginal eﬀects are reported below the estimates in square brackets, and are multiplied by 100. In Panel B, we report
estimates from a matched-sample for the diﬀerence in the number of misstatements between ﬁrms with positive AM
membership and ﬁrms with zero AM accounts. We report the means for both the AM-sample (number of accounts
greater than zero) and the matched sample, as well as the diﬀerence, the number of matched pairs, the t-statistic, and
the p-value. We detail our matching procedure in section 3.1

80

Table 2.3 (cont’d): AAER Misstatements and AshleyMadison Membership

Panel A: Regression

Active AM Account

Dummy(AM>0)

RSST Accruals

Change in receivables

Change in inventory

% Soft assets

Change in cash sales

Change in ROA

Change in cash margin

Actual issuance

Abnormal change in employees

Existence of operating leases

Market adjusted stock return t

Market adjusted stock return t−1

Log # of employees

Industry FE
Year FE

Pseudo R-Square
Observations

Panel B: Matched Sample

Misstatement

(1)

0.390 ***
[0.658]
(4.18)

0.139 ***
[0.234]
(4.1)
-1.435
[-2.422]
(-1.31)
2.156
[3.64]
(1.47)

1.931 ***
[3.259]
(6.65)

0.000 ***
[0.000]
(-3.96)

-0.010 ***
[-0.016]
(-2.92)
0.000 ***
[0.000]
(4.52)
0.802 *
[1.353]
(1.87)

(2)

0.628 ***

[1.06]
(7.98)

0.139 ***
[0.235]
(4.38)
-1.420
[-2.397]
(-1.24)
2.149
[3.629]
(1.46)

1.977 ***
[3.338]
(7.13)

0.000 ***
[0.000]
(-3.95)

-0.009 ***
[-0.016]
(-2.79)
0.000 ***
[0.000]
(4.31)
0.790 *
[1.334]
(1.79)

(3)

0.383 ***
[0.656]
(4.02)

0.139 ***
[0.238]
(4.72)
-1.595
[-2.73]
(-1.4)
2.153
[3.684]
(1.45)

1.945 ***
[3.328]
(6.66)

0.000 ***
[0.000]
(-3.81)

-0.010 ***
[-0.018]
(-3.04)
0.000 ***
[0.000]
(4.1)
0.817 *
[1.398]
(1.88)
-0.071
[-0.121]
(-0.55)
-0.449
[-0.768]
(-0.38)

(4)

0.612 ***
[1.048]
(7.49)

0.140 ***

[0.24]
(4.81)
-1.571
[-2.69]
(-1.32)
2.142
[3.666]
(1.44)

1.990 ***
[3.407]
(7.09)

0.000 ***
[0.000]
(-3.82)

-0.010 ***
[-0.017]
(-2.81)
0.000 ***
[0.000]
(4.08)
0.808 *
[1.383]
(1.81)
-0.065
[-0.111]
(-0.54)
-0.466
[-0.797]
(-0.39)

Yes
Yes

0.194
27775

Yes
Yes

0.194
27775

Mean (AM) Mean (Matched)

1.186

0.678

Yes
Yes

0.196
27325

Number of Matched Pairs

4132

Yes
Yes

0.196
27325

∆Mean
0.508

81

(5)

0.249 *
[0.447]
(1.87)

-0.515 *
[-0.924]
(-1.82)

-2.365 ***
[-4.241]
(-3.49)
3.336 ***
[5.982]
(3.1)

2.407 ***
[4.317]
(5.7)

0.000 ***
[0.000]
(-5.29)
0.115
[0.207]
(0.47)

0.000 ***
[0.000]
(4.92)
0.720
[1.291]
(1.04)
-0.287 *
[-0.515]
(-1.84)

-2.555 ***
[-4.583]
(-2.79)
0.297
[0.532]
(0.92)
0.306
[0.549]
(1.26)

0.119 ***
[0.213]
(2.63)
Yes
Yes

(6)

0.469 ***
[0.842]
(4.66)
-0.535 *
[-0.959]
(-1.85)

-2.374 ***
[-4.256]
(-3.35)
3.364 ***

[6.03]
(3.23)

2.451 ***
[4.394]

(6)

0.000 ***
[0.000]
(-5.2)
0.119
[0.213]
(0.48)

0.000 ***
[0.000]
(4.84)
0.724
[1.297]
(1.00)
-0.301 *
[-0.54]
(-1.92)

-2.603 ***
[-4.667]
(-2.78)
0.286
[0.513]
(0.88)
0.295
[0.529]
(1.24)

0.127 ***
[0.227]
(4.00)
Yes
Yes

0.219
22557

t-stat
1.780

0.220
22557

P-Value
0.075

Table 2.4: KLD Ethics Ratings and AshleyMadison Membership

In this table we report OLS estimates for KLD ratings of ﬁrm behavior on the number of active AshleyMadison (AM) accounts. KLD ratings are annual company
performance indicators with respect to meeting stakeholder needs regarding environmental, social, and governance factors. The indicators are developed by MSCI
analysts who provide research for institutional investors. The KLD data are described in greater detail in section 2.2. As the dependent variable we use the number
of positive ratings minus the number of negative ratings within a given KLD category. Our regressor of interest is the natural logarithm of one plus the number of
active AM accounts for a given ﬁrm year. All regressors are lagged one year relative to our KLD measures. All other variables are deﬁned in the Internet Appendix.
The t-statistics, calculated from standard errors clustered at the ﬁrm level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance (two-sided)
at the 1% 5%, and 10% level is denoted by *, **, and ***, respectively.

(5)

(6)

(7)

(8)

(9)

(10)

Human Rights Human Rights Product Quality Product Quality Cash/Stock Sharing Cash/Stock Sharing

0.0303 ***

(2.594)

-0.0342
(-1.416)
-0.00513
(-1.145)
-0.0480
(-1.465)
0.0106
(1.261)
1.632
(1.630)
-0.00952
(-1.201)
0.0801 ***

(7.601)
0.110 **
(2.304)

Yes
Yes
Yes
10,935
0.241

0.0285 *
(1.798)
-0.0235
(-1.058)
-0.00247
(-0.588)
-0.0427
(-1.453)
0.0102
(1.277)
1.779 *
(1.896)
0.000185
(0.0240)
0.0772 ***

(7.663)
0.0820 *
(1.844)

Yes
Yes
Yes
10,935
0.227

Panel A: Regression

VARIABLES

Active AM Account

Dummy(AM>0)

Book Leverage

Tobin’s Q (t-1)

EBITDA/Assets (t-1)

Annual Return

Volatility

Log # of Employee

Market Cap (t-1)

Tangibility (t-1)

(1)

(2)

(3)

Business Ethics Business Ethics Tax Disputes

0.0331 ***

(4.672)

-0.0235
(-1.599)

-0.00442 **

(-2.023)
-0.00967
(-0.553)
-0.0234 **
(-2.508)
1.101
(1.255)

0.0262 ***

(4.383)
0.00514
(0.834)

-0.0917 **
(-2.506)

0.0241 ***

(2.735)
-0.0234
(-1.577)

-0.00564 **

(-2.524)
-0.0171
(-0.954)
-0.0235 **
(-2.468)
1.320
(1.500)

0.0321 ***

(5.063)
0.0104 *
(1.721)

-0.0948 ***

(-2.587)

0.0190 ***

(2.759)

0.00344
(0.264)
-0.000185
(-0.127)
-0.01000
(-1.117)
0.00231
(0.855)
0.829 *
(1.777)
0.00553 *
(1.815)

0.0144 ***

(3.855)
0.0173
(0.681)

(4)

Tax Disputes

0.0166 **
(2.025)
0.00247
(0.191)
-0.000238
(-0.162)
-0.0128
(-1.414)
0.00302
(1.148)
0.918 **
(1.990)

0.00813 ***

(2.666)

0.0164 ***

(4.305)
0.0126
(0.504)

-0.0155 **
(-2.333)

0.000621
(0.0552)
-0.000766
(-0.498)
0.0110
(1.191)

-0.0109 **
(-2.029)
-0.0782
(-0.175)

-0.0168 ***

(-3.798)
-0.00470
(-1.026)
0.0738 ***

(2.966)

-0.00332
(-0.517)
0.000357
(0.0316)
-0.000469
(-0.307)
0.0140
(1.517)

-0.0115 **
(-2.135)
-0.160
(-0.360)

-0.0199 ***

(-4.257)
-0.00703
(-1.552)
0.0777 ***

(3.128)

-0.0141 *
(-1.919)

-0.00297
(-0.196)
0.00113
(0.511)
0.0137
(0.950)
0.00360
(0.666)
-0.868
(-1.448)

-0.0132 ***

(-2.670)
0.00897
(1.627)
0.0736 **
(2.127)

-0.0120 **
(-2.082)
-0.00317
(-0.304)
0.00136
(0.695)
0.0162
(1.221)
0.00325
(0.542)
-0.948 **
(-1.993)

