Journal of Accounting and Economics 55 (2013) 183–205

Contents lists available at SciVerse ScienceDirect

Journal of Accounting and Economics

journal homepage: www.elsevier.com/locate/jae

The spillover effect of fraudulent financial reporting on peer

firms’ investments$, $$
Anne Beatty a,n, Scott Liao b,1, Jeff Jiewei Yu c,2
a
Fisher College of Business, The Ohio State University, United States
b
Rotman School of Management, University of Toronto, Canada
c
Cox School of Business, Southern Methodist University, United States

a r t i c l e i n f o abstract

Article history: We investigate how high-profile accounting frauds affect peer firms’ investment.
Received 5 May 2011 We document that peers react to the fraudulent reports by increasing investment during
Received in revised form fraud periods. We show that this finding is not driven by frauds that have a higher
7 January 2013
ex ante likelihood of detection or by an association between fraud and investment
Accepted 21 January 2013
Available online 8 February 2013
booms. In addition, we find that peers’ investments increase in fraudulent earnings
overstatements, and in industries with higher investor sentiment, lower cost of capital
JEL classification: and higher private benefits of control. We also find evidence consistent with equity
G30 analysts potentially facilitating the spillover effect.
M41
& 2013 Elsevier B.V. All rights reserved.

Keywords:
Fraudulent financial reporting
Spillover effect
Investment efficiency

1. Introduction

How accounting information affects investment is a fundamentally important question. A growing literature studies
how a firm’s accounting quality affects its own investment. However, there is little systematic evidence on how one firm’s
accounting information affects other firms’ real investment behavior (Leuz and Wysocki, 2008). We focus on ‘‘the adverse
effects of bad accounting’’ (Francis, 2001) by examining the spillover effect of fraudulent reporting on peers’ investments.
Within an industry one firm’s disclosures can have a spillover effect on other firms’ investments. Bushman and Smith
(2001) argue that ‘‘managers can identify promising new investment opportunities on the basis of the high profit margins
reported by other firms.’’ Therefore, distortion of the investment opportunity signal by fraudulent accounting may lead to
sub-optimal investments by industry peers. In addition, Gigler (1994) demonstrates theoretically that a threat of increased
competitor output in the product market is required to make favorable information disclosure credible in the capital

$
We would like to thank the anonymous reviewer, Anil Arya Hans Christiansen, Bill Cready, Merle Erickson, Pingyang Gao, Weili Ge, S.P. Kothari (the
editor), Chad Larson, Christian Leuz, Doug Skinner, Abbie Smith, Ro Verrecchia, workshop participants at the University of Chicago, New York University
Summer Camp and Southern Methodist University and the participants of the 2009 CAPANA Conference, the 2009 AAA Annual Meeting, the 2010 Lone
Star Conference and the 2011 Washington University Dopuch Conference for their valuable comments.
$$
Data availability: All data are available from public sources.
n
Corresponding author. Tel.: þ1 614 292 5418.
E-mail addresses: [email protected] (A. Beatty), [email protected] (S. Liao), [email protected] (J.J. Yu).
1
Tel.: þ1 416 946 8599.
2
Tel.: þ1 214 768 8321.

0165-4101/$ - see front matter & 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jacceco.2013.01.003
184 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

market, suggestive of increased investments in response to peers’ favorable accounting information. Consistent with this
possibility, Kumar and Langberg (2010) develop a rational expectations model where peers increase investments as an
equilibrium response to the industry leader’s inflated productivity report.
Based on these arguments and theories, we hypothesize that peers will invest more during the scandal period than
during the pre-scandal period in response to the industry leader’s overstated earnings. To test our hypothesis, we examine
the effect of 35 accounting frauds on 2305 industry peers’ investments during the scandal period. We focus on industry
leaders’ high-profile frauds (Fortune 500 firms accused of frauds in SEC Accounting and Auditing Enforcement Releases)
because their financial reports are highly visible and likely to be used to evaluate market conditions and investment
opportunities. We use a difference-in-differences design where peers share the fraud firm’s 3-digit SIC code and control
firms share the fraud firm’s 2-digit SIC code.3 We find significantly greater capital expenditures by the peer relative to the
control firms during the fraud period compared to the three-preceding-year control period.4
We conduct several sensitivity analyses to ensure the interpretation of our results. As an alternative to identifying
control firms based on SIC codes, we use peers’ secondary segments as an alternative control group. One advantage of this
identification strategy is that a firm can serve as its own control thereby circumventing the concern that firms with the
same 2-digit SIC codes may have different growth opportunities or other firm characteristics from the peers. We find that
the spillover effect on peers’ investments continues to hold using this alternative control group.
We take two steps to alleviate concerns that our results might be driven by the examination of ex post detected frauds. First,
we test whether the results differ based on the Dechow et al. (2011) ex ante fraud detection probability. If ex post fraud detection
drives the results, then we would expect a more pronounced peer investment effect when the ex ante probability of fraud
detection is higher. In contrast, if peers respond less to financial reports when the expectation of fraud is higher, then the results
should be more pronounced when the ex ante fraud detection probability is lower. Our finding of the latter possibility provides
support for examining peers’ behavior during the period that is only discovered in retrospect to be a fraud period.
Second, we examine the effect of the expected fraud detection probability on peers’ investments when an industry
leader has a large reported revenue increase, potentially due to earnings management when there is no ex post fraud
detection. The results are similar to those for our fraud sample. Specifically, when the expected likelihood of fraud is
higher, peers are less likely to respond to increased reported revenue by industry leaders compared to when the expected
likelihood of fraud is lower. This suggests that our results are not driven by the examination of detected frauds.
We also conduct several additional tests to mitigate reverse causality or omitted variables concerns. First, we examine
the lead–lag relation between fraud and investment. Our finding that fraud precedes increases in peers’ investments,
combined with our failure to find peer investment increases prior to the fraud, supports the inference that the increases in
peers’ investments result from fraud rather than the reverse. Next, we investigate whether the spillover effect differs
between competitive and concentrated industries, and between high and low growth industries based on Wang and
Winton’s (2010) finding that firms’ fraud propensity is insensitive to industry investment booms in concentrated
industries, but that firms in competitive industries have a pro-cyclical propensity to commit frauds. If industry booms
cause both frauds and increased peer investments, then we should observe stronger results in higher growth and in more
competitive industries. The lack of significant differences in the relation between fraud and peer investment in high versus
low growth or in competitive versus concentrated industries further mitigates reverse causality concerns.5
After documenting that the industry leader’s accounting fraud affects peer investment during the scandal period, we
then consider cross-sectional variations in the extent of the spillover effect. Kumar and Langberg (2010) theorize that the
spillover effect should increase with the magnitude of falsified productivity, especially in industries with higher investor
sentiment, lower cost of capital and higher private benefits of control. Accordingly, we hypothesize that the spillover effect
on peers’ investment will be positively associated with the magnitude of the scandal firm’s earnings overstatement and
will be stronger in industries with higher investor sentiment, lower cost of capital and higher private benefits of control.
We further examine whether the spillover effect may be partially fueled by propagation of industry-wide information
contained in analysts’ recommendations. Cotter and Young (2007) find that analysts are not able to distinguish true frauds
from fictitious frauds. Akhigbe et al. (2006) argue that because analysts’ recommendations reflect overall industry
prospects, industry rivals often experience valuation effects in the same direction as the revised firm. If analysts help
transmit distorted industry investment signals, then we expect the unexplained overlap in analysts’ coverage of scandal
firms and peers will strengthen the spillover effect of distorted signals on peers’ investment decisions.6 In addition, we
directly examine whether equity analysts’ recommendations are more favorable for peers during the scandal period.

3
Our sample of 5454 test and control firms includes more than 56% of COMPUSTAT firms that report positive assets, market-to-book, total debt and
capital expenditures during the sample period 1999–2009.
4
We confirm the fraud period using subsequent earnings restatements. We use the terms ‘‘fraud period’’ and ‘‘scandal period’’ interchangeably to
denote the period of fraudulent financial reporting prior to detection.
5
These tests also address concerns that peer investment changes lead to fraud detection because Wang and Winton (2010) use ex post detected
frauds to test Povel et al.’s (2007) predictions that fraud propensity increases during investment booms in competitive vs. concentrated industries. Their
evidence suggests that ex post detected frauds are an order preserving proxy for unobservable fraud propensity.
6
Unexplained overlap is the residual from a model designed to remove the economic similarity between the scandal firms and peer/control firms.
More details are provided in Section 4.2.
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 185

Finally, we argue that peers’ investments during the fraud period resulting from distorted signals by industry leaders should
produce a weaker association between future cash flows and investments compared to investments in previous periods.
We find that peers’ investments are positively associated with the magnitude of the scandal firm’s earnings
overstatements during the scandal period. In addition, we find that peers’ investments are greater in industries where
the investor sentiment is high, the industry cost of capital is low and managers’ private benefits of control are large.
We also document that our investment findings hold only when the unexplained overlap in analyst coverage between scandal
firms and peers is high. In addition, we find that equity analysts’ recommendations are more favorable during the scandal period
for peers with high unexplained overlap in analyst coverage with scandal firms. Finally, we find a lower correlation between
peers’ scandal period investments and future performance, which persists for at least three years after the investment.
In supplemental analyses, we find that peers’ insiders purchase more shares during the scandal period. Further, we find
the spillover effect of leaders’ restated earnings on peer investment to be similar to that of overstated earnings. Finally, we
find that the peer investment spillover effect only holds for revenue recognition scandals.7
Our paper makes several contributions. By examining the spillover effect of fraudulent financial reporting on real
investment behavior, we address an issue identified as under-explored by Leuz and Wysocki (2008). Our findings suggest
that a leading firm’s distorted accounting information is important in peers’ investment prior to detection of misreporting.
We contribute to the fraud dynamics literature by analyzing the lead–lag and contemporaneous associations between
fraud and peer investment and by examining how the association differs based on industry competition and growth. These
analyses complement Durnev and Mangen’s (2009) finding that industry competitors’ investments are lower in the year
after restatement announcements by focusing on the causal link between fraud and peer investment. In addition, we
extend their findings by considering analysts’ role in transmitting the distorted investment opportunity signal.
Furthermore, we contribute to the literature documenting consequences of financial statement misreporting. Francis
(2001) calls for research on ‘‘the adverse effects of bad accounting.’’ Sadka (2006) also argues that the existing literature
understates the economic consequences of accounting fraud. We respond to their call for research by providing systematic
evidence that accounting frauds distort industry competitors’ investment decisions and result in investments that
generate lower future cash flows. Our study, however, is subject to some limitations. First, despite our efforts in addressing
endogeneity, the relation between peer investments and industry leaders’ reporting should be interpreted with cautions.
Second, the analysis of the effect of analysts overlap on the investment spillover is subject to an alternative interpretation
that analyst overlap may be an alternative proxy for more closely related peer firms.
The rest of the paper is organized as follows. Section 2 provides background information. We develop our hypotheses in
Section 3. We describe our research design and sample in Section 4. We present our results in Section 5, robustness checks
in Section 6, and conclude in Section 7.

2. Background

2.1. Anecdotal evidence

On June 26, 2002, the SEC filed a complaint charging WorldCom with ‘‘a massive accounting fraud totaling more than $3.8
billion.’’ WorldCom later admitted that from 1999 through 2002, the company materially overstated its reported earnings by about
$9 billion in the fraud. In a WorldCom case study, Sidak (2003) concludes that ‘‘WorldCom’s false internet traffic reports and
accounting fraud encouraged [competitor’s] overinvestment in long-distance capacity and Internet backbone capacity.’’ World-
Com’s overstated earnings distorted the economic gains of acquiring new customers causing other firms to overinvest. AT&T Labs
reported in 2001 that rivals made investments relying on WorldCom’s fraudulent reports. The Eastern Management Group also
argued that a substantial fraction of the $90 billion invested by other carriers was due to WorldCom’s faulty projections.

