ArticlePDF Available

Financial statement comparability and accounting fraud

Authors:

Abstract and Figures

We examine the association between financial statement comparability and the likelihood of accounting fraud. Prior research documents a negative association between the quality of a firm's reporting environment and accounting fraud. We build on this literature and show that poor financial statement comparability is associated with a greater likelihood of accounting fraud. We also find that accounting comparability declines over time as the year of fraud detection approaches that the association between comparability and fraud becomes more negative over this time. We also find that financial statement comparability improves after fraud detection, consistent with the notion that managers improve their financial reporting quality after fraud. This article is protected by copyright. All rights reserved
Content may be subject to copyright.
Financial Statement Comparability and Accounting Fraud
BELEN BLANCO
Adelaide Business School
The University of Adelaide
Adelaide
Australia
SANDIP DHOLE
Department of Accounting
Monash University
Melbourne
Australia
FERDINAND A. GUL
Department of Accounting
Deakin University/ Sunway University
Melbourne/ Kuala Lumpur
Australia/ Malaysia
This version: May 2022
†Corresponding author: Sandip Dhole, Monash Business School, Department of Accounting, Monash University,
900 Dandenong Road, Caulfield East, VIC 3145. Australia. Email: sandip.dhole@monash.edu. Telephone: +61
(03) 9903 2912.
We thank Peter Pope (editor), an anonymous referee, Debarati Basu, Jean Canil, Pamela Kent, Gladys Lee, Yan
Li, Brian Miller, Sagarika Mishra, Srinivasan Rangan, Grant Richardson, Qingbo Yuan, and workshop
participants at the Indian Institute of Management Bangalore, the annual AFAANZ Conference (Adelaide 2017)
and the EAA Annual Congress (Valencia 2017) for their helpful comments and suggestions.
Electronic copy available at: https://ssrn.com/abstract=4207137
Financial Statement Comparability and Accounting Fraud
Abstract: We examine the association between financial statement comparability and the
likelihood of accounting fraud. Prior research documents a negative association between the
quality of a firm’s reporting environment and accounting fraud. We build on this literature and
show that poor financial statement comparability is associated with a greater likelihood of
accounting fraud. We also find that accounting comparability declines over time as the year of
fraud detection approaches that the association between comparability and fraud becomes more
negative over this time. We also find that financial statement comparability improves after
fraud detection, consistent with the notion that managers improve their financial reporting
quality after fraud.
Keywords: Accounting Fraud, Financial Statement Comparability, Reputation Repair
JEL Classification: M41, M48
Data availability: The data used for this study are available from public sources identified in
the text.
Electronic copy available at: https://ssrn.com/abstract=4207137
1
1. INTRODUCTION
This study examines the association between financial statement comparability
(comparability, hence) and accounting fraud (hereafter, fraud). Fraud broadly refers to
managers employing financial reporting tactics that violate generally accepted accounting
principles (GAAP) to mislead users of financial statements. Firms committing fraud are
therefore liable to be sued under various law provisions. Specifically, Amiram et al. (2018)
suggest characterizing financial reporting fraud as “allegations brought under Section 17(a) of
the 1933 Securities Act (fraudulent interstate transactions) or Section 10(b) of the 1934
Securities Exchange Act (manipulative and deceptive devices).” Alternatively, Amiram et al.
(2018) note that fraud may be defined by “class action lawsuits that allege violations of SEC
1
Rule 10b-5.” We focus on fraud since there have been increasing concerns in many media
outlets about the increasing trends of fraud. Fraud is costly to organizations and economies.
Indeed, a 2016 report by the Association of Certified Fraud Examiners estimates that the
median loss caused by financial statement fraud was $975,000 per fraud scheme between
January 2014 and October 2015.
2
Frauds also trigger regulatory investigations, prosecutions
and financial penalties. In some cases, they also lead to criminal charges against concerned
parties (Feroz et al., 1991).
Given the topical nature of fraud, it is not surprising that much prior research has
examined issues relating to fraud. A key focus of this literature is to identify determinants of
fraud. There is evidence in the literature that fraud firms generally tend to have poor corporate
governance (Beasley et al., 2000; Crutchley et al., 2007). Lennox & Pittman (2010) show that
1
Securities and Exchange Commission
2
See http://www.acfe.com/rttn2016/costs.aspx for details.
Electronic copy available at: https://ssrn.com/abstract=4207137
2
having a Big 5 auditor can reduce the likelihood of fraud, consistent with the idea that improved
monitoring can reduce fraud.
While corporate governance is an important determinant of fraud, other factors could
also be associated with fraud. For example, Amiram et al. (2015) design a financial statement
error measure and show that this measure could predict fraud. Because preparing financial
statements involves estimating future cash flow realizations, managers could have incentives
to estimate these future realizations opportunistically. When managers misuse their reporting
discretion, it adversely affects the firm’s information environment, creating a favorable
environment for fraud (Ndofor et al., 2015).
In this study, we extend this line of investigation by focusing on an important aspect of
the financial reporting environment comparability. We study comparability
3
because it is an
important qualitative characteristic of financial statements, according to accounting standard-
setters.
4
The FASB Conceptual Framework notes that comparability is important since it aids
in valuation. Consistent with this argument, texts on financial statement analysis highlight the
importance of benchmarking a firm’s financial statement ratios with those of its peers to
evaluate firm performance (see Penman, 2013).
High comparability could make it difficult for managers to commit fraud because it
could be more easily detected. Specifically, if a firm’s financial statements are highly
comparable to those of another firm, external users of financial information could infer the
firm’s performance based on that of its peers, given a particular economic scenario. In other
words, high comparability could significantly reduce the information asymmetry between
3
The FASB defines comparability as “the quality of information that enables users to identify similarities in and
differences between two sets of economic phenomena.” (FASB Conceptual Framework, 2006, p.30).
4
See the Financial Accounting Standards Board (FASB) Conceptual Framework (2010), page 32.
Electronic copy available at: https://ssrn.com/abstract=4207137
3
managers and users (Cheng & Wu, 2018; Kim et al., 2013). If a firm with highly comparable
financial statements reports significantly different financial results than its peers under similar
economic conditions, it could raise questions about its financial reporting. This would likely
make fraud easier to detect. Indeed, Jackson et al. (2017) find that lower earnings co-movement
creates incentives for managers to issue biased earnings reports, thus increasing its likelihood
of being the subject of an SEC’s Accounting and Auditing Enforcement Release (AAER).
However, it is not evident that higher comparability would reduce the likelihood of
fraud. Indeed, one could argue that higher comparability could pressure managers to engage in
opportunistic reporting and fraud to maintain the illusion of comparable financial performance
and avoid negative short-term market responses. Consistent with this idea, Fung (2015) finds
that a firm is more likely to commit fraud if its performance is worse than that of industry peers.
Using a propensity score-matched sample of firms, we find a negative association between
comparability and fraud.
Recent research (King & Nielsen, 2019), however, raises concerns about propensity-
score matching, arguing that it could introduce bias. To ensure that the matching procedure
does not drive our results, we use entropy balancing as an alternate approach to generate a
matched sample. Entropy balancing is a superior matching technique (Hainmueller, 2012).
Many recent accounting studies use this technique as an alternative research design (Chen et
al., 2022; Ege et al., 2019; Elnahas et al., 2021). Following these studies, we replicate our main
results using entropy balancing and find that they hold for two of the three comparability
proxies used in the study.
We extend our analysis by examining how the association between financial statement
comparability and fraud changes as the year of fraud approaches. Specifically, we examine
whether comparability becomes steadily worse, closer to the discovery of fraud. Our analysis
Electronic copy available at: https://ssrn.com/abstract=4207137
4
is motivated by prior studies (Dechow et al., 2011; Amiram et al., 2015) that show that financial
statement characteristics could be used to predict financial misstatements. It is important to
understand whether the association between comparability and the likelihood of fraud changes
as the year of fraud approaches, as it could potentially provide a timely warning to users. Our
results suggest that the association between fraud and comparability is more negative in the
year before fraud than four years before. Further, we find that fraud firms experience a
significant decline in comparability from four years before the discovery of fraud to the year
before the fraud. These results suggest that worsening comparability could be a potential red
flag of impending fraud.
Next, we test the consequences of fraud on financial statement comparability. Prior
research (Fich & Shivdasani, 2007; Karpoff et al., 2008a, b) finds that there are serious labor
market and reputational costs of fraud. Once a firm is accused of fraud, we conjecture that
managers would likely take remedial steps to minimize the market’s perception of the firm and
repair lost reputation (Chakravarthy et al., 2014). Farber (2005) finds that firms strengthen their
corporate governance after fraud, which leads to improvements in financial reporting quality
three years after fraud. Since earnings quality is positively associated with comparability (De
Franco et al., 2011), we expect there to be improvements in financial statement comparability
over time after fraud. We find that comparability improves four years after fraud, suggesting
that managers take remedial action to improve the quality of information after fraud.
We perform some important additional tests. First, we re-examine our research question
considering a regulatory change in the US in 2007. On November 15, 2007, the SEC eliminated
the 20-F reconciliation requirement for foreign filers that report under International Financial
Reporting Standards (IFRS). The SEC’s decision raises concern among practitioners that it
might impair financial statement comparability (CFA Institute, 2007). Prior research finds
evidence that the 20-F reconciliations have information content (Chen & Sami, 2008; Harris &
Electronic copy available at: https://ssrn.com/abstract=4207137
5
Muller, 1999). To the extent that the elimination of the 20-F reconciliation adversely affects
the comparability of these foreign filers, it is possible that they would have greater
opportunities to engage in fraud after the elimination of the 20-F reconciliation requirement.
We find evidence that this is indeed so.
We also study how financial distress moderates the association between comparability
and fraud. Prior research by Kothari et al. (2009) and Koch (2002) shows that managers have
incentives to withhold bad news. Thus, it is probable that managers of financially distressed
firms might commit fraud to avoid disclosing the bad news to investors. If this is so, the
association between comparability and fraud would be more negative for financially distressed
firms. We find evidence consistent with this expectation.
We also investigate the moderating effect of corporate governance on the association
between comparability and fraud. Prior research (Beasley et al. 2000; Crutchley et al. 2007)
shows that firms with poor (good) corporate governance are more (less) likely to commit fraud.
This would suggest that the negative association between comparability and fraud is stronger
for firms with poor corporate governance. We find results consistent with this argument using
board size and board independence as proxies for corporate governance.
We make important contributions to the literature. First, to the best of our knowledge,
ours is the first study to provide empirical evidence on the association between accounting
fraud and financial statement comparability. Our result that firms with less comparable
financial statements are more likely to commit fraud is an important contribution. It highlights
the importance of comparability in reducing the likelihood of fraud. Our results should be of
importance to users and regulators.
Second, we find that comparability worsens, and the association between comparability
and fraud is more negative in the year before fraud than four years before. This suggests that
Electronic copy available at: https://ssrn.com/abstract=4207137
6
declining comparability could be a potential indicator of fraud. This result contributes to the
literature on fraud detection.
We also contribute to the literature on reputation repair by showing that there is an
improvement in financial statement comparability after fraud. Our result thus shows
information quality improves after fraud.
While our study is related to Sohn (2016), we believe that there are significant
differences between our study and Sohn (2016). First, unlike Sohn (2016), we focus on
accounting fraud (which is a violation of accounting rules) rather than earnings management
(which is within the scope of GAAP). This enables us to draw more accurate inferences about
the relation between opportunistic manipulation of financial statements and financial statement
comparability. This is because we focus explicitly on instances of fraud identified by the SEC.
Therefore, our research design choice allows us to avoid using accruals-based measures of
earnings management, which have their limitations (Dechow & Skinner, 2000; McNichols,
2001).
Second, we provide definite empirical evidence on the direction of the association
between financial misreporting and financial statement comparability. Specifically, we show
that there is a negative association between financial statement comparability and the
likelihood of fraud. In contrast, Sohn (2016) shows that there is a negative (positive) association
between accounting (real) earnings management and comparability. Thus, based on Sohn
(2016), the nature of the association between total earnings management and comparability
(i.e., whether the association is positive or negative) cannot be ascertained.
Third, unlike Sohn (2016), we show that financial statement comparability declines
steadily in the years leading up to the year of the fraud for firms committing fraud. This is an
important distinction and is an important contribution of our study, as identified above.
Electronic copy available at: https://ssrn.com/abstract=4207137
7
Finally, we also extend our analysis by showing that firms committing fraud
experience significant improvements in comparability four years after fraud. This analysis is
important, as it shows that firms improve their financial reporting practices after being
investigated for fraud. Thus, our study documents an important consequence of fraud for
financial reporting.
The rest of the paper is organized as follows: we discuss our hypotheses in Section 2;
Section 3 discusses the research methodology; we describe our data in Section 4; Section 5
presents the empirical results; Section 6 discusses some additional tests. Section 7 concludes.
2. HYPOTHESES
An extensive literature examines the antecedents and consequences of accounting fraud
(see Amiram et al., 2018 for an excellent review of this literature). Since fraud has high
economic costs, as mentioned above, an important focus of the extant literature is on the
determinants of fraud. Much of the emphasis in the literature is on the association between
governance mechanisms and fraud (see, for example, Beasley et al., 2000; Crutchley et al.,
2007 and Lennox & Pittman, 2010). Generally, these studies find a negative association
between the strength of governance and the likelihood of fraud.
While corporate governance can influence fraudulent financial reporting, managers
could have other incentives to commit fraud. For instance, a firm’s information environment
could also create incentives for managers to commit fraud. Indeed, Ndofor et al. (2015) find
that a higher level of information asymmetry between managers and shareholders increases the
likelihood of fraud. In this study, we extend this line of research by focusing on the association
between financial statement comparability and fraud.
Electronic copy available at: https://ssrn.com/abstract=4207137
8
Comparable financial statements aid in financial decision making and valuation, as they
allow users to compare the performance of one firm with that of another. Thus, greater
comparability reduces the information asymmetry between management and shareholders
(Kim et al., 2013), making it easier to compare the reported financial performance of a company
with its peers. Thus, if a company’s reported performance differs significantly from its peers,
it could raise a red flag for investors. Based on this line of reasoning, we argue that greater
comparability would make it difficult for managers to commit fraud. We state our hypothesis
below:
5
H1A: Firms with less comparable financial statements are more likely to commit fraud.
However, it is not clear whether there will be a negative association between
comparability and fraud. Indeed, higher comparability could create incentives for opportunistic
reporting (Bratten et al., 2016; Kedia et al., 2015) and fraud (Fung 2015) if it creates pressure
on management to maintain a given level of operating performance. H1A is, therefore, a
refutable hypothesis.
Next, we study how comparability changes in the years leading up to the discovery of
the fraud and how the association between comparability and fraud changes over this period.
Our analysis is motivated by prior work by Amiram et al. (2015) and Dechow et al. (2011),
who show that financial statement data can be used to predict future financial misstatement and
fraud. The idea is that if financial reporting quality declines before the discovery of fraud, it
should also be reflected in worsening comparability and thus a more negative association
between comparability and fraud over time. Accordingly, we hypothesize that:
5
We state all hypotheses in the alternate form.
Electronic copy available at: https://ssrn.com/abstract=4207137
9
H1B: The association between financial statement comparability and the likelihood of
fraud becomes more negative as the firm approaches the year of fraud.
Finally, we focus on the consequences of fraud and study whether there is a post-fraud
change in financial statement comparability. Prior research finds that there are severe labor
market consequences of accounting fraud. Fich and Shivdasani (2007) show that fraud is costly
for outside directors. Specifically, outside directors find it challenging to get directorships after
fraud, especially if they bear greater responsibility for monitoring fraud. Karpoff et al. (2008a)
find that individuals identified as responsible parties for fraud tend to lose their current
employment and suffer severe financial consequences arising out of restrictions on
employment, shareholdings in the firm, and SEC penalties. These studies thus show that
reputation loss can be severe for firms committing fraud. Consistent with firms suffering
reputation loss, Karpoff et al. (2008b) show that the reputational penalty on firms targeted by
SEC enforcement actions for financial misrepresentation is more than 7.5 times the costs
imposed by the legal and regulatory systems.
Firms also suffer adverse market consequences of fraud. Feroz et al. (1991) report that
disclosure of alleged reporting violations is associated with a significant negative abnormal
return, suggesting that the stock market penalizes firms for fraud. Graham et al. (2008) show
that financial misreporting has bond market consequences also. Specifically, loans initiated
after financial restatements tend to be of shorter maturities, have higher spreads, have a higher
likelihood of being secured, and tend to have more covenants.
Given the above evidence, we argue that firms committing fraud will likely have strong
incentives to repair their lost reputation and minimize the market impact of the fraud. Farber
(2005) finds that firms committing fraud improve their corporate governance and that three
years after fraud, their governance characteristics are comparable to non-fraud firms.
Electronic copy available at: https://ssrn.com/abstract=4207137
10
Chakravarthy et al. (2014) show that managers significantly increase the number of reputation-
building actions following restatements targeted at different stakeholder groups and that the
earnings response coefficient for firms that engage in more (fewer) reputation-building
measures is higher (lower) after the restatement. We argue that the improvement in financial
reporting quality after fraud would also make financial statements more comparable. We
formally state our hypothesis below:
H2: Financial statement comparability improves after fraud.
3. RESEARCH METHODOLOGY
(i) Measuring Fraud
Following prior research (Erickson et al., 2006; Johnson et al., 2009; Lennox &
Pittman, 2010), we define FRAUD as a dummy variable equal to 1 if the firm is the subject of
a fraud-related SEC AAER. We use the AAER dataset of Dechow et al. (2011) to identify fraud
cases. Note that not all AAERs relate to fraud. Some relate to the auditor issues, such as the
auditor not being a PCAOB registrant (for example, AAER Number 2679) and auditor
independence (for example, AAER Number 1491). Some others relate to bribes (for instance,
SEC AAER Number 3884) and insider trading (for example, SEC AAER Number 3853). Some
other AAERs (for example, SEC AAER Number 1609) have no detail. We exclude these
AAERs from our definition of fraud, consistent with prior literature (for example, Brown et al.,
2020).
6
We define FRAUD as 1 in the last year of the AAER, following prior research (Glancy
& Yadav, 2011).
7
6
In an untabulated sensitivity test, we find that our main results do not change if we include all AAERs in our
definition of fraud.
7
We test the robustness of our results by using two alternate definitions of fraud, defined as: (1) the first year of
the AAER, and (2) the full period of the AAER. We describe these results in Section 6.
Electronic copy available at: https://ssrn.com/abstract=4207137
11
(ii) Measuring Financial Statement Comparability
We use the De Franco et al. (2011) approach to define Comparability. De Franco et al.
(2011) view comparability as the extent to which economic events map into firms’ earnings.
Two firms have comparable earnings if their reported earnings are similar under similar
economic conditions. In the De Franco et al. (2011) framework, economic events are measured
by firms’ stock returns. This approach is consistent with Barth et al. (2012), who view returns
as a measure of economic outcome because they measure “change in equity value that reflect
investors’ capital allocation decisions”. The idea that stock returns capture economic events is
well articulated by Kothari (2001), who notes that “the information set reflected in prices is
richer than that in contemporaneous accounting earnings. In an efficient market, price changes
instantaneously incorporate the present value of the revisions in the market's expectations of
future net cash flows”.
De Franco et al. (2011) essentially measure accounting system comparability. This view
of comparability is consistent with the FASB’s view of comparability. De Franco et al. (2011)
start with the following regression model:
  