-0.0151 ***

(-4.572)
0.00705 *
(1.932)

0.0759 ***

(3.579)

EA FE
Year FE
Industry(SIC2) FE
Observations
R-squared
Panel B: Matched Sample

Yes
Yes
Yes
3,168
0.219

Yes
Yes
Yes
8,203
0.208

Mean (AM) Mean (Matched)

Business Ethics
Tax Disputes
Human Rights
Product Quality
KLD_sharing

5.323
3.512
-3.952
-1.084
22.969

3.387
1.171
-2.343
-0.874
13.864

Yes
Yes
Yes
10,935
0.241

Yes
Yes
Yes
10,935
0.227

Yes
Yes
Yes
8,203
0.202

∆Mean
1.935
2.342
-1.609
-0.21
9.105

Yes
Yes
Yes
14,649
0.151

Number of Matched Pairs

620
1452
2859
2859
1933

Yes
Yes
Yes
14,655
0.140

P-Value
0.0887
0.0104
0.0843
0.8659
0.0003

Yes
Yes
Yes
14,649
0.149

t-stat
1.71
2.57
-1.73
-0.17
3.67

82

Table 2.5: Tax Policy and AshleyMadison Membership

In Panel A, the dependent variable is a dummy variable equal to one if the proportion of tax havens among countries
mentioned in Exhibit 21 of a ﬁrm’s 10-K ﬁling exceeds 50% (75%, 90%) (see Dyreng and Lindsey (2009) for a detailed
description). Countries are identiﬁed as tax havens if they are deﬁned as such by three of the four following sources: (1)
the Organization for Economic Cooperation and Development (OECD), (2) the U.S. Stop Tax Havens Abuse Act, (3) The
International Monetary Fund (IMF), and (4) the Tax Research Organization. We deﬁne mostly_txh50 (mostly_txh75,
mostly_txh90) as a dummy variable equal to one if more than 50% (75%, 90%) of the countries mentioned in 10-K
ﬁlings are listed as tax havens, and zero otherwise. All other variables are deﬁned in the Internet Appendix All reported
estimates are marginal eﬀects (multiplied by 100) from logit regression speciﬁcations. z-statistics, computed from
standard errors clustered by year, are reported in parentheses below coeﬃcient estimates. All speciﬁcations include
industry, year, and EA ﬁxed eﬀects transformations. In Panel B, we report OLS estimates for the dependent variable
Eﬀective Tax Rate, calculated using income tax divided by pretax income excluding special items. All other variables
are deﬁned in the Internet Appendix.
t-statistics, computed from standard errors clustered by year, are reported in
parentheses below coeﬃcient estimates. All speciﬁcations include industry, year, and EA ﬁxed eﬀects transformations.
Statistical signiﬁcance (two-sided) at the 1% 5%, and 10% level is denoted by *, **, and ***, respectively.

Panel A: Use of Tax Havens

(4)

(5)

(6)

(1)

(2)

(3)

MostlyTxh50 MostlyTxh50 MostlyTxh75 MostlyTxh75 MostlyTxh90 MostlyTxh90

Active AM Account

Dummy(AM>0)

Institutional Investor

HHI (SIC4)

Market Cap (t-1)

Log # of Employee

EBITDA/Assets

Tobin’s Q (t-1)

Family Firm

GIndex

0.0136***
(4.062)

0.0263*
(1.843)
-0.0803
(-1.558)
-0.0139
(-1.619)
-0.0505***
(-7.866)
0.0608**
(2.375)
0.0127***
(3.082)
0.00812
(0.648)

-0.00400***

(-2.823)

EA FE
Year FE
Industry(SIC2) FE
Observations
Pseudo R2

Yes
Yes
Yes
10,169
0.3353

0.0146***
(4.444)

0.0121
(0.955)
-0.0496
(-1.189)
-0.0206**
(-2.354)
-0.0462***
(-11.74)
0.0946***
(2.893)
0.0118***
(2.727)
0.0131
(1.134)
-0.00216*
(-1.702)

Yes
Yes
Yes
10,169
0.3426

0.0138***
(3.346)
0.0123
(1.007)
-0.0460
(-1.131)
-0.0193**
(-2.456)
-0.0441***
(-10.26)
0.0873***
(3.063)
0.0113**
(2.512)
0.0119
(1.036)
-0.00199*
(-1.697)

Yes
Yes
Yes
10,169
0.3417

0.0147***
(3.494)
0.0117
(0.968)
-0.0496
(-1.207)
-0.0199**
(-2.548)
-0.0437***
(-10.23)
0.0894***
(3.104)
0.0118***
(2.680)
0.0133
(1.167)
-0.00194
(-1.636)

Yes
Yes
Yes
10,169
0.3421

0.00757
(1.575)
0.0258*
(1.722)
-0.0806
(-1.547)
-0.0132
(-1.423)
-0.0474***
(-9.804)
0.0563*
(1.952)
0.0128***
(3.155)
0.00838
(0.659)

-0.00371**
(-2.520)

Yes
Yes
Yes
10,169
0.3348

0.0145***
(4.470)

0.0127
(1.029)
-0.0458
(-1.123)
-0.0200***
(-2.581)
-0.0467***
(-9.095)
0.0925***
(3.294)
0.0112**
(2.487)
0.0117
(1.024)
-0.00222*
(-1.843)

Yes
Yes
Yes
10,169
0.3422

83

Table 2.5 (cont’d): Tax Policy and AshleyMadison Membership

Panel B: Eﬀective Corporate Tax Rate

(1)

Active AM Account

Dummy(AM>0)

-0.208 ***

(-2.64)

-0.307
(-1.57)
-0.988 **
(-2.16)

-0.960 **
(-2.08)

(2)

(3)

(4)

(5)

(6)

-0.203 ***

(-2.51)

-0.378
(-1.59)

-0.203 ***

(-2.51)

-0.379
(-1.59)

-32.615 *** -32.989 ***

-32.657 ***

-32.779 ***

-32.655 ***

-32.776 ***

MostlyTxh50

MostlyTxh75

MostlyTxh90

international

Institutional Investor

R&D/Assets

HHI (SIC4)

Market Cap (t-1)

Log # of Employee

EBITDA/Assets

Family Firm

Tobin’s Q (t-1)

Gindex

-1.625 ***

(-2.43)

-1.639 ***

(-2.46)

-0.738 ***

-0.741 ***

-0.745 ***

-0.742 ***

(-2.79)
-0.574
(-0.93)

(-10.36)
0.282
(0.6)

0.561 ***

(3.86)

-0.499 ***

(-3.42)

31.439 ***

(21.13)
0.852
(1.33)
-0.135
(-1.2)
0.023
(0.8)

(-2.8)
-0.564
(-0.92)

(-10.43)
0.316
(0.68)

0.510 ***

(3.61)

-0.571 ***

(-3.79)

31.575 ***

(20.84)
0.870
(1.36)
-0.129
(-1.15)
0.020
(0.7)

(-2.96)
-0.580
(-0.94)

(-10.36)
0.277
(0.58)

0.564 ***

(3.91)

-0.502 ***

(-3.48)

31.490 ***

(20.95)
0.846
(1.31)
-0.137
(-1.22)
0.026
(0.89)

(-2.95)
-0.593
(-0.97)

(-10.4)
0.225
(0.47)

0.552 ***

(3.82)

-0.520 ***

(-3.6)

31.557 ***

(20.78)
0.864
(1.34)
-0.136
(-1.21)
0.029
(0.99)

-1.616 **
(-2.31)

-0.749 ***

(-2.98)
-0.582
(-0.94)

(-10.36)
0.276
(0.57)

0.563 ***

(3.9)

-0.501 ***

(-3.47)

31.503 ***

(20.92)
0.849
(1.32)
-0.136
(-1.21)
0.026
(0.89)

-1.632 ***

(-2.35)

-0.746 ***

(-2.97)
-0.595
(-0.97)

(-10.41)
0.224
(0.47)

0.551 ***

(3.81)

-0.519 ***

(-3.59)

31.570 ***

(20.75)
0.867
(1.35)
-0.135
(-1.2)
0.029
(0.99)

EA FE
Year FE
Industry(SIC2) FE
Observations
Pseudo R2

Yes
Yes
Yes
7970
0.344

Yes
Yes
Yes
7970
0.344

Yes
Yes
Yes
7970
0.344

Yes
Yes
Yes
7970
0.344

Yes
Yes
Yes
7970
0.344

Yes
Yes
Yes
7970
0.344

84

Table 2.6: AshleyMadison Membership and the Choice of Internal vs. External CEO

In this table we report the marginal eﬀects estimates from logit speciﬁcations of choosing an internal CEO (1) vs.
external CEO (0) on the number of active AshleyMadison (AM) accounts and a dummy variable equal to 1 if active
AshleyMadison (AM) accounts > 0. The data on CEOs come from Boardex for 2003-2014. We deﬁne a CEO as
internal if he/she was employed at a given company for at least one full year before being appointed as CEO. Our
regressor of interest is the natural logarithm of one plus the number of active AM accounts for a given ﬁrm year.
Speciﬁcations 1-4 include year ﬁxed eﬀects, column 3 includes industry (2 digit sic code) ﬁxed eﬀects, and column 4
includes industry and EA ﬁxed eﬀects. All regressors are lagged one year relative to our CEO appointment variables.
All variables are deﬁned in the Internet Appendix. The t-statistics, calculated from standard errors clustered at the ﬁrm
level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance (two-sided) at the 1% 5%, and
10% level is denoted by *, **, and ***, respectively.