2.2. Related research

2.2.1. Accounting information and firms’ own investment behavior

Prior literature examines the direct effect of a firm’s accounting quality on its own investment efficiency. Several
studies including Biddle and Hilary (2006), Biddle et al. (2009) and Beatty et al. (2010) examine how differences in
accounting quality in the absence of misstatements affect firms’ investments. Kedia and Philippon (2009) and Sadka (2006)
predict that fraud firms will overinvest in equilibrium where bad managers, who fraudulently boost reported accounting
numbers, mimic good managers’ investments to maintain consistency between reported performance and investments.8
Kumar and Langberg (2009) extend this literature by developing a rational expectations equilibrium model where fraud
and overinvestment jointly occur. In their model, agency conflicts arise because privately informed managers derive
personal benefits from larger investments, and shareholders cannot credibly pre-commit to induce manager truth-telling
through ex post inefficient investment policies that are renegotiable in an active takeover market. They show that the
optimal renegotiation-proof contract induces insider misreporting resulting in rational overinvestment in some states.

7
This finding is consistent with prior theory papers, e.g., Darrough and Stoughton (1990) and Gigler (1994).
8
Sadka (2006) further predicts negative externalities induced by fraud firms’ falsified financial reports.
186 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

2.2.2. Truthful versus non-truthful disclosure

The theory-based literature has posited alternative ways that ‘‘good news’’ by one firm might affect the investment
activity of other firms. Specifically, given the key assumption that firms report truthfully, ‘‘good-news’’ disclosures can be
negatively associated with peer investment. In the absence of truthful disclosure, however, there is little support for a
negative association.
Bagnoli and Watts (2010) note that the prior literature on the interaction between disclosures and product market
competition requires that firms make unbiased disclosures. With truthful disclosure, under Cournot competition, reporting
good news about firm specific information will induce competitors to reduce production. Beyond the difficulty of
classifying competition (Cournot vs. Bertrand) in reality, Verrecchia (2001) identified the reliance on truthful reporting as
an issue of concern with this literature by saying ‘‘some have questioned the assumption that if the manager chooses to
release her private information, then she does so truthfully.’’ He further notes that, while this restriction is descriptive in
many accounting settings, there are instances where it is difficult to assess the integrity of the manager’s disclosure.
Beyer et al. (2010) discuss two types of models that relax the assumption of truthful reporting and argue that ‘‘How
much investors can learn from corporate disclosures when management may misrepresent its information depends to a
large extent on whether misrepresentation is costless (cheap-talk models) or costly (costly state falsification models).’’ In
cheap-talk models there is no direct disclosure cost so there must be an indirect cost for the disclosure to be credible. For
example, Gigler (1994) shows that, when firms would like to disclose high product demand to mislead the capital market,
the disclosure is made credible to the capital market only when there is a product market cost of disclosing high demand.
Gigler’s results suggest that biased reporting would not be credible in the capital market if it also induced a production
market benefit (such as entry deterrence) to the firm that makes the (biased) report.9
These papers suggest that the circumstances under which competitors can induce production cuts using biased
disclosure are limited. In contrast, we rely on theories (e.g., Kumar and Langberg (2010), Gigler (1994)) with relatively
relaxed assumptions that build on more realistic frameworks.10 In summary, given our focus on biased disclosures, our
assumption that the disclosure of fraudulent ‘‘good news’’ will induce greater investment by other firms in the industry
seems reasonable and consistent with both anecdotal evidence and analyses that explicitly model investment distortions
in industries when firms manipulate information to overstate their productivity.

2.2.3. Accounting information and spillover effects

Gleason et al. (2008) find that when accounting restatements adversely affect the restating firms’ shareholder wealth they
also induce share price declines among non-restating industry peers. This suggests that accounting restatements make
investors reassess non-restating firms’ previous financial statements. This study provides initial evidence of restatements
contagion effects, but does not investigate their real effects.
Kumar and Langberg (2010) argue that outsiders’ posterior beliefs on industry growth potential and their investment
responses depend on the incumbent firm’s disclosure. Privately informed managers of leading firms have a strong incentive to
manipulate outsiders’ beliefs by inflating their performance to attract investment. Therefore, along the perfect Bayesian
equilibrium investment path, inflated performance reports by the leading firm could lead to a run-up in the market expectation
of industry productivity that diverges from the true state, resulting in over-investments by competitors entering the industry to
exploit the new investment opportunities.
Durnev and Mangen (2009) investigate whether restatement announcements are associated with systematic changes in
peers’ investments. They find that peers significantly lower their investment growth in the year after a competitor’s
restatement announcement. Assuming that these announcements are exogenous shocks, they infer that peers learn from
the new information and modify their investment strategies accordingly.

2.2.4. The effect of investment on the incidence of fraud

Povel et al. (2007) model a firm’s fraud decision based on investors’ priors about the economy and the cost of
monitoring and predict a non-monotonic relationship between optimistic priors and the incidence of fraud. Specifically,
they predict that low priors lead to low fraud incentives because uncertain investors will monitor firms carefully even if
public reports are positive. When priors are fairly optimistic the fraud incentive is high because investors do not carefully
monitor firms with confirming positive public reports, but they do monitor firms with inconsistent negative public reports.
However, fraud incentives are low again when priors are so optimistic that investors do not even monitor firms with
negative public reports.
Wang and Winton (2010) test this model by comparing the association between investment booms and fraud
propensity in competitive versus concentrated industries. Their results using either industry level or firm specific data
suggest a more positive association between fraud propensity and investment booms in competitive than in concentrated
industries, although the likelihood of frauds does not increase with investment booms on average.

9
While Bagnoli and Watts (2010) costly state falsification model shows that biased reporting may induce production cuts in equilibrium with
rational investors, their assumption that misreporting costs are private information and there are multiple misreporting firms in Cournot duopoly
competition is highly stylized.
10
For example, Kumar and Langberg’s model assumes that managers have private benefits from controlling larger investments and have private
information on investments, and that shareholders cannot credibly pre-commit to investment that is inefficient ex post due to active takeover markets.
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 187

2.2.5. Analysts’ role in the spillover effect

Cotter and Young (2007) find that analysts are not able to distinguish true from fictitious frauds suggesting that they
may help transmit the distorted signals. Furthermore, Jensen (2005) argues that analysts pressure managers for higher
growth to justify overvalued equity, leading managers to increase their capital investments. Finally, Akhigbe et al. (2006)
argue that because analysts’ recommendations reflect overall industry prospects, industry rivals are likely to experience
similar valuation effects.

3. Hypothesis development

3.1. Association between fraud and peer investment

High-profile firms’ fraudulent financial reporting can have real effects on the investments of industry peers if they rely
on the fraudulent financial reports to mitigate product market uncertainty and to distinguish between promising and
inauspicious investments. We hypothesize that when a high-profile firm materially inflates its reported financial
performance, the overstated investment prospects will encourage peers to make more investments than they would
absent the misleading information.

H1a. Investment by peers will be greater in the scandal period compared to investment in the pre-scandal period.

Based on Kumar and Langberg (2010), we hypothesize that the magnitude of peers’ investments will be positively
associated with the amount of overstated profits.

H1b. Scandal period investment by peers is positively associated with the magnitude of the earnings overstatement by the
scandal firms.

Kumar and Langberg (2010) also contend that strategic information manipulation by one firm is more likely to generate
a dynamic externality on industry-wide investment distortion ‘‘when the cost of capital is low or when the agency costs [of
private control benefits] are high or when investors have optimistic a priori assessment of the growth potential of the
innovation.’’ Under these conditions, inflated reports are more likely to cause a run-up in outsiders’ posterior beliefs that
diverge from the true industry productivity. More optimistic market expectations in turn dampen the information content
of subsequent signals and prevent peers from uncovering the true productivity, thereby increasing peers’ reliance on the
lead firm’s profit reports in making investment decisions. Therefore, we hypothesize that the spillover effect of fraudulent
earnings on peers’ investment will be stronger when these three conditions are met.

H1c. Scandal period investment by peers is greater when the cost of capital is low, managers’ private benefit of control is
high, and investor industry sentiment is high.

Based on Jensen’s (2005) argument, pressure from analysts for higher growth to justify overvalued equity leads
managers to make greater capital investments. We argue that analysts may therefore play an information transmission
role that facilitates the real spillover effect. For hypotheses H1a–c, we also predict that the magnitude of the effect will
depend on the extent of overlap between analysts’ coverage of the scandal firm and of the industry peers. That is, we
expect that when the analysts covering the scandal firm cover more peers, the investment spillover effects should be
stronger.

3.2. Industry effects of analyst recommendations

Based on findings in Cotter and Young (2007) that analysts cannot distinguish true from fictitious frauds and in Akhigbe
et al. (2006) that analysts’ recommendations reflect overall industry prospects, we hypothesize that analysts’ recommen-
dations will be more favorable for scandal firms’ peers.

H2. Equity Analyst’s recommendations for peers will be more favorable in the scandal period compared to the pre-scandal
period.

3.3. Efficiency of peers’ investments

Fraudulent financial reporting provides misleading information about product market demand and profitability. To the
extent that peers invest based on the falsified scandal period information, we hypothesize that the investments will have a
weaker association with future performance compared to previous investments.

H3. The association between peers’ investments and future performance is weaker during the scandal period compared to
the pre-scandal period.
188 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

4. Research design

To identify fraudulent reporting firms, we adopt a strategy similar to Karaoglu et al. (2006) by focusing on accounting
frauds of a group of high-profile firms.11 We define these high-profile scandal firms as those that were accused of
accounting fraud by the SEC from 1999 to 2009, and were in the Fortune 500. These scandal firms are more likely to be
leading firms in their industries due to their size and visibility. For each industry classified using 3-digit SIC codes, if we
identified more than one high-profile scandal firm, we only include the first firm that commit a fraud in the sample
period.12 We also exclude financial institutions (SIC code 6000-6999) because financial institutions’ investment behaviors
are different from other industries.
To identify scandal periods, we start with the periods stated in SEC Accounting and Auditing Enforcement Releases.
Then we use historical 10 K filings to verify that the scandal firm actually restated reported earnings during the period
after the fraud is detected. We identify 35 scandal firms from 35 distinct industries. We provide descriptive statistics about
the magnitude of the frauds in Panel A of Table 1. Similar to previous studies, revenue recognition related frauds are the
predominant type of fraud, making up 40% of our sample. The mean restatement amount in earnings in our sample is
slightly over $423 million for revenue recognition related frauds and $282 million for other frauds and represents nearly
16% of the average pre-scandal period revenues for revenue recognition related frauds and 7.6% for other frauds.
We define peers as firms with the same 3-digit SIC code as the scandal firms. Our sample includes 2305 peers
distributed among 35 industries. Although this includes only 14% of the 250 3-digit SIC codes represented in COMPUSTAT,
our peer sample encompasses 24% of all non-financial COMPUSTAT firms and 44% of total non-financial COMPUSTAT
assets due to larger than average sample industries. The number and size of our sample peers suggest that a relatively
small number of frauds could have a large economic impact.
To investigate how the scandal firm’s fraudulent reporting affects peers’ investments during the scandal period vis-a -
vis the pre-scandal period, we employ a difference-in-differences approach to control for industry and time effects, where
the pre-scandal period is defined as the three years before the onset of the scandal period. We use firms with the same
2-digit SIC code as the scandal firm (excluding peers) as the control group.13 Our control sample of 3149 firms when
combined with our test sample of 2305 firms covers 56% of all non-financial COMPUSTAT firms and 80% of total non-
financial COMPUSTAT assets. We assume that firms with the same 2-digit SIC codes share similar overall growth
opportunities but non-peers are less likely to rely on scandal firms’ financial reports for their investment decisions than
peers.14 To address concerns with this approach that firms with the same 2-digit SIC codes may be different in firm
characteristics, we employ a second identification strategy by using peers’ segments that operate in other industries as
controls for the segments in the same industry as the fraud firm.
The observations in our models are clustered in the time periods surrounding the 35 sample frauds, with an average of
nearly 100 firms per scandal year. We address the resulting lack of independence by clustering the model standard errors
by time. In addition, sample firms could appear during each of the three pre-scandal period years and in each scandal
period year. We also check the sensitivity of our results to clustering by both year and firm. We find that two-way
clustering does not change the results.15
To provide additional support for our hypothesized relation between the fraud firms’ overstated earnings and peers’
investment incentives, we examine peers’ exit and entry into the industry, defined using the 3-digit SIC codes, during the
scandal period. The results of these analyses are reported in Panel B of Table 1. Consistent with an incentive for peers to
increase their investments during the scandal period, we find a significant decline in the ratio of firms exiting the industry
to the number of firms in the industry, and a significant increase in the ratio of firms entering the industry to the number
of firms in the industry. These results suggest that competitors have incentives to increase investment as a result of
the fraud.