(1)
In equation (1), Earnings is the quarterly income before extraordinary items of firm i
in quarter t, deflated by the beginning of the quarter market value of equity. , which
measures firm i’s quarterly stock return, is a proxy for economic events. Equation (1) is
estimated for each firm-year over the past 12 quarters. The estimated coefficients (
) and (
󰆹)
from the regression capture firm i’s accounting function. Similarly,
and
󰆹 measure the
accounting function of firm j from the same two-digit SIC industry as firm i.
Electronic copy available at: https://ssrn.com/abstract=4207137
12
To gauge the distance in the accounting systems between firm i and firm j in mapping
economic events, the predicted earnings for firms i and j are calculated, given the economic
events of firm i (Returnit), using each firm’s estimated accounting functions as follows:
󰇛󰇜

(2)


(3)
In equations (2) and (3) above, E(Earningsiit) is the expected earnings for firm i given
firm its accounting function and return in quarter t, and E(Earningsijt) is that for firm j, given
firm its accounting function and firm i’s return in quarter t. The mean absolute difference
between the expected earnings of the past 12 quarters measures how comparable earnings are.
It is multiplied by minus 1 to measure comparability:

 󰇛󰇜

(4)
De Franco et al.’s (2011) measure is the most common measure of comparability used
in the literature.
8
Following De Franco et al. (2011), we construct four proxies of comparability:
CompAcctM4 (based on the top four values of CompAcct for a firm in an industry),
CompAcctM10 (based on the top 10 values of CompAcct for a firm in an industry),
CompAcctInd (based on the mean value of CompAcct for a firm in an industry) and
CompAcctIndMed (based on the median value of CompAcct for a firm in an industry). We use
CompAcctM4 as one of our three primary proxies of comparability.
8
An analysis of the comparability research published in leading accounting journals to date suggests that more
than 70 per cent of the published articles use the De Franco et al. (2011) measure or some modification of it.
Electronic copy available at: https://ssrn.com/abstract=4207137
13
Our second measure of comparability is based on earnings covariance (Francis et al.,
2014). Unlike the accounting system comparability measure, the earnings covariance (EComp)
measure captures comparability by the extent to which the earnings of two firms co-vary. We
construct this measure following a similar empirical approach as above. Specifically, we
capture pairwise differences in the covariance in earnings and estimate the earnings
comparability (EComp) measure as the average of the top four values of EComp for a firm in
an industry ECompM4.
Our final measure of comparability is the disclosure quality (DQ) measure of Chen et
al. (2015).
9
DQ considers the extensiveness of a firm’s financial statement disclosures.
Specifically, it is measured as the total number of non-missing line items in a firm’s financial
statements relative to the maximum number of line items that the firm can disclose. The higher
the ratio, the greater the disclosure quality. DQ can be regarded as a measure of comparability
because the number of line items that a firm can report depends on the industry to which it
belongs. For example, internet firms would typically not have any inventory, resulting in
missing inventory line-item values in their financial statements. Chen et al. (2015) construct
two proxies of disclosure quality, one based on the Balance Sheet and the other based on the
Income Statement and then generate the DQ measure as the average of the Balance Sheet and
Income Statement measures. Recent research (Caylor et al., 2022) shows that DQ is positively
correlated with the accounting comparability measure of De Franco et al. (2011).
Comparability is related to the concept of financial statement uniformity (Caylor et al.,
2022). However, unlike uniformity which measures the extent to which the financial reporting
of a firm converges with that of the industry, comparability compares similarities in the
reporting of individual firms (Caylor et al., 2022).
9
We thank Bin Miao and Shuping Chen for sharing the data.
Electronic copy available at: https://ssrn.com/abstract=4207137
14
(iii) Empirical Models
To test H1A (the association between financial reporting comparability and the
likelihood of fraud), we estimate the following logit model following Erickson et al. (2006):
󰇛 󰇜
  
   
  
  



(5)
We define FRAUD and Comparability as above. Based on prior literature (Erickson et
al. 2006), we include the following control variables: the annual stock return volatility,
calculated as the standard deviation of the daily stock returns (Volatility), as a proxy of
uncertainty. We control for Volatility, as monitoring executives is more difficult for firms
operating in less predictable environments (Demsetz & Lehn, 1985), and executives would
have more opportunities to commit fraud. Size is the natural logarithm of total assets; Lev is the
total debt divided by total assets. BTM is the book value of shareholder’s equity, divided by the
market value of equity. ROA is net income divided by the closing balance of assets.
SalesGrowth is the percentage change in sales from the prior year to the current year.
EarningsPrice is the net income per share, divided by stock price at the end of the year.
AltmanZ is the proxy for the risk of financial distress calculated based on Altman (1968).
Financing is a proxy for a firm’s needs for external financing. It is a dummy variable equal to
1 if the firm’s variable FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated
Electronic copy available at: https://ssrn.com/abstract=4207137
15
as (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current Assetst−1. BoardSize
is the size of the board. Big4 is a dummy variable equal to 1 if the external auditor is a Big4
auditor, and zero otherwise. We control for BoardSize, since it is difficult for a larger board to
monitor the manager’s decisions, thus potentially creating more incentives for fraud (Anderson
et al. 2004). Finally, we include M&AIndicator defined as a dummy variable equal to 1 if the
company has an acquisition that contributed to sales, and zero otherwise.
The coefficient of interest in equation (5) is , which measures the relation between
financial statement comparability (Comparability) and fraud (FRAUD). If financial statement
comparability reduces the likelihood of fraud, then would be negative.
We test H1B by replacing Comparability with lagged values of comparability in
equation (5) above. Specifically, we use four lags of comparability. H1B predicts that
becomes more negative over time. Thus, empirical evidence that decreases from four years
before fraud to one year before fraud would be consistent with the prediction of H1B.
H2 tests whether financial reporting comparability improves after fraud. Following
Brochet et al. (2013) and Dhole et al. (2015), we estimate the following model to test H2:

  
  
  
  



(6)
In equation (6), PostXFRAUD is a dummy variable set equal to 1 if the observation is
from “X” periods after fraud, and 0 otherwise. It thus measures the time since fraud. Financial
statement comparability may improve for the fraud firms a few years after fraud. Indeed, Farber
Electronic copy available at: https://ssrn.com/abstract=4207137
16
(2005) finds that firms accused of fraud strengthen their corporate governance and that three
years after fraud, their corporate governance characteristics are similar to those of Non-
FRAUD firms. To allow for the possibility that improvements in comparability can occur over
time, we estimate equation (6) using different time horizons after fraud. Specifically, X can be
1, 2, 3, or 4 periods after fraud. Thus, Post1FRAUD is a dummy variable equal to 1 if the
observation is from 1 period after fraud, and zero otherwise. We control for Volatility, as
monitoring executives is more difficult for firms operating in less predictable environments
(Demsetz & Lehn, 1985), and executives would have more opportunities to commit fraud. We
also control for Lev, defined as the total debt divided by total assets. All other control variables
that could affect Comparability are as described above.
The coefficient of interest in equation (6) is , which measures whether Comparability
improves X years after fraud. If financial statement comparability increases after a firm is
charged with accounting fraud, then would be positive and significant.
4. DATA
(i) Sample Selection
Our initial sample is based on the intersection of firms in the Compustat North America
Industrial Annual, CRSP, Execucomp and the Dechow et al. (2011) AAER databases. Our
sample covers the period 2000-2018. Our initial sample consists of 28,722 non-missing firm-
year observations of all regression variables. Recall that our primary independent variable of
interest in H1 is comparability. We argue that lower comparability increases the likelihood of
fraud. However, it is possible that firms with high comparability could be systematically
different from those with lower comparability. This could introduce self-selection bias in our
sample. Our results may not be attributed to comparability alone if this is the case. Not
accounting for this endogeneity is problematic. Hence, we generate a propensity score-matched
Electronic copy available at: https://ssrn.com/abstract=4207137
17
(PSM) sample of firms based on observable firm characteristics by estimating the following
logit model:
    


(7)
In equation (7), High is a dummy variable equal to 1 if the value of the comparability
proxy is greater than the median of the distribution; zero otherwise. Following prior research
(Rosenbaum & Rubin, 1983), for each firm with greater than median comparability for the full
sample, we identify one control firm with lower than median comparability with the closest
propensity score within a caliper of 0.001. Before constructing the matched sample, we
compare the observable firm characteristics of the high and low comparability firms and find
that they are significantly different. We report these statistics in the online appendix. We match
firms on comparability based on the following covariates: Volatility, Size, Lev, and BTM. These
variables are as defined above. We identify these determinants based on prior research (Chen
et al., 2020; De Franco et al., 2011)
We also include industry-fixed effects in the determinants model based on Cascino &
Gassen (2015). Finally, we include year-fixed effects to control for year-specific effects on
accounting comparability. We report the estimation results of Equation (7) in the online
appendix.
To ensure that the control firms identified by the propensity score matching are similar
in terms of the observable firm characteristics, we conduct t-tests of differences in those
characteristics between the high and low comparability firms. We find that the differences are
not significant. We report these statistics in the online appendix. Our propensity score
matching yields a final maximum regression sample of 8,996 firm-year observations for
Electronic copy available at: https://ssrn.com/abstract=4207137
18
CompAcctM4. We follow a similar approach to generate propensity score-matched samples for
the other two comparability proxies (ECompM4 and DQ).
10
(ii) Descriptive Statistics
Table 1 presents descriptive statistics for the main variables of interest. The mean
FRAUD across all firm years represents 1.9 per cent. This shows that fraud is relatively rare.
The mean (median) values of CompAcctM4, CompAcctM10, CompAcctInd, and
CompAcctIndMed are -0.519 (-0.210), -0.743 (-0.320), -3.541 (-3.010), and -2.604 (-1.780)
respectively. The mean (median) value of ECompM4 is 0.458 (0.466) and those of DQ are
0.753 (0.767). The Table shows that the mean value of Size is 6.248 (translating to mean total
assets of $516 million), the average leverage is 0.166, and BTM is 0.557. The mean AltmanZ
score for our sample firms is 4.784, suggesting that our sample firms are at low risk of financial
distress, on average. These descriptive statistics reported in Table 1 are consistent with prior
studies that use the Compustat data (Francis et al. 2014; Amiram et al., 2015). This lends
confidence in the representativeness of our sample.
[Insert Table 1 about here]
In Table 2, we compare the descriptive statistics of FRAUD and Non-FRAUD firms.
Recall that FRAUD firms are those that are the subject of fraud-related SEC AAERs in any of
our sample years. In contrast, Non-FRAUD firms are those firms not charged with. From the
Table, we see that Comparability measures are lower for FRAUD firms than Non-FRAUD
firms, for all the fraud measures, except for CompAcctInd and ECompM4. These descriptive
statistics present preliminary evidence in support of H1A. We also observe that FRAUD firms
10
We also define high and low comparability by year and replicate our main results (Table 4) by estimating the
model on a propensity score matched sample based on this alternate definition of comparability. We find that our
results remain unchanged. We present these results in the online appendix.
Electronic copy available at: https://ssrn.com/abstract=4207137
19
are larger, have poorer performance, have larger boards of directors, and are more complex in
general, consistent with prior research (Lennox & Pittman, 2010).
[Insert Table 2 about here]
We present the distribution of fraud cases by year in Panel A of Table 3. The Table
shows that there are many fraud cases between 2006 and 2009. This could be because this
period coincided with greater regulatory oversight because of the Global Financial Crisis. We
also observe a jump in 2003 and 2004, which may be associated with the Sarbanes-Oxley Act
(SOX). There are no fraud cases after 2016.
Panel B of Table 3 presents the distribution of fraud by industry. We observe that
FRAUD firms are more likely to be in the Food and Kindred Products; Industrial Machinery
and Equipment; Electronic and Other Electric Equipment; Instruments and Related Products;
and Business Services industries. The industry distribution is consistent with prior studies
(Brazel et al., 2009).
[Insert Table 3 about here]
5. RESULTS
(i) Financial Statement Comparability and the Likelihood of Accounting Fraud
Table 4 presents results for the test of H1A. Recall that H1A tests whether low
comparability is associated with a higher likelihood of fraud. As discussed above, we estimate
equation (5) for a PSM sample to test H1A. H1A predicts that the coefficient of Comparability
is negative. The Table shows that this is indeed so. We present the CompAcctM4, ECompM4
and DQ results in Columns 1-3 respectively of Table 4. The coefficient of Comparability is
negative for all three measures of comparability (coefficient = -0.466, -2.640, and -4.086
respectively; p-value = 0.051, 0.096, and 0.018 respectively). These results suggest that the
likelihood of fraud is higher for firms with less comparable financial statements, consistent
Electronic copy available at: https://ssrn.com/abstract=4207137
20
with H1A. Our results suggest that poor comparability creates opportunities for managers to
commit fraud. We find that size is positively associated with the likelihood of fraud (coefficient
= 0.722, p-value=0.000).
11
[Insert Table 4 about here]
(ii) Entropy Balancing
Recent research (King & Nielsen, 2019) questions the validity of the propensity score
matching technique, arguing that it can often lead to increased imbalance, model dependence,
and bias. Hainmueller (2012) propose entropy balancing as a superior balancing technique.
Entropy balancing weights control firms on different moments of control variables in the
model. It is a superior approach because it achieves covariate balance by construction. Many
recent accounting studies (Chen et al., 2022; Ege et al., 2019; Elnahas et al., 2021) use entropy
balancing as an additional balancing measure to address the issue of endogeneity. We follow
these studies and replicate the results of Hypothesis 1a using entropy balancing.
We conduct entropy balancing based on the covariates identified above for the
propensity-score matched sample. The entropy balancing assigns weights to achieve covariate
balancing. Specifically, it balances weights based on how well the high and low comparability
observations match on the mean and variance of these covariates (because we match on two
moments).
12
Like our propensity score matched design, we perform the entropy balancing for
the whole sample period.
13
By construction, there is no difference in the covariates for the high
11
We only discuss results for the first column, for the sake of brevity.
12
Note that we convert the continuous comparability variables into dichotomous proxies (High) to implement the
propensity score matching and entropy balancing. Our approach is based on the traditional propensity score
matching and entropy balancing research designs. However, recent research (Fong et al., 2018) provides
alternative approaches to match on continuous treatment variables. This is a potential limitation of our study.
13
We also replicate the entropy balancing results by defining high and low comparability by year. The results are
consistent with those reported in Table 5 (significant for CompAcctM4 and ECompM4, but not significant for
DQ). We tabulate these results in the online appendix.
Electronic copy available at: https://ssrn.com/abstract=4207137
21
and low comparability observations after entropy balancing. We present these results in the
online appendix.
We report these results in Table 5 using two moments to balance covariates. As above,
we present results for three comparability proxies CompAcctM4 (Column 1), ECompM4
(Column 2), and DQ (Column 3). We find that the coefficient of Comparability is negative in
Columns 1 and 2 (coefficient = -0.300 and -1.405 respectively; p-value = 0.056 and 0.021
respectively).
While our results remain significant for CompAcctM4 and ECompM4, we observe that
the negative association between comparability and the likelihood of fraud does not hold for
the third proxy, DQ (p-value = 0.304). Interestingly, the coefficient of DQ is the most
statistically significant of the three comparability proxies in the PSM design (p-value = 0.018
in Table 4). This shows that empirical results obtained using a PSM research design can differ
from those using entropy balancing.
14
[Insert Table 5 about here]
While entropy balancing is a superior covariate balancing technique than propensity-
score matching (PSM), this approach may not always produce results that completely match
with PSM. It is, therefore, necessary to understand why the results may differ.
First, we observe significant differences in the covariates between the pre-PSM/pre-
entropy balancing sample and the post-PSM sample for both the treated and control samples.
Although the entropy balancing sample also differs from the pre-PSM/pre-entropy balancing
sample in terms of covariates, the differences in covariates are smaller. We report these
statistics in the online appendix. This result shows that the characteristics of the three samples
are significantly different.
14
We also replicate the results by balancing on three moments. We find that our results are qualitatively similar.
Electronic copy available at: https://ssrn.com/abstract=4207137
22
Second, we explore the differences between the treatment and control samples obtained
under the PSM and entropy balancing designs. Our PSM approach uses a caliper of 0.001 to
match control and treatment observations. This stringent caliper results in a significant loss of
information and sample size (our PSM sample consists of 8,996 firm-year observations in
Table 4, whereas the entropy balancing sample consists of 18,069 firm-year observations in
Table 5 for CompAcctM4). In contrast, there is no information or sample loss in the entropy
balancing because it finds a set of weights that satisfies the balance conditions and remains as
close as possible (in an entropy sense) to uniform base weights, and retains efficiency. The
difference in approach leads to different samples for the post-matching regression analysis.
To illustrate this, we compare the descriptive statistics of the covariates in the post-PSM
sample with those of the post-entropy balancing sample. All covariates except Lev (for the
treatment sample) differ significantly in these samples. We tabulate these results in the online
appendix. These sample differences could account for the difference in the PSM and entropy
balancing results.
Finally, McMullin & Schonberger (2022) note that while entropy balancing has many
advantages as a matching procedure, they highlight some key challenges of implementing this
technique when using panel data, especially in situations where the assignment to treatment
and control groups is not clear (such as in our setting, where the assignment is based on
comparability). McMullin & Schonberger (2022) identify two main issues of entropy balancing
in these settings the assignment of extreme weights to some control observations and
covariate balancing across years in a pooled sample. The assignment of extreme weights to
control sample observations (likely to occur when there are systematic differences in treatment
and control covariates) raises concerns about how reproducible the estimate of the treatment
effect is with alternate samples. The problem with covariate balancing across years in a pooled
Electronic copy available at: https://ssrn.com/abstract=4207137
23
sample is that entropy balancing could produce biased estimates of a treatment effect if there
are significant changes to the observations.
(iii) The Association between Financial Statement Comparability and Fraud over Time
As a preliminary step to test H1B, we plot financial statement comparability for the
fraud and matched non-fraud firms over time. We present the graph in Figure 1. The Figure
shows that the fraud firms have significantly lower comparability than the non-fraud firms over
the 9 years surrounding the fraud (year 0), on average. Figure 1 shows that fraud firms
experience a significant decline in comparability as the year of fraud approaches. Specifically,
the average comparability (CompAcctM4) decreases from -0.483 four years before fraud to -
0.578 in the year of the fraud. Untabulated results suggest that the mean difference in
comparability four and one years before fraud is significant (p-value = 0.046). After that,
comparability generally improves to -0.478 four years after fraud. We also note that
comparability seems to return to the pre-fraud levels four years after fraud. In contrast, we do
not observe a similar pattern in the comparability of the non-fraud firms over this period. These
results are interesting and provide preliminary evidence of changes in comparability between
fraud and matched non-fraud firms over time. The Figure shows that, while the fraud firms
have lower comparability than the non-fraud firms, on average, their comparability difference
becomes smaller with time after the fraud, suggesting that fraud firms experience
improvements in comparability after the detection of the fraud.
[Insert Figure 1 about here]
We present the empirical test results of H1B in Table 6. Columns 1-4 of the Table show
results for models using the first, second, third, and fourth lags of Comparability (measured as
CompAcctM4), respectively. The Table shows that the coefficient of lagged comparability is
only significant in Column 1 (coefficient=-0.368; p-value=0.091), consistent with Table 4. It
Electronic copy available at: https://ssrn.com/abstract=4207137
24
is not significant in any of the other columns. Further, statistical tests show that these
coefficient estimates are significantly different the p-values for differences between Columns
are less than 0.01. Specifically, we see that the association between comparability in the year
before fraud with the likelihood of fraud is significantly more negative than that four years
before fraud, consistent with the prediction of H1B. Together with the univariate result
described above, this result is important because it suggests that comparability declines over
time before fraud. Further, the association between comparability and fraud becomes more
negative over time before the fraud is eventually discovered. The coefficients of the control
variables are consistent with Table 4. In Panel B of Table 6, we report the coefficients of
Comparability measured using ECompM4 and DQ, respectively. For brevity, we only report
the coefficients of the comparability proxies. While we do not find any significant coefficient
of the lags of ECompM4, the results for DQ are comparable to those reported in Panel A.
Specifically, we find that the coefficient of the first lag of DQ is negative (coefficient = -1.937;
p-value = 0.086). Together, these results show that comparability seems to worsen closer to the
year of the detection of fraud. Thus, worsening comparability could be a potential red flag
indicating impending fraud.
[Insert Table 6 about here]
(iv) Financial Statement Comparability after Accounting Fraud
We present the results of H2 in Table 7. H2 tests whether there is an improvement in
financial statement comparability after fraud. As mentioned above, we test for changes in
comparability 1, 2, 3, and 4 years after fraud. The coefficient of interest in Table 7 is the
coefficient of PostXFRAUD (). In Panel A of Table 7, we report results for CompAcctM4
and in Panel B, those for ECompM4 and DQ. From Panel A, we see that the coefficient of
PostXFRAUD is significant only four years after fraud (coefficient = 0.328; p-value = 0.021),
Electronic copy available at: https://ssrn.com/abstract=4207137
25
suggesting that comparability improves four years after fraud. Panel B also shows similar
results the coefficient of PostXFRAUD is significant only four years after fraud (coefficient
= 0.896 and 1.867 respectively; p-value = 0.085 and 0.056 respectively). Our result is consistent
with Farber (2005), who finds that firms committing fraud experience improvements in
corporate governance after fraud.
[Insert Table 7 about here]
We next discuss the results of some additional tests.
6. ADDITIONAL TESTS
This section describes some additional tests that we have performed to understand
further the association between comparability and the likelihood of accounting fraud. In our
first set of tests, we identify two firm-specific factors likely to affect comparability and fraud
and test the moderating effect of these variables on the association between comparability and
fraud. We then test the robustness of our main results to alternate measures of financial
statement comparability and fraud. Finally, we consider whether poor current comparability
could be a consequence of fraud.
(i) Financial Distress and the Association between Financial Statement Comparability and
Fraud
We first examine how financial distress affects the association between comparability
and fraud. Prior research (Kothari et al. 2009) suggests that managers have incentives to
withhold the disclosure of bad news. Thus, firms approaching financial distress would likely
have incentives to hide this information from market participants. Indeed, declining operating
performance could create incentives for managers to commit fraud. Consistent with this idea,
Maksimovic & Titman (1991) argue that financially distressed firms are more likely to commit
fraud. Based on this discussion, we argue that if financially distressed firms have incentives to
Electronic copy available at: https://ssrn.com/abstract=4207137
26
commit fraud, the association between comparability and fraud would be more negative for
such firms. We test this conjecture in Table 8.
We measure financial distress using the Altman (1968) Z-score and divide our sample
into a high likelihood of distress (Distress) sample (Z-score less than 1.81) and a low likelihood
of distress (Non-Distress) sample (all other firms). We then estimate equation (7) for each
subsample. We present results using CompAcctM4 in Panel A of Table 8. We see that β1 is
negative for the High Distress sample (coefficient = -1.864; p-value = 0.032) but insignificant
for the Low Distress sample (p-value = 0.432). This suggests that the negative association
between comparability and fraud is more pronounced for financially distressed firms.
[Insert Table 8 about here]
We replicate the results in Panel A with the other two comparability proxies
EcompM4 and DQ respectively and report the coefficients on EcompM4 and DQ in Panel B.
Consistent with our results above, we find that the coefficients on EcompM4 and DQ are
negative for the High Distress sample (coefficient = -3.908 and -4.743 respectively; p-value =
0.012 and 0.011 respectively), but not for the Low Distress sample (p-value = 0.734 and 0.401
respectively).
(ii) Corporate Governance and the Association between Financial Statement Comparability
and Fraud
We also examine how corporate governance affects the association between
comparability and fraud. Prior research (Beasley et al. 2000; Crutchley et al. 2007) argues that
there is a negative association between corporate governance and the likelihood of fraud,
suggesting that firms with poor (good) corporate governance are more (less) likely to commit
fraud. This is consistent with the idea that strong corporate governance reduces firms’ risk-
taking behavior (Ali et al., 2022). Prior research (for example, Campbell & Yeung, 2017;
Endrawes et al., 2020) also finds that strong corporate governance is a determinant of
Electronic copy available at: https://ssrn.com/abstract=4207137
27
comparability. Based on this research, we argue that the negative association between fraud
and comparability is likely to be stronger for firms with poor corporate governance. We present
these results in Table 9.
Following prior research, we use two proxies for corporate governance board size and
the number of independent directors. Using the same approach as Table 8, we first divide our
sample into large (small) board size and more (less) independent directors on the board
samples, respectively, based on whether the value of board size (independent directors) is
greater than the industry-year median. We present the results for the board of directors in Panel
A (using CompAcctM4 as the comparability proxy) of Table 9. Columns 1-2 of Panel A Table
9 present results for board size and 3-4 for independent directors. We see from Column 1 (3)
that β1 is negative (coefficient = -1.278, and -1.692 respectively; p-value = 0.013, and 0.011
respectively). In contrast, β1 is not significant in Columns 2 and 4 (p-value = 0.129, and 0.914
respectively). This is consistent with our expectation above.
[Insert Table 9 about here]
As in Table 8, we present results for the other comparability proxies (ECompM4 and
DQ) in Panel B. We find similar results as above. Specifically, the coefficient of Comparability
is negative for both the low board size (coefficient = -3.983 and -5.929 respectively; p-value =
0.011 and 0.013 respectively) and low board independence (coefficient = -3.398 and -5.018
respectively; p-value = 0.016 and 0.014 respectively) samples. It is not significant for the large
board size and high board independence samples. This shows that the negative association
between comparability and the likelihood of fraud is more pronounced for firms with weaker
corporate governance.
(iii) The Effect of the Elimination of the 20-F Reconciliation Requirement for Foreign Filers
on the Association between Fraud and Financial Statement Comparability
Electronic copy available at: https://ssrn.com/abstract=4207137
28
Although we conduct our primary analyses using a PSM sample, we conduct an
additional test to control for potential endogeneity in comparability. Specifically, we test our
main result using a regulatory change in the US. On November 15, 2007, the SEC eliminated
the 20-F reconciliation requirements for foreign filers using IFRS. Practitioners criticize this
decision on the grounds that eliminating the 20-F reconciliation would impair financial
statement comparability (CFA Institute 2007). Indeed, some evidence shows that comparability
has been impaired after the regulatory change (Byard et al., 2017).
Prior research shows that the 20-F reconciliations have information content (Harris &
Mueller, 1999; Chen & Sami, 2008; Kang et al., 2012). To the extent that the elimination of
the 20-F reconciliation causes a loss of valuable information, this regulatory event could create
incentives for managers of foreign filers to commit fraud. We empirically test this conjecture
by using the logit model below:
󰇛 󰇜
 