(1)

(2)

(3)

(4)

(5)

0.0874***
(11.60)

0.0844***
(9.576)

0.178***
(13.91)

0.235***
(7.890)

Active AM Account

Dummy(AM>0)

Dummy: Institutional Investor

Shares held by insiders

HHI (SIC4)

Log Market Cap(t-1)

Log # of Employee

Family Firm

ROA

Governance Index (Gompers, Ishii, Metrick)

Founder is director

Tobin’s Q (t-1)

∆OROA

∆OROS

Year FE
2-digit SIC FE
EA FE
Observations
Pseudo-R2

0.178***
(9.112)
-0.451***
(-9.677)
-0.319**
(-2.548)
0.918***
(7.500)
0.0868***
(5.146)
0.0534**
(2.246)
0.105
(1.604)
-0.554***
(-2.644)
-0.0112**
(-2.546)
0.0189
(0.450)
-0.0751**
(-2.211)
-0.00630**
(-2.403)
-0.0494
(-0.356)

Yes
Yes
Yes
766
0.445

-0.399***
(-8.460)
-0.182*
(-1.899)
0.871***
(6.666)
0.0154
(0.789)
0.0515**
(2.392)
0.133**
(2.414)
-0.212
(-1.017)
-0.0246***
(-4.077)
-0.0510
(-1.003)
-0.0524*
(-1.662)
-0.00441
(-1.519)
-0.169
(-1.264)

Yes
Yes
Yes
766
0.502

0.0954***
(3.138)
-0.645***
(-7.869)
0.218***
(4.490)

-0.0815***
(-11.21)
0.0450***
(5.712)
0.0735
(1.478)
-0.107
(-1.253)
-0.00614
(-1.463)
0.00779
(0.306)
0.0568***
(4.427)

0.0507
(1.536)
-0.638***
(-7.851)
0.205***
(4.025)

-0.0800***
(-8.556)
0.0458***
(5.605)
0.0783
(1.537)
-0.188*
(-1.849)
-0.00563
(-1.352)
0.0138
(0.524)
0.0777***
(4.628)

0.215***
(4.114)
-0.558***
(-5.133)
0.335***
(5.034)

-0.0748***
(-4.234)
-0.0275**
(-2.453)
-0.0974
(-1.461)
0.137
(1.047)
-0.0142**
(-1.978)
-0.0396
(-1.284)
0.00770
(0.380)

-0.0172*** -0.00708***
(-5.916)
0.0767
(0.400)

(-2.776)
-0.276*
(-1.759)

Yes

Yes

955
0.077

955
0.083

Yes
Yes

850
0.206

85

Table 2.7: Determinants of AshleyMadison Membership

In this table we report estimates for determinants of the number of active AshleyMadison (AM) accounts at the ﬁrm-level. We use Tobit speciﬁcations because the
dependent variable, the natural logarithm of one plus the number of active AM accounts, is truncated at zero and continuous to the right of zero. Industry covariates
are deﬁned using four-digit SIC codes and geography covariates are deﬁned at the zipcode level. Detailed variable deﬁnitions are provided in the appendix. All
speciﬁcations have year ﬁxed eﬀects, speciﬁcations (2-6) include industry (three-digit SIC) ﬁxed eﬀects, and speciﬁcations (3-6) include EA ﬁxed eﬀects. The
t-statistics, calculated from standard errors clustered at the ﬁrm level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance (two-sided) at
the 1% 5%, and 10% level is denoted by *, **, and ***, respectively. We also report sigma and pseudo r-squared from the Tobit regressions. In unreported analyses
we ﬁnd qualitatively similar and statistically signiﬁcant results using a linear probability model speciﬁcation.

VARIABLES

Log Market Cap

Firm Age

Log # of Employee

Volatility-3 Factor adjusted

Population Density

Population

Median Population Age

Avg Income per Household

HHI (SIC4)

Market to Book (SIC4)

R&D intensity (SIC4)

Sales growth rate (SIC4)

sigma

Observations
Year FE
Industry FE
EA FE
Pseudo-R2

Active AM Accounts Active AM Accounts Active AM Accounts Active AM Accounts Active AM Accounts Active AM Accounts

(3)

(4)

(5)

(1)

(2)

(6)

0.214***
(8.60)
0.003
(0.99)
0.306***
(13.66)
0.999
(0.30)

0.148***
(5.58)
0.003
(1.08)
0.409***
(13.78)
3.621
(1.22)

0.122***
(39.68)
0.005***
(6.53)
0.420***
(89.94)
3.180***
(4.68)

1.503***
(46.06)

28,374
Yes

.12

1.368***
(45.42)

28,374
Yes
Yes

.174

1.314***
(179.15)

27,824
Yes
Yes
Yes
.198

86

0.118***
(36.00)
0.005***
(6.25)
0.425***
(87.99)
3.008***
(4.15)

-0.044
(-0.91)
-0.002
(-0.38)
0.659***
(5.13)
0.003
(0.06)
1.313***
(171.34)

27,754
Yes
Yes
Yes
.198

0.121***
(35.65)
0.004***
(6.07)
0.420***
(88.13)
3.092***
(4.07)
0.024
(0.00)
0.036***
(9.97)

-0.028***
(-41.44)
-4.300***
(-10.05)

1.314***
(168.38)

27,792
Yes
Yes
Yes
.198

0.117***
(33.40)
0.005***
(5.94)
0.425***
(86.28)
2.915***
(3.73)
1.965
(0.16)
0.037***
(9.85)

-0.028***
(-40.37)
-5.093***
(-11.57)
-0.046
(-0.91)
-0.002
(-0.35)
0.673***
(5.05)
0.002
(0.06)
1.312***
(164.61)

27,722
Yes
Yes
Yes
.198

APPENDIX C

PRODUCT COMPETITION AND CORPORATE FRAUD

87

Variable
AAER Misstatement

SCAC

Fraud
SIC3 HHI

TNIC HHI

Avg Similarity Score

Precision
Proﬁt Margin
Top 4 Concentration

Age
Analyst Num
Inst Ownership
Assets
Capex
Book Leverage
ROA
EFN
RSST Accruals
Dummy Security Issue
Change AR
Change Inventory
Pct Soft Assets
Change in Cash Sales
Change in ROA
Change in Employee

R&D

R&D (dummy)
NCOM SIC3
Ind R2

RPE Flag
NCOMP TNIC

Table 3.1: Variable Description

1

NcompT N IC

1

×

(1−score)2)0.5

Deﬁnitions
Equal to 1 for ﬁrm-years for which ﬁrms have settled with the SEC for
corporate Fraud.
Securities and Class Action Equal to 1 for all ﬁrm-years for which ﬁrms
settle a securities class action lawsuit for an alleged 10B-5 fraud allegation.
Equal to 1 for all ﬁrm-years with an AAER or SCA.
Herﬁndahl-Hirschman index based on ﬁrm sales and three-digit SIC code
industry classiﬁcations .
Herﬁndahl-Hirschman index based on ﬁrm sales Text-based Network In-
dustry classiﬁcations (TNIC) from Hoberg and Phillips.
Mean Hoberg and Phillips Similarity Score for all rivals within each ﬁrm-
year’s TNIC.
Deﬁned as (
Average EBITDA/sales ratio for ﬁrms within each three-digit SIC code.
Proportion of sales within a three-digit SIC code attributable to the four
largest ﬁrms within an industry.
Number of years the ﬁrm has been in Compustat.
Number of analysts covering the ﬁrm in each year from IBES (0 if missing).
Percentage of shares outstanding held by 13-F institutions.
Total Assets.
Capital Expenditures / log Assets.
(Total Long-Term Debt +Debt in Current Liabilities)/ log Assets
Net Income / Assets
Equity Finance Needed deﬁned as ROA/(1 âĹŠ ROA.
See Rachardson et. al (2005)
An indicator variable equal to 1 if the ﬁrm issued securities during year.
Change in Accounts Receivable/Total Assets.
Change in Inventory/Total Assets.
(Total Assets - PP&E- Cash and Cash Equivalent)/Total Assets.
Percentage change in Cash Sales - Change in Accounts Receivable.
Change in Return on Assets.
Percentage change in the number of employees - percentage change in
assets.
Research and Development scaled by assets. Missing observations are ﬁlled
with either the ﬁrm average, if a time series exists, or the industry average
if not.
Equal to 1 if R&D is missing and 0 otherwise.
Number of competitors within the three-digit SIC Code.
Following Wang and Winton (2014), we ﬁrst regress each ﬁrm’s daily stock
returns on the weighted-average daily market return and the weighted-
average daily industry return. Then, we take the average r-squared for each
ﬁrm in each three-digit SIC code.
See Page 8.
Number of competitors according to Text-based Network Industry classiﬁ-
cations (TNIC) from Hoberg and Phillips.