4.1. Investment model

To test our hypotheses that peers increase their investment in the periods when the leading firm committed a fraud, we
run the following OLS regression.16
CAPEX ¼ b0 þ b1 nPEER þ b2 nSCAN þ b3 nRESTATEðI_FACTORÞ þ b4 nPEERnSCAN
þ b5 nPEERnSCAN nRESTATEðPEERnSCAN nI_FACTORÞ

11
We follow prior literature (e.g., Wang and Winton (2010)) and focus on ex post detected frauds.
12
We exclude the remaining scandal firms from ‘‘peer firms’’ and ‘‘the control group’’ defined below.
13
As a robustness check, we also assign firms that share the scandal firm’s 2-digit SIC as peer firms and assign those that share the scandal firm’s
one-digit SIC code as control firms. Our results continue to hold.
14
We compare peer and control firms’ size and market-to-book ratios in the pre-scandal periods. We delete the top 2% of size and market-to-book
ratios for the control firms to make these two groups comparable.
15
In fact, two-way clustering increases t-values of our variables of interest in most regressions.
16
We measure investment using only capital expenditures due to limited R&D data availability across all fraud industries and at the segment level.
However, using available R&D data in our measure of investment does not change our results.
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 189

Table 1
Descriptive statistics for restatements and industry exits and entries.

Panel A: Restatement amounts

Revenue recognition related restatements Other restatements

Restatement Restatement amount/average Restatement Restatement amount/average

amount ($000,000) pre-scandal period Sales amount ($000,000) pre-scandal period sales

Median 49.00 0.021 128.25 0.018

Mean 423.65 0.160 282.20 0.076
Maximum 2794 1.214 2063 0.429
Minimum 0.69 0.001 1.12 0.001
Standard
834.66 0.358 565.79 0.125
deviation
N 14 21

Panel B: Industry exits and entries

Variable
Exit ratio Entry ratio
Coefficients (clustered-t stats) Coefficients (clustered-t stats)

Intercept 0.057 0.144

(7.74)nnn (8.65)nnn
PEER 0.015  0.029
(2.87)nnn (  3.02)nnn
SCAN 0.009  0.032
(0.92) (  1.60)
PEERnSCAN  0.021 0.032
(  2.14)nn (1.89)n
N 310 310
Adjusted R-squared 0.014 0.011

Note: Standard errors are clustered at the year level. nnn, nn and n represent 1%, 5% and 10% significance levels, respectively. Exit is
defined as the ratio of the number of firms that exited the industry to the total number of firms in the industry and entry is
defined as the ratio of the number of firms that enter the industry to the total number of firms in that industry.
Variable Definition:

PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for control firms
that have the same 2-digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and zero for 3 years prior to
the scandal period.

þ b6 nSIZE þ b7 nMTB þ b8 nLEV þ b9 nCFO þ b10 nRATING þ b11 nSG

þ b12 nCAPEX_S þ b13 nCOMOVE þ e ð1Þ
Variable definitions:

CAPEX: the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged properties, plants and equipment (COMPUSTAT
‘‘ppent’’).
PEER: an indicator variable equal to one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for
control firms that have the same 2-digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SCAN: an indicator variable equal to one for years during which fraudulent firms committed frauds (unless explicitly
defined otherwise), and zero for 3 years prior to the scandal period.
RESTATE: tercile rankings of the ratio of fraudulent firms’ total restatement to the average revenues of the pre-scandal
period.
I_FACTOR: an indicator variable for firms in the industries (3-digit SIC) that have investor sentiments higher than the
median of all industries, median earnings-to-price ratio lower than the median of all industries and higher counts of
M&A activities than the annual median.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’  ‘‘ceq’’þ‘‘prcc_f’’n‘‘csho’’) to book value of total
assets (COMPUSTAT ‘‘at’’).
LEV: long term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
CAPEX_S: fraudulent firms’ CAPEX.
190 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

COMOVE: tercile rankings of the co-movement of change in market-to-book ratios between the fraudulent firms and
sample or control firms in the pre-scandal period (at the 3-digit SIC code level). The co-movement is measured as b in
the regression DMTB¼ a þ bDMTB_Sþ e, where DMTB is defined as annual change in MTB and DMTB_S represents
fraudulent firms’ change in MTB.

Our baseline model to test H1a excludes the variables designed to capture cross-sectional variation in the extent of the
spillover effect (i.e., 3-way interaction term as well as the RESTATE or I_FACTOR variable) from model (1). Based on H1a, we
expect the coefficient on PEERnSCAN to be positive.
To address concerns that examining ex post detected frauds may introduce selection bias or that peers only increase
investment when the scandal firms’ earnings management is easy to detect, we first partition our sample based on the
scandal firms’ F-scores (Dechow et al., 2011) capturing earnings management and the ex ante likelihood of fraud detection.
If our results depend on the likelihood of ex post fraud detection, then we would expect the effect of the fraud on peer
investment to be more pronounced for firms with a higher ex ante fraud detection probability. In contrast, if peers are less
likely to respond to financial reports when there is a higher expectation that they are fraudulent, then the results
(PEERnSCAN) should be more pronounced for firms with a lower ex ante fraud detection probability. Additionally, we
investigate the effect of the expected probability of fraud detection on peers’ investment behavior when an industry leader
has a large increase in reported revenue, which can be a result of earnings management, but there is no ex post fraud
detection. That is, to address the selection bias we employ a non-scandal sample where earnings management is likely.
If increased investment arises from the fraudulent signals rather than the propensity to commit fraud increasing with
the investment booms, then we expect the increased investment should not precede the fraud. Also, the increased
investment in the fraud periods should not be greater in competitive versus concentrated industries based on Wang and
Winton (2010), and the increased investment should not be greater for high growth versus low growth industries, where
peer industry sales growth, measured at the inception of the fraud, captures an industry boom.17
Specifically, we examine: whether the increase in peers’ investment in the year prior to the inception of the scandal period
differs from their investment in the three prior years, whether the increase in peers’ investment during the scandal period
differs in competitive versus concentrated industries, and whether the increase in peers’ investment during the scandal period
differs in high growth versus low growth industries. When SCAN is set equal to one in the one-period prior to fraud inception,
we do not expect to find a significant coefficient on PEERnSCAN. Similarly, once SCAN is set equal to one in the years after fraud
inception, if increased peer investment reflects a reaction to the fraud rather than frauds following investment booms, then the
coefficient on PEERnSCAN for firms in competitive industries should not exceed that for firms in concentrated industries and the
coefficient on PEERnSCAN for firms in high growth industries should not exceed that for firms in low growth industries.
To test H1b, we collect the restatement data and use the total restatement in earnings during the scandal period (scaled
by the average sales in the pre-scandal period), we then use the tercile ranking of this variable (i.e., RESTATE) as our proxy
to test whether peers’ investment is affected by leading firms’ misstatements. Based on H1b we expect a positive
coefficient on PEERnSCANnRESTATE.
To test H1c, we use investor sentiment measured as Baker and Wurgler (2006) to proxy for investors’ expectations of
industry productivity. That is, for each year and industry at the 3-digit SIC code level, we take the first principal component of
the following variables using factor analysis: IPO_count, IPO_ret, Turnover, and MTB_diff. We measure IPO_count as the total
number of IPOs, IPO_ret as the average first-day returns of IPOs, Turnover as the average ratio of share volume to the number
of shares outstanding, and MTB_diff as the difference of the market-to-book ratios between divided payers and non-payers.18
We measure the industry cost of capital by calculating the median earnings-to-price ratio for each year and industry at the 3-
digit SIC code level. Finally, to capture the private benefits of control, we count the number of merger and acquisitions for the
industry-year, assuming that management can accrue more private benefits of control by empire building.
To capture the overall effect of the three Kumar and Langberg (2010) variables, we construct a composite variable
I_FACTOR, measured as an indicator variable for firms in the industries (defined by 3-digit SIC codes) that have investor
sentiments higher than the annual median of all industries, earnings-to-price ratio (measured at the median of the
industry-year) lower than the annual median of all industries and higher counts of M&A activities than the annual
median.19 Based on H1c, we expect PEERnSCANnI_FACTOR to have a positive coefficient.
To ensure that we are not merely capturing mimicking behavior we also control for the scandal firm’s investment
(CAPEX_S). Following prior literature (e.g., Biddle and Hilary (2006), Biddle et al. (2009), and Beatty et al. (2010)), we also
control for firm size, market-to-book ratio, existence of S&P rating, leverage, sales growth and operating cash flows.20 We
expect that capital expenditures increase with sales growth and market-to-book ratios, as investment tends to be higher
when the firm has more growth opportunities. Finally, to ensure that we are not just capturing the similarity of growth

17
We also partition the sample using peer firms’ change in sales growth from the year before the fraud inception to the beginning year of the fraud
period. The results are similar.
18
In the primary analysis, we exclude the ratio of total equity issues to the sum of equity and public debt issues to avoid losing four industries.
However, including this variable does not change the main results.
19
I_FACTOR might also capture industry booms. To mitigate this concern, we examine the trend of the components of I_FACTOR over time and find
that these components are quite constant before fraud and then increase (or decrease for cost of capital) after the inception of the fraud period as
predicted by Kumar and Langberg (2010) if these factors fuel frauds. This pattern is not consistent with the reverse causality.
20
We control for whether the firm is rated to control for firms’ access to the public debt market.
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 191

opportunities between the peers and the scandal firms, we control for the co-movement in growth opportunities (proxied
by MTB) between the scandal firms and either peers or control firms measured in the pre-scandal periods.