 

 
  
 
 
 


(8)
Electronic copy available at: https://ssrn.com/abstract=4207137
29
In equation (8) above, REC is a dummy variable equal to 1 for years on or after 2008;
0 otherwise, FOREIGN is a dummy variable equal to 1 if the firm is a foreign filer; 0 otherwise.
The coefficient of interest in the above model is . It shows the incremental association
between comparability and the likelihood of fraud by foreign filers, after the elimination of the
20-F reconciliation. If the elimination of the 20-F reconciliation lowered comparability and
created opportunities and incentives for affected foreign filers, we would expect the coefficient
to be negative.
We present the results for this test in Table 10. We report results for all three measures
of comparability
15
CompAcctM4 in Column 1, ECompM4 in Column 2 and DQ in Column
3. We see that the coefficient of REC*FOREIGN*Comparability is negative for all the
comparability proxies (coefficient = -0.827, -2.745 and -3.958 respectively; p-value = 0.000 in
all Columns), suggesting that the association between comparability and fraud is more negative
for foreign filers after the elimination of the 20-F reconciliation requirement, consistent with
the notion that the rule change reduces accounting comparability of these filers and creates
opportunities for fraud.
[Insert Table 10 about here]
(iv)Alternate Measure of Financial Statement Comparability
We test the robustness of our results above to three other comparability proxies
CompAcctM10, CompAcctInd and CompAcctIndMed, which are based on the comparability
measurement approach of De Franco et al. (2011). As mentioned above, CompAcctM10 is
calculated as the average of the ten highest firm-pair values of comparability, CompAcctInd is
based on the mean of the firm-pair values of comparability for a firm in an industry and
CompAcctIndMed is based on the median of the firm-pair values of comparability for a firm in
15
Our results are robust to the other Comparability measures.
Electronic copy available at: https://ssrn.com/abstract=4207137
30
an industry. We present these results in Table 11, with CompAcctM10 (in Column 1),
CompAcctInd (in Column 2) and CompAcctIndMed (in Column 3). We find that the coefficient
of comparability is negative in all 3 columns (coefficient = -0.447, -0.196 and -0.206
respectively; p-value = 0.039, 0.099 and 0.060 respectively). This is consistent with the results
reported in Table 4.
[Insert Table 11 about here]
(v) Alternate Measure of Fraud
In all the tests reported above, we define Fraud as a dummy variable equal to 1 for the
last year of the AAER; zero otherwise. We now consider two alternate definitions of Fraud
defined as above, based on the first year of the AAER and the full period of the AAER. We re-
estimate equation (5) using these alternate definitions of fraud and present our results in Table
12. In Panel A, we present the results of equation (5), with fraud defined as the first year of the
AAER. As in Table 4, we report results for all three comparability proxies CompAcctM4
(Column 1), ECompM4 (Column 2) and DQ (Column 3). We find that the coefficient of
comparability is negative for CompAcctM4 and DQ comparability proxies (coefficient = -0.184
and -3.637 respectively; p-value = 0.041 and 0.039 respectively).
[Insert Table 12 about here]
In Panel B, we present the results of equation (5) with fraud defined over the full period
of the AAER. As in Panel A, we present results for all three comparability proxies. Consistent
with Panel A, we find that the coefficient of comparability is negative for CompAcctM4 and
DQ comparability proxies (coefficient = -0.192 and -4.086 respectively; p-value = 0.030 and
0.018 respectively). The results of Table 12 thus show that our results are not sensitive to the
definition of fraud.
Electronic copy available at: https://ssrn.com/abstract=4207137
31
(vi) Is Lower Comparability a Consequence of Accounting Fraud?
Recall that H1A tests whether firms that have lower financial statement comparability
are more likely to commit fraud. In other words, H1A investigates whether lower comparability
creates incentives and opportunities that increase the likelihood of fraud. It is, however,
plausible that lower comparability is a consequence of fraud. That is, the negative association
that we have documented in Table 4 is because of fraud. We now investigate this possibility.
Specifically, we estimate the following model to test this conjecture:

   
   
  
  