88

This table reports summary statistics of ﬁrm characteristics at the ﬁrm-year level. Variable deﬁnitions are provided in
the Table 3.1. Our sample spans 1996 through 2011.

Table 3.2: Summary Statistics

Std. Dev.

10th Percentile

90th Percentile

VARIABLES
AAER Misstatement
SCAC
Fraud
Avg Similarity Score
Avg Top5 Similarity
Avg Top10 Similarity
Avg Top15 Similarity
Avg Score Precision
Sum Similarity
Product Market Fluidity
SIC3 HHI
SIC3 Proﬁt Margin
TNIC HHI
NCOMP TNIC
NCOMP SIC3
RSST accruals
Change AR
Change Inventory
Pct Soft Assets
Change in Cash Sales
ROA
Change in ROA
Change in employee
Dummy Security Issue
Whited-Wu Index
Book Leverage
Capex
R&D
R&D dummy
Age
Inst Ownership
Number of Analysts
Stock Industry Return R2
Relative Perf Eval Flag
Ln Asset

No. Obs Mean
55,381
0.014
0.006
55,381
0.019
55,381
55,381
0.03
0.08
55,381
0.066
55,381
0.059
55,381
55,381
1.002
2.847
55,381
7.182
50,402
55,381
0.176
-0.039
55,381
0.235
55,381
74.204
55,381
55,381
121.607
0.024
51,487
0.01
55,381
55,060
0.006
0.541
55,377
0.195
51,888
51,497
-0.005
-0.007
54,671
-0.08
54,053
0.92
55,381
54,954
-0.196
0.299
55,237
0.06
55,381
55,381
0.069
0.627
55,381
15.353
53,295
0.516
43,018
55,381
5.837
0.342
53,238
0.677
55,179
55,381
5.618

89

0.119
0.074
0.135
0.023
0.058
0.05
0.047
0.103
4.999
3.292
0.145
0.272
0.197
90.52
170.694

0.24
0.065
0.049
0.245
0.71
0.195
0.175
0.469
0.272
0.198
0.294
0.093
0.117
0.484
11.825
0.315
7.008
0.173
0.467
1.937

0.012
0.017
0.014
0.013
0.924
0.074
3.292
0.062
-0.346
0.064

-0.182
-0.045
-0.028
0.175
-0.214
-0.205
-0.149
-0.365

-0.389

0
0
0

5
6

1

0
0
0
0
4

0

0

0.068

0.121

3.155

0
0
0

0.055
0.156
0.135
0.123
1.053
7.659
11.685
0.332
0.156
0.518
204
351
0.22
0.07
0.05
0.852
0.574
0.141
0.12
0.241

1

0.012
0.733
0.14
0.184

1
35
0.922
15
0.58
1

8.181

Table 3.3: Product Market Diﬀerentiation and Corporate Fraud

This table reports OLS estimates for the incidence of fraud on the average similarity of each ﬁrmâĂŹs rivals. Our
proxy for corporate fraud includes a combination of AAER misstatements from the AAER dataset and Securities
Class Actions from the Stanford University Lawsuit Database. The speciﬁcation in Column 1 does not include control
variables. The speciﬁcation in Column 2 includes controls used in Dechow et al.
In Columns 3-5 we
include our full set of controls as described in Section II and Column 3 also includes Institutional Ownership. All
speciﬁcations are run at the ﬁrm-year level, include year ﬁxed eﬀects, and include explanatory variables are lagged by
one year. Column 4 also includes three-digit SIC code (SIC3) ﬁxed eﬀects, Column 5 adds year × SIC3 ﬁxed eﬀects.
The t-statistics, calculated from standard errors clustered at SIC3 level, are reported in parentheses below coeﬃcient
estimates. Statistical signiﬁcance (two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***, respectively.

(2011).

(1)
Fraud

(2)
Fraud

(3)
Fraud

(4)
Fraud

(5)
Fraud

-0.113**
(-2.089)

-0.180***
(-3.517)

-0.220***
(-3.915)
0.009
(0.740)
-0.000
(-0.011)
0.001
(0.227)
0.007
(0.792)
0.005
(0.859)
-0.003
(-0.636)
0.023
(1.460)
0.026
(1.258)
0.022***
(4.186)
0.004**
(2.057)
-0.017***
(-3.336)
-0.004*
(-1.868)
-0.009***
(-3.040)
-0.001
(-0.258)
-0.009
(-0.822)
0.007**
(2.124)
0.001
(1.205)
0.005***
(3.245)

-0.163***
(-3.244)
-0.012
(-1.143)
-0.004
(-1.135)
0.001
(0.392)

0.033*
(1.892)
0.002
(0.572)
0.023*
(1.808)
0.027
(1.225)
0.019***
(3.733)
0.005**
(2.367)
-0.018***
(-5.212)
-0.003
(-1.414)
-0.007***
(-2.974)
0.001
(0.275)

-0.171***
(-3.946)
-0.011
(-1.008)
-0.003
(-0.914)
0.001
(0.267)

-0.000
(-0.057)
0.001
(0.342)
0.016
(1.296)
0.020
(0.937)
0.019***
(3.825)
0.005**
(2.227)
-0.021***
(-5.887)
-0.003
(-1.456)
-0.008***
(-3.513)
0.002
(0.737)
0.017
(1.515)

0.002
(0.945)
0.006***
(2.913)

0.001
(0.756)
0.007***
(3.546)

28,912
0.014
Year

38,916
37,144
0.034
0.079
Year+SIC3 Year×SIC3

0.002
(0.538)
0.022*
(1.791)
0.016
(0.756)
0.019***
(4.241)
0.005**
(2.250)
-0.023***
(-6.132)
-0.004**
(-2.101)
-0.010***
(-3.413)
0.003
(1.327)

0.006***
(4.617)

39,519
0.015
Year

90

Avg. Similarity Score

R&D

R&D dummy

Ln number analysts

Inst Ownership

Whited-Wu Index

RSST accruals

Change AR

Change Inventory

Pct. Soft Assets

Change in Cash Sales

Change in ROA

Change in employee

Ln Age

Dummy Security Issue

Stock Industry Return R2

Relative Perf Eval Flag

Ln NCOMP TNIC

Ln Asset

Observations
R-squared
FE

50,526
0.005
Year

Table 3.4: Product Diﬀerentiation and Fraud by Complexity Quartiles

This table reports OLS estimates for the incidence of fraud on the average similarity of each ﬁrm’s rivals split into
complexity quartiles. Our proxy for corporate fraud includes a combination of AAER misstatements from the AAER
dataset and Securities Class Actions from the Stanford University Lawsuit Database. We deﬁne complexity as the
number of unique SIC codes spanned by a ﬁrm’s set of competitors according to the TNIC developed by Hoberg and
Phillips, 2016. Panel A reports competitor and fraud classiﬁcations for each quartile. Panel B reports OLS estimates
for each quartile including our full set of control variables described in Section II, and Panel C includes Institutional
Ownership. The t-statistics, calculated from standard errors clustered at the three-digit SIC code (SIC3) level, are
reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance (two-sided) at the 10% 5%, and 1% level
is denoted by *, **, and ***, respectively.

Complexity

Panel A
Unique SICs in TNIC
Competitors in TNIC
% Fraud
Avg Similarity Score

Panel B
Avg Similarity Score

Observations
R-squared

FE
Controls

Panel C
Avg Similarity Score

Inst Ownership

Low
Q1

3.3
13
1.6
2.7

Q2

8.2
49
1.7
2.8

Q3

13
117
1.9
3.3

High
Q4

22.5
150
2.1
3.5

-0.168*** -0.197**
(-3.418)
(-2.029)

-0.201
(-1.508)

-0.683***
(-5.053)

9,995
0.016

Year
Full

9,628
0.017

Year
Full

9,018
0.024

Year
Full

8,503
0.026

Year
Full

-0.191***
(-3.217)
0.010
(0.769)

-0.169
(-1.631)
0.004
(0.225)

-0.166
(-1.029)
0.009
(0.729)

-0.677***
(-4.081)
0.011
(1.012)

Observations
R-squared
FE
Controls

7,707
0.015
Year
Full

7,469
0.021
Year
Full

7,048
0.023
Year
Full

6,688
0.023
Year
Full

91

Table 3.5: IPOs and Acquisitions of Rivals as Change to Information Environment

This table reports OLS estimates for the association between fraud and rival IPOs (M&A) activity. The speciﬁcations
are the same as model (4) of Table 3, but also include rival ﬁrm IPO (M&A) activity in Panel A (Panel B). For each
ﬁrm-year, include the natural log of the number of ﬁrms that compete with ﬁrm i and that underwent an IPO or were
acquired in year t, and an interaction term Ln Num Competitor IPO × Avg Similarity Score or Ln Num Competitor
Target × Avg Similarity Score. In Column 2 of Panel A (B), we control for IPO (M&A) Size ($) which is the sum of
all-capital raised by IPO rivals (total market capitalization of Target rivals). In Columns 4 and 5 of Panel A (Panel B),
we split the data by high and low non-IPO (non-acquired) similarity scores in year t-1. All speciﬁcations include year
ﬁxed eﬀects and all control variables are lagged one year. Columns 3-5 also include three-digit SIC code (SIC3) ﬁxed
eﬀects. The t-statistics, calculated from standard errors clustered at the SIC3 level, are reported in parentheses below
coeﬃcient estimates. Statistical signiﬁcance (two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***,
respectively.