4.2. Analyst coverage effects

We test the effects of analyst coverage by estimating model (1) for sub-samples based on analyst coverage overlap. We
classify industries based on the extent to which peer and control firms are covered by the same analysts as the scandal
firms. Because firms with more economic similarity are more likely to be covered by the same analysts, we adopt
the following procedure to remove this component from the ratio of overlapped analysts. First, we run the following
regression: Overlap ¼ a þ b1nComove_returnþ b2nSIZE_m þ b3nMTB_m þ b4nLEV_m þ b5nSG_mþ e for each industry-year,
where Overlap is measured as the ratio of the number of peer/control firms that have at least one analyst also covering
the scandal firm to the total number of peer/control firms that have any analyst coverage at the 3-digit SIC code level,
Comove_return is the R-squared of the regression of peer/control firms’ daily returns on scandal firms’ returns, measured
annually, and SIZE_m (MTB_m, LEV_m, and SG_m) is measured as the industry median SIZE (MTB, LEV and SG). Industries
that have higher regression residual (unexplained overlap) than the median are defined as ‘‘High Overlap’’; otherwise,
‘‘Low Overlap’’.21 We expect the H1a, b, and c results to be stronger in the ‘‘High Overlap’’ sub-sample.22
We test hypothesis 2 using the following ordered probit model:
Recom ¼ b0 þ b1 nPEER þ b2 nSCAN þ b3 nPEERnSCAN þ b4 nSIZEþ b5 nMTB
þ b6 nLEV þ b7 nCFO þ b8 nRATING þ b9 nSG þ b10 nCAPEX_S
þ b10 nCOMOVE þ e ð2Þ
Variable definitions:
Recom: the median value of all analysts’ recommendations during a year. one represents strong buy and five represents
strong sell.
All other variables are as defined above.
We expect the coefficient on PEERnSCAN to be negative based on H2; that is, if analysts cannot see through the
fraudulent reporting, they may overestimate the overall industry prospects and thereby provide better recommendations.
We further test the effects of analyst coverage by estimating model (2) for sub-samples based on the overlap in analyst
coverage in the same way as for model (1). We expect the recommendations spillover effect to be stronger when the
overlap between peer and scandal firms is higher.

4.3. The association of current investment with future performance

To test whether the investment made during the scandal period is suboptimal compared to the pre-scandal period, we
run the following panel data regression.
CFOt þ i ¼ b0 þ b1 nPEER þ b2 nSCAN þ b3 nPEERnSCAN þ b4 nCAPEX þ b5 nCAPEX nPEER þ b6 nCAPEX nSCAN
þ b7 nCAPEX nPEERnSCAN þ b8 nSIZE þ b9 nMTB þ b10 nLEV þ b11 nRATING þ e i ¼ 1,2,3 ð3Þ
Variable definitions:

CFOt þ i: one year (or two, three-year) ahead CFO, where CFO is defined as cash flow from operations (COMPUSTAT
‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).

All other variables are as defined above.

Based on H3, we predict the coefficient on CAPEXnPEERnSCAN to be negative. That is, we expect that capital expenditures
made in the scandal period will generate lower cash flows compared to those made in the pre-scandal period.

5. Results

5.1. Descriptive statistics

We compare the descriptive statistics between the peers and the control group in Table 2. We find that there is no
significant difference between our control group firms and peers in SIZE, MTB and SG in the pre-scandal period, suggesting
that they are comparable in terms of asset-in-place and investment opportunities. However, peers have significantly
higher capital expenditures than the control group even in the three-year period before the scandal period. However, the
difference becomes insignificant in subsequent multivariate analyses. In the scandal period, we find that peers’ capital

21
The coefficient on Comove_return is consistently significantly positive, suggesting that the ratio of overlapped analysts captures economic
similarity. The yearly model R-squared ranges from 0.17 to 0.40.
22
As a robustness check, we also put the predicted value of Overlap from this prediction model in the investment regression as a control variable, and
all results continue to hold.
192 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

Table 2
Descriptive statistics of peers and control group firms.

Variables Pre-scandal period Scandal period

Peers Control firms Peers Control firms

Mean Mean (t-stat Peer-Control) Mean Mean (t-stat Peer-Control)

CAPEX 0.400 0.343 0.449 0.319

(5.54)nnn (14.38)nnn
SIZE 4.789 4.803 4.742 4.971
( 0.29) ( 5.10)nnn
MTB 2.207 2.249 2.876 2.131
( 0.84) (13.82)nnn
LEV 0.166 0.174 0.171 0.186
(  1.83)n ( 3.75)nnn
CFO 0.280 0.008  0.038 0.006
(3.63)nnn ( 7.47)nnn
RATING 0.218 0.223 0.219 0.245
( 0.58) ( 3.11)nnn
SG 0.136 0.135 0.081 0.105
(0.17) ( 2.02)nn
Number of observations 3170 6091 4428 7361

nnn nn
, and n represent 1%, 5% and 10% significance levels, respectively.
Variable definition:

CAPEX: the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged properties, plants and equipment (COMPUSTAT ‘‘ppent’’).
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’–‘‘ceq’’ þ‘‘prcc_f’’n‘‘csho’’) to book value of total assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).

expenditures are higher than the control group, consistent with our hypotheses. However, we also observe that peers and
control group are significantly different in other firm characteristics and so it is important that we control for these
variables in the regressions.
We show the Pearson correlation coefficients in Table 3. The correlations suggest that peers tend to have larger capital
expenditures in scandal periods.

5.2. Ex ante likelihood of fraud detection results

Consistent with H1a, the first model in Table 4 shows that on average capital expenditures are greater during the
scandal period for peers compared to control firms. The average effect documented in model 1 indicates approximately a
14% increase in the average capital expenditures of peers during the scandal period relative to mean peer capital
expenditures during the pre-scandal period. The results in model 2 indicate that this increase in capital expenditures is
only significant for the subsample where scandal firms have low F-scores rather than high F-scores. The findings reflect an
approximately 22% increase in average investment for peers with a lower ex ante likelihood of fraudulent reporting, but an
insignificant increase in capital expenditures for firms with a higher ex ante fraud likelihood. This result is not consistent
with the concern that peers increase investment in response to earnings management that is particularly easy to detect. In
contrast, this finding seems to suggest that peers do not increase investments when the fraud can be easily detected.
Further, we conduct a pseudo-scandal analysis by constructing a sample of pseudo-scandal leading firms who were not
accused of fraud ex post and corresponding peers to test our assumption that firms use (leading) peers’ earnings information
to make investment decisions. We use Fortune 100 firms with no fraud record during our sample period as industry leaders.
We determine shock years by first identifying the year with the highest lead firm revenue growth in excess of the median
peer’s revenue growth for that year. Within the distribution of highest lead firm excess revenue growth years we select those
above the median as being shock years. Based on this process we obtain 31 shock periods for our lead firms. We identify peer
and control groups in the same fashion as our main analysis, i.e., 3-digit vs. 2-digit SIC codes.
In Appendix A, we report that the distributions of the F-score (Dechow et al., 2011) measures of fraud detection likelihood
between our scandal and pseudo-scandal samples are very similar. Specifically, the 25th, median and 75th percentiles for the
scandal sample (pseudo-scandal sample) are 0.96 (0.96), 1.21 (1.14) and 1.58 (1.81), respectively. In Appendix B, we investigate
the effect of the expected fraud detection likelihood on peers’ investment in response to pseudo-scandal leading firms’ financial
reporting. We find results similar in spirit to those for our fraud sample. Specifically, peers respond to increased reported
revenue by industry leaders only for the subsample where the leading firms have a low ex ante fraud detection likelihood
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 193

Table 3
Pearson correlation (and P-value) for investment model variables, for both scandal and pre-scandal periods.

(2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
PEER SCAN SIZE MTB LEV CFO RAT SG COM RES I_FACTOR

(1) CAPEX 0.101 0.005  0.177 0.325  0.173  0.167  0.129 0.288 0.075 0.072 0.157
(0.001) (0.470) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
(2) PEER 0.034  0.029 0.077  0.028  0.030  0.019  0.012 0.119 0.375 0.346
(0.001) (0.001) (0.001) (0.001) (0.001) (0.005) (0.001) (0.001) (0.001) (0.001)
(3) SCAN 0.020 0.032 0.025  0.041 0.017  0.042 0.078 0.186 0.147
(0.004) (0.001) (0.001) (0.001) (0.011) (0.001) (0.001) (0.001) (0.001)
(4) SIZE  0.265 0.277 0.409 0.656  0.016  0.031 0.051  0.111
(0.001) (0.001) (0.001) (0.001) (0.017) (0.001) (0.001) (0.001)
(5) MTB  0.102  0.485  0.120 0.106  0.041 0.062 0.122
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
(6) LEV 0.049 0.372  0.032  0.069 0.001  0.093
(0.001) (0.001) (0.001) (0.001) (0.977) (0.001)
(7) CFO 0.162 0.057 0.082  0.013  0.025
(0.001) (0.001) (0.001) (0.068) (0.001)
(8) RATING  0.004  0.065 0.041  0.095
(0.593) (0.001) (0.001) (0.001)
(9) SG 0.064  0.011 0.024
(0.001) (0.120) (0.001)
(10) COMOVE 0.105 0.044
(0.001) (0.001)
(11) RESTATE 0.469
(0.001)

nnn nn
, and n represent 1%, 5% and 10% significance levels, respectively.
Variable definition:

CAPEX: the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged properties, plants and equipment (COMPUSTAT ‘‘ppent’’).
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for control firms that have the same
2-digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and zero for 3 years prior to the scandal period.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’–‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book value of total assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RESTATE: tercile rankings of the amount of overstatement by the scandal firms.
COMOVE: tercile rankings of the co-movement of change in market-to-book ratios between the fraudulent firms and sample or control firms in the
pre-scandal period (at the 3-digit SIC code level). The co-movement is measured as b in the regression DMTB¼ a þ bDMTB_S þ e, where DMTB is
defined as annual change in MTB and DMTB_S represents fraudulent firms’ change in MTB.
I_FACTOR: an indicator variable for firms in the industries (defined by 3-digit SIC codes) that have investor sentiments higher than the median of all
industries, earnings-to-price ratio (measured at the median of the industry) lower than the median of all industries and higher counts of M&A
activities than the median.

(i.e., low F-scores) but no significant difference for those where leading firms have a high ex ante likelihood that the large
revenue growth reflects fraudulent reporting. These findings are supportive of H1a, and mitigate the concern that examining
detected frauds induces our results.

5.3. Lead-lag results

In Panel A of Table 5, we report the results of the investment models where we set the dummy variable SCANt equal to 1
for the period that is one year prior to, the year of, the first year following, and the second year following fraud inception
respectively. In addition, we examine the one year following the end of the fraud period. In models (3) and (4), we find
results consistent with H1a that peers make more investments in the first or second years following the fraud inception than
in the pre-scandal period. The economic magnitude of this incremental investment is similar across these two models and is
large: compared to peers’ capital expenditure level in the pre-scandal period, there is an approximately 20% (i.e., 0.054/0.286
and 0.06/0.278)23 increase in investment. Interestingly, the coefficient on SCANt, which captures the effect of the scandal
period on our control firms’ investments, is insignificant. In addition, the coefficient on PEER, which captures the difference in

23
The denominator is the sum of the intercept and the coefficient on PEER, representing the investment level for the peer firms in the pre-scandal
period.
194 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

Table 4
The effect of fraud score on peers’ capital expenditure.