(9)
The variables in equation (9) are as defined above. We estimate equation (9) using all
measures of comparability identified above. If lower comparability is a consequence of fraud,
the coefficient of Fraud will be negative.
We present the results in Table 13. We see that the coefficient of FRAUD is not
significant for any of the three measures of comparability (p-value = 0.342, 0.297, and 0.101,
respectively). These results suggest that there is no evidence that lower comparability is a
consequence of fraud, thus making the alternate explanation for our results less plausible. The
coefficients on Volatility and Lev are negative (coefficient = -10.621 and -0.394; p-value=0.000
Electronic copy available at: https://ssrn.com/abstract=4207137
32
for both), and that on Size is positive (coefficient = 0.178; p-value = 0.000), consistent with
prior research (De Franco et al. 2011; Dhole et al. 2015).
16
[Insert Table 13 about here]
7. CONCLUSION
We examine the association between financial statement comparability and the
likelihood of fraud. Accounting fraud is a serious corporate offence and is costly for
organizations, capital markets and concerned employees. Based on prior research that finds
that a poor information environment increases the likelihood of fraud, we examine how
financial statement comparability, an important qualitative characteristic of financial
statements, relates to fraud. High comparability significantly reduces information asymmetry
between managers and external parties. Therefore, managers of firms with highly comparable
financial statements would presumably be less likely to commit fraud. Consistent with our
expectations, we find that fraud is more likely when firms have less comparable financial
statements. We also that the association between comparability and fraud becomes
significantly more negative over time in the years leading up to the fraud.
These results should be of interest to managers, shareholders, auditors, and regulators,
as they highlight how financial statement comparability can be used to identify fraud. Our result
that lower comparability is associated with a greater likelihood of fraud speaks to the
importance of comparability as a qualitative characteristic of financial statements in potentially
helping external parties detect fraudulent financial reporting.
We next examine changes in comparability after fraud. Prior research suggests that
managers seek to improve corporate governance practices in the wake of fraud. Managers are
also known to engage in reputation-building activities after restatements. Building on this
16
Again, we only present results for the first column, for the sake of brevity.
Electronic copy available at: https://ssrn.com/abstract=4207137
33
literature, we argue that managers would have strong incentives to improve the quality of
information conveyed by financial statements after fraud. Consequently, we expect there to be
improvements in financial statement comparability after fraud. Consistent with our
expectations, we find that comparability improves significantly four years after fraud. These
results are important, as they show that managers remedy their financial reporting policies after
fraud.
Electronic copy available at: https://ssrn.com/abstract=4207137
34
REFERENCES
Ali, S., Liu, B., & Su, J. J. (2022). Does corporate governance have a differential effect on downside
and upside risk? Journal of Business Finance & Accounting, Forthcoming.
https://doi.org/10.1111/jbfa.12606
Altman, E. I. (1968). Financial Ratios, Discriminant Analysis and the Prediction of Corporate
Bankruptcy. The Journal of Finance, 23(4), 589609.
Amiram, D., Bozanic, Z., Cox, J. D., Dupont, Q., Karpoff, J. M., & Sloan, R. (2018). Financial reporting
fraud and other forms of misconduct: A multidisciplinary review of the literature. Review of
Accounting Studies, 23(2), 732783.
Amiram, D., Bozanic, Z., & Rouen, E. (2015). Financial statement errors: Evidence from the
distributional properties of financial statement numbers. Review of Accounting Studies, 20(4),
15401593.
Anderson, R. C., Mansi, S. A., & Reeb, D. M. (2004). Board characteristics, accounting report integrity,
and the cost of debt. Journal of Accounting and Economics, 37(3), 315342.
Barth, M. E., Landsman, W. R., Lang, M., & Williams, C. (2012). Are IFRS-based and US GAAP-
based accounting amounts comparable? Journal of Accounting and Economics, 54(1), 6893.
Beasley, M. S., Carcello, J. V., Hermanson, D. R., & Lapides, P. D. (2000). Fraudulent financial
reporting: Consideration of industry traits and corporate governance mechanisms. Accounting
Horizons, 14(4), 441454.
Bratten, B., Payne, J. L., & Thomas, W. B. (2016). Earnings Management: Do Firms Play “Follow the
Leader”? Contemporary Accounting Research, 33(2), 616643.
Brazel, J. F., Jones, K. L., & Zimbelman, M. F. (2009). Using nonfinancial measures to assess fraud
risk. Journal of Accounting Research, 47(5), 11351166.
Brochet, F., Jagolinzer, A. D., & Riedl, E. J. (2013). Mandatory IFRS adoption and financial statement
comparability. Contemporary Accounting Research, 30(4), 13731400.
Electronic copy available at: https://ssrn.com/abstract=4207137
35
Brown, N. C., Crowley, R. M., & Elliott, W. B. (2020). What Are You Saying? Using topic to Detect
Financial Misreporting. Journal of Accounting Research, 58(1), 237291.
https://doi.org/10.1111/1475-679x.12294
Byard, D., Mashruwala, S., & Suh, J. (2017). Does the 20-F Reconciliation Affect Investors’ Perception
of Comparability Between Foreign Private Issuers (FPIs) and U.S. Firms? Accounting
Horizons, 31(2).
Campbell, J. L., & Yeung, P. E. (2017). Earnings comparability, accounting similarities, and stock
returns: Evidence from peer firms’ earnings restatements. Journal of Accounting, Auditing &
Finance, 32(4), 480509.
Cascino, S., & Gassen, J. (2015). What drives the comparability effect of mandatory IFRS adoption?
Review of Accounting Studies, 20(1), 242282.
Caylor, M. L., Chambers, D. J., & Mutlu, S. (2022). Financial reporting uniformity: Its relation to
comparability and its impact on financial statement users. Journal of Business Finance &
Accounting, Forthcoming. https://doi.org/10.1111/jbfa.12608
CFA Institute. (2007). Comment Letter to the SEC dated October 2, 2007. Re: File no. S7-13-07.
http://https://www.sec.gov/comments/s7-13-07/s71307-125.pdf
Chakravarthy, J., DeHaan, E., & Rajgopal, S. (2014). Reputation repair after a serious restatement. The
Accounting Review, 89(4), 13291363.
Chen, J. Z., M-H, C., Chin, C.-L., & Lobo, G. J. (2020). Do Firms That Have a Common Signing
Auditor Exhibit Higher Earnings Comparability? The Accounting Review, 95(3), 115143.
Chen, L. H., & Sami, H. (2008). Trading volume reaction to the earnings reconciliation from IAS to US
GAAP. Contemporary Accounting Research, 25(1), 1553.
Chen, S., Miao, B., & Shevlin, T. (2015). A New Measure of Disclosure Quality: The Level of
Disaggregation of Accounting Data in Annual Reports. Journal of Accounting Research, 53(5),
10171054. https://doi.org/10.1111/1475-679x.12094
Chen, S., Miao, B., & Valentine, K. (2022). Corporate Control Contests and the Asymmetric Disclosure
of Bad News: Evidence from Peer Firm Disclosure Response to Takeover Threat. The
Accounting Review, 97(1), 123146. https://doi.org/10.2308/tar-2018-0619
Electronic copy available at: https://ssrn.com/abstract=4207137
36
Cheng, J.-C., & Wu, R.-S. (2018). Internal capital market efficiency and the diversification discount:
The role of financial statement comparability. Journal of Business Finance & Accounting,
45(56), 572603.
Crutchley, C. E., Jensen, M. R. H., & Marshall, B. B. (2007). Climate for scandal: Corporate
environments that contribute to accounting fraud. Financial Review, 42(1), 5373.
De Franco, G., Kothari, S. P., & Verdi, R. S. (2011). The benefits of financial statement comparability.
Journal of Accounting Research, 49(4), 895931.
Dechow, P. M., Ge, W., Larson, C. R., & Sloan, R. G. (2011). Predicting Material Accounting
Misstatements. Contemporary Accounting Research, 28(1), 1782.
Dechow, P. M., & Skinner, D. J. (2000). Earnings management: Reconciling the views of accounting
academics, practitioners, and regulators. Accounting Horizons, 14(2), 235250.
Demsetz, H., & Lehn, K. (1985). The structure of corporate ownership: Causes and consequences.
Journal of Political Economy, 93(6), 11551177.
Dhole, S., Lobo, G. J., Mishra, S., & Pal, A. M. (2015). Effects of the SEC’s XBRL Mandate on the
Financial Reporting Comparability. International Journal of Accounting Information Systems,
19, 2944.
Ege, M., Glenn, J. L., & Robinson, J. R. (2019). Unexpected SEC Resource Constraints and Comment
Letter Quality. Contemporary Accounting Research, 37(1), 3367.
https://doi.org/10.1111/1911-3846.12505
Elnahas, A. M., Jain, P. K., & McInish, T. H. (2021). Mixed-signal stock splits. Journal of Business
Finance & Accounting. https://doi.org/10.1111/jbfa.12570
Endrawes, M., Feng, Z., Lu, M., & Shan, Y. (2020). Audit committee characteristics and financial
statement comparability. Accounting & Finance, 60(3), 23612395.
Erickson, M., Hanlon, M., & Maydew, E. L. (2006). Is there a link between executive equity incentives
and accounting fraud? Journal of Accounting Research, 44(1), 113143.
Farber, D. B. (2005). Restoring trust after fraud: Does corporate governance matter? The Accounting
Review, 80(2), 539561.
Electronic copy available at: https://ssrn.com/abstract=4207137
37
Feroz, E. H., Park, K. J., & Pastena, V. (1991). The financial and market effects of the SEC’s accounting
and auditing enforcement releases. Journal of Accounting Research, 29(Supplement), 107142.
Fich, E. M., & Shivdasani, A. (2007). Financial fraud, director reputation, and shareholder wealth.
Journal of Financial Economics, 86(2), 306336.
Fong, C., Hazlett, C., & Imai, K. (2018). Covariate balancing propensity score for a continuous
treatment: Application to the efficacy of political advertisements. The Annals of Applied
Statistics, 12(1), 156177.
Francis, J. R., Pinnuck, M. L., & Watanabe, O. (2014). Auditor style and financial statement
comparability. The Accounting Review, 89(2), 605633.
Fung, M. K. (2015). Cumulative prospect theory and managerial incentives for fraudulent financial
reporting. Contemporary Accounting Research, 32(1), 5575.
Glancy, F. H., & Yadav, S. B. (2011). A computational model for financial reporting fraud detection.
Decision Support Systems, 50(3), 595601. https://doi.org/10.1016/j.dss.2010.08.010
Graham, J. R., Li, S., & Qiu, J. (2008). Corporate misreporting and bank loan contracting. Journal of
Financial Economics, 89(1), 4461.
Hainmueller, J. (2012). Entropy Balancing for Causal Effects: A Multivariate Reweighting Method to
Produce Balanced Samples in Observational Studies. Political Analysis, 20(1), 2546.
https://doi.org/10.1093/pan/mpr025
Harris, M. S., & Muller, K. A. (1999). The market valuation of IAS versus US-GAAP accounting
measures using Form 20-F reconciliations. Journal of Accounting and Economics, 26(1), 285
312.
Jackson, A. B., Rountree, B. R., & Sivaramakrishnan, K. (2017). Earnings co-movements and earnings
manipulation. Review of Accounting Studies, 22(3), 13401365.
Johnson, S. A., Ryan, H. E., & Tian, Y. S. (2009). Managerial incentives and corporate fraud: The
sources of incentives matter. Review of Finance, 13(1), 115145.
Kang, T., Krishnan, G. V., Wolfe, M. C., & Yi, H. S. (2012). The impact of eliminating the 20-F
reconciliation requirement for IFRS filers on earnings persistence and information uncertainty.
Accounting Horizons, 26(4), 741765.
Electronic copy available at: https://ssrn.com/abstract=4207137
38
Karpoff, J. M., Lee, D. S., & Martin, G. S. (2008a). The consequences to managers for financial
misrepresentation. Journal of Financial Economics, 88(2), 193215.
Karpoff, J. M., Lee, D. S., & Martin, G. S. (2008b). The cost to firms of cooking the books. Journal of
Financial and Quantitative Analysis, 43(03), 581611.
Kedia, S., Koh, K., & Rajgopal, S. (2015). Evidence on contagion in earnings management. The
Accounting Review, 90(6), 23372373.
Kim, S., Kraft, P., & Ryan, S. G. (2013). Financial statement comparability and credit risk. Review of
Accounting Studies, 18(3), 783823.
King, G., & Nielsen, R. (2019). Why Propensity Scores Should Not Be Used for Matching. Political
Analysis, 27(4), 435454. https://doi.org/10.1017/pan.2019.11
Koch, A. S. (2002). Financial Distress and the Credibility of Management Earnings Forecasts. GSIA
Working Paper No. 2000-10. Available at SSRN: Https://Ssrn.Com/Abstract=415580 or
Http://Dx.Doi.Org/10.2139/Ssrn.415580.
Kothari, S. P. (2001). Capital markets research in accounting. Journal of Accounting and Economics,
31(1), 105231.
Kothari, S. P., Shu, S., & Wysocki, P. D. (2009). Do managers withhold bad news? Journal of
Accounting Research, 47(1), 241276.
Lennox, C., & Pittman, J. A. (2010). Big Five audits and accounting fraud. Contemporary Accounting
Research, 27(1), 209247.
Maksimovic, V., & Titman, S. (1991). Financial policy and reputation for product quality. The Review
of Financial Studies, 4(1), 175200.
McMullin, J. L., & Schonberger, B. (2022). When good balance goes bad: A discussion of common
pitfalls when using entropy balancing. Journal of Financial Reporting, Forthcoming.
McNichols, M. F. (2001). Research design issues in earnings management studies. Journal of
Accounting and Public Policy, 19(4), 313345.
Ndofor, H. A., Wesley, C., & Priem, R. L. (2015). Providing CEOs with opportunities to cheat the
effects of complexity-based information asymmetries on financial reporting fraud. Journal of
Management, 41(6), 17741797.
Electronic copy available at: https://ssrn.com/abstract=4207137
39
Penman, S. H. (2013). Financial Statement Analysis and Security Valuation (5th ed.). McGraw-Hill.
Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational
studies for causal effects. Biometrika, 70(1), 4155.
Sohn, B. C. (2016). The effect of accounting comparability on the accrual-based and real earnings
management. Journal of Accounting and Public Policy, 35(5), 513539.
Electronic copy available at: https://ssrn.com/abstract=4207137
40
Appendix A: Variable Definitions
Variable
Dependent Variable:
FRAUD
Independent Variable:
Comparability
PostXFRAUD
Other Variables:
Volatility
Size
Lev
BTM
ROA
SalesGrowth
EarningsPrice
AltmanZ
BoardSize
Electronic copy available at: https://ssrn.com/abstract=4207137
41
Big4
M&A Indicator
Financing
REC
FOREIGN
Electronic copy available at: https://ssrn.com/abstract=4207137
42
Figure 1. Changes in Comparability over Time
The graphs above plot the change in financial statement comparability (CompAcctM4) around the identification
of fraud (0 on the horizontal axis). Separate plots are shown from FRAUD firms (in red) and Non-FRAUD firms
(in blue). The graphs are based on a propensity score matched sample which matches firms based on their
accounting comparability. The sample period is 2000-2018.
-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
-4 -3 -2 -1 0 1 2 3 4
Accounting Comparability Over Time
No Fraud Fraud
Electronic copy available at: https://ssrn.com/abstract=4207137
43
Table 1
Descriptive Statistics
N
p25
Mean
Median
p75
St.Dev
Fraud
8,996
0.000
0.019
0.000
0.000
0.136
CompAcctM4
8,996
-0.500
-0.519
-0.210
-0.090
0.932
CompAcctM10
8,954
-0.760
-0.743
-0.320
-0.150
1.229
CompAcctInd
5,254
-4.070
-3.541
-3.010
-2.190
2.195
CompAcctIndMed
5,247
-2.970
-2.604
-1.780
-1.110
2.508
ECompM4
7,289
0.301
0.458
0.466
0.624
0.225
DQ
7,776
0.706
0.753
0.767
0.807
0.084
Volatility
8,996
0.021
0.032
0.028
0.039
0.016
Size
8,996
4.805
6.248
6.186
7.613
2.003
Lev
8,996
0.000
0.166
0.120
0.280
0.176
BTM
8,996
0.270
0.557
0.452
0.709
0.467
ROA
8,996
-0.019
0.015
0.040
0.090
0.173
SalesGrowth
8,996
-0.022
0.150
0.082
0.210
0.566
EarningsPrice
8,996
-0.019
-0.030
0.034
0.059
0.297
AltmanZ
8,996
1.716
4.784
3.445
5.969
7.85
BoardSize
8,996
2.079
2.177
2.197
2.303
0.244
Big4
8,996
1.000
0.784
1.000
1.000
0.411
M&AIndicator
8,996
0.000
0.427
0.000
1.000
0.495
Financing
8,996
0.000
0.098
0.000
0.000
0.298
The Table above is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a
dummy variable equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically
related to fraud, and zero otherwise. Fraud is defined as one in the last year of the AAER. CompAcctM4,
CompAcctM10, CompAcctInd, and CompAcctIndMed are four proxies of financial statement comparability, based
on De Franco et al. (2011); ECompM4 measures earnings covariance following Francis et al. (2014) and De Franco
et al. (2011); DQ measures disclosure quality, calculated following the approach of Chen et al. (2015).. Volatility
is the annual stock return return volatility, calculated as the standard deviation of daily stock returns, measured
over the fiscal year; Size is the natural logarithm of total assets; Lev is the ratio of total debt (long-term and short-
term) to total assets; BTM is the book value of shareholder’s equity, divided by the market value of equity; ROA
is net income divided by year-end assets; SalesGrowth is the annual percentage change in sales; EarningsPrice is
the ratio of earnings price and share price at the start of the year; AltmanZ is the Altman (1968) Z-Score. It is is
defined as: Z=1.2 (Working Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3 (Earnings before
Interest and Taxes/Total Assets) +0.6 (Market Value of Equity/Book Value of Liabilities) +1.0(Sales/Total
Assets). BoardSize is the size of the board, defined as the number of directors on the company’s board; Big4 is a
dummy variable equal to 1, if the firm is audited by a Big 4 auditor; zero otherwise (it is also coded as 1 for Arthur
Anderson clients for 2000 and 2001); M&AIndicator is a dummy variable equal to 1, if the firm has made an
acquisition during the year; zero otherwise; Financing is a dummy variable set equal to 1 if the firm’s variable
FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated as: (Cash from Operationst - Average
Capital Expenditurest−3 to t−1)/Current Assetst−1. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
44
Table 2
Descriptive Statistics Comparison Fraud vs Non-Fraud Firms
Fraud
Non-Fraud
N
Mean
Median
N
Mean
Median
t-test
Wilcoxon
CompAcctM4
158
-0.578
-0.250
8,838
-0.454
-0.230
(0.058)
(0.548)
CompAcctM10
158
-0.791
-0.375
8,796
-0.652
-0.360
(0.072)
(0.679)
CompAcctInd
158
-3.615
-3.125
5,096
-3.445
-3.090
(0.082)
(0.415)
CompAcctIndMed
158
-2.572
-1.795
5,089
-2.485
-1.870
(0.615)
(0.621)
ECompM4
112
0.437
0.431
7,137
0.458
0.466
(0.317)
(0.338)
DQ
133
0.722
0.744
7,643
0.754
0.767
(0.000)
(0.000)
Volatility
158
0.032
0.029
8,838
0.032
0.029
(0.451)
(0.505)
Size
158
6.838
6.708
8,838
6.070
5.924
(0.000)
(0.000)
Lev
158
0.171
0.155
8,838
0.157
0.105
(0.296)
(0.064)
BTM
158
0.562
0.496
8,838
0.541
0.472
(0.487)
(0.546)
ROA
158
0.043
0.032
8,838
0.017
0.037
(0.039)
(0.895)
SalesGrowth
158
0.152
0.081
8,838
0.161
0.082
(0.843)
(0.745)
EarningsPrice
158
0.002
0.024
8,838
0.000
0.033
(0.851)
(0.235)
AltmanZ
158
4.095
3.122
8,838
4.696
3.460
(0.297)
(0.477)
BoardSize
158
2.204
2.197
8,838
2.162
2.197
(0.032)
(0.049)
Big4
158
0.875
1.000
8,838
0.775
1.000
(0.003)
(0.003)
M&AIndicator
158
0.566
1.000
8,838
0.418
0.000
(0.002)
(0.000)
Financing
158
0.057
0.000
8,838
0.098
0.000
(0.000)
(0.000)
The Table above is based on a propensity score-matched sample of observations from 2000-2018. CompAcctM4,
CompAcctM10, CompAcctInd, and CompAcctIndMed are four proxies of financial statement comparability, based
on De Franco et al. (2011); ECompM4 measures earnings covariance following Francis et al. (2014) and De Franco
et al. (2011); DQ measures disclosure quality, calculated following the approach of Chen et al. (2015). Volatility
is the annual stock return return volatility, calculated as the standard deviation of daily stock returns, measured
over the fiscal year; Size is the natural logarithm of total assets; Lev is the ratio of total debt (long-term and short-
term) to total assets; BTM is the book value of shareholder’s equity, divided by the market value of equity; ROA
is net income divided by year-end assets; SalesGrowth is the annual percentage change in sales; EarningsPrice is
the ratio of earnings price and share price at the start of the year; AltmanZ is the Altman (1968) Z-Score. It is is
defined as: Z=1.2 (Working Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3 (Earnings before
Interest and Taxes/Total Assets) +0.6 (Market Value of Equity/Book Value of Liabilities) +1.0(Sales/Total
Assets). BoardSize is the size of the board, defined as the number of directors on the company’s board; Big4 is a
dummy variable equal to 1, if the firm is audited by a Big 4 auditor; zero otherwise (it is also coded as 1 for Arthur
Anderson clients for 2000 and 2001); M&AIndicator is a dummy variable equal to 1, if the firm has made an
acquisition during the year; zero otherwise; Financing is a dummy variable set equal to 1 if the firm’s variable
FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated as: (Cash from Operationst - Average
Capital Expenditurest−3 to t−1)/Current Assetst−1. We also present the variable definitions in Appendix A. The
sample is split into fraud and non-fraud subsamples.
Electronic copy available at: https://ssrn.com/abstract=4207137
45
Table 3
Fraud Cases by Year and Industry
Panel A: By Year
Year
Fraud
2000
11
2001
13
2002
17
2003
19
2004
20
2005
8
2006
10
2007
11
2008
7
2009
7
2010
6
2011
6
2012
10
2013
5
2014
2
2015
3
2016
3
Total
158
The Table above is based on a propensity score-matched sample of observations from 2000-2018. The Table
presents the number of fraud cases by year. Fraud is a dummy variable equal to 1 if the firm is the subject of a
US Securities and Exchange (SEC) AAER specifically related to fraud, and zero otherwise. Fraud is defined as
one in the last year of the AAER.
Electronic copy available at: https://ssrn.com/abstract=4207137
46
Panel B: By Industry
Two-digit SIC Code
Industry
Fraud
13
Oil and Gas Extraction
3
20
Food and Kindred Products
10
23
Apparel and Other Textile Products
1
33
Primary Metal Industries
2
34
Fabricated Metal Products
1
35
Industrial Machinery and Equipment
16
36
Electronic and Other Electric Equipment
20
37
Transportation Equipment
6
38
Instruments and Related Products
10
44
Water Transportation
2
48
Communications
3
49
Electric, Gas, and Sanitary Services
5
50
Wholesale Trade Durable Goods
3
51
Wholesale Trade Nondurable Goods
6
55
Automotive Dealers and Service Stations
1
56
Apparel and Accessory Stores
2
59
Miscellaneous Retail
9
62
Security and Commodity Brokers
3
64
Insurance Agents, Brokers, and Service
3
73
Business Services
52
Total
158
The Table above is based on a propensity score-matched sample of observations from 2000-2018. The Table
presents the number of fraud cases by industry (measured by the two-digit SIC code). Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one in the last year of the AAER.
Electronic copy available at: https://ssrn.com/abstract=4207137
47
Table 4
Financial Statement Comparability and Accounting Fraud
CompAcctM4
ECompM4
DQ
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Comparability
-0.466*
-2.640*
-4.086**
(0.051)
(0.096)
(0.018)
Volatility
-19.521
-5.397
-4.121
(0.218)
(0.557)
(0.645)
Size
0.722***
0.384***
0.288***
(0.000)
(0.000)
(0.000)
Lev
-1.526
0.394
-0.622
(0.229)
(0.568)
(0.367)
BTM
0.475
0.122
0.303
(0.331)
(0.691)
(0.263)
ROA
1.913
1.638
0.698
(0.241)
(0.120)
(0.457)
SalesGrowth
0.158
0.174
0.125
(0.601)
(0.406)
(0.569)
EarningsPrice
-1.539
-0.647
-0.340
(0.140)
(0.306)
(0.562)
AltmanZ
0.007
-0.002
-0.026
(0.841)
(0.933)
(0.165)
BoardSize
-0.836
-0.621
-0.248
(0.394)
(0.257)
(0.615)
Big4
-0.635
-0.554*
-0.037
(0.267)
(0.084)
(0.904)
M&AIndicator
0.426
0.324
0.071
(0.185)
(0.138)
(0.713)
Financing
0.000
0.000
0.000
(0.992)
(0.992)
(0.961)
Firm Fixed Effects
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Constant
-16.650***
-7.202***
-4.662**
(0.000)
(0.000)
(0.022)
Observations
8,996
7,289
7,776
Wald
58.200
90.690
79.620
(p-value)
(0.001)
(0.000)
(0.000)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below.
󰇛 󰇜
     
    
  


Electronic copy available at: https://ssrn.com/abstract=4207137
48
It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one in the last year of the AAER. CompAcctM4 is a proxy of financial
statement comparability, based on De Franco et al. (2011); ECompM4 measures earnings covariance following
Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated following the
approach of Chen et al. (2015). Volatility is the annual stock return return volatility, calculated as the standard
deviation of daily stock returns, measured over the fiscal year; Size is the natural logarithm of total assets; Lev is
the ratio of total debt (long-term and short-term) to total assets; BTM is the book value of shareholder’s equity,
divided by the market value of equity; ROA is net income divided by year-end assets; SalesGrowth is the annual
percentage change in sales; EarningsPrice is the ratio of earnings price and share price at the start of the year;
AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working Capital/Total Assets) +1.4 (Retained
Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total Assets) +0.6 (Market Value of Equity/Book
Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size of the board, defined as the number of
directors on the company’s board; Big4 is a dummy variable equal to 1, if the firm is audited by a Big 4 auditor;
zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000 and 2001); M&AIndicator is a dummy
variable equal to 1, if the firm has made an acquisition during the year; zero otherwise; Financing is a dummy
variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated
as: (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current Assetst−1. We include firm and year
fixed effects in the model. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
49
Table 5
Financial Statement Comparability and Accounting Fraud
CompAcctM4
ECompM4
DQ
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Comparability
-0.300*
-1.405**
-2.025
(0.056)
(0.021)
(0.304)
Volatility
-35.600***
-29.650**
-35.080***
(0.000)
(0.002)
(0.000)
Size
0.171*
0.185*
0.128
(0.095)
(0.097)
(0.249)
Lev
-0.855
-0.261
-0.955
(0.321)
(0.758)
(0.290)
BTM
0.852**
0.762**
0.786**
(0.003)
(0.020)
(0.007)
ROA
0.565
0.085
-0.148
(0.553)
(0.919)
(0.836)
SalesGrowth
-0.124
-0.128
-0.187
(0.282)
(0.326)
(0.145)
EarningsPrice
-0.272
-0.261
-0.163
(0.508)
(0.524)
(0.661)
AltmanZ
-0.019
-0.009
-0.023
(0.360)
(0.593)
(0.313)
BoardSize
-1.129
-1.257
-0.798
(0.122)
(0.121)
(0.279)
Big4
-0.585
-1.050**
-0.636
(0.185)
(0.021)
(0.164)
M&AIndicator
0.201
0.464**
0.176
(0.285)
(0.035)
(0.392)
Financing
0.000
0.000
0.000
(0.875)
(0.921)
(0.896)
Firm Fixed Effects
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Constant
0.937
2.153
2.081
(0.558)
(0.213)
(0.354)
Observations
18,069
14,770
15,392
Wald
104.700
119.400
98.610
(p-value)
(0.000)
(0.000)
(0.000)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below.
󰇛 󰇜
     
    
  