Low Non-IPO Rival Score

High Non-IPO Rival Score

(4)

Fraud

(4)

Fraud

(5)

Fraud

(5)

Fraud

Low Non-M&A Rival Score High Non-M&A Rival Score

0.005***
(2.82)

-0.000
(-0.02)
18,858
0.050

0.004
(1.44)

(-0.72)
18,279
0.037

0.103***
(3.141)

0.002
(-2.829)
18,672
0.052
Full
Yes
Yes

-0.000
(-0.024)

(1.380)
18,449
0.039
Full
Yes
Yes

Panel A : Rival IPO

Avg Similarity Score

Ln Num Competitor IPO
Avg Score × Ln Num Comp IPO

IPO Size ($)

Observations
R-squared

Panel B: Rival M&A

Avg Similarity Score

Ln Num Competitor Target
Avg Score × Ln Num Comp Target

Ln Target MarketCap

Observations
R-squared
Controls
SIC3 FE
Year FE

(1)

(2)

(3)

Fraud

Fraud

Fraud

-0.149***
(-3.488)
0.009***
(4.767)
-0.119***
(-3.370)
-0.000

37,144
0.017

-0.149***
(-3.473)
0.009***
(4.909)
-0.123***
(-3.460)

(-0.498)
37,144
0.017

-0.134***
(-3.382)
0.008***
(4.565)
-0.109***
(-2.660)
-0.000

37,144
0.034

(1)

(2)

(3)

Fraud

Fraud

Fraud

-0.185***
(-4.060)
0.052***
(3.373)
-0.844***
(-2.693)
-0.001

37,144
0.017
Full
No
Yes

-0.185***
(-4.061)
0.061**
(2.436)
-0.874**
(-2.593)

(-0.588)
37,144
0.017
Full
No
Yes

-0.162***
(-3.927)
0.048***
(3.380)
-0.685**
(-2.430)
-0.008***

37,144
0.035
Full
Yes
Yes

92

Table 3.6: Product Market Diﬀerentiation and Corporate Fraud (Controlling for Alternative Measures of Competition)

This table reports OLS estimates for the incidence of fraud on Average Similarity Score, while controlling for alternative
measures of competition. Our proxy for corporate fraud includes a combination of misstatements from the AAER
dataset and Securities Class Actions from the Stanford University Lawsuit Database. Column 1 includes sales based
Herﬁndahl-Hirschman Index (HHI) according to three digit SIC code (SIC3). Column 2 also includes the number
of competitors (logged) in the same SIC3. Column 3 also includes the proﬁt margin and an industry concentration
measure. In Column 4 we include the sales based HHI according to the ﬁrmâĂŹs TNIC. Column 5 also includes the
number of competitors within a ﬁrm’s TNIC. Column 6 also includes the sum similarity score. The speciﬁcations
include the full set of controls as described in Section II. All speciﬁcations are run at the ﬁrm-year level, include year
ﬁxed eﬀects, and explanatory variables lagged by one year. The t-statistics, calculated from standard errors clustered
at SIC3 level, are reported in parentheses below coeﬃcient estimates Statistical signiﬁcance (two-sided) at the 10%
5%, and 1% level is denoted by *, **, and ***, respectively.

(1)
Fraud

(2)
Fraud

(3)
Fraud

(4)
Fraud

(5)
Fraud

(6)
Fraud

-0.169*** -0.173***
(-3.938)
(-3.712)

0.006
(0.985)
0.002
(1.113)

0.000

37,335
0.034
Yes
Yes

0.006
(1.091)

(1.299)

(1.062)

(0.123)

37,335
0.034
Yes
Yes

Avg Similarity Score

SIC3 HHI

NCOMP_SIC3

SIC3 PM sale

-0.158***
(-3.318)
0.017
(0.831)
0.016**

Top 4 Concentration

0.041**

TNIC HHI

NCOMP_TNIC

Product Market Fluidity

Sum Similarity

(-3.250)
0.009
(0.393)

-0.160*** -0.158*** -0.160***
(-3.305)
(-3.519)
0.033
(1.458)
0.016***
(2.164)
-0.007

(2.699)

(-0.719)

(2.140)

-0.002
(-0.276)
0.002

0.000

Observations
R-squared
SIC3 FE
Year FE

37,335
0.034
Yes
Yes

37,335
0.034
Yes
Yes

37,335
0.034
Yes
Yes

37,335
0.034
Yes
Yes

93

Table 3.7: Alternative Measures of Competition and Corporate Fraud

This table reports OLS estimates for the incidence of fraud on commonly used industry-level proxies for competition.
Our proxy for corporate fraud includes a combination of AAER misstatements from the AAER dataset and Securities
Class Actions from the Stanford University Lawsuit Database. Our measures of competition include: sales based HHI
according to TNIC, sales based HHI according to SIC3, sum similarity, product market ﬂuidity, average (SIC3) proﬁt
margin, top-4 sales concentration and number of competitors constructed using three-digit SIC code. Columns 1-7
include the full set of controls as described in Section II. The ﬁrm-year is the unit of observation in this analysis.
All speciﬁcations include year ﬁxed eﬀects, and control variables lagged by one year. The t-statistics, calculated
from standard errors clustered at the SIC3 level, are reported in parentheses below coeﬃcient estimates. Statistical
signiﬁcance (two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***, respectively.

(1)
Fraud

(2)
Fraud

(3)
Fraud

(4)
Fraud

(5)
Fraud

(6)
Fraud

(7)
Fraud

0.002
(0.397)

-0.001
(-0.108)

TNIC HHI

SIC3 HHI

Sum Similarity

Product Market Fluidity

SIC3 Proﬁt Margin

Top 4 Concentration

Ln NCOMP SIC3

-0.000
(-0.770)

0.000
(0.800)

-0.008
(-1.272)

0.006
(0.719)

Observations
R-squared
Controls
FE

37,335
0.014
Full
Year

37,335
0.014
Full
Year

37,335
0.014
Full
Year

36,180
0.014
Full
Year

37,335
0.014
Full
Year

37,335
0.014
Full
Year

0.003*
(1.871)

37,335
0.015
Full
Year

94

Table 3.8: Product Market Diﬀerentiation and Corporate Fraud âĂŞ Standardized Variables

This table reports OLS estimates for the incidence of fraud on standardized RHS variables. Our proxy for corporate
fraud includes a combination of AAER misstatements and Securities Class Actions from the Stanford University
Lawsuit Database. The speciﬁcation in Column 1 does not include control variables. The speciﬁcation in Column
2 includes control variables used in Dechow et al. (2011). In Columns 3-6 we include our full set of controls as
described in Section II and Column 3 also includes Institutional Ownership. The unit of observation in this analysis
is the ﬁrm-year. All speciﬁcations include year ﬁxed eﬀects, and control variables are lagged by one year. Column
4 includes three-digit SIC code (SIC3) ﬁxed eﬀects, Column 5 year x SIC3 ﬁxed eﬀects, and Column 6 ﬁrm ﬁxed
eﬀects. All continuous RHS variables are standardized. The t-statistics, calculated from standard errors clustered at
SIC3 level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance (two-sided) at the 10% 5%,
and 1% level is denoted by *, **, and ***, respectively.