Variable Sign Model 1 Model 2

Coefficient (clustered-t) Coefficient (clustered-t)

Intercept ? 0.253 0.204

(15.18)nnn (13.53)nnn
PEER ? 0.018 0.037
(1.06) (2.07)n
SCAN ?  0.006  0.002
(  0.47) ( 0.14)
PEERnSCAN þ 0.054 0.086
(2.60)nn (3.18)nnn
F Score ? 0.057
(5.17)nnn
PEERnSCANn F Score þ  0.078
( 3.09)nnn
SIZE ?  0.009  0.009
( 3.11)nnn
(  3.48)nnn
MTB þ 0.044 0.044
(9.17)nnn (8.85)nnn
LEV ?  0.260  0.246
(  10.17)nnn ( 9.59)nnn
CFO ?  0.059  0.051
(  1.63) ( 1.29)
RATING ?  0.011  0.009
(  0.88) ( 0.67)
SG þ 0.279 0.277
(19.17)nnn (20.78)nnn
CAPEX_S þ 0.065 0.077
(2.73)nnn (3.06)nnn
COMOVE þ 0.032 0.028
(5.40)nnn (5.09)nnn
N 21,050 20,192
R-squared 0.2084 0.2097

nnn nn
, and n represent 1%, 5% and 10% significance levels, respectively.
Variable definition:

CAPEX: the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged properties, plants and equipment (COMPUSTAT ‘‘ppent’’).
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for control firms that have the same 2-
digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and zero for 3 years prior to the scandal period.
FScore: An indicator variable that equals one if the scandal firms’ F score is greater than the median, where F score is defined following Dechow et al.
(2011).
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’  ‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book value of total assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
COMOVE: tercile rankings of the co-movement of change in market-to-book ratios between the fraudulent firms and sample or control firms in the
pre-scandal period (at the 3-digit SIC code level). The co-movement is measured as b in the regression DMTB ¼ a þ bDMTB_Sþ e, where DMTB is
defined as annual change in MTB and DMTB_S represents fraudulent firms’ change in MTB.

capital expenditures between the peers and the control group during the pre-scandal period, is not significant. This suggests
that the control firms are serving the desired role in our difference-in-differences design.
In contrast, the coefficient on PEERnSCANt is insignificant in models (1) and (2), suggesting that firms are reacting to the
fraudulent reports after they are issued, and that the increase in investment is a response to the fraud rather than higher capital
expenditures of the peers providing an incentive for leading firms to commit fraud. Finally, in model (5), the coefficient on
PEERnSCAN is not significant suggesting that the investment after the fraud period reverts to the pre-fraud period levels.
In Panel B of Table 5, we examine whether our results differ for competitive versus concentrated industries and for high
versus low sales growth firms, based on Wang and Winton’s (2010) argument that fraud in competitive industries is more
likely to be driven by investment booms. We find no significant difference in the coefficient on PEERnSCAN for firms in
competitive versus concentrated industries or in high versus low growth firms. Both panels of Table 5 seem to suggest that
our results are not merely driven by reverse causality or by omitted factors affecting both industry investment boom and
fraud propensity, but are more consistent with our spillover hypothesis.
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 195

Table 5
Causality analysis of peer capital expenditures and lead firms’ frauds.

Panel A: Lead–lag analysis

Variable Prediction Model 1 Model 2 Model 3 Model 4 Model 5

SCAN  1 (one year SCAN0 (the year of SCAN þ 1 (the first year SCAN þ 2 (the second year SCANPS1 (one year after
before fraud fraud inception) after fraud inception) after fraud inception) the end of fraud period)
inception)
Coefficient Coefficient Coefficient Coefficient Coefficient
(clustered-t) (clustered-t) (clustered-t) (clustered-t) (clustered-t)

Intercept ? 0.266 0.253 0.267 0.261 0.271

(12.57)nnn (15.96)nnn (18.17)nnn (9.31)nnn (16.09)nnn
PEER ? 0.034 0.020 0.019 0.017 0.021
(2.86)nnn (1.64) (1.59) (1.20) (1.63)
SCANt ? 0.001 0.005 0.007 0.000  0.014
(0.01) (0.54) (0.34) (0.00) (  0.82)
PEERnSCANt þ  0.029 0.010 0.054 0.060  0.014
( 1.22) (0.58) (2.23)nn (2.17)nn (  0.71)
SIZE ?  0.014  0.011  0.012  0.011  0.010
( 2.64)nnn ( 2.80)nnn (  2.99)nnn ( 2.10)nn (  1.54)
MTB þ 0.050 0.047 0.046 0.044 0.040
(8.18)nnn (8.94)nnn (9.96)nnn (6.40)nnn (8.25)nnn
LEV ?  0.296  0.261  0.298  0.285  0.286
( 7.63)nnn ( 7.00)nnn (  2.68)nnn ( 7.79)nnn (  8.17)nnn
CFO ?  0.034  0.053  0.059  0.069  0.035
( 1.22) ( 1.27) (  2.68)nn ( 1.44) (  1.06)
RATING ? 0.002  0.010 0.005  0.001  0.017
(0.07) (  0.43) (0.24) ( 0.07) (  0.56)
SG þ 0.292 0.294 0.275 0.304 0.304
(17.07)nnn (12.72)nnn (13.54)nnn (20.35)nnn (19.54)nnn
CAPEX_S þ 0.048 0.047 0.050 0.059 0.048
(4.99)nnn (3.79)nnn (3.17)nnn (2.83)nn (3.38)nnn
COMOVE þ 0.033 0.037 0.037 0.033 0.032
(4.69)nnn (4.49)nnn (4.18)nnn (3.21)nnn (3.77)nnn
N 10,971 13,017 12,945 12,426 12,951
R-squared 0.2020 0.1960 0.1970 0.2171 0.1995

Note: nnn, nn and n represent 1%, 5% and 10% significance, respectively, based on a one- or two-tailed test, as appropriate. Standard errors are clustered
by fiscal year. In Model 1, the pre-scandal period is from 4 years to 2 years before the scandal; in other models, the pre-scandal period is from 3
years to 1 year before the scandal.

Panel B: Partitions by competitiveness and revenue growth

Variable Prediction Competitive industries Concentrated industries High growth Low growth
Coefficient (clustered-t) Coefficient (clustered-t) Coefficient (clustered-t) Coefficient (clustered-t)

Intercept ? 0.204 0.289 0.203 0.265

(9.67)nnn (14.03)nnn (7.12)nnn (16.60)nnn
PEER ? 0.075  0.043 0.009  0.036
(5.33)nnn ( 2.93)nnn (0.68) ( 1.90)n
SCAN ? 0.024  0.019  0.000 0.000
(2.61)nnn ( 1.95)n (  0.03) (0.00)
PEERnSCAN þ 0.032 0.029 0.063 0.059
(1.87)nn (1.51)n (3.55)nnn (3.23)nnn
SIZE ?  0.007  0.009  0.004  0.009
( 2.00)nn ( 2.88)nnn (  0.80) ( 2.33)nn
MTB þ 0.040 0.046 0.051 0.038
(10.09)nnn (11.32)nnn (12.01)nnn (6.36)nnn
LEV ?  0.221  0.304  0.323  0.203
( 7.43)nnn ( 10.25)nnn ( 11.65)nnn ( 4.81)nnn
CFO ?  0.091  0.031  0.086  0.077
( 2.29)nn ( 0.93) ( 1.82)n ( 1.75)n
RATING ? 0.001  0.027  0.009  0.023
(0.06) ( 2.26)nn (  0.75) ( 1.41)
SG þ 0.311 0.256 0.277 0.271
(16.67)nnn (15.34)nnn (17.07)nnn (14.95)nnn
CAPEX_S þ 0.063 0.047 0.062 0.068
(7.28)nnn (3.18)nnn (3.65)nnn (2.41)nn
196 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

Table 5 (continued )

Panel B: Partitions by competitiveness and revenue growth

Variable Prediction Competitive industries Concentrated industries High growth Low growth
Coefficient (clustered-t) Coefficient (clustered-t) Coefficient (clustered-t) Coefficient (clustered-t)

COMOVE þ 0.035 0.022 0.050 0.008

(4.86)nnn (3.85)nnn (7.99)nnn (1.27)
N 10,415 10,635 10,557 10,493
R-squared 0.2335 0.1829 0.2416 0.1668

Test of coefficient equality on PEERnSCAN: Test of coefficient equality on PEERnSCAN:

w2 ¼ 0.0144 p-value ¼0.9039 w2 ¼ 0.3364 p-value¼ 0.5743

Note: nnn, nn and n represent 1%, 5% and 10% significance, respectively, based on a one- or two-tailed test, as appropriate. Standard errors are clustered by
fiscal year. ‘‘Competitive Industries’’ vs. ‘‘Concentrated Industries’’ partition is based on the peer industries’ ratio of top 5 firms in revenue to the total
revenue of each year-industry, classified by 3-digit SIC code. Both peer and control industries are classified as ‘‘Competitive Industries’’ if the peer
industries’ ratio is lower than the annual median; otherwise as ‘‘Concentrated Industries’’. ‘‘High Growth’’ vs. ‘‘Low Growth’’ is defined by peer industries’
sales growth (SG) at the beginning year of the scandal period. Peer and control industries are classified as ‘‘High Growth’’ if the peer industries’ median
sales growth is higher than the median; otherwise, classified as ‘‘Low Growth’’.
Variable definition:

CAPEX (dependent variable): the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged properties, plants and equipment (COMPUSTAT
‘‘ppent’’).
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for control firms that have the same
2-digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and zero for 3 years prior to the scandal period.
RESTATE: tercile rankings of the ratio of fraudulent firms’ total restatement to the average revenues of the pre-scandal period.
I_FACTOR: an indicator variable for firms in the industries (defined by 3-digit SIC codes) that have investor sentiments higher than the median of all
industries, earnings-to-price ratio (measured at the median of the industry) lower than the median of all industries and higher counts of M&A
activities than the median.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’  ‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book value of total assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
CAPEX_S: fraudulent firms’ CAPEX.
COMOVE: tercile rankings of the co-movement of change in market-to-book ratios between the fraudulent firms and sample or control firms in the
pre-scandal period (at the 3-digit SIC code level). The co-movement is measured as b in the regression DMTB ¼ a þ bDMTB_Sþ e, where DMTB is
defined as annual change in MTB and DMTB_S represents fraudulent firms’ change in MTB.

5.4. Restatement amount and industry condition results

The results of our tests of H1b and H1c are reported in Table 6. The first two models are conducted at the firm level
using firms that share 2-digit SIC codes with the scandal firms as control firms. The second set of models is conducted at
the segment level, using the segments of the same peer that operate in other industries as controls. Specifically, in
our segment level analysis, PEER¼1 for the segment of the peer that share 3-digit SIC codes with the scandal firms, and
PEER ¼0 for other peer segments.
In model (1), consistent with H1b, we find that peers’ capital investment is increasing in the amount that leading firms
overstate their earnings during the fraudulent period. This spillover effect is economically significant. Moving from the bottom
tercile to the top tercile in scandal firms’ misstatement, peer firms’ investment increases by 0.088, representing a 32% increase in
the investment from the pre-scandal period. In model (2), we find that the coefficient on PEERnSCANnI_FACTOR is significantly
positive, suggesting that the increase in peers’ capital expenditures is greater when the industry has a higher investor sentiment, a
lower cost of capital and larger private benefits of control. This result is consistent with both H1c and Kumar and Langberg (2010).
Again, this effect is significant economically: firms in the high I_FACTOR industries on average invest more than low I_FACTOR
industries by 0.085, representing 33% of the investment in the pre-scandal period. The results of our segment level analysis and of
our firm level analysis are quite similar. The similarity of the results provides comfort that our 2-digit SIC code control firms
perform in the desired way.24
Other than the main variables of interest, our control variables also load as predicted. For example, capital expenditures
increase with growth opportunities. We also find that peer/control firms’ investments increase with scandal firms’
investments, suggesting that firms also directly mimic industry leaders’ investment behavior besides the information

24
While not tabulated, the average effect (PEERnSCAN) using segment data is also significant.
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 197

Table 6
Cross-sectional analysis of the information spillover effect on peer firms’ capital expenditures.