It is based on an entropy balanced (based on two moments) sample of observations from 2000-2018. Fraud is a
Electronic copy available at: https://ssrn.com/abstract=4207137
50
dummy variable equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically
related to fraud, and zero otherwise. Fraud is defined as one in the last year of the AAER. CompAcctM4 is a proxy
of financial statement comparability, based on De Franco et al. (2011); ECompM4 measures earnings covariance
following Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated following
the approach of Chen et al. (2015). Volatility is the annual stock return return volatility, calculated as the standard
deviation of daily stock returns, measured over the fiscal year; Size is the natural logarithm of total assets; Lev is
the ratio of total debt (long-term and short-term) to total assets; BTM is the book value of shareholder’s equity,
divided by the market value of equity; ROA is net income divided by year-end assets; SalesGrowth is the annual
percentage change in sales; EarningsPrice is the ratio of earnings price and share price at the start of the year;
AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working Capital/Total Assets) +1.4 (Retained
Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total Assets) +0.6 (Market Value of Equity/Book
Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size of the board, defined as the number of
directors on the company’s board; Big4 is a dummy variable equal to 1, if the firm is audited by a Big 4 auditor;
zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000 and 2001); M&AIndicator is a dummy
variable equal to 1, if the firm has made an acquisition during the year; zero otherwise; Financing is a dummy
variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated
as: (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current Assetst−1. We include firm and year
fixed effects in the model. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
51
Table 6
Financial Statement Comparability and Fraud in the Years before Fraud
Panel A: Comparability defined as CompAcctM4
Comparability in year
t-1
t-2
t-3
t-4
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Comparability
-0.368*
-0.068
-0.211
-0.149
(0.091)
(0.729)
(0.130)
(0.191)
Volatility
-18.460
-21.830
-13.970
-19.280
(0.257)
(0.207)
(0.440)
(0.334)
Size
0.659***
0.698***
0.755***
0.848***
(0.000)
(0.000)
(0.000)
(0.000)
Lev
-1.477
-1.209
-1.189
-1.531
(0.256)
(0.377)
(0.406)
(0.326)
BTM
0.407
0.357
0.446
0.225
(0.415)
(0.511)
(0.435)
(0.718)
ROA
1.678
1.170
0.508
0.093
(0.311)
(0.529)
(0.808)
(0.967)
SalesGrowth
0.124
0.102
0.131
0.158
(0.707)
(0.776)
(0.712)
(0.703)
EarningsPrice
-1.697*
-1.433
-1.436
-1.630
(0.088)
(0.284)
(0.323)
(0.258)
AltmanZ
0.0041
0.0022
0.014
-0.029
(0.915)
(0.959)
(0.755)
(0.607)
BoardSize
-0.715
-0.720
-0.401
-1.241
(0.485)
(0.500)
(0.720)
(0.300)
Big4
-0.557
-0.753
-0.922
-1.029
(0.344)
(0.224)
(0.165)
(0.141)
M&AIndicator
0.402
0.373
0.307
0.614
(0.663)
(0.492)
(0.681)
(0.418)
Financing
0.000
0.000
0.000
0.000
(0.923)
(0.939)
(0.749)
(0.868)
Firm Fixed Effects
Yes
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Yes
Constant
-9.527**
-9.359**
-10.711**
-8.996**
(0.005)
(0.007)
(0.003)
(0.014)
Observations
8,364
7,869
7,367
6,824
Wald
55.670
53.130
50.860
48.450
(p-value)
(0.000)
(0.002)
(0.004)
(0.001)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below.
Electronic copy available at: https://ssrn.com/abstract=4207137
52
󰇛 󰇜
    
    
   


It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one in the last year of the AAER. CompAcctM4 is a proxy of financial
statement comparability, based on De Franco et al. (2011); Volatility is the annual stock return return volatility,
calculated as the standard deviation of daily stock returns, measured over the fiscal year; Size is the natural
logarithm of total assets; Lev is the ratio of total debt (long-term and short-term) to total assets; BTM is the book
value of shareholder’s equity, divided by the market value of equity; ROA is net income divided by year-end
assets; SalesGrowth is the annual percentage change in sales; EarningsPrice is the ratio of earnings price and
share price at the start of the year; AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working
Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total
Assets) +0.6 (Market Value of Equity/Book Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size
of the board, defined as the number of directors on the company’s board; Big4 is a dummy variable equal to 1, if
the firm is audited by a Big 4 auditor; zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000
and 2001); M&AIndicator is a dummy variable equal to 1, if the firm has made an acquisition during the year;
zero otherwise; Financing is a dummy variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and
zero otherwise. FreeCash is calculated as: (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current
Assetst−1. We include firm and year fixed effects in the model. We also present the variable definitions in Appendix
A.
Electronic copy available at: https://ssrn.com/abstract=4207137
53
Panel B: Other Proxies of Comparability
Comparability
Measure
EcompM4
DQ
Comparabilityt-1
-1.087
-1.937*
(0.194)
(0.086)
Comparabilityt-2
-0.372
-0.983
(0.496)
(0.267)
Comparabilityt-3
-0.219
-0.638
(0.628)
(0.487)
Comparabilityt-4
-0.021
-0.372
(0.793)
(0.623)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the coefficient of Comparability of the model below.
󰇛 󰇜
    
    
   


It is based on a propensity score-matched sample of observations from 2000-2018. ECompM4 measures earnings
covariance following Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated
following the approach of Chen et al. (2015).
Electronic copy available at: https://ssrn.com/abstract=4207137
54
Table 7
Financial Statement Comparability after Fraud
Panel A: Comparability defined as CompAcctM4
CompAcctM4
CompAcctM4
CompAcctM4
CompAcctM4
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Post1FRAUD
-0.087
(0.926)
Post2FRAUD
-0.002
(0.832)
Post3FRAUD
0.182
(0.174)
Post4FRAUD
0.328**
(0.021)
Volatility
-10.852***
-10.860***
-10.858***
-10.864***
(0.000)
(0.000)
(0.000)
(0.000)
Size
0.182***
0.183***
0.180***
0.179***
(0.000)
(0.000)
(0.000)
(0.000)
Lev
-0.396***
-0.398***
-0.389***
-0.401***
(0.000)
(0.000)
(0.000)
(0.000)
BTM
-0.056*
-0.057*
-0.053*
-0.052*
(0.061)
(0.059)
(0.066)
(0.071)
ROA
0.012
0.014
0.018
0.013
(0.812)
(0.801)
(0.782)
(0.808)
SalesGrowth
-0.008
-0.007
-0.009
-0.009
(0.347)
(0.362)
(0.321)
(0.319)
EarningsPrice
0.459***
0.461***
0.451***
0.462***
(0.000)
(0.000)
(0.000)
(0.000)
AltmanZ
-0.001
-0.001
-0.001
-0.001
(0.689)
(0.688)
(0.687)
(0.691)
BoardSize
-0.073
-0.071
-0.072
-0.073
(0.266)
(0.287)
(0.281)
(0.269)
Big4
-0.051
-0.054
-0.059
-0.052
(0.203)
(0.197)
(0.184)
(0.201)
M&AIndicator
0.079***
0.081***
0.086***
0.083***
(0.000)
(0.000)
(0.000)
(0.000)
Financing
0.061
0.063
0.066
0.062
(0.214)
(0.201)
(0.194)
(0.210)
Firm Fixed Effects
Yes
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Yes
Constant
-0.913**
-0.984**
-0.869**
-0.938**
(0.009)
(0.006)
(0.014)
(0.007)
Observations
8,996
8,996
8,996
8,996
Adjusted R2
0.100
0.113
0.090
0.093
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the model below.
Electronic copy available at: https://ssrn.com/abstract=4207137
55
     
    
  


It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one in the last year of the AAER. CompAcctM4 is a proxy of financial
statement comparability, based on De Franco et al. (2011); Volatility is the annual stock return return volatility,
calculated as the standard deviation of daily stock returns, measured over the fiscal year; Size is the natural
logarithm of total assets; Lev is the ratio of total debt (long-term and short-term) to total assets; BTM is the book
value of shareholder’s equity, divided by the market value of equity; ROA is net income divided by year-end
assets; SalesGrowth is the annual percentage change in sales; EarningsPrice is the ratio of earnings price and
share price at the start of the year; AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working
Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total
Assets) +0.6 (Market Value of Equity/Book Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size
of the board, defined as the number of directors on the company’s board; Big4 is a dummy variable equal to 1, if
the firm is audited by a Big 4 auditor; zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000
and 2001); M&AIndicator is a dummy variable equal to 1, if the firm has made an acquisition during the year;
zero otherwise; Financing is a dummy variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and
zero otherwise. FreeCash is calculated as: (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current
Assetst−1. We include firm and year fixed effects in the model. We also present the variable definitions in Appendix
A.
Electronic copy available at: https://ssrn.com/abstract=4207137
56
Panel B: Other Proxies of Comparability
Comparability
Measure
ECompM4
DQ
Post1FRAUD
-0.168
-1.964
(0.756)
(0.597)
Post2FRAUD
-0.067
-0.892
(0.834)
(0.587)
Post3FRAUD
0.027
0.002
(0.327)
(0.765)
Post4FRAUD
0.896*
1.867*
(0.085)
(0.056)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the coefficient of Comparability of the model below.
     
    
  


It is based on a propensity score-matched sample of observations from 2000-2018. ECompM4 measures earnings
covariance following Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated
following the approach of Chen et al. (2015).
Electronic copy available at: https://ssrn.com/abstract=4207137
57
Table 8
Financial Distress Financial Statement Comparability and Fraud
Panel A: Comparability defined as CompAcctM4
Financial Distress
High
Low
Variable
Coef.
(p-value)
Coef.
(p-value)
Comparability
-1.864**
-0.312
(0.032)
(0.432)
Volatility
-42.276
-23.285
(0.529)
(0.320)
Size
2.321***
0.526**
(0.000)
(0.014)
Lev
-25.271***
1.898
(0.000)
(0.178)
BTM
-0.069
0.734
(0.982)
(0.328)
ROA
12.187
1.782
(0.187)
(0.287)
SalesGrowth
-1.692
0.392
(0.198)
(0.322)
EarningsPrice
-4.016
-3.726
(0.236)
(0.163)
AltmanZ
-0.085
0.023
(0.523)
(0.521)
BoardSize
-1.899
-0.426
(0.598)
(0.721)
Big4
1.011
-0.696
(0.596)
(0.314)
M&AIndicator
0.598
0.711*
(0.602)
(0.089)
Financing
0.000
0.000
(0.997)
(0.998)
Firm Fixed Effects
Yes
Yes
Year Fixed Effects
Yes
Yes
Constant
-26.498***
-15.997***
(0.001)
(0.000)
Observations
1,902
7,094
Wald
50.321
45.896
(p-value)
(0.000)
(0.006)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below. The model is estimated for the high
financial distress (Altman Z < 1.81) and low financial distress (Altman Z > 1.81) samples.
Electronic copy available at: https://ssrn.com/abstract=4207137
58
󰇛 󰇜
     
    
  


It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one in the last year of the AAER. CompAcctM4 is a proxy of financial
statement comparability, based on De Franco et al. (2011); Volatility is the annual stock return return volatility,
calculated as the standard deviation of daily stock returns, measured over the fiscal year; Size is the natural
logarithm of total assets; Lev is the ratio of total debt (long-term and short-term) to total assets; BTM is the book
value of shareholder’s equity, divided by the market value of equity; ROA is net income divided by year-end
assets; SalesGrowth is the annual percentage change in sales; EarningsPrice is the ratio of earnings price and
share price at the start of the year; AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working
Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total
Assets) +0.6 (Market Value of Equity/Book Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size
of the board, defined as the number of directors on the company’s board; Big4 is a dummy variable equal to 1, if
the firm is audited by a Big 4 auditor; zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000
and 2001); M&AIndicator is a dummy variable equal to 1, if the firm has made an acquisition during the year;
zero otherwise; Financing is a dummy variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and
zero otherwise. FreeCash is calculated as: (Cash from Operationst - Average Capital Expenditurest−3 to
t−1)/Current Assetst−1. We include firm and year fixed effects in the model. We also present the variable definitions
in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
59
Panel B: Other Proxies of Comparability
Financial Distress
Comparability
Measures
High
Low
ECompM4
-3.908**
-0.072
(0.012)
(0.734)
DQ
-4.743**
-1.002
(0.011)
(0.401)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show the coefficients of Comparability from the estimation of the model below. The model is
estimated for the high financial distress (Altman Z < 1.81) and low financial distress (Altman Z > 1.81) samples.
󰇛 󰇜
     
    
  


It is based on a propensity score-matched sample of observations from 2000-2018. ECompM4 measures earnings
covariance following Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated
following the approach of Chen et al. (2015).
Electronic copy available at: https://ssrn.com/abstract=4207137
60
Table 9
Corporate Governance Financial Statement Comparability and Fraud
Panel A: Comparability defined as CompAcctM4
Board Size
Board Independence
Low
High
Low
High
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Comparability
-1.278**
-0.397
-1.692**
0.032
(0.013)
(0.129)
(0.011)
(0.914)
Volatility
18.984
-9.387
8.693
-4.321
(0.395)
(0.711)
(0.712)
(0.899)
Size
1.745***
0.621**
1.212***
0.632**
(0.000)
(0.015)
(0.000)
(0.013)
Lev
-3.632
-1.265
-1.214
-3.277
(0.123)
(0.548)
(0.654)
(0.117)
BTM
-0.044
1.479**
-0.856
1.217*
(0.964)
(0.036)
(0.423)
(0.064)
ROA
1.336
5.989*
0.957
1.689
(0.618)
(0.073)
(0.683)
(0.527)
SalesGrowth
-0.647
-0.013
0.227
-0.327
(0.328)
(0.953)
(0.493)
(0.638)
EarningsPrice
-1.818
-1.428
-1.538
-1.466
(0.327)
(0.364)
(0.534)
(0.421)
AltmanZ
0.069
-0.056
0.048
-0.016
(0.409)
(0.521)
(0.325)
(0.811)
BoardSize
-0.299
1.657
-0.553
-1.856
(0.923)
(0.532)
(0.716)
(0.327)
Big4
-1.287
-0.798
0.889
-1.212
(0.121)
(0.427)
(0.328)
(0.102)
M&AIndicator
1.786**
1.012*
1.326**
0.922**
(0.008)
(0.074)
(0.018)
(0.038)
Financing
0.000
0.000
0.000
0.000
(0.998)
(0.997)
(0.956)
(0.986)
Firm Fixed Effects
Yes
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Yes
Constant
-23.427***
-21.957***
-22.488***
-20.337***
(0.000)
(0.000)
(0.000)
(0.000)
Observations
4,201
4,795
3,660
5,336
Wald
55.232
49.498
59.995
53.557
(p-value)
(0.001)
(0.006)
(0.000)
(0.005)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below. The model is estimated for the strong and
weak corporate governance samples, measured by Board Size (Columns 1 and 2) and Board Independence
(Columns 3 and 4).
Electronic copy available at: https://ssrn.com/abstract=4207137
61
󰇛 󰇜
     
    
  


It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one in the last year of the AAER. CompAcctM4 is a proxy of financial
statement comparability, based on De Franco et al. (2011); Volatility is the annual stock return return volatility,
calculated as the standard deviation of daily stock returns, measured over the fiscal year; Size is the natural
logarithm of total assets; Lev is the ratio of total debt (long-term and short-term) to total assets; BTM is the book
value of shareholder’s equity, divided by the market value of equity; ROA is net income divided by year-end
assets; SalesGrowth is the annual percentage change in sales; EarningsPrice is the ratio of earnings price and
share price at the start of the year; AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working
Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total
Assets) +0.6 (Market Value of Equity/Book Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size
of the board, defined as the number of directors on the company’s board; Big4 is a dummy variable equal to 1, if
the firm is audited by a Big 4 auditor; zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000
and 2001); M&AIndicator is a dummy variable equal to 1, if the firm has made an acquisition during the year;
zero otherwise; Financing is a dummy variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and
zero otherwise. FreeCash is calculated as: (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current
Assetst−1. We include firm and year fixed effects in the model. We also present the variable definitions in Appendix
A.
Electronic copy available at: https://ssrn.com/abstract=4207137
62
Panel B: Other Proxies of Comparability
Board Size
Board Independence
Low
High
Low
High
ECompM4
-3.983**
0.164
-3.398**
0.169
(0.011)
(0.599)
(0.016)
(0.526)
DQ
-5.929**
0.253
-5.018**
0.217
(0.013)
(0.599)
(0.014)
(0.596)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show the coefficient of comparability from the estimation of the model below. The model is
estimated for the strong and weak corporate governance samples, measured by Board Size (Columns 1 and 2)
and Board Independence (Columns 3 and 4).
󰇛 󰇜
     
    
  


It is based on a propensity score-matched sample of observations from 2000-2018. ECompM4 measures earnings
covariance following Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated
following the approach of Chen et al. (2015).
Electronic copy available at: https://ssrn.com/abstract=4207137
63
Table 10
Financial Statement Characteristics, Comparability and Fraud 20-F Reconciliation
CompAcctM4
ECompM4
DQ
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Comparability
-0.103
-0.218**
-0.281**
(0.115)
(0.047)
(0.039)
REC
0.624**
0.734**
0.768**
(0.019)
(0.008)
(0.006)
FOREIGN
1.668**
1.848**
1.925**
(0.027)
(0.006)
(0.005)
REC*FOREIGN
-1.403
-1.287
-1.297
(0.112)
(0.174)
(0.171)
REC*Comparability
0.073
0.086
0.083
(0.250)
(0.203)
(0.204)
FOREIGN*Comparability
0.489***
0.709***
0.726***
(0.000)
(0.000)
(0.000)
REC* FOREIGN *Comparability
-0.827***
-2.745***
-3.958***
(0.000)
(0.000)
(0.000)
Volatility
-16.112
-19.428
-19.355
(0.796)
(0.884)
(0.884)
Size
0.749***
0.413***
0.413***
(0.000)
(0.000)
(0.000)
Lev
-1.990*
-1.287**
-1.285**
(0.092)
(0.037)
(0.037)
BTM
0.172
-0.059
-0.062
(0.490)
(0.836)
(0.827)
ROA
1.277**
0.579
0.567
(0.011)
(0.391)
(0.401)
SalesGrowth
0.017
0.002
0.002
(0.815)
(0.953)
(0.956)
EarningsPrice
-0.611
-0.460
-0.461
(0.157)
(0.403)
(0.398)
AltmanZ
-0.023
-0.017
-0.017
(0.177)
(0.338)
(0.334)
BoardSize
-0.566
-1.454**
-1.450**
(0.192)
(0.002)
(0.002)
Big4
-0.411
-0.032
-0.033
(0.886)
(0.920)
(0.919)
M&AIndicator
0.366**
0.626***
0.624***
(0.034)
(0.001)
(0.001)
Financing
0.000
0.000
0.000
(0.993)
(0.993)
(0.994)
Firm Fixed Effects
Yes
Yes
Yes
Constant
-6.173***
-5.261***
-5.297***
(0.000)
(0.000)
(0.000)
Observations
8,996
7,289
7,776
Wald
59.180
92.745
81.023
(p-value)
(0.000)
(0.000)
(0.000)
Electronic copy available at: https://ssrn.com/abstract=4207137
64
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below.
󰇛 󰇜  
 
  
   
   