(1)
Fraud

(2)
Fraud

(3)
Fraud

(4)
Fraud

(5)
Fraud

-0.003** -0.004***
(-2.089)
(-3.502)

0.000
(0.550)
0.001*
(1.805)
0.001
(0.762)
0.005***
(4.227)
0.003**
(2.236)
-0.004***
(-6.131)
-0.002**
(-2.105)
-0.007***
(-3.402)
0.003
(1.479)

Avg. Similarity Score

R&D

R&D dummy

Ln number analysts

Inst Ownership

Whited-Wu Index

RSST accruals

Change AR

Change Inventory

Pct. Soft Assets

Change in Cash Sales

Change in ROA

Change in employee

Ln Age

Dummy Security Issue

Stock Industry Return R2

Relative Perf Eval Flag

Ln NCOMP TNIC

Ln Asset

Observations
R-squared
FE

0.011***
(4.607)

39,465
0.015
Year

50,526
0.005
Year

-0.005***
(-3.915)
0.001
(0.740)
-0.000
(-0.011)
0.001
(0.227)
0.002
(0.792)
0.001
(0.859)
-0.001
(-0.636)
0.001
(1.460)
0.001
(1.258)
0.005***
(4.186)
0.003**
(2.057)
-0.003***
(-3.336)
-0.002*
(-1.868)
-0.007***
(-3.040)
-0.001
(-0.258)
-0.002
(-0.822)
0.003**
(2.124)
0.002
(1.205)
0.009***
(3.245)

28,912
0.014
Year

95

-0.004***
(-3.230)
-0.001
(-1.134)
-0.004
(-1.171)
0.001
(0.378)

0.007*
(1.889)
0.000
(0.567)
0.002*
(1.817)
0.001
(1.228)
0.005***
(3.735)
0.004**
(2.359)
-0.003***
(-5.182)
-0.001
(-1.421)
-0.005***
(-2.967)
0.001
(0.284)

-0.004***
(-3.946)
-0.001
(-1.008)
-0.003
(-0.914)
0.001
(0.267)

-0.000
(-0.057)
0.000
(0.342)
0.001
(1.296)
0.001
(0.937)
0.005***
(3.825)
0.003**
(2.227)
-0.004***
(-5.887)
-0.001
(-1.456)
-0.006***
(-3.513)
0.002
(0.737)
0.003
(1.515)

0.002
(0.945)
0.011***
(2.913)

0.002
(0.756)
0.014***
(3.547)

(6)
Fraud

-0.002*
(-1.691)
-0.001
(-0.906)
0.008
(1.306)
0.009***
(3.995)

0.000
(0.554)
0.000
(0.537)
-0.000
(-0.275)
0.000
(0.219)
0.003**
(2.214)
0.002*
(1.776)
-0.003***
(-3.775)
-0.000
(-0.343)
-0.010*
(-1.859)
0.003
(0.855)
0.004*
(1.800)
-106.330
(-0.000)
0.002
(0.740)
0.018***
(4.653)

37,144
0.034
Year Sic3 Year × Sic3 Year Gvkey

36,380
0.437

38,910
0.078

Table 3.9: Product Market Diﬀerentiation and Corporate Fraud Alternate Similarity Scores (Standardized)

This table reports estimates for the incidence of fraud on alternative standardized similarity scores of each ﬁrm’s
rivals using ordinary least squares (OLS) regressions. Our proxy for corporate fraud includes a combination of AAER
misstatements from the AAER dataset and Securities Class Actions from the Stanford University Lawsuit Database.
All variables of interest (not controls) are standardized for purposes of comparison. In Column 1, we report results for
the main dependent variable used throughout our analysis. In Columns 2-4, we replace Average Similarity Score with
a ﬁrm’s product market similarity score averaged across its closest 5, 10, and 15 competitors, respectively. In Column
5, we replace Average Similarity Score with the Precision measure outlined in section III. In Columns 6-8, Average
Similarity Score is replaced with the (natural log of) number of a ﬁrm’s rivals in the top 75th, 90th and 95th percentile
of similarity scores, respectively in the full cross section of ﬁrms in year t. All speciﬁcations are run at the ﬁrm-year
level, include year ﬁxed eﬀects, and explanatory variables are lagged by one year. The t-statistics, calculated from
standard errors clustered at SIC3 level, are reported in parentheses below coeﬃcient estimates. Statistical signiﬁcance
(two-sided) at the 10% 5%, and 1% level is denoted by *, **, and ***, respectively.

(1)
Fraud

(2)
Fraud

(3)
Fraud

(4)
Fraud

(5)
Fraud

(6)
Fraud

(7)
Fraud

(8)
Fraud

Std Avg Similarity Score

Std Avg Top5 Similarity

Std Avg Top10 Similarity

Std Avg Top15 Similarity

Std Precision

Std Ln NCOMP TNIC 75th

Std Ln NCOMP TNIC 90th

Std Ln NCOMP TNIC 95th

cline2-9
Observations
R-squared
Controls
FE

-0.005***
(-4.248)

-0.007***
(-3.753)

-0.007***
(-3.383)

-0.006***
(-3.202)

-0.005***
(-4.231)

-0.004*
(-1.778)

-0.004**
(-2.265)

37,144
0.016
Full
Year

37,144
0.016
Full
Year

37,144
0.016
Full
Year

37,144
0.016
Full
Year

37,144
0.016
Full
Year

37,144
0.015
Full
Year

37,144
0.015
Full
Year

-0.004**
(-2.314)

37,144
0.016
Full
Year

96

BIBLIOGRAPHY

97

BIBLIOGRAPHY

Ades, Alberto, and Rafael Di Tella, 1999, Rents, Competition, and Corruption, The American Economic Review 89,

982–993.

Agarwal, Vikas, Gerald D. Gay, and Leng Ling, 2014, Window Dressing in Mutual Funds, The Review of Financial

Studies 27, 3133–3170.

Andergassen, Rainer, 2016, Managerial compensation, product market competition and fraud, International Review of

Economics & Finance 45, 1–15.

Bailey, Elizabeth E., and Constance E. Helfat, 2003, External management succession, human capital, and ﬁrm

performance: an integrative analysis, Managerial and Decision Economics 24, 347–369.

Bain, Joe S., 1951, Relation of Proﬁt Rate to Industry Concentration: American Manufacturing, 1936-1940, The

Quarterly Journal of Economics 65, 293–324.

Balakrishnan, Karthik, and Daniel A. Cohen, 2013, Competition and Financial Accounting Misreporting, Working

Paper .

Barber, Brad M., Xing Huang, and Terrance Odean, 2016, Which risk factors matter to investors? Evidence from

mutual fund ﬂows .

Barber, Brad M., and Terrance Odean, 2000, Trading Is Hazardous to Your Wealth: The Common Stock Investment

Performance of Individual Investors, The Journal of Finance 55, 773–806.

Barber, Brad M., and Terrance Odean, 2008, All That Glitters: The Eﬀect of Attention and News on the Buying

Behavior of Individual and Institutional Investors, Review of Financial Studies 21, 785–818.

Barclay, Michael J., Neil D. Pearson, and Michael S. Weisbach, 1998, Open-end mutual funds and capital-gains taxes,

Journal of Financial Economics 49, 3–43.

Bauguess, Scott W., John Cooney, and Kathleen Weiss Hanley, 2018, Investor Demand for Information in Newly Issued

Securities, Working Paper .

Bergstresser, Daniel, John M. R. Chalmers, and Peter Tufano, 2009, Assessing the Costs and Beneﬁts of Brokers in the

Mutual Fund Industry, The Review of Financial Studies 22, 4129–4156.

Bergstresser, Daniel, and Thomas Philippon, 2006, CEO incentives and earnings management, Journal of Financial

Economics 80, 511–529.

Berk, Jonathan B., and Richard C. Green, 2004, Mutual Fund Flows and Performance in Rational Markets, Journal of

Political Economy 112, 1269–1295.

Billett, Matthew T., Jon A. Garﬁnkel, and Edward S. O’Neal, 1998, The cost of market versus regulatory discipline in

banking, Journal of Financial Economics 48, 333–358.

Bliss, Robert R., and Mark J. Flannery, 2002, Market Discipline in the Governance of U.S. Bank Holding Companies:

Monitoring vs. Inﬂuencing, Review of Finance 6, 361–396.

Brinn, Tony, Michael J. Peel, and Roydon Roberts, 1994, Audit Fee Determinants Of Independent & Subsidiary

Unquoted Companies In The Uk – An Exploratory Study, The British Accounting Review 26, 101–121.

Brown, Stephen, William Goetzmann, Bing Liang, and Christopher Schwarz, 2008, Mandatory Disclosure and Oper-

ational Risk: Evidence from Hedge Fund Registration, The Journal of Finance 63, 2785–2815.

98

Buzby, Stephen L., 1975, Company Size, Listed Versus Unlisted Stocks, and the Extent of Financial Disclosure, Journal

of Accounting Research 13, 16–37.

Cheng, Ing-Haw, 2011, Corporate Governance Spillovers, Working Paper .

Chevalier, Judith, and Glenn Ellison, 1997, Risk Taking by Mutual Funds as a Response to Incentives, Journal of

Political Economy 105, 1167–1200.

Choi, Stephen J., Karen K. Nelson, and A. C. Pritchard, 2009, The Screening Eﬀect of the Private Securities Litigation

Reform Act, Journal of Empirical Legal Studies 6, 35–68.

Cline, Brandon N., Ralph A. Walkling, and Adam S. Yore, 2016, The Agency Costs of Managerial Indiscretions: Sex,

Lies, and Firm Value, Working Paper .