Variable Sign Firm level analysis controls share 2-digit SIC codes Segment level analysis controls are other segments

Model 1 Model 2 Model 1 Model 2

Coefficient (clustered-t) Coefficient (clustered-t) Coefficient (clustered-t) Coefficient (clustered-t-)

Intercept ? 0.255 0.249 0.167 0.170

(15.60)nnn (16.85)nnn (8.06)nnn (5.40)nnn
PEER ? 0.021 0.008 0.070 0.070
(1.36) (0.53) (3.49)nnn (3.44)nnn
SCAN ?  0.006  0.015  0.011  0.015
(  0.51) (  1.08) (  0.75) (  1.03)
RESTATE ?  0.002  0.009
(  0.29) ( 1.08)
I_FACTOR ? 0.049 0.021
(2.97)nn (1.91)n
PEERnSCAN þ  0.011 0.001  0.000 0.009
(  0.46) (0.02) (  0.00) (0.36)
PEERnSCANnRESTATE þ 0.044 0.053
(2.88)nnn (2.51)nn
PEERnSCANnI_FACTOR þ 0.085 0.121
(3.07)nnn (3.18)nnn
SIZE ?  0.009  0.008  0.0166  0.016
( 3.55)nnn (  3.80)nnn ( 4.76)nnn (  4.73)nnn
MTB þ 0.044 0.043 0.042 0.036
(9.15)nnn (8.77)nnn (4.84)nnn (4.69)nnn
LEV ?  0.259  0.250  0.100  0.103
( 10.01)nnn (  10.09)nnn ( 3.38)nnn (  3.60)nnn
CFO ?  0.057  0.064  0.193  0.194
( 1.61) (  1.83)n ( 2.61)nn (  2.53)nn
RATING ?  0.011  0.010 0.036 0.040
(  0.93) ( 0.84) (5.13)nnn (5.84)nnn
SG þ 0.280 0.280 0.121 0.126
(19.16)nnn (19.48)nnn (5.46)nnn (5.97)nnn
CAPEX_S þ 0.062 0.048 0.053 0.037
(2.76)nnn (2.45)nn (3.49)nnn (3.57)nnn
COMOVE þ 0.031 0.029
(5.43)nnn (4.95)nnn
N 21,050 21,050 3179 3179
R-squared 0.2092 0.2134 0.1789 0.1895

Note: nnn, nn and n represent 1%, 5% and 10% significance, respectively, based on a one- or two-tailed test, as appropriate. Standard errors are clustered by
fiscal year.
Variable definition:

CAPEX (dependent variable): the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged properties, plants and equipment (COMPUSTAT
‘‘ppent’’).
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for control firms that have the same 2-
digit SIC codes as the fraudulent firms (but different 3-digit SIC codes). For segment level analysis, PEER¼ 1 for segments of the peer that share the
same 3-digit SIC codes as the scandal firm, and PEER¼ 0 for segments of the peer that operate in other industries.
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and zero for 3 years prior to the scandal period.
RESTATE: tercile rankings of the ratio of fraudulent firms’ total restatement to the average revenues of the pre-scandal period.
I_FACTOR: an indicator variable for firms in the industries (defined by 3-digit SIC codes) that have investor sentiments higher than the median of all
industries, earnings-to-price ratio (measured at the median of the industry) lower than the median of all industries and higher counts of M&A
activities than the median.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’  ‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book value of total assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
CAPEX_S: fraudulent firms’ CAPEX.
COMOVE: tercile rankings of the co-movement of change in market-to-book ratios between the fraudulent firms and sample or control firms in the
pre-scandal period (at the 3-digit SIC code level). The co-movement is measured as b in the regression DMTB¼ a þ bDMTB_S þ e, where DMTB is
defined as annual change in MTB and DMTB_S represents fraudulent firms’ change in MTB.

spillover effect that we document. Finally, we also find that capital expenditures increase with the co-movement in MTB
between peer/control firms and scandal firms, suggesting that peer/control firms’ investment is more likely when their
growth opportunities are co-moving with scandal firms.
198 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

Table 7
The effects of overlapping analysts on peers’ investment.

Variables Model 1 Model 2 Model 3

High Overlap_R Low High Low Overlap_R High Overlap_R Low Overlap_R
Overlap_R Overlap_R
Coefficient Coefficient Coefficient Coefficient Coefficient Coefficient
(clustered-t) (clustered-t) (clustered-t) (clustered-t) (clustered-t) (clustered-t)

Intercept 0.235 0.274 0.236 0.278 0.239 0.265

(11.53)nnn (12.25)nnn (12.67)nnn (13.61)nnn (11.53)nnn
PEER  0.017 0.061  0.014 0.066  0.043 0.055
( 1.01) (1.42) (  1.10) (1.59) (  3.92)nnn (1.49)
SCAN  0.037 0.009  0.037 0.012  0.052 0.001
( 3.27)nnn (0.42) (  3.42)nnn (0.79) (  4.42)nnn (0.04)
RESTATE 0.001  0.008
(0.14) ( 0.72)
I_FACTOR 0.083 0.038
(4.62)nn (1.55)
PEERnSCAN 0.126  0.063 0.039  0.016 0.057  0.042
(4.46)nnn (  1.16) (1.80)nn ( 0.24) (4.90)nnn ( 0.93)
PEERnSCANnRESTATE 0.057  0.033
(2.77)nn (  1.70)n
PEERnSCANnI_FACTOR 0.085  0.038
(2.25)nn ( 0.45)
SIZE  0.004  0.014  0.004  0.013  0.003  0.013
(  1.00) ( 3.63)nnn
nnn
( 1.39) (  3.61) (  1.31) (  3.22nnn
MTB 0.043 0.044 0.042 0.044 0.041 0.044
(9.15)nnn (7.26)nnn (8.88)nnn (7.18)nn (8.46)nnn (7.35)nnn
LEV  0.260  0.256  0.261  0.257  0.240  0.259
( 7.71)nnn (  8.11)nnn (  7.76)nnn (  7.88)nnn (  7.26)nnn ( 7.85)nnn
CFO  0.099  0.013  0.098  0.016  0.103  0.019
( 2.25)nn ( 0.55) (  2.25)nn ( 0.70) (  2.36)nn ( 0.83)
RATING  0.026 0.005  0.028 0.005  0.024 0.005
( 2.43)nn (0.21) (  2.54)nn (0.21) (  2.16)nn (0.20)
SG 0.297 0.262 0.298 0.262 0.299 0.26
(14.52)nnn (18.74)nnn (14.20)nnn (18.92)nnn (14.59)nnn (18.68)nnn
CAPEX_S 0.078 0.051 0.071 0.055 0.053 0.041
(4.24)nnn (2.43)nn (4.27)nnn (2.67)nn (3.80)nnn (1.97)nn
COMOVE 0.034 0.029 0.036 0.030 0.023 0.033
(7.57)nnn (3.35)nnn (8.64)nnn (3.57)nnn (3.84)nnn (3.09)nnn
N 10,522 10,528 10,522 10,528 10,522 10,528

R-squared 0.2379 0.1742 0.2400 0.1747 0.2459 0.1755

Test of Equality of PEERnSCAN:/ ¼ 47.89 PEERnSCANnRESTATE:/ ¼14.06 PEERnSCANnI_FACTOR:/ ¼10.96

Coefficients p-value ¼ 0.001 p-value ¼0.001 p-value¼ 0.001

Note: nnn, nn and n represent 1%, 5% and 10% significance, respectively, based on a one- or two-tailed test, as appropriate. Standard errors are clustered by
fiscal year.
Variable definition:

Overlap_R: the residual from the following regression: Overlap ¼ a þ b1Comove_return þ b2SIZE_mþ b3MTB_m þ b4LEV_mþ b5SG_m þ e for each
industry-year,
where
Overlap is measured as the ratio of the number of firms that have at least one analyst also covering the scandal firm to the total number of firms that
have any analyst coverage at the 3-digit SIC code level,
Comove_return is the R-squared of the regression of peer/control firms’ daily returns on scandal firms’ returns, measured annually, and SIZE_m is
measured as the industry median size, etc.
High Overlap_R: Industries with above the median Overlap_R
Low Overlap_R: Industries with below the median Overlap_R
CAPEX (dependent variable): the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged properties, plants and equipment (COMPUSTAT
‘‘ppent’’).
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for control firms that have the same
2-digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and zero for 3 years prior to the scandal period.
RESTATE: tercile rankings of the ratio of fraudulent firms’ total restatement to the average revenues of the pre-scandal period.
I_FACTOR: an indicator variable for firms in the industries (defined by 3-digit SIC codes) that have investor sentiments higher than the median of all
industries, earnings-to-price ratio (measured at the median of the industry) lower than the median of all industries and higher counts of M&A
activities than the median.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 199

MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’  ‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book value of total assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
CAPEX_S: fraudulent firms’ CAPEX.
COMOVE: tercile rankings of the co-movement of change in market-to-book ratios between the fraudulent firms and sample or control firms in the
pre-scandal period (at the 3-digit SIC code level). The co-movement is measured as b in the regression DMTB¼ a þ bDMTB_S þ e, where DMTB is
defined as annual change in MTB and DMTB_S represents fraudulent firms’ change in MTB.

5.5. Analyst coverage overlap results

Table 7 reports the results of repeating our investment tests for a sub-sample of firms in industries with a high degree
of unexplained analyst overlap versus those with a low degree of unexplained overlap. We find that the results
documented in previous sections on H1a, b, and care primarily concentrated in the high-overlap sub-sample, but not in the
low-overlap subsample. These results support our hypothesis that information intermediaries play an important role in
transmitting information from scandal firms to peers, although it is also possible that analyst overlap may help identify
more closely related ‘‘peer firms,’’ so cautions should be applied when interpreting the findings.
In Table 8 we directly examine whether analysts’ recommendations help transmit the distorted fraud signals. For our
overall sample we find more favorable recommendations during the scandal period for peers. Further, we only find this
result for our high unexplained overlap sample. These findings again suggest that the spillover effect on peer investments
may be facilitated by analysts’ information intermediary roles.

5.6. Future cash flow results

In Table 9 Model 1, we find that capital investments made by peers in the pre-scandal period have a positive correlation
with future cash flows for at least three years after the investment is made. Consistent with H3, the coefficients on
CAPEXnPEERnSCAN are significantly negative and greater in absolute magnitude than those on CAPEXnPEER in all models,
consistent with the positive correlation being offset for investments made by peers in the scandal period, supporting the
notion that the investment made by peers during the scandal period are likely sub-optimal.
One competing explanation for the observed weaker association between peer investment and future cash flows is that
after a fraud is detected, contracting parties may impose reputational penalties on all firms in the fraud industry, resulting
in declining future performance for peers. To disentangle this competing hypothesis, we investigate the investment-cash
flow relation before vs. after the fraud is detected. Specifically, we allow CAPEXnPEERnSCAN to vary with whether the cash
flow is observed after the scandal period since frauds are detected after the scandal period. While our hypothesis predicts
no difference between the investment-cash flow relation for cash flows before and after fraud detection, the reputational
penalty story predicts a more negative investment-cash flow relation for cash flows observed after fraud detection than for
those before fraud detection. In Model 2 of Table 9, we find results consistent with our hypothesis but inconsistent with
this alternative explanation.