It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one in the last year of the AAER. CompAcctM4 is a proxy of financial
statement comparability, based on De Franco et al. (2011); ECompM4 measures earnings covariance following
Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated following the
approach of Chen et al. (2015). REC is a dummy variable equal to 1 for years on or after 2008; 0 otherwise;
FOREIGN is a dummy variable equal to 1 if the firm is a foreign filer; zero otherwise; Volatility is the annual
stock return return volatility, calculated as the standard deviation of daily stock returns, measured over the fiscal
year; Size is the natural logarithm of total assets; Lev is the ratio of total debt (long-term and short-term) to total
assets; BTM is the book value of shareholder’s equity, divided by the market value of equity; ROA is net income
divided by year-end assets; SalesGrowth is the annual percentage change in sales; EarningsPrice is the ratio of
earnings price and share price at the start of the year; AltmanZ is the Altman (1968) Z-Score. It is is defined as:
Z=1.2 (Working Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3 (Earnings before Interest and
Taxes/Total Assets) +0.6 (Market Value of Equity/Book Value of Liabilities) +1.0(Sales/Total Assets). BoardSize
is the size of the board, defined as the number of directors on the company’s board; Big4 is a dummy variable
equal to 1, if the firm is audited by a Big 4 auditor; zero otherwise (it is also coded as 1 for Arthur Anderson
clients for 2000 and 2001); M&AIndicator is a dummy variable equal to 1, if the firm has made an acquisition
during the year; zero otherwise; Financing is a dummy variable set equal to 1 if the firm’s variable FreeCash is
less than -0.5, and zero otherwise. FreeCash is calculated as: (Cash from Operationst - Average Capital
Expenditurest−3 to t−1)/Current Assetst−1. We include firm and year fixed effects in the model. We also present the
variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
65
Table 11
Financial Statement Comparability and Accounting Fraud Other Measures of Comparability
CompAcctM10
CompAcctInd
CompAcctIndMed
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Comparability
-0.447**
-0.196*
-0.206*
(0.039)
(0.099)
(0.060)
Volatility
-19.72
42.511**
40.961**
(0.215)
(0.025)
(0.032)
Size
0.727***
0.443**
0.454**
(0.000)
(0.012)
(0.011)
Lev
-1.551
-0.014
-0.116
(0.222)
(0.992)
(0.935)
BTM
0.507
0.540
0.545
(0.300)
(0.206)
(0.200)
ROA
1.902
-3.583
-3.509
(0.245)
(0.226)
(0.234)
SalesGrowth
0.167
-0.446
-0.439
(0.577)
(0.560)
(0.563)
EarningsPrice
-1.472
2.269
2.309
(0.163)
(0.182)
(0.171)
AltmanZ
0.00923
-0.033
-0.030
(0.788)
(0.548)
(0.577)
BoardSize
-0.866
0.789
0.738
(0.378)
(0.456)
(0.489)
Big4
-0.621
0.127
0.104
(0.278)
(0.879)
(0.902)
M&AIndicator
0.434
0.974**
0.980**
(0.179)
(0.015)
(0.015)
Financing
0.000
0.000
0.000
(0.961)
(0.873)
(0.954)
Firm Fixed Effects
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Constant
-16.600***
-18.070***
-17.770***
(0.000)
(0.000)
(0.000)
Observations
8954
5,254
5,247
Wald
57.810
57.510
57.420
(p-value)
(0.001)
(0.001)
(0.001)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below.
󰇛 󰇜
    
    
   


It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
Electronic copy available at: https://ssrn.com/abstract=4207137
66
and zero otherwise. Fraud is defined as one in the last year of the AAER. CompAcctM10, CompAcctInd, and
CompAcctIndMed are four proxies of financial statement comparability, based on De Franco et al. (2011);
Volatility is the annual stock return return volatility, calculated as the standard deviation of daily stock returns,
measured over the fiscal year; Size is the natural logarithm of total assets; Lev is the ratio of total debt (long-term
and short-term) to total assets; BTM is the book value of shareholder’s equity, divided by the market value of
equity; ROA is net income divided by year-end assets; SalesGrowth is the annual percentage change in sales;
EarningsPrice is the ratio of earnings price and share price at the start of the year; AltmanZ is the Altman (1968)
Z-Score. It is is defined as: Z=1.2 (Working Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3
(Earnings before Interest and Taxes/Total Assets) +0.6 (Market Value of Equity/Book Value of Liabilities)
+1.0(Sales/Total Assets). BoardSize is the size of the board, defined as the number of directors on the company’s
board; Big4 is a dummy variable equal to 1, if the firm is audited by a Big 4 auditor; zero otherwise (it is also
coded as 1 for Arthur Anderson clients for 2000 and 2001); M&AIndicator is a dummy variable equal to 1, if the
firm has made an acquisition during the year; zero otherwise; Financing is a dummy variable set equal to 1 if the
firm’s variable FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated as: (Cash from Operationst
- Average Capital Expenditurest−3 to t−1)/Current Assetst−1. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
67
Table 12
Financial Statement Comparability and Fraud
*, **, and *** represent statistical significance at the 10%, 5% and 1% levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below.
Panel A: Fraud defined as the First Year of the AAER
CompAcctM4
ECompM4
DQ
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Comparability
-0.184**
-0.530
-3.637**
(0.041)
(0.292)
(0.039)
Volatility
1.428
6.987
-2.934
(0.862)
(0.451)
(0.746)
Size
0.361***
0.381***
0.290***
(0.000)
(0.000)
(0.000)
Lev
-0.453
0.335
-0.599
(0.482)
(0.633)
(0.391)
BTM
0.196
0.155
0.356
(0.465)
(0.621)
(0.195)
ROA
1.753*
1.641
0.786
(0.059)
(0.125)
(0.421)
SalesGrowth
0.196
0.166
0.133
(0.283)
(0.432)
(0.547)
EarningsPrice
-0.591
-0.610
-0.304
(0.285)
(0.371)
(0.635)
AltmanZ
-0.0242
-0.00163
-0.0321*
(0.203)
(0.932)
(0.086)
BoardSize
-0.702
-0.550
-0.244
(0.146)
(0.320)
(0.629)
Big4
0.0657
-0.430
0.113
(0.830)
(0.198)
(0.724)
M&AIndicator
0.101
0.364*
0.0730
(0.596)
(0.099)
(0.713)
Financing
0.000
0.000
0.000
(0.876)
(0.749)
(0.967)
Firm Fixed Effects
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Constant
-4.143**
-7.395***
-2.911
(0.008)
(0.000)
(0.135)
Observations
8,996
7,289
7,776
Wald
104.40
94.390
87.350
(p-value)
(0.000)
(0.000)
(0.000)
Electronic copy available at: https://ssrn.com/abstract=4207137
68
󰇛 󰇜
    
    
   


It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one in the first year of the AAER. CompAcctM4 is a proxy of financial
statement comparability, based on De Franco et al. (2011); ECompM4 measures earnings covariance following
Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated following the
approach of Chen et al. (2015). Volatility is the annual stock return return volatility, calculated as the standard
deviation of daily stock returns, measured over the fiscal year; Size is the natural logarithm of total assets; Lev is
the ratio of total debt (long-term and short-term) to total assets; BTM is the book value of shareholder’s equity,
divided by the market value of equity; ROA is net income divided by year-end assets; SalesGrowth is the annual
percentage change in sales; EarningsPrice is the ratio of earnings price and share price at the start of the year;
AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working Capital/Total Assets) +1.4 (Retained
Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total Assets) +0.6 (Market Value of Equity/Book
Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size of the board, defined as the number of
directors on the company’s board; Big4 is a dummy variable equal to 1, if the firm is audited by a Big 4 auditor;
zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000 and 2001); M&AIndicator is a dummy
variable equal to 1, if the firm has made an acquisition during the year; zero otherwise; Financing is a dummy
variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated
as: (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current Assetst−1. We include firm and year
fixed effects in the model. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
69
Panel B: Fraud defined as the Full Period of the AAER
CompAcctM4
ECompM4
DQ
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
Comparability
-0.192**
-0.606
-4.086**
(0.030)
(0.223)
(0.018)
Volatility
0.210
5.604
-4.121
(0.979)
(0.542)
(0.645)
Size
0.349***
0.387***
0.288***
(0.000)
(0.000)
(0.000)
Lev
-0.431
0.424
-0.622
(0.497)
(0.539)
(0.367)
BTM
0.174
0.126
0.303
(0.511)
(0.683)
(0.263)
ROA
1.611*
1.610
0.698
(0.074)
(0.126)
(0.457)
SalesGrowth
0.181
0.168
0.125
(0.315)
(0.422)
(0.569)
EarningsPrice
-0.588
-0.634
-0.340
(0.256)
(0.314)
(0.562)
AltmanZ
-0.0181
-0.00197
-0.0257
(0.323)
(0.918)
(0.165)
BoardSize
-0.663
-0.615
-0.248
(0.160)
(0.261)
(0.615)
Big4
-0.00260
-0.556*
0.0366
(0.993)
(0.083)
(0.904)
M&AIndicator
0.0973
0.320
0.0715
(0.603)
(0.142)
(0.713)
Financing
0.000
0.000
0.000
(0.976)
(0.895)
(0.984)
Firm Fixed Effects
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Constant
-4.056**
-7.141***
-2.500
(0.008)
(0.000)
(0.192)
Observations
8,996
7,289
7,776
Wald
98.720
92.680
82.540
(p-value)
(0.000)
(0.000)
(0.000)
*, **, and *** represent statistical significance at the 10%, 5% and 1% levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below.
󰇛 󰇜
    
    
   


Electronic copy available at: https://ssrn.com/abstract=4207137
70
It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one over the full period of the AAER. CompAcctM4 is a proxy of financial
statement comparability, based on De Franco et al. (2011); ECompM4 measures earnings covariance following
Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated following the
approach of Chen et al. (2015). Volatility is the annual stock return return volatility, calculated as the standard
deviation of daily stock returns, measured over the fiscal year; Size is the natural logarithm of total assets; Lev is
the ratio of total debt (long-term and short-term) to total assets; BTM is the book value of shareholder’s equity,
divided by the market value of equity; ROA is net income divided by year-end assets; SalesGrowth is the annual
percentage change in sales; EarningsPrice is the ratio of earnings price and share price at the start of the year;
AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working Capital/Total Assets) +1.4 (Retained
Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total Assets) +0.6 (Market Value of Equity/Book
Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size of the board, defined as the number of
directors on the company’s board; Big4 is a dummy variable equal to 1, if the firm is audited by a Big 4 auditor;
zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000 and 2001); M&AIndicator is a dummy
variable equal to 1, if the firm has made an acquisition during the year; zero otherwise; Financing is a dummy
variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated
as: (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current Assetst−1. We include firm and year
fixed effects in the model. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
71
Table 13: Is Poor Comparability a Consequence of Fraud?
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below.
      
    
  