Cohen, Lauren, and Dong Lou, 2012, Complicated ﬁrms, Journal of Financial Economics 104, 383–400.

Cooper, Michael J., Orlin Dimitrov, and P. Raghavendra Rau, 2001, A Rose.com by Any Other Name, The Journal of

Finance 56, 2371–2388.

Cooper, Michael J., Huseyin Gulen, and P. Raghavendra Rau, 2005, Changing Names with Style: Mutual Fund Name

Changes and Their Eﬀects on Fund Flows, The Journal of Finance 60, 2825–2858.

Dahlquist, Magnus, and Jose Vicente Martinez, 2015, Investor Inattention: A Hidden Cost of Choice in Pension Plans?,

European Financial Management 21, 1–19.

Dangl, Thomas, Youchang Wu, and Josef Zechner, 2008, Market Discipline and Internal Governance in the Mutual

Fund Industry, The Review of Financial Studies 21, 2307–2343.

Dechow, Patricia M., Weili Ge, Chad R. Larson, and Richard G. Sloan, 2011, Predicting Material Accounting

Misstatements, Contemporary Accounting Research 28, 17–82.

Dechow, Patricia M., Richard G. Sloan, and Amy P. Sweeney, 1996, Causes and Consequences of Earnings Manipula-
tion: An Analysis of Firms Subject to Enforcement Actions by the SEC*, Contemporary Accounting Research 13,
1–36.

Del Guercio, Diane, Larry Y Dann, and M. Megan Partch, 2003, Governance and boards of directors in closed-end

investment companies, Journal of Financial Economics 69, 111–152.

Del Guercio, Diane, and Paula A. Tkac, 2002, The Determinants of the Flow of Funds of Managed Portfolios: Mutual

Funds vs. Pension Funds, The Journal of Financial and Quantitative Analysis 37, 523–557.

DellaVigna, Stefano, and Joshua M. Pollet, 2009, Investor Inattention and Friday Earnings Announcements, The

Journal of Finance 64, 709–749.

DeMarzo, Peter M., Michael J. Fishman, and Kathleen M. Hagerty, 1998, The Optimal Enforcement of Insider Trading

Regulations, Journal of Political Economy 106, 602–632.

Dichev, Ilia D., John R. Graham, Campbell R. Harvey, and Shiva Rajgopal, 2013, Earnings quality: Evidence from the

ﬁeld, Journal of Accounting and Economics 56, 1–33.

Dimmock, Stephen G., and William C. Gerken, 2012, Predicting fraud by investment managers, Journal of Financial

Economics 105, 153–173.

Ding, Bill, and Russ Wermers, 2012, Mutual Fund Performance and Governance Structure: The Role of Portfolio

Managers and Boards of Directors, Working Paper .

Donelson, Dain C., Antonis Kartapanis, John McInnis, and Christopher G. Yust, 2017, Measuring Financial Reporting

Fraud Using Public and Private Enforcement .

99

Dyck, Alexander, Adair Morse, and Luigi Zingales, 2010, Who Blows the Whistle on Corporate Fraud?, The Journal

of Finance 65, 2213–2253.

Dyck, I. J. Alexander, Adair Morse, and Luigi Zingales, 2013, How Pervasive is Corporate Fraud?, Working Paper .

Dyreng, Scott D., and Bradley P. Lindsey, 2009, Using Financial Accounting Data to Examine the Eﬀect of Foreign
Operations Located in Tax Havens and Other Countries on U.S. Multinational Firms’ Tax Rates, Journal of
Accounting Research 47, 1283–1316.

Edmans, Alex, Mirko S. Heinle, and Chong Huang, 2016, The Real Costs of Financial Eﬃciency When Some

Information Is Soft, Review of Finance 20, 2151–2182.

Evans, Richard B., and Rudiger Fahlenbrach, 2012, Institutional Investors and Mutual Fund Governance: Evidence

from Retail-Institutional Fund Twins, The Review of Financial Studies 25, 3530–3571.

Fama, Eugene F., and Michael C. Jensen, 1983, Separation of Ownership and Control, The Journal of Law & Economics

26, 301–325.

Fang, Vivian W., Allen H. Huang, and Jonathan M. Karpoﬀ, 2016, Short Selling and Earnings Management: A

Controlled Experiment, The Journal of Finance 71, 1251–1294.

Farber, David B., 2005, Restoring Trust after Fraud: Does Corporate Governance Matter?, The Accounting Review 80,

539–561.

Fee, C. Edward, and Shawn Thomas, 2004, Sources of gains in horizontal mergers: evidence from customer, supplier,

and rival ﬁrms, Journal of Financial Economics 74, 423–460.

Feng, Mei, Weili Ge, Shuqing Luo, and Terry Shevlin, 2011, Why do CFOs become involved in material accounting

manipulations?, Journal of Accounting and Economics 51, 21–36.

Fiordelisi, Franco, and Ornella Ricci, 2014, Corporate culture and CEO turnover, Journal of Corporate Finance 28,

66–82.

Flanagan, Patrick, 1995, The ABCs of changing corporate culture, Management Review 84, 57+.

Franco, Gus De, S. P. Kothari, and Rodrigo S. Verdi, 2011, The Beneﬁts of Financial Statement Comparability, Journal

of Accounting Research 49, 895–931.

Garoupa, Nuno, 2002, The Theory of Optimal Law Enforcement, Journal of Economic Surveys 11, 267–295.

Garrett, Jace, Rani Hoitash, and Douglas F. Prawitt, 2014, Trust and Financial Reporting Quality, Journal of Accounting

Research 52, 1087–1125.

Gaspar, Jose-Miguel, Massimo Massa, and Pedro Matos, 2006, Favoritism in Mutual Fund Families? Evidence on

Strategic Cross-Fund Subsidization, The Journal of Finance 61, 73–104.

Giannetti, Mariassunta, and Tracy Yue Wang, 2016, Corporate Scandals and Household Stock Market Participation,

The Journal of Finance 71, 2591–2636.

Gormley, Todd A., David A. Matsa, and Todd Milbourn, 2013, CEO compensation and corporate risk: Evidence from

a natural experiment, Journal of Accounting and Economics 56, 79–101.

Graham, John R., Campbell R. Harvey, Jillian A. Popadak, and Shiva Rajgopal, 2015, Corporate Culture: Evidence

from the Field .

Graham, John R., Campbell R. Harvey, and Shiva Rajgopal, 2005, The economic Implications of Corporate Financial

Reporting, Journal of Accounting and Economics 40, 3–73.

100

Greenspan, Alan, 2008, The Age of Turbulence: Adventures in a New World (Penguin).

Griﬃn, John M., Samuel A. Kruger, and Gonzalo Maturana, 2016, Do Personal Ethics Inﬂuence Corporate Ethics?,

Working Paper .

Griﬃn, Paul A., Joseph A. Grundfest, and Michael A. Perino, 2004, Stock Price Response to News of Securities Fraud

Litigation: An Analysis of Sequential and Conditional Information, Abacus 40, 21–48.

Grossman, Sanford, 1976, On the Eﬃciency of Competitive Stock Markets Where Trades Have Diverse Information,

The Journal of Finance 31, 573–585.

Grossman, Sanford J., and Oliver D. Hart, 1982, Corporate Financial Structure and Managerial Incentives, University

of Chicago Press 107–14.

Guercio, Del, Diane, Jonathan Reuter, and Paula Tkac, 2010, Demand for Financial Advice, Broker Incentives, and

Mutual Fund Market Segmentation, Working Paper .

Guiso, Luigi, Paola Sapienza, and Luigi Zingales, 2008, Trusting the Stock Market, The Journal of Finance 63,

2557–2600.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales, 2015, The value of corporate culture, Journal of Financial Economics

117, 60–76.

Gurun, Umit G., Noah Stoﬀman, and Scott E. Yonker, 2018, Trust Busting: The Eﬀect of Fraud on Investor Behavior,

The Review of Financial Studies 31, 1341–1376.

Hartzell, Jay C., and Laura T. Starks, 2003, Institutional Investors and Executive Compensation, The Journal of Finance

58, 2351–2374.

Healy, Paul M., and James M. Wahlen, 1999, A Review of the Earnings Management Literature and Its Implications

for Standard Setting, Accounting Horizons 13, 365–383.

Heﬂebower, R. B., 1957, Barriers to New Competition, The American Economic Review 47, 363–371.

Herﬁndahl, Orris, 1950, Concentration in the steel industry, Columbia Univ., Ph. D. thesis.

Hermalin, Benjamin E., and Michael S. Weisbach, 2012, Information Disclosure and Corporate Governance, The

Journal of Finance 67, 195–233.

Hirshleifer, David, and Siew Hong Teoh, 2003, Limited attention, information disclosure, and ﬁnancial reporting,

Journal of Accounting and Economics 36, 337–386.