6. Supplemental analysis and robustness checks

6.1. Peer managers’ information

We argue that managers of peers may engage in sub-optimal investments during the scandal period because they are
misled by the rosy prospects portrayed in the scandal firm’s financial reports. As a result, managers of peers may be
optimistic about their investment returns and choose to increase their insider holdings during the scandal period to
benefit from the expected stock price jumps.
To test this prediction, we investigate peers’ insider trading behavior. We follow Piotroski and Roulstone (2005) by measuring
the insider purchase ratio as shares purchased over the sum of shares purchased and shares sold. The insider trading data is
collected from Thomson Financial that collects from SEC filings (forms 3, 4 and 5). In Table 10, we show results consistent with
our prediction: insiders in peers tend to purchase more shares in the scandal periods compared to pre-scandal periods.

6.2. Different types of scandals/restatements

We also investigate whether the spillover effect on peer investments depends on the types of accounting frauds: revenue
recognition versus others. Based on Gigler’s (1994) model that credible capital market disclosures must induce competitors
200 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

Table 8
Ordered probit model of analyst recommendations.

Variable Sign Overall High overlap_R Low overlap_R

Coefficient (clustered z-stats) Coefficient (clustered z-stats) Coefficient (clustered z-stats)

PEER ? 0.086 0.183  0.062

(1.66)n (1.80)n (  0.72)
SCAN ? 0.004 0.125  0.077
(0.06) (1.71)n ( 1.10)
PEERnSCAN –  0.142  0.297 0.009
( 1.72)nn (  2.31)nn (0.07)
SIZE ? 0.085 0.076 0.092
(5.65)nnn (3.76)nnn (0.07)
MTB ?  0.041  0.035  0.055
( 4.09)nnn (  5.06)nnn ( 2.77)nn
LEV ?  0.325  0.383  0.213
( 4.46)nnn (  3.69)nnn ( 1.85)n
CFO ?  0.202  0.096  0.286
( 1.96)nn ( 0.73) ( 1.84)n
RATING ?  0.006 0.037  0.049
( 0.21) (0.79) ( 1.64)
SG ?  0.615  0.600  0.636
(  16.80)nnn (  12.89)nnn ( 13.61)nnn
CAPEX_S ?  0.035  0.049  0.078
( 1.42) (  2.25)nn ( 1.56)
COMOVE ?  0.055  0.096  0.020
(  2.90)nnn (  0.66)
( 1.88)n
N 11,278 5907 5371
Pseudo R-squared 0.0302 0.0330 0.0298

Test of equality of coefficient PEERnSCAN w2 ¼3.13 p-value¼ 0.076

Note: nnn, nn and n represent 1%, 5% and 10% significance, respectively, based on a one- or two-tailed test, as appropriate. Standard errors are clustered by
fiscal year.
Variable definition:

Overlap_R: the residual from the following regression: Overlap ¼ a þ b1Comove_return þ b2SIZE_mþ b3MTB_m þ b4LEV_mþ b5SG_m þ e for each
industry-year,
where
Overlap is measured as the ratio of the number of firms that have at least one analyst also covering the scandal firm to the total number of firms that
have any analyst coverage at the 3-digit SIC code level,
Comove_return is the R-squared of the regression of peer/control firms’ daily returns on scandal firms’ returns, measured annually, and SIZE_m is
measured as the industry median size, etc.
High Overlap_R: Industries with above the median Overlap_R
Low Overlap_R: Industries with below the median Overlap_R
Recom (Dependent Variable): the median value of all analysts’ recommendations during a year. 1 represents strong buy and 5 represents strong sell.
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for control firms that have the same
2-digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and zero for 3 years prior to the scandal period.
RESTATE: tercile rankings of the ratio of fraudulent firms’ total restatement to the average revenues of the pre-scandal period.
I_FACTOR: an indicator variable for firms in the industries (defined by 3-digit SIC codes) that have investor sentiments higher than the median of all
industries, earnings-to-price ratio (measured at the median of the industry) lower than the median of all industries and higher counts of M&A
activities than the median.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’  ‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book value of total assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
CAPEX_S: fraudulent firms’ CAPEX.
COMOVE: tercile rankings of the co-movement of change in market-to-book ratios between the fraudulent firms and sample or control firms in the
pre-scandal period (at the 3-digit SIC code level). The co-movement is measured as b in the regression DMTB ¼ a þ bDMTB_Sþ e, where DMTB is
defined as annual change in MTB and DMTB_S represents fraudulent firms’ change in MTB.

to enter the product market and therefore increase investments, we investigate whether our results are driven by frauds
involving revenue recognition issues that are more likely related to product market demand. In Table 11, we find that peer
investments increase in the scandal periods only when the restatements are related to revenue recognition. Similarly, we
find that the effects of the magnitude of restatements and macro-variables on the increase in peer investments are only
significant for revenue recognition driven frauds.
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 201

Table 9
Association between investment and future cash flows.

Variables Pred Model 1 Model 2

Dependent Dependent Dependent Dependent Dependent Dependent

Variable: CFOt þ 1 Variable: CFOt þ 2 Variable: CFOt þ 3 Variable: CFOt þ 1 Variable: CFOt þ 2 Variable: CFOt þ 3
Coefficient Coefficient Coefficient Coefficient Coefficient Coefficient
(clustered-t stats) (clustered-t stats) (clustered-t stats) (clustered-t stats) (clustered-t stats) (clustered-t stats)

Intercept ?  0.160  0.161  0.163  0.158  0.158  0.154

( 5.92)nnn (  5.69)nnn ( 5.47)nnn (  6.16)nnn (  5.86)nnn ( 5.31)nnn
PEER ?  0.006 -0.018  0.025  0.006  0.019  0.027
(  0.40) (  1.08) ( 1.82)n (  0.40) (  1.24) ( 2.02)nn
SCAN ?  0.019  0.018  0.030  0.003  0.002  0.008
(  0.96) (  0.92) ( 1.44) (  0.20) ( 0.14) ( 0.43)
PEERnSCAN ?  0.009  0.005 0.026  0.010  0.004 0.027
(  0.36) (  0.18) (1.01) (  0.38) ( 0.16) (1.02)
CAPEX þ 0.027 0.006  0.005 0.027 0.006  0.004
(1.12) (0.21) (  0.17) (1.14) (0.21) ( 0.15)
CAPEXnPEER ? 0.033 0.069 0.091 0.034 0.069 0.088
(1.17) (2.26)nn (2.50)nn (1.17) (2.24)nn (2.54)nn
CAPEXnSCAN ? 0.012 0.043 0.078 0.012 0.042 0.077
(0.46) (1.18) (2.13)nn (0.42) (1.15) (2.17)nn
CAPEXnPEERnSCAN   0.045  0.073  0.111  0.071  0.107  0.171
( 1.59)n (  2.03)nn ( 2.61)nn (  1.70)nn (  2.41)nn ( 2.33)nn
Post ?  0.044  0.028  0.036
(  1.54) ( 0.98) ( 1.31)
CAPEXnPEERnSCANnPost ? 0.074 0.054 0.069
(1.60) (1.39) (1.34)
SIZE ? 0.055 0.054 0.053 0.055 0.054 0.053
13.23)nnn (15.91)nnn (11.64)nnn (13.78)nnn (15.82)nnn (11.60)nnn
MTB ?  0.034  0.031  0.031  0.035  0.031  0.031
( 8.60)nnn (  8.75)nnn ( 10.78)nnn (  8.70)nnn (  9.22)nnn (  12.07)nnn
LEV ?  0.036  0.028 0.017  0.036  0.027 0.018
( 3.35)nnn (  1.72)n (0.99) (  3.26)nnn (  1.74)n (1.01)
RATING ?  0.098  0.091  0.091  0.097  0.091  0.091
( 12.67)nnn ( 12.62)nnn ( 8.51)nnn (  12.41)nnn (  12.34)nnn ( 8.55)nnn
N 19,279 17,701 16,326 19,279 17,701 16,326
R-squared 0.2570 0.2205 0.1980 0.2597 0.2220 0.1998

Note: nnn, nn and n represent 1%, 5% and 10% significance, respectively, based on a one- or two-tailed test, as appropriate. Standard errors are clustered by
fiscal year.
Variable definition:

CFOt þ 1(dependent variable): one year (or two, three-year) ahead CFO, where CFO is defined as cash flow from operations (COMPUSTAT ‘‘oancf’’)
divided by lagged total assets (COMPUSTAT ‘‘at’’).
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for control firms that have the same
2-digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and zero for 3 years prior to the scandal period.
CAPEX: the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged properties, plants and equipment (COMPUSTAT ‘‘ppent’’).
Post: an indicator variable if CFO of interest is observed after the scandal period.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’  ‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book value of total assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
RATING: an indicator variable for firms with S&P credit ratings.

6.3. Robustness checks

While not tabulated, we also investigate the effect of each macro-variable composing I_FACTOR separately and find that
the tenor of the results for each variable (i.e., investor sentiment, industry cost of equity, and private benefits of control) is
the same as using our composite measure. We also allow COMOVE to interact with PEER, SCAN and PEERnSCAN and the
main results continue to hold. Further, we also partition cash flow analysis by industry growth, and we do not find that our
results are driven by the high growth industries.
To examine whether peers respond differentially to the fraudulent earnings compared to the restated earnings of the
fraud firms, we compare the coefficients on overstated earnings and original earnings in the investment regression. If the
coefficients do not differ, then this suggests that peers cannot see through the frauds and therefore behave as if the
falsified earnings are true earnings. We find that (untabulated) the coefficient on PEERnSCAN interacted with overstated
earnings does not differ statistically or economically from that on PEERnSCAN interacted with the originally reported
202 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

Table 10
Determinants of insider trading.

Variables Predictions Coefficients Clustered-t stats

Intercept ? 0.429 16.45nnn

PEER ?  0.034  1.94n
SCAN ?  0.085  7.28nnn
PEERnSCAN þ 0.052 2.36nn
SIZE ?  0.016  3.39nnn
MTB ?  0.051  14.53nnn
LEV ? 0.159 4.59nnn
CFO ?  0.301  8.59nnn
RATING ? 0.032 1.60
RET ?  0.076  10.37nnn
N 17,514
R-squared 0.0425

Note: nnn, nn, and n represent 1%, 5% and 10% significance levels, respectively.
Variable definition:

NET_PURCHASE (dependent variable): net purchase in shares (total purchase-total sale) by insiders,
divided by the sum of purchase and sale.
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms,
and zero for control firms that have the same 2-digit SIC codes as the fraudulent firms (but different 3-
digit SIC codes).
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and
zero for 3 years prior to the scandal period.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’  ‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book
value of total assets (COMPUSTAT ‘‘at’’).
LEV: total debts (COMPUSTAT ‘‘dltt’’þ‘‘dlc’’) divided by total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
RET: compounded market-adjusted returns of the 12-month period before the fiscal year end.

earnings. This indicates that peers do not discount the overstated earnings when making capital expenditures, suggesting
that peers do not seem to distinguish fraudulent versus restated earnings of the leading firms.
Finally, we also investigate whether the peers’ equity issuance is affected by industry leaders’ fraudulent reporting.
Based on our previous findings that analysts’ recommendations tend to be more favorable in the scandal periods, it would
be consistent to find that peers also increase equity issuance to fuel the increased investments. We find (untabulated) that
peers’ equity increases in the scandal periods, and the increase in equity issuance is more pronounced when the
restatement is large and when the I_FACTOR is high.