CompAcctM4
ECompM4
DQ
Variable
Coef.
(p-value)
Coef.
(p-value)
Coef.
(p-value)
FRAUD
-0.049
-0.028
-0.010
(0.471)
(0.297)
(0.101)
Volatility
-10.621***
0.310
-0.198**
(0.000)
(0.329)
(0.007)
Size
0.178***
0.013***
0.003***
(0.000)
(0.001)
(0.000)
Lev
-0.394***
0.0524**
-0.107***
(0.000)
(0.035)
(0.000)
BTM
-0.058*
-0.014
0.000
(0.055)
(0.144)
(0.906)
ROA
0.012
0.090***
-0.010
(0.806)
(0.001)
(0.123)
SalesGrowth
-0.007
0.005
-0.000
(0.360)
(0.112)
(0.762)
EarningsPrice
0.450***
-0.112***
-0.017***
(0.000)
(0.000)
(0.000)
AltmanZ
-0.000
0.001*
0.001***
(0.722)
(0.050)
(0.000)
BoardSize
-0.071
-0.010
-0.011**
(0.262)
(0.625)
(0.029)
Big4
-0.058
0.028**
0.010***
(0.169)
(0.015)
(0.000)
M&AIndicator
0.070***
-0.004
0.001
(0.000)
(0.525)
(0.651)
Financing
0.054
0.099
0.018*
(0.266)
(0.184)
(0.097)
Firm Fixed Effects
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Constant
-0.904**
-0.248**
-0.584***
(0.012)
(0.005)
(0.000)
Observations
8,996
7,289
7,776
Adjusted R2
0.396
0.267
0.679
Electronic copy available at: https://ssrn.com/abstract=4207137
72
It is based on a propensity score-matched sample of observations from 2000-2018. Fraud is a dummy variable
equal to 1 if the firm is the subject of a US Securities and Exchange (SEC) AAER specifically related to fraud,
and zero otherwise. Fraud is defined as one in the first year of the AAER. CompAcctM4 is a proxy of financial
statement comparability, based on De Franco et al. (2011); ECompM4 measures earnings covariance following
Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure quality, calculated following the
approach of Chen et al. (2015). Volatility is the annual stock return return volatility, calculated as the standard
deviation of daily stock returns, measured over the fiscal year; Size is the natural logarithm of total assets; Lev is
the ratio of total debt (long-term and short-term) to total assets; BTM is the book value of shareholder’s equity,
divided by the market value of equity; ROA is net income divided by year-end assets; SalesGrowth is the annual
percentage change in sales; EarningsPrice is the ratio of earnings price and share price at the start of the year;
AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working Capital/Total Assets) +1.4 (Retained
Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total Assets) +0.6 (Market Value of Equity/Book
Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size of the board, defined as the number of
directors on the company’s board; Big4 is a dummy variable equal to 1, if the firm is audited by a Big 4 auditor;
zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000 and 2001); M&AIndicator is a dummy
variable equal to 1, if the firm has made an acquisition during the year; zero otherwise; Financing is a dummy
variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated
as: (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current Assetst−1. We include firm and year
fixed effects in the model. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Internet Appendix
In this Internet Appendix, we present additional descriptive statistics and estimation
results for our paper titled “Financial Statement Comparability and Accounting Fraud”.
Treatment and Control Variable Comparisons
Table IA1 presents the descriptive statistics for the treatment and control groups for the
base sample, propensity score-matched sample (PSM) and entropy balance sample (EB). The
base sample is the pre-PSM\pre-EB sample. Firm-year observations with comparability values
greater (lower) than the sample median constitute the treatment (control) sample. Table IA1 is
divided into three Panels: Panel A defines treatment and control samples based on the
distribution of CompAcctM4, Panel B ECompM4, and Panel C DQ. We compare the descriptive
statistics of the covariates used in the PSM and EB models: Volatility, Size, Lev, and BTM. We
define the variables in the variable appendix of the paper. Panel A of Table IA1 shows that the
covariates in the base sample are significantly different from those in the PSM and EB samples.
Specifically, we find that the base sample firms are larger, more leveraged, and have lower
book-to-market ratios and return volatility than those in the PSM sample for the treatment
observations. Further, the base sample firms are larger and have lower leverage than those in
the EB sample. We also note that the covariates for the treatment sample are significantly
different for the PSM and EB samples. Further, the base sample firms have higher ROA than
those in the PSM and EB samples.
The control sample comparisons also show that the covariates differ significantly in the
three samples. Specifically, we note that the firms in the base sample are smaller, have higher
book-to-market ratios than those in the PSM and EB samples. Further, the base sample firms
have higher return volatility than those in the PSM and EB samples. Finally, we observe that
the covariates are significantly different for the PSM and EB samples.
Electronic copy available at: https://ssrn.com/abstract=4207137
We present the descriptive statistics based on ECompM4 and DQ in Panels B and C of
Table IA1. We note that the base sample observations differ significantly from the PSM and
EB samples in Panels B and C, consistent with Panel A. However, there are some differences
in the distribution of covariates in these panels. For example, we find that the PSM firms have
lower book-to-market ratios than the base sample firms for the treatment group in Panel B. This
suggests that different comparability proxies could lead to different distributions of the
covariates in the PSM and entropy balancing samples.
[Insert Table IA1 here]
Treatment and Control Group Covariates in Base Sample
Table IA2 compares the covariates in the treatment and control groups, defined based
on whether comparability is greater than or less than the sample median, for the base sample.
We present the results in three Panels: CompAcctM4 (Panel A), ECompM4 (Panel B), and DQ
(Panel C). The Table shows that the covariates differ significantly between treatment and
control samples for all definitions of comparability (only ROA is not significantly different for
DQ).
[Insert Table IA2 here]
PSM Regression
Table IA3 presents the results of the logit regression model to generate the propensity
scores. Table IA3 has three columns Column 1 presents results for CompAcctM4, Column 2
ECompM4, and Column 3 DQ. The Table shows that the coefficient of Volatility is negative
in Columns 1 and 3 (coefficient = -53.220 and -11.511 respectively; p-value = 0.000 and 0.001
respectively). This suggests that return volatility is negatively associated with comparability.
Electronic copy available at: https://ssrn.com/abstract=4207137
We also find that the coefficient of Size is positive in Columns 1 and 2 (coefficient = 0.182 and
0.142 respectively; p-value = 0.000 and 0.000 respectively) and those on Lev (coefficient = -
1.391 and -2.647 respectively; p-value = 0.000 and 0.000 respectively) and BTM (coefficient
= -1.003 and -0.294 respectively; p-value = 0.000 and 0.000 respectively) are negative in
Columns 1 and 3. Surprisingly, we find that the coefficient of ROA is negative in Columns 2
and 3 (coefficient = -0.178 and -0.691 respectively; p-value = 0.041 and 0.004 respectively)
[Insert Table IA3 here]
PSM Covariate Comparison
Table IA4 compares the covariates of the treatment and control groups for the PSM
sample. We present the results in three Panels: CompAcctM4 (Panel A), ECompM4 (Panel B),
DQ (Panel C). Table IA4, Panel A shows that none of the covariates is significantly different
between the treatment and control samples, except for ROA (p-value = 0.000). We generally
observe similar results in Panels B and C.
[Insert Table IA4 here]
Entropy Balancing Covariate Comparison
Table IA5 compares the covariates of the treatment and control groups for the PSM
sample. We present the results in three Panels: CompAcctM4 (Panel A), ECompM4 (Panel B),
DQ (Panel C). The Table shows that the covariates are almost equal for all three definitions of
comparability. This is by design because entropy balancing assigns weights to balance the
covariates in the treatment and control samples.
[Insert Table IA5 here]
Electronic copy available at: https://ssrn.com/abstract=4207137
Generating PSM by Constructing Treatment and Control Samples by Year
The results reported in our study are based on a PSM sample where the treatment
(control) group is constructed by choosing comparability observations that are greater (lower)
than the sample median. We do not define median by year. We now check the sensitivity of
our results by constructing the treatment and control samples based on the yearly median of
comparability. We present these results in Table IA5
Covariate Comparison in Entropy Balancing Sample
Panel A: Entropy Balancing based on CompAcctM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
Size
6.781
2.086
6.780
2.086
0.001
BTM
0.553
0.394
0.553
0.394
0.000
Lev
0.194
0.179
0.194
0.179
0.000
ROA
0.046
0.135
0.045
0.139
0.000
Volatility
0.026
0.015
0.026
0.015
0.000
Panel B: Entropy Balancing based on ECompM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
Size
6.696
2.202
6.696
2.202
0.000
BTM
0.707
0.676
0.707
0.676
0.000
Lev
0.213
0.196
0.213
0.196
0.000
ROA
-0.011
0.214
-0.011
0.214
0.000
Volatility
0.032
0.020
0.032
0.020
0.000
Panel C: Entropy Balancing based on DQ
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
Size
5.585
1.999
5.585
1.999
0.000
BTM
0.611
0.643
0.611
0.643
0.000
Lev
0.158
0.183
0.158
0.183
0.000
ROA
-0.090
0.620
-0.091
0.623
0.000
Volatility
0.037
0.021
0.037
0.021
0.000
Electronic copy available at: https://ssrn.com/abstract=4207137
The Table is based on an entropy balanced (based on two moments) sample of observations from 2000-2018. The
sample sizes in Panels A, B, and C are 18,069, 14,770, and 15,392 firm-year observations respectively. The
propensity scores are generated by classifying observations into high and low comparability samples based on the
median of the distribution. We use three measures of comparability: CompAcctM4 a proxy of financial statement
comparability, based on De Franco et al. (2011); ECompM4 earnings covariance following Francis et al. (2014)
and De Franco et al. (2011); DQ disclosure quality, calculated following the approach of Chen et al. (2015)) is
above the median for the sample; zero otherwise. The covariates are Size: the natural logarithm of total assets;
BTM: the book value of shareholder’s equity, divided by the market value of equity; Lev: the ratio of total debt
(long-term and short-term) to total assets; ROA: net income divided by year-end assets; and Volatility: the annual
stock return volatility, calculated as the standard deviation of daily stock returns, measured over the fiscal year.
We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Table IA6. The Table has 3 columns - CompAcctM4 (Column 1), ECompM4 (Column
2), and DQ (Column 3). We observe that the coefficient of Comparability is negative in all
three columns (coefficient = -0.390, -2.142, and -3.738 respectively; p-value = 0.058, 0.084,
and 0.029 respectively). This shows that our results are not sensitive to the construction of the
treatment and control samples.
[Insert Table IA5
Covariate Comparison in Entropy Balancing Sample
Panel A: Entropy Balancing based on CompAcctM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
Size
6.781
2.086
6.780
2.086
0.001
BTM
0.553
0.394
0.553
0.394
0.000
Lev
0.194
0.179
0.194
0.179
0.000
ROA
0.046
0.135
0.045
0.139
0.000
Volatility
0.026
0.015
0.026
0.015
0.000
Panel B: Entropy Balancing based on ECompM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
Size
6.696
2.202
6.696
2.202
0.000
BTM
0.707
0.676
0.707
0.676
0.000
Lev
0.213
0.196
0.213
0.196
0.000
ROA
-0.011
0.214
-0.011
0.214
0.000
Volatility
0.032
0.020
0.032
0.020
0.000
Panel C: Entropy Balancing based on DQ
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
Size
5.585
1.999
5.585
1.999
0.000
BTM
0.611
0.643
0.611
0.643
0.000
Lev
0.158
0.183
0.158
0.183
0.000
ROA
-0.090
0.620
-0.091
0.623
0.000
Volatility
0.037
0.021
0.037
0.021
0.000
Electronic copy available at: https://ssrn.com/abstract=4207137
The Table is based on an entropy balanced (based on two moments) sample of observations from 2000-2018. The
sample sizes in Panels A, B, and C are 18,069, 14,770, and 15,392 firm-year observations respectively. The
propensity scores are generated by classifying observations into high and low comparability samples based on the
median of the distribution. We use three measures of comparability: CompAcctM4 a proxy of financial statement
comparability, based on De Franco et al. (2011); ECompM4 earnings covariance following Francis et al. (2014)
and De Franco et al. (2011); DQ disclosure quality, calculated following the approach of Chen et al. (2015)) is
above the median for the sample; zero otherwise. The covariates are Size: the natural logarithm of total assets;
BTM: the book value of shareholder’s equity, divided by the market value of equity; Lev: the ratio of total debt
(long-term and short-term) to total assets; ROA: net income divided by year-end assets; and Volatility: the annual
stock return volatility, calculated as the standard deviation of daily stock returns, measured over the fiscal year.
We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Table IA6 here]
Generating Entropy Balanced Sample by Constructing Treatment and Control Samples
by Year
We next check the sensitivity of our results by constructing the treatment and control
samples based on the yearly median of comparability and then generating an entropy balanced
sample based on two moments. We present these results in Table IA7. The Table has 3 columns
- CompAcctM4 (Column 1), ECompM4 (Column 2), and DQ (Column 3). We observe that the
coefficient of Comparability is negative in columns 1 and 2 (coefficient = -0.323, and -1.252
respectively; p-value = 0.050, and 0.058 respectively). These results are consistent with those
reported in Table 5.
[Insert Table IA7 here]
Electronic copy available at: https://ssrn.com/abstract=4207137
Table IA1
Comparing Base, PSM and Entropy Balancing Samples
Panel A: Treatment and Control Samples defined with respect to CompAcctM4
High Comparability:
Full Sample
PSM Sample
Entropy Balancing
Sample
Full and PSM Sample
Full and Entropy Balancing
Sample
PSM and Entropy Balancing
Sample
Variable
Mean
Std Dev
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Difference
p-value
Difference
p-value
Size
6.816
1.958
6.081
1.882
6.781
2.086
0.735
(0.000)
0.035
(0.095)
-0.700
(0.000)
BTM
0.433
0.295
0.549
0.365
0.553
0.394
-0.116
(0.000)
-0.119
(0.000)
-0.003
(0.000)
Lev
0.166
0.169
0.156
0.167
0.194
0.179
0.010
(0.001)
-0.029
(0.000)
-0.039
(0.187)
ROA
0.054
0.138
0.025
0.131
0.046
0.135
0.029
(0.004)
0.008
(0.000)
-0.021
(0.000)
Volatility
0.026
0.012
0.032
0.015
0.026
0.015
-0.006
(0.000)
0.000
(0.012)
0.006
(0.000)
Low Comparability:
Full Sample
PSM Sample
Entropy Balancing
Sample
Full and PSM Sample
Full and Entropy Balancing
Sample
PSM and Entropy Balancing
Sample
Variable
Mean
Std Dev
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Difference
p-value
Difference
p-value
Size
5.824
2.054
6.085
1.955
6.780
2.086
-0.261
(0.000)
-0.956
(0.088)
-0.695
(0.000)
BTM
0.650
0.604
0.533
0.359
0.553
0.394
0.117
(0.000)
0.097
(0.071)
-0.020
(0.000)
Lev
0.190
0.193
0.158
0.176
0.194
0.179
0.032
(0.000)
-0.004
(0.605)
-0.037
(0.000)
ROA
-0.077
0.306
0.010
0.131
0.045
0.139
-0.087
(0.004)
-0.122
(0.000)
-0.035
(0.000)
Volatility
0.036
0.018
0.031
0.013
0.026
0.015
0.005
(0.000)
0.011
(0.015)
0.005
(0.000)
The Table above is based on a sample of observations from 2000-2018. There are 18,069, 8,996, and 18,069 firm-year observations in the full, PSM and entropy balancing samples respectively.
The Table presents comparisons of the covariates used to generate the propensity score-matched (PSM) sample and entropy balancing (EB) sample and the base sample (pre-PSM/entropy
balancing). The treatment (High Comparability) and control (Low Comparability) samples are based on the median of CompAcctM4, the measure of comparability constructed following De
Franco et al. (2011). The covariates are Size: the natural logarithm of total assets; BTM: the book value of shareholder’s equity, divided by the market value of equity; Lev: the ratio of total debt
(long-term and short-term) to total assets; ROA: net income divided by year-end assets; and Volatility: the annual stock return volatility, calculated as the standard deviation of daily stock returns,
measured over the fiscal year. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Panel B: Treatment and Control Samples defined with respect to ECompM4
High Comparability:
Full Sample
PSM Sample
Entropy Balancing
Sample
Full and PSM Sample
Full and Entropy Balancing
Sample
PSM and Entropy
Balancing Sample
Variable
Mean
Std Dev
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Difference
p-value
Difference
p-value
Size
6.470
2.059
6.222
2.022
6.696
2.202
0.247
(0.000)
-0.226
(0.661)
-0.474
(0.000)
BTM
0.529
0.493
0.539
0.502
0.707
0.676
-0.010
(0.329)
-0.178
(0.256)
-0.168
(0.048)
Lev
0.188
0.187
0.176
0.176
0.213
0.196
0.012
(0.000)
-0.024
(0.000)
-0.036
(0.053)
ROA
-0.019
0.222
-0.019
0.223
-0.011
0.214
0.001
(0.000)
-0.007
(0.504)
-0.008
(0.198)
Volatility
0.031
0.016
0.032
0.013
0.032
0.020
-0.001
(0.002)
0.000
(0.821)
0.001
(0.059)
Low Comparability:
Full Sample
PSM Sample
Entropy Balancing
Sample
Full and PSM Sample
Full and Entropy Balancing
Sample
PSM and Entropy
Balancing Sample
Variable
Mean
Std Dev
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Difference
p-value
Difference
p-value
Size
6.171
2.087
6.213
2.151
6.696
2.202
-0.042
(0.000)
-0.525
(0.010)
-0.483
(0.001)
BTM
0.550
0.475
0.537
0.476
0.707
0.676
0.013
(0.431)
-0.157
(0.119)
-0.170
(0.633)
Lev
0.177
0.181
0.179
0.184
0.213
0.196
-0.001
(0.000)
-0.035
(0.001)
-0.034
(0.005)
ROA
-0.009
0.213
-0.018
0.210
-0.011
0.214
0.009
(0.004)
0.002
(0.338)
-0.007
(0.368)
Volatility
0.031
0.016
0.032
0.017
0.032
0.020
-0.001
(0.913)
-0.001
(0.715)
0.000
(0.107)
The Table above is based on a sample of observations from 2000-2018. There are 14,770, 7,289, and 14,770 firm-year observations in the full, PSM and entropy balancing samples respectively.
The Table presents comparisons of the covariates used to generate the propensity score-matched (PSM) sample and entropy balancing (EB) sample and the base sample (pre-PSM/entropy
balancing). The treatment (High Comparability) and control (Low Comparability) samples are based on the median of ECompM4, the measure of comparability constructed following De Franco
et al. (2011) and Francis et al. (2014). The covariates are Size: the natural logarithm of total assets; BTM: the book value of shareholder’s equity, divided by the market value of equity; Lev: the
ratio of total debt (long-term and short-term) to total assets; ROA: net income divided by year-end assets; and Volatility: the annual stock return volatility, calculated as the standard deviation of
daily stock returns, measured over the fiscal year. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Panel C: Treatment and Control Samples defined with respect to DQ
High Comparability:
Full Sample
PSM Sample
Entropy Balancing
Sample
Full and PSM Sample
Full and Entropy Balancing
Sample
PSM and Entropy
Balancing Sample
Variable
Mean
Std Dev
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Difference
p-value
Difference
p-value
Size
5.846
1.835
6.557
1.895
5.585
1.999
-0.711
(0.002)
0.261
(0.000)
0.972
(0.000)
BTM
0.515
0.420
0.598
0.568
0.611
0.643
-0.084
(0.329)
-0.096
(0.256)
-0.012
(0.646)
Lev
0.107
0.157
0.248
0.196
0.158
0.183
-0.141
(0.000)
-0.051
(0.000)
0.090
(0.011)
ROA
-0.021
0.283
0.013
0.195
-0.090
0.620
-0.034
(0.000)
0.069
(0.000)
0.103
(0.850)
Volatility
0.031
0.015
0.037
0.017
0.037
0.021
-0.006
(0.002)
-0.006
(0.138)
0.000
(0.685)
Low Comparability:
Full Sample
PSM Sample
Entropy Balancing
Sample
Full and PSM Sample
Full and Entropy Balancing
Sample
PSM and Entropy Balancing
Sample
Variable
Mean
Std Dev
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Difference
p-value
Difference
p-value
Size
6.168
2.020
6.568
2.057
5.585
1.999
-0.400
(0.000)
0.583
(0.010)
0.983
(0.846)
BTM
0.563
0.545
0.577
0.511
0.611
0.643
-0.014
(0.035)
-0.048
(0.174)
-0.034
(0.005)
Lev
0.206
0.182
0.230
0.187
0.158
0.183
-0.024
(0.000)
0.048
(0.001)
0.073
(0.335)
ROA
-0.015
0.231
0.014
0.208
-0.091
0.623
-0.029
(0.004)
0.076
(0.000)
0.105
(0.183)
Volatility
0.034
0.017
0.035
0.017
0.037
0.021
-0.001
(0.537)
-0.003
(0.052)
-0.002
(0.033)
The Table above is based on a sample of observations from 2000-2018. There are 15,392, 7,776, and 15,392 firm-year observations in the full, PSM and entropy balancing samples respectively
The Table presents comparisons of the covariates used to generate the propensity score-matched (PSM) sample and entropy balancing (EB) sample and the base sample (pre-PSM/entropy
balancing). The treatment (High Comparability) and control (Low Comparability) samples are based on the median of DQ, the measure of disclosure quality constructed following Chen et al.
(2015). The covariates are Size: the natural logarithm of total assets; BTM: the book value of shareholder’s equity, divided by the market value of equity; Lev: the ratio of total debt (long-term
and short-term) to total assets; ROA: net income divided by year-end assets; and Volatility: the annual stock return volatility, calculated as the standard deviation of daily stock returns, measured
over the fiscal year. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Table IA2
Covariate Comparison in Base Sample
Panel A: Treatment and Control Samples based on CompAcctM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Size
6.816
1.958
5.824
2.054
0.992
(0.000)
BTM
0.433
0.295
0.650
0.604
-0.216
(0.000)
Lev
0.166
0.169
0.190
0.193
-0.025
(0.000)
ROA
0.054
0.138
-0.077
0.306
0.131
(0.000)
Volatility
0.026
0.012
0.036
0.018
-0.011
(0.000)
Panel B: Treatment and Control Samples based on ECompM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Size
6.470
2.059
6.171
2.087
0.299
(0.000)
BTM
0.529
0.493
0.550
0.475
-0.021
(0.002)
Lev
0.188
0.187
0.177
0.181
0.011
(0.000)
ROA
-0.019
0.222
-0.009
0.213
-0.010
(0.002)
Volatility
0.031
0.016
0.031
0.016
0.001
(0.006)
Panel C: Treatment and Control Samples based on DQ
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Size
5.846
1.835
6.168
2.020
-0.323
(0.000)
BTM
0.515
0.420
0.563
0.545
-0.048
(0.000)
Lev
0.107
0.157
0.206
0.182
-0.099
(0.000)
ROA
-0.021
0.283
-0.015
0.231
-0.006
(0.119)
Volatility
0.031
0.015
0.034
0.017
-0.003
(0.000)
The Table is based on a sample of observations from 2000-2018. The sample sizes in Panels A, B, and C are
18,069, 14,770, and 15,392 firm-year observations respectively. We assign observations into treatment and control
groups based on whether comparability is greater than or less than the sample median respectively. We use three
measures of comparability: CompAcctM4 a proxy of financial statement comparability, based on De Franco et
al. (2011); ECompM4 earnings covariance following Francis et al. (2014) and De Franco et al. (2011); DQ
disclosure quality, calculated following the approach of Chen et al. (2015)) is above the median for the sample;
zero otherwise. The covariates are Size: the natural logarithm of total assets; BTM: the book value of shareholder’s
equity, divided by the market value of equity; Lev: the ratio of total debt (long-term and short-term) to total assets;
ROA: net income divided by year-end assets; and Volatility: the annual stock return volatility, calculated as the
standard deviation of daily stock returns, measured over the fiscal year. We also present the variable definitions
in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Table IA3
Models to Generate Propensity Scores
CompAcctM4
ECompM4
DQ
Variable
Coef.
Coef.
Coef.
(p-value)
(p-value)
(p-value)
Volatility
-53.220***
2.271
-11.511***
(0.000)
(0.107)
(0.001)
Size
0.182***
0.142***
0.036
(0.000)
(0.000)
(0.227)
Lev
-1.391***
0.002
-2.647***
(0.000)
(0.983)
(0.000)
BTM
-1.003***
0.038
-0.294***
(0.000)
(0.303)
(0.000)
ROA
2.415***
-0.178**
-0.691***
(0.000)
(0.041)
(0.004)
Industry Fixed Effects
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Constant
3.575
-0.227
1.830***
(0.752)
(0.173)
(0.001)
Observations
18,069
14,770
15,392
Pseudo R2
0.264
0.106
0.590
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above show results of the estimation of the model below.
    


It is based on a sample of observations from 2000-2018. High is a dummy variable equal to 1 if the value of
comparability (measured as CompAcctM4 a proxy of financial statement comparability, based on De Franco et
al. (2011); ECompM4 earnings covariance following Francis et al. (2014) and De Franco et al. (2011); DQ
disclosure quality, calculated following the approach of Chen et al. (2015)) is above the median for the sample;
zero otherwise. Volatility is the annual stock return return volatility, calculated as the standard deviation of daily
stock returns, measured over the fiscal year; Size is the natural logarithm of total assets; Lev is the ratio of total
debt (long-term and short-term) to total assets; BTM is the book value of shareholder’s equity, divided by the
market value of equity; and ROA is net income divided by year-end assets. We include industry and year fixed
effects in the model. We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Table IA4
Covariate Comparison in PSM Sample
Panel A: Propensity Score Matched Sample based on CompAcctM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Size
6.081
1.882
6.085
1.955
-0.004
(0.923)
BTM
0.549
0.365
0.533
0.359
0.016
(0.122)
Lev
0.156
0.167
0.158
0.176
-0.002
(0.544)
ROA
0.025
0.131
0.010
0.131
0.014
(0.000)
Volatility
0.032
0.015
0.031
0.013
0.000
(0.609)
Panel B: Propensity Score Matched Sample based on ECompM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Size
6.222
2.022
6.213
2.151
0.009
(0.795)
BTM
0.539
0.502
0.537
0.476
0.002
(0.783)
Lev
0.176
0.176
0.179
0.184
-0.003
(0.433)
ROA
-0.019
0.223
-0.018
0.210
-0.001
(0.697)
Volatility
0.032
0.013
0.032
0.017
0.001
(0.516)
Panel C: Propensity Score Matched Sample based on DQ
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
p-value
Size
6.557
1.895
6.568
2.057
-0.012
(0.912)
BTM
0.598
0.568
0.577
0.511
0.021
(0.459)
Lev
0.248
0.196
0.230
0.187
0.017
(0.132)
ROA
0.013
0.195
0.014
0.208
-0.001
(0.903)
Volatility
0.037
0.017
0.035
0.017
0.002
(0.023)
The Table is based on a propensity score-matched sample of observations from 2000-2018. The sample sizes in
Panels A, B, and C are 8,996, 7,289, and 7,776 firm-year observations respectively. The propensity scores are
generated by classifying observations into high and low comparability samples based on the median of the
distribution. We use three measures of comparability: CompAcctM4 a proxy of financial statement
comparability, based on De Franco et al. (2011); ECompM4 earnings covariance following Francis et al. (2014)
and De Franco et al. (2011); DQ disclosure quality, calculated following the approach of Chen et al. (2015)) is
above the median for the sample; zero otherwise. The covariates are Size: the natural logarithm of total assets;
BTM: the book value of shareholder’s equity, divided by the market value of equity; Lev: the ratio of total debt
(long-term and short-term) to total assets; ROA: net income divided by year-end assets; and Volatility: the annual
stock return volatility, calculated as the standard deviation of daily stock returns, measured over the fiscal year.
We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Table IA5
Covariate Comparison in Entropy Balancing Sample
Panel A: Entropy Balancing based on CompAcctM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
Size
6.781
2.086
6.780
2.086
0.001
BTM
0.553
0.394
0.553
0.394
0.000
Lev
0.194
0.179
0.194
0.179
0.000
ROA
0.046
0.135
0.045
0.139
0.000
Volatility
0.026
0.015
0.026
0.015
0.000
Panel B: Entropy Balancing based on ECompM4
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
Size
6.696
2.202
6.696
2.202
0.000
BTM
0.707
0.676
0.707
0.676
0.000
Lev
0.213
0.196
0.213
0.196
0.000
ROA
-0.011
0.214
-0.011
0.214
0.000
Volatility
0.032
0.020
0.032
0.020
0.000
Panel C: Entropy Balancing based on DQ
High Comparability
Low Comparability
Variable
Mean
Std Dev
Mean
Std Dev
Difference
Size
5.585
1.999
5.585
1.999
0.000
BTM
0.611
0.643
0.611
0.643
0.000
Lev
0.158
0.183
0.158
0.183
0.000
ROA
-0.090
0.620
-0.091
0.623
0.000
Volatility
0.037
0.021
0.037
0.021
0.000
The Table is based on an entropy balanced (based on two moments) sample of observations from 2000-2018. The
sample sizes in Panels A, B, and C are 18,069, 14,770, and 15,392 firm-year observations respectively. The
propensity scores are generated by classifying observations into high and low comparability samples based on the
median of the distribution. We use three measures of comparability: CompAcctM4 a proxy of financial statement
comparability, based on De Franco et al. (2011); ECompM4 earnings covariance following Francis et al. (2014)
and De Franco et al. (2011); DQ disclosure quality, calculated following the approach of Chen et al. (2015)) is
above the median for the sample; zero otherwise. The covariates are Size: the natural logarithm of total assets;
BTM: the book value of shareholder’s equity, divided by the market value of equity; Lev: the ratio of total debt
(long-term and short-term) to total assets; ROA: net income divided by year-end assets; and Volatility: the annual
stock return volatility, calculated as the standard deviation of daily stock returns, measured over the fiscal year.
We also present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Table IA6
Financial Statement Comparability and Accounting Fraud
CompAcctM4
ECompM4
DQ
Coef.
Coef.
Coef.
Variable
(p-value)
(p-value)
(p-value)
Comparability
-0.390*
-2.142*
-3.738*
(0.058)
(0.084)
(0.029)
Volatility
4.050
12.940
18.750
(0.798)
(0.511)
(0.282)
Size
0.742***
0.890**
0.770**
(0.000)
(0.000)
(0.000)
Lev
-1.085
-0.965
-0.308
(0.407)
(0.546)
(0.834)
BTM
0.103
0.153
0.095
(0.841)
(0.812)
(0.865)
ROA
1.754
0.937
1.254
(0.300)
(0.674)
(0.513)
SalesGrowth
0.127
0.298
0.153
(0.711)
(0.464)
(0.735)
EarningsPrice
-1.308
-0.910
-1.370
(0.261)
(0.571)
(0.323)
AltZ
0.000
-0.003
-0.004
(0.997)
(0.951)
(0.917)
BoardSize
-1.553
-1.223
-1.483
(0.102)
(0.292)
(0.165)
Big4
-0.230
-1.117*
-0.864
(0.690)
(0.098)
(0.166)
M&A
0.472
0.823*
0.400
(0.150)
(0.044)
(0.269)
Financing
0.000
0.000
0.000
(0.972)
(0.984)
(0.985)
Firm Fixed Effects
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Constant
-15.780***
-17.766**
-16.681**
(0.000)
(0.000)
(0.000)
Observations
8,996
7,289
7,776
Wald
66.160
56.790
48.180
(p-value)
(0.001)
(0.000)
(0.000)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above shows the results of the estimation of the model below.
󰇛 󰇜
     