Hoberg, Gerard, and Gordon Phillips, 2010, Product Market Synergies and Competition in Mergers and Acquisitions:

A Text-Based Analysis, The Review of Financial Studies 23, 3773–3811.

Hoberg, Gerard, and Gordon Phillips, 2016, Text-Based Network Industries and Endogenous Product Diﬀerentiation,

Journal of Political Economy 124, 1423–1465.

Hoberg, Gerard, Gordon Phillips, and Nagpurnanand Prabhala, 2014, Product Market Threats, Payouts, and Financial

Flexibility, The Journal of Finance 69, 293–324.

Holmstrom, Bengt, 1982, Moral Hazard in Teams, The Bell Journal of Economics 13, 324–340.

Houge, Todd, and Jay Wellman, 2005, Fallout from the Mutual Fund Trading Scandal, Journal of Business Ethics 62,

129–139.

Hsu, Charles, Xi Li, Zhiming Ma, and Gordon M. Phillips, 2017, Does Product Market Competition Inﬂuence Analyst

Coverage and Analyst Career Success?, Working Paper .

101

Hubbard, R. Glenn, Michael F. Koehn, Stanley I. Ornstein, Marc Van Audenrode, and Jimmy Royer, 2010, The Mutual

Fund Industry: Competition and Investor Welfare (Columbia University Press).

Ivkovic, Zoran, James Poterba, and Scott Weisbenner, 2005, Tax-Motivated Trading by Individual Investors, The

American Economic Review 95, 1605–1630.

Jayaraman, Sudarshan, and Todd Milbourn, 2009, Does equity-based CEO compensation really increase litigation risk,

Working Paper .

Jia, Yuping, Laurence Van Lent, and Yachang Zeng, 2014, Masculinity, Testosterone, and Financial Misreporting,

Journal of Accounting Research 52, 1195–1246.

Jickling, Mark, and Edward V. Murphy, 2010, Who Regulates Whom? An Overview of U.S. Financial Supervision,

CRS Report for Congress .

Johnson, William C., Wenjuan Xie, and Sangho Yi, 2014, Corporate fraud and the value of reputations in the product

market, Journal of Corporate Finance 25, 16–39.

Karaevli, Ayse, and Edward J. Zajac, 2013, When Do Outsider CEOs Generate Strategic Change? The Enabling Role

of Corporate Stability, Journal of Management Studies 50, 1267–1294.

Karpoﬀ, Jonathan M., Allison Koester, D. Scott Lee, and Gerald S. Martin, 2017, Proxies and Databases in Financial

Misconduct Research, The Accounting Review 92, 129–163.

Karpoﬀ, Jonathan M., D. Scott Lee, and Gerald S. Martin, 2008, The Cost to Firms of Cooking the Books, The Journal

of Financial and Quantitative Analysis 43, 581–611.

Kedia, Simi, and Thomas Philippon, 2009, The Economics of Fraudulent Accounting, The Review of Financial Studies

22, 2169–2199.

Khorana, Ajay, Peter Tufano, and Lei Wedge, 2007, Board structure, mergers, and shareholder wealth: A study of the

mutual fund industry, Journal of Financial Economics 85, 571–598.

Kim, Yongtae, Myung Seok Park, and Benson Wier, 2012, Is Earnings Quality Associated with Corporate Social

Responsibility?, The Accounting Review 87, 761–796.

Koh, Ping-Sheng, and David M. Reeb, 2015, Missing R&D, Journal of Accounting and Economics 60, 73–94.

Kotter, John P., 1995, Leading Change: Why Transformation Eﬀorts Fail.

Lowry, Michelle, and G. William Schwert, 2002, IPO Market Cycles: Bubbles or Sequential Learning?, The Journal

of Finance 57, 1171–1200.

Malmendier, Ulrike, and Geoﬀrey Tate, 2005, CEO Overconﬁdence and Corporate Investment, The Journal of Finance

60, 2661–2700.

Martinez Peria, Maria Soledad, and Sergio L. Schmukler, 2001, Do Depositors Punish Banks for Bad Behavior?

Market Discipline, Deposit Insurance, and Banking Crises, The Journal of Finance 56, 1029–1051.

Mattingly, James E., and Shawn L. Berman, 2006, Measurement of Corporate Social Action Discovering Taxonomy

in the Kinder Lydenburg Domini Ratings Data, Business & Society 45, 20–46.

Mironov, Maxim, 2015, Should one hire a corrupt CEO in a corrupt country?, Journal of Financial Economics 117,

29–42.

Nalebuﬀ, Barry J., and Joseph E. Stiglitz, 1983, Prizes and Incentives: Towards a General Theory of Compensation

and Competition, The Bell Journal of Economics 14, 21–43.

102

Nanda, Vikram, Z. Jay Wang, and Lu Zheng, 2004, Family Values and the Star Phenomenon: Strategies of Mutual

Fund Families, The Review of Financial Studies 17, 667–698.

Nickell, Stephen J., 1996, Competition and Corporate Performance, Journal of Political Economy 104, 724–746.

Parrino, Robert, 1997, CEO turnover and outside succession A cross-sectional analysis, Journal of Financial Economics

46, 165–197.

Popadak, Jillian A., 2013, A Corporate Culture Channel: How Increased Shareholder Governance Reduces Firm Value,

Working Paper .

Qian, Meijun, and Basak Tanyeri, 2017, Litigation and mutual-fund runs, Journal of Financial Stability 31, 119–135.

Rangan, Srinivasan, 1998, Earnings management and the performance of seasoned equity oﬀerings, Journal of

Financial Economics 50, 101–122.

Rashes, Michael S., 2001, Massively Confused Investors Making Conspicuously Ignorant Choices (MCI-MCIC), The

Journal of Finance 56, 1911–1927.

Reynolds, J. Kenneth, and Jere R Francis, 2000, Does size matter? The inﬂuence of large clients on oﬃce-level auditor

reporting decisions, Journal of Accounting and Economics 30, 375–400.

Schein, Edgar H., 1992, Organizational Culture and Leadership (John Wiley & Sons).

Schmidt, Klaus M., 1997, Managerial Incentives and Product Market Competition, The Review of Economic Studies

64, 191–213.

Schneider, Benjamin, 1987, The People Make the Place, Personnel Psychology 40, 437–453.

Schrand, Catherine M., and Sarah L. C. Zechman, 2012, Executive overconﬁdence and the slippery slope to ﬁnancial

misreporting, Journal of Accounting and Economics 53, 311–329.

Shleifer, Andrei, 2004, Does Competition Destroy Ethical Behavior?, Working Paper .

Sialm, Clemens, and Laura Starks, 2012, Mutual Fund Tax Clienteles, The Journal of Finance 67, 1397–1422.

Sialm, Clemens, Laura T. Starks, and Hanjiang Zhang, 2015, Deﬁned Contribution Pension Plans: Sticky or Discerning

Money?, The Journal of Finance 70, 805–838.

Sialm, Clemens, and T. Mandy Tham, 2015, Spillover Eﬀects in Mutual Fund Companies, Management Science 62,

1472–1486.

Sirri, Erik R., and Peter Tufano, 1998, Costly Search and Mutual Fund Flows, The Journal of Finance 53, 1589–1622.

Song, Moon H, and Ralph A Walkling, 2000, Abnormal returns to rivals of acquisition targets: A test of the ‘acquisition

probability hypothesis’, Journal of Financial Economics 55, 143–171.

Sun, Yang, 2014, The Eﬀect of Index Fund Competition on Money Management Fees, Working Paper .

Teoh, Siew Hong, Ivo Welch, and T.j. Wong, 1998, Earnings Management and the Long-Run Market Performance of

Initial Public Oﬀerings, The Journal of Finance 53, 1935–1974.

Thaler, Richard H., and Cass R. Sunstein, 2008, Nudge: Improving Decisions about Health, Wealth, and Happiness

(Yale University Press).

Thompson, Robert B., and Hillary A. Sale, 2003, Securities Fraud as Corporate Governance: Reﬂections upon

Federalism, Vanderbilt Law Review 56, 859–910.

Tirole, Jean, 1988, The Theory of Industrial Organization (MIT Press).

103

Tirole, Jean, 2010, The Theory of Corporate Finance (Princeton University Press).

Tufano, Peter, and Matthew Sevick, 1997, Board structure and fee-setting in the U.S. mutual fund industry, Journal of

Financial Economics 46, 321–355.

Wang, Tracy Yue, and Andrew Winton, 2014, Product Market Interactions and Corporate Fraud, Working Paper .

Wang, Tracy Yue, Andrew Winton, and Xiaoyun Yu, 2010, Corporate Fraud and Business Conditions: Evidence from

IPOs, The Journal of Finance 65, 2255–2292.

Whited, Toni M., and Guojun Wu, 2006, Financial Constraints Risk, The Review of Financial Studies 19, 531–559.

Wu, Kai, 2017, The Economic Consequences of Mutual Fund Advisory Misconduct, Working Paper .

104