7. Conclusions

We examine whether a firm’s accounting quality affects peers’ investments. This real spillover effect remains largely
unexplored in the literature (Leuz and Wysocki, 2008). Focusing on accounting frauds conducted by high-profile firms that
are visible and likely used as benchmarks, we find peers’ investments are greater during the scandal period using
difference-in-differences estimation. This association does not appear to be driven by frauds that are more likely to be
detected or an increase in fraud during investment booms. We also find that peers’ investments are positively associated
with the amount of the earnings overstatement, and are greater when their industry has higher investor sentiment, a
lower cost of capital and higher private benefits of control. Furthermore, we find that this spillover effect holds only for
firms when the unexplained overlap in analyst coverage between scandal firms and peers is high. We also provide
evidence that is consistent with an equity analysts transmission mechanism by showing that analysts’ recommendation
are more favorable for peers during the scandal period only when unexplained overlap in analyst coverage is high,
although this finding may be subject to an alternative interpretation.
We find that peers’ scandal period investments have weaker associations with future cash flows, suggesting that
investments made during the scandal period may be less efficient. Finally, our results suggest that peers’ insiders rely on the
falsified prospects portrayed in the scandal firm’s fraudulent financial reports, as manifested by their increased ownership.
Our findings are consistent with distorted accounting signals generated by high-profile scandal firms on average increasing
investment by industry peers. Our paper contributes to the accounting and investment efficiency literatures by showing that
accounting information not only plays an important role in firms’ own investment, it also affects other firms’ investments.
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 203

Table 11
The differential effects of restatement types on peers’ investment.

Variables Prediction Model 1 Model 2 Model 3

Coefficients (clustered-t stats) Coefficients (clustered-t stats) Coefficients (clustered-t stats)

Intercept ? 0.249 0.249 0.251

(14.78)nnn (14.45)nnn (16.31)nnn
PEER ? 0.016 0.019 0.009
(0.86) (1.15) (0.56)
SCAN ?  0.009  0.009  0.016
( 0.64) (  0.74) ( 1.09)
REV_REC ? 0.019 0.015  0.002
(2.13)nn (1.73)n ( 0.15)
RESTATE ?  0.001
(  0.18)
I_FACTOR ? 0.051
(2.45)nn
PEERnSCAN þ  0.010  0.003 0.001
( 0.33) (  0.10) (0.03)
PEERnSCANnRESTATE þ  0.021
(  2.50)nn
PEERnSCANnI_FACTOR þ  0.024
( 0.39)
PEERnSCANnREV_REC þ 0.089
(3.57)nnn
PEERnSCANnRESTATEn REV_REC þ 0.076
(4.24)nnn
PEERnSCANnI_FACTORn REV_REC þ 0.123
(2.14)nn
SIZE ?  0.009  0.009  0.009
( 3.63)nnn (  3.52)nnn ( 3.95)nnn
MTB þ 0.044 0.043 0.043
(9.18)nnn (9.13)nnn
LEV ?  0.250  0.251  0.278
( 10.05)nnn (  10.02)nnn ( 9.73)nnn
CFO ?  0.056  0.056  0.063
( 1.60) (  1.61) ( 1.78)n
RATING ?  0.008  0.008  0.009
( 0.74) (  0.72) ( 0.78)
SG þ 0.281 0.281 0.280
(19.45)nnn (19.38)nnn (19.48)nnn
CAPEX_S þ 0.056 0.053 0.045
(3.01)nnn (3.11)nnn (2.81)nnn
COMOVE þ 0.029 0.032 0.029
(5.50)nnn (5.95)nnn (4.78)nnn
N 21,050 21,050 21,050
R-squared 0.2108 0.2122 0.2134

Note: nnn, nn and n represent 1%, 5% and 10% significance, respectively, based on a one- or two-tailed test, as appropriate. Standard errors are clustered by
fiscal year.
Variable definition:

CAPEX (dependent variable): the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged properties, plants and equipment (COMPUSTAT
‘‘ppent’’).
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for control firms that have the same
2-digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SCAN: an indicator variable equal one for years during which fraudulent firms committed frauds, and zero for 3 years prior to the scandal period.
REV_REC: an indicator variable for restatements caused by revenue recognition related frauds.
RESTATE: tercile rankings of the ratio of fraudulent firms’ total restatement to the average revenues of the pre-scandal period.
I_FACTOR: an indicator variable for firms in the industries (defined by 3-digit SIC codes) that have investor sentiments higher than the median of all
industries, earnings-to-price ratio (measured at the median of the industry) lower than the median of all industries and higher counts of M&A
activities than the median.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’–‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book value of total assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
CAPEX_S: fraudulent firms’ CAPEX.
COMOVE: tercile rankings of the co-movement of change in market-to-book ratios between the fraudulent firms and sample or control firms in the
pre-scandal period (at the 3-digit SIC code level). The co-movement is measured as b in the regression DMTB¼ a þ bDMTB_S þ e, where DMTB is
defined as annual change in MTB and DMTB_S represents fraudulent firms’ change in MTB.
204 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205

Appendix A. Comparison of F-Scores between scandal firms and leading pseudo-scandal firms

See Table A1.

Appendix B. Relation between peer capital expenditures and leading firms’ earnings growth in non-scandal periods

See Table B1.

Table A1

Scandal firms Leading pseudo-scandal firms

25th Percentile 0.96 0.96

Median 1.21 1.14
75th Percentile 1.58 1.81
N 35 31

Note: We use Fortune 100 firms with no fraud record during our sample period as industry leaders. We determine shock years by
first identifying the year with the highest lead firm revenue growth in excess of the median peer’s revenue growth for that year.
Within the distribution of highest lead firm excess revenue growth years we select those above the median as being shock years.
F-scores are measured as average F-scores (following Dechow et al. (2011)) of either the scandal period or the pseudo-scandal
period for non-fraud firms.

Table B1
.

Variable Sign Coefficient Clustered-t

Intercept ? 0.017 (2.39)nn

PEER ?  0.006 ( 1.52)
SHOCK ?  0.000 ( 0.02)
PEERnSHOCK þ 0.013 (2.48)nn
FScore ? 0.011 (1.75)n
PEERnSHOCK nFScore ?  0.015 ( 2.49)nn
SIZE ? 0.002 (1.95)n
MTB þ 0.010 (7.06)nnn
LEV ? 0.064 (6.21)nnn
CFO ? 0.072 (5.91)nnn
RATING ?  0.005 ( 1.18)
SG þ 0.068 (11.12)nnn
LEAD_CAPX þ 0.040 (1.58)
N 6773
R-squared 0.1626

Variable definition:

CAPX (dependent variable): the ratio of capital expenditure (COMPUSTAT ‘‘capx’’) to lagged assets.
PEER: an indicator variable equal one for firms in the same 3-digit SIC codes as the fraudulent firms, and zero for
control firms that have the same 2-digit SIC codes as the fraudulent firms (but different 3-digit SIC codes).
SHOCK: an indicator variable equal one for years during which the lead firm experience very high real revenue
growth, and zero for 3 years prior to the shock period.
FScore: an indicator variable that equals one for lead firms whose F-scores following Dechow et al. (2011) are higher
than the median of the sample; zero otherwise.
SIZE: the natural log of lagged total assets (COMPUSTAT ‘‘at’’).
MTB: lagged ratio of market value of total assets (COMPUSTAT ‘‘at’’–‘‘ceq’’ þ ‘‘prcc_f’’n‘‘csho’’) to book value of total
assets (COMPUSTAT ‘‘at’’).
LEV: long-term debt (COMPUSTAT ‘‘dltt’’) divided by total assets (COMPUSTAT ‘‘at’’), measured at the beginning of
the year.
CFO: cash flow from operations (COMPUSTAT ‘‘oancf’’) divided by lagged total assets (COMPUSTAT ‘‘at’’).
RATING: an indicator variable for firms with S&P credit ratings.
SG: change in revenues (COMPUSTAT ‘‘revt’’) divided by lagged total assets.
LEAD_CAPX: leading firms’ CAPX.
 A. Beatty et al. / Journal of Accounting and Economics 55 (2013) 183–205 205

References

Akhigbe, A., Madura, J., Newman, M., 2006. Industry effects of analyst stock revisions. The Journal of Financial Research 26, 181–198.
Bagnoli, M., Watts, S., 2010. Oligopoly, disclosure, and earnings management. The Accounting Review 84 (4), 1191–1214.
Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of market returns. Journal of Finance 61, 1645–1680.
Beatty, A., Liao, S., Weber, J., 2010. The role of accounting quality in reducing investment inefficiency in the presence of private information and direct
monitoring. The Accounting Review 85, 1215–1238.
Beyer, A., Cohen, D., Lys, T., Walther, B., 2010. The financial reporting environment: review of the recent literature. Journal of Accounting and Economics
50 (2–3), 296–343.
Biddle, G., Hilary, G., 2006. Accounting quality and firm-level capital investment. The Accounting Review 81, 963–982.
Biddle, G., Hilary, G., Verdi, R., 2009. How does financial reporting quality improve investment efficiency? Journal of Accounting and Economics 48 (2–3),
112–131.
Bushman, R., Smith, A., 2001. Financial accounting information and corporate governance. Journal of Accounting and Economics 32 (1–3), 237–333.
Cotter, J., Young S., 2007. Do analysts anticipate accounting fraud? Working Paper. Baruch College.
Darrough, M.N., Stoughton, M., 1990. Financial disclosure policy in an entry game. Journal of Accounting and Economics 12, 480–511.
Dechow, P., Ge, W., Larson, C., Sloan, R., 2011. Predicting material accounting misstatements. Contemporary Accounting Research 28 (1), 17–82.
Durnev, A., Mangen, C., 2009. Corporate investments: learning from restatements. Journal of Accounting Research 47 (3), 679–720.
Francis, J., 2001. Discussion of empirical research on accounting choice. Journal of Accounting and Economics 31 (1–3), 309–320.
Gigler, F.B., 1994. Self-enforcing voluntary disclosures. Journal of Accounting Research 32, 224–240.
Gleason, C., Jenkins, N., Johnson, W.B., 2008. The contagion effects of accounting restatements. The Accounting Review 83, 83–110.
Jensen, M., 2005. The agency costs of overvalued equity. Financial Management 34 (1), 5–19.
Karaoglu, E., Sandino T., Beatty R., 2006. Benchmarking against the performance of high profile ‘‘scandal’’ firms. Working Paper. University of Southern
California.
Kedia, S., Philippon, T., 2009. The economics of fraudulent accounting. Review of Financial Studies 22 (6), 2169–2199.
Kumar, P., Langberg, N., 2009. Corporate fraud and investment distortions in efficient capital markets. RAND Journal of Economics 40 (1), 144–172.
Kumar, P., Langberg N., 2010. Innovation and Investment Bubbles. Working Paper. University of Houston.
Leuz, C., Wysocki P., 2008. Economic consequences of financial reporting and disclosure regulation: a review and suggestions for future research. Working
Paper. University of Chicago and MIT Sloan School of Management.
Piotroski, J., Roulstone, D., 2005. Do insider trades reflect both contrarian beliefs and superior knowledge about future cash-flow realizations? Journal of
Accounting and Economics 39, 55–82.
Povel, P., Singh, R., Winton, A., 2007. Booms, busts, and fraud. The Review of Financial Studies 20, 1219–1254.
Sadka, G., 2006. The economic consequences of accounting fraud in product markets: theory and a case from a US telecommunications industry
(Worldcom). American Law and Economics Review 8 (3), 439–475.
Sidak, J., 2003. The failure of good intentions: the WorldCom fraud and the collapse of American telecommunications after deregulation. Yale Journal on
Regulation 20 (2), 207–267.
Verrecchia, R., 2001. Essays on disclosure. Journal of Accounting and Economics 32, 97–180.
Wang, T., Winton A., 2010. Competition and Corporate Fraud Waves. Working Paper. University of Minnesota.