    
  


It is based on a propensity score-matched sample of observations from 2000-2018. We generate the treatment and
control samples based on whether the value of comparability is greater than or less than the yearly median of the
sample. Fraud is a dummy variable equal to 1 if the firm is the subject of a US Securities and Exchange (SEC)
AAER specifically related to fraud, and zero otherwise. Fraud is defined as one in the last year of the AAER.
CompAcctM4 is a proxy of financial statement comparability, based on De Franco et al. (2011); ECompM4
measures earnings covariance following Francis et al. (2014) and De Franco et al. (2011); DQ measures disclosure
Electronic copy available at: https://ssrn.com/abstract=4207137
quality, calculated following the approach of Chen et al. (2015). Volatility is the annual stock return return
volatility, calculated as the standard deviation of daily stock returns, measured over the fiscal year; Size is the
natural logarithm of total assets; Lev is the ratio of total debt (long-term and short-term) to total assets; BTM is
the book value of shareholder’s equity, divided by the market value of equity; ROA is net income divided by year-
end assets; SalesGrowth is the annual percentage change in sales; EarningsPrice is the ratio of earnings price and
share price at the start of the year; AltmanZ is the Altman (1968) Z-Score. It is is defined as: Z=1.2 (Working
Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3 (Earnings before Interest and Taxes/Total
Assets) +0.6 (Market Value of Equity/Book Value of Liabilities) +1.0(Sales/Total Assets). BoardSize is the size
of the board, defined as the number of directors on the company’s board; Big4 is a dummy variable equal to 1, if
the firm is audited by a Big 4 auditor; zero otherwise (it is also coded as 1 for Arthur Anderson clients for 2000
and 2001); M&AIndicator is a dummy variable equal to 1, if the firm has made an acquisition during the year;
zero otherwise; Financing is a dummy variable set equal to 1 if the firm’s variable FreeCash is less than -0.5, and
zero otherwise. FreeCash is calculated as: (Cash from Operationst - Average Capital Expenditurest−3 to t−1)/Current
Assetst−1. We include firm and year fixed effects in the model. We also present the variable definitions in Appendix
A.
Electronic copy available at: https://ssrn.com/abstract=4207137
Table IA7
Financial Statement Comparability and Accounting Fraud
CompAcctM4
ECompM4
DQ
Coef.
Coef.
Coef.
(p-value)
(p-value)
(p-value)
Comparability
-0.323**
-1.252*
-1.460
(0.050)
(0.058)
(0.487)
Volatility
-26.720**
-19.990**
-25.940**
(0.003)
(0.048)
(0.008)
Size
0.186*
0.188
0.149
(0.092)
(0.118)
(0.217)
Lev
-0.883
-0.485
-0.903
(0.314)
(0.577)
(0.342)
BTM
0.778**
0.629*
0.770**
(0.009)
(0.058)
(0.012)
ROA
0.659
0.357
0.0114
(0.485)
(0.675)
(0.988)
SalesGrowth
-0.155
-0.154
-0.236*
(0.198)
(0.244)
(0.077)
EarningsPric
-0.111
-0.212
0.0433
(0.786)
(0.618)
(0.896)
AltZ
-0.022
-0.018
-0.028
(0.296)
(0.344)
(0.241)
BoardSize
-1.300*
-1.415*
-0.916
(0.080)
(0.092)
(0.236)
Big4
-0.518
-0.967**
-0.549
(0.269)
(0.047)
(0.269)
M&A
0.288
0.606**
0.221
(0.146)
(0.009)
(0.315)
Financing
0.000
0.000
0.000
(0.974)
(0.912)
(0.936)
Firm Fixed Effects
Yes
Yes
Yes
Year Fixed Effects
Yes
Yes
Yes
Constant
2.671*
3.871**
3.188
(0.096)
(0.028)
(0.159)
Observations
18,069
14,770
15,392
Wald
72.440
88.500
61.660
(p-value)
(0.000)
(0.000)
(0.000)
*, **, and *** represent statistical significance at the 10%, 5% and 1% Levels of significance respectively; based
on one-tailed tests. The numbers in parentheses are p-values.
The Table above shows the results of the estimation of the model below.
󰇛 󰇜
     
    
  


It is based on an entropy-balanced sample of observations from 2000-2018. We generate the treatment and control
samples based on whether the value of comparability is greater than or less than the yearly median of the sample.
The entropy balancing is based on two moments. Fraud is a dummy variable equal to 1 if the firm is the subject
of a US Securities and Exchange (SEC) AAER specifically related to fraud, and zero otherwise. Fraud is defined
as one in the last year of the AAER. CompAcctM4 is a proxy of financial statement comparability, based on De
Franco et al. (2011); ECompM4 measures earnings covariance following Francis et al. (2014) and De Franco et
al. (2011); DQ measures disclosure quality, calculated following the approach of Chen et al. (2015). Volatility is
the annual stock return return volatility, calculated as the standard deviation of daily stock returns, measured over
Electronic copy available at: https://ssrn.com/abstract=4207137
the fiscal year; Size is the natural logarithm of total assets; Lev is the ratio of total debt (long-term and short-term)
to total assets; BTM is the book value of shareholder’s equity, divided by the market value of equity; ROA is net
income divided by year-end assets; SalesGrowth is the annual percentage change in sales; EarningsPrice is the
ratio of earnings price and share price at the start of the year; AltmanZ is the Altman (1968) Z-Score. It is is defined
as: Z=1.2 (Working Capital/Total Assets) +1.4 (Retained Earnings/Total Assets) +3.3 (Earnings before Interest
and Taxes/Total Assets) +0.6 (Market Value of Equity/Book Value of Liabilities) +1.0(Sales/Total Assets).
BoardSize is the size of the board, defined as the number of directors on the company’s board; Big4 is a dummy
variable equal to 1, if the firm is audited by a Big 4 auditor; zero otherwise (it is also coded as 1 for Arthur
Anderson clients for 2000 and 2001); M&AIndicator is a dummy variable equal to 1, if the firm has made an
acquisition during the year; zero otherwise; Financing is a dummy variable set equal to 1 if the firm’s variable
FreeCash is less than -0.5, and zero otherwise. FreeCash is calculated as: (Cash from Operationst - Average
Capital Expenditurest−3 to t−1)/Current Assetst−1. We include firm and year fixed effects in the model. We also
present the variable definitions in Appendix A.
Electronic copy available at: https://ssrn.com/abstract=4207137
... 62 The costs of nondisclosure would have to be weighed against the risks of false or misleading statements where disclosure was made. et al. 2023), 63 and facilitate an assessment of performance across firms (Campbell and Young 2017), including peers (Blanco et al. 2023). Greater comparability can result in an increase in analyst coverage and improvements in forecasts (De Franco et al. 2011). ...
... Greater comparability can result in an increase in analyst coverage and improvements in forecasts (De Franco et al. 2011). Comparability can also facilitate fraud detection (Blanco et al. 2023). ...
... The benefits of comparative data explain some of the opposition. The approach increases transparency for competitors, acquirers, and regulators (Blanco et al. 2023). The accompanying reduction in the application of judgment also reduces the ability of companies to consider disclosure through a filter of "management concerns." ...
... Detecting and preventing financial statement fraud is essential to maintaining the credibility and integrity o f financial reporting. This is consistent with the findings of Blanco et al. [58], which indicate that the detection of fraud improves the quality and comparability of financial statements. Examples of falsification of financial statements in Kazakhstan and Japan presented in the article led to serious consequences for the company and investors. ...
Article
Full-text available
To present and compare the financial reporting control situation between Kazakhstan and Japan, and to disclose methods of detecting and preventing distortions in financial statements that can be used to ensure the integrity and accuracy of doing business. A literature study using bibliometric analysis, including the identification of causes, ways, and methods to minimize distortions in companies' financial statements, a case study of companies in financial distress in Kazakhstan and Japan, and the use of the Beneish model for early identification of potential risks. The topics of scientific research on distortions in financial statements cover many aspects. Similar methods of assessing financial statements for distortions, as well as the Beneish model, can be applied in Kazakhstan and Japan. The main differences used in the internal assessment of reports include a different approach to the internal control system and the transparency system of financial data. Although it may still be challenging and somewhat unfair to compare two countries due to differences in economic development and systemic internal controls, there are shared methods for detecting financial statement misstatements used by companies in both. One such method is the Beneish M-Score, which has demonstrated its applicability in analyzing companies from both countries. The article outlines how to identify distortions in financial statements and take measures to prevent them using internal control methods, the influence of third-party factors, and the Beneish methodology.
... Building on this foundation, recent studies provide complementary insights. halteh and tiwari (2023) highlight that indicators of financial distress can be used to preemptively identify firms at high risk of fraud, emphasizing the importance of early detection mechanisms. in a related vein, Blanco et al. (2023) show that firms with lower financial statement comparability-often linked to weak reporting quality-are more prone to accounting fraud, especially when under financial pressure, as reduced transparency impairs external monitoring. ...
Article
Full-text available
This study examines the relationship between earnings manipulation and cash holdings in non-financial firms across G7 countries from 2006 to 2022, using 111,640 firm-year observations from 9,766 listed companies. Earnings manipulators are identified using the Beneish M-Score. The analysis explores how manipulation relates to cash-holding practices across institutional settings. While prior studies mainly focused on single-country contexts, this study applies a unified detection approach in a cross-country setting, offering broader insights into how governance and culture influence corporate liquidity policies. Results show that manipulators hold significantly more cash than non-manipulators in the US, UK, Canada, France, and Italy, but not in Germany and Japan. This variation reflects firm-level factors such as overvaluation and financial distress, and country-level traits like ownership concentration, strength of accounting and auditing enforcement, and individualism. In France and Italy, precautionary cash accumulation is linked to moderate enforcement and concentrated ownership. In contrast, the US, UK, and Canada exhibit strong enforcement and individualistic cultures, encouraging cash hoarding to manage litigation and governance pressures. Overall, the results underscore the interplay between firm incentives and institutional environments in shaping fraudulent firms’ liquidity strategies.
... If commonly owned firms' accounting numbers are more comparable, it is easier for CIIs to detect earnings management. Therefore, comparable reporting practices constrain managers from managing earnings (Kim, Kraft, and Ryan 2013;Sohn 2016;Blanco, Dhole, and Gul 2023). In addition, more comparable financial reporting can make it easier for CIIs to understand and evaluate a firm's performance relative to that of other firms in the industry. ...
Article
We examine how common institutional investors (CIIs) facilitate the financial reporting comparability (FRC) of US firms. Common ownership increases FRC of firms that are directly owned by CIIs (via a direct effect) and has positive spillover effects on other firms in the same industry. We find spillover effects in two types of firms: (1) those that are commonly owned by different institutional investors but are connected through common firms, and (2) those that do not have any common ownership. These results suggest that the effect of common ownership goes beyond commonly owned firms and extends to non‐commonly owned firms. Furthermore, we find two mechanisms for the direct and spillover effects of common ownership on reporting comparability: firms' hiring of common auditors and their adoption of similar accounting practices. Overall, we provide comprehensive evidence on how common institutional ownership benefits the comparability of financial reporting in the United States.
Thesis
Full-text available
ADVERTIMENT. La consulta d'aquesta tesi queda condicionada a l'acceptació de les següents condicions d'ús: La difusió d'aquesta tesi per mitjà del servei TDX (www.tdx.cat) i a través del Dipòsit Digital de la UB (diposit.ub.edu) ha estat autoritzada pels titulars dels drets de propietat intelꞏlectual únicament per a usos privats emmarcats en activitats d'investigació i docència. No s'autoritza la seva reproducció amb finalitats de lucre ni la seva difusió i posada a disposició des d'un lloc aliè al servei TDX ni al Dipòsit Digital de la UB. No s'autoritza la presentació del seu contingut en una finestra o marc aliè a TDX o al Dipòsit Digital de la UB (framing). Aquesta reserva de drets afecta tant al resum de presentació de la tesi com als seus continguts. En la utilització o cita de parts de la tesi és obligat indicar el nom de la persona autora. ADVERTENCIA. La consulta de esta tesis queda condicionada a la aceptación de las siguientes condiciones de uso: La difusión de esta tesis por medio del servicio TDR (www.tdx.cat) y a través del Repositorio Digital de la UB (diposit.ub.edu) ha sido autorizada por los titulares de los derechos de propiedad intelectual únicamente para usos privados enmarcados en actividades de investigación y docencia. No se autoriza su reproducción con finalidades de lucro ni su difusión y puesta a disposición desde un sitio ajeno al servicio TDR o al Repositorio Digital de la UB. No se autoriza la presentación de su contenido en una ventana o marco ajeno a TDR o al Repositorio Digital de la UB (framing). Esta reserva de derechos afecta tanto al resumen de presentación de la tesis como a sus contenidos. En la utilización o cita de partes de la tesis es obligado indicar el nombre de la persona autora. WARNING. On having consulted this thesis you're accepting the following use conditions: Spreading this thesis by the TDX (www.tdx.cat) service and by the UB Digital Repository (diposit.ub.edu) has been authorized by the titular of the intellectual property rights only for private uses placed in investigation and teaching activities. Reproduction with lucrative aims is not authorized nor its spreading and availability from a site foreign to the TDX service or to the UB Digital Repository. Introducing its content in a window or frame foreign to the TDX service or to the UB Digital Repository is not authorized (framing). Those rights affect to the presentation summary of the thesis as well as to its contents. In the using or citation of parts of the thesis it's obliged to indicate the name of the author.
Article
Full-text available
This investigation delves into the ramifications of beta convergence on the stock markets of the Pacific Alliance countries, scrutinizing data extracted from their annual financial statements. Beta convergence, denoting the gradual mitigation of risk and return disparities in international financial markets, is inexorably entwined with the process of market integration. Employing sophisticated econometric methodologies, the study discerns nuanced convergence or divergence patterns within crucial financial ratios, including the Quick Ratio, Return on Equity (ROE), Return on Assets (ROA), Working Capital, Cash Conversion Cycle, and Debt-to-Equity Ratio, thereby furnishing invaluable insights into corporate performance. The findings therefore contribute decisively to well-informed decision-making processes for discerning investors, analysts, and executives maneuvering within the dynamic and intricately interlinked global financial milieu. Resumen Esta investigación profundiza en las ramificaciones de la convergencia beta en las bolsas de valores de los países de la Alianza del Pacífico, utilizando datos extraídos de las cuentas anuales de las empresas cotizadas. La convergencia beta, que denota la mitigación gradual de las disparidades de riesgo y rendimiento en los mercados financieros internacionales, está inexorablemente entrelazada con el proceso de integración del mercado bursátil. Empleando metodologías econométricas, el estudio discierne patrones matizados de convergencia o divergencia dentro de índices financieros cruciales, que incluyen Quick Ratio, Return on Equity (ROE), Return on Assets (ROA), Fondo de Maniobra, Ciclo de Caja y deuda a capital. Ratio, proporcionando así información invaluable acerca del desempeño corporativo. Por tanto, los hallazgos contribuyen decisivamente a procesos de toma de decisiones bien informados para inversores, analistas y ejecutivos exigentes que maniobran dentro del entorno financiero global dinámico e intrincadamente interrelacionado. Palabras clave: Convergencia beta, Disparidades de riesgo y rentabilidad, Integración de mercados, Alianza del Pacífico. Clasificación JEL: F15, G12, O4.
Article
Full-text available
We investigate whether corporate governance has differential effects on downside and upside risk. Intuitively, strong corporate governance should decrease the downside risk but increase the upside risk. However, using a large panel of 1,164 non‐financial Australian firms from 2001 to 2013, we find that strong corporate governance relates negatively not only to downside risk but also to upside risk. These findings are robust to alternative risk‐taking and corporate governance proxies, and to alternative sample specifications. We also show that our main results are unaffected due to endogeneity bias by using firm and industry‐year fixed effects, lagged independent variables, generalized method of moments, and entropy balancing estimation techniques. In additional analyses, we document that our main results are homogeneous across different industries and for firms with varying levels of boardroom gender diversity. However, we find that our main results are driven by firms with lower ownership concentration as well as by older and more mature firms. Finally, we document that while reducing risk‐taking (downside and upside risk), corporate governance reduces firm value. From a regulatory perspective, these findings raise questions on the design of monitoring‐focused corporate governance recommendations and have implications for risk management. This article is protected by copyright. All rights reserved
Article
Full-text available
We use a machine learning technique to assess whether the thematic content of financial statement disclosures (labeled topic) is incrementally informative in predicting intentional misreporting. Using a Bayesian topic modeling algorithm, we determine and empirically quantify the topic content of a large collection of 10‐K narratives spanning 1994 to 2012. We find that the algorithm produces a valid set of semantically meaningful topics that predict financial misreporting, based on samples of Securities and Exchange Commission (SEC) enforcement actions (Accounting and Auditing Enforcement Releases [AAERs]) and irregularities identified from financial restatements and 10‐K filing amendments. Our out‐of‐sample tests indicate that topic significantly improves the detection of financial misreporting by as much as 59% when added to models based on commonly used financial and textual style variables. Furthermore, models that incorporate topic significantly outperform traditional models when detecting serious revenue recognition and core expense errors. Taken together, our results suggest that the topics discussed in annual report filings and the attention devoted to each topic are useful signals in detecting financial misreporting.
Article
Full-text available
Financial statement comparability enables weighing the similarities and differences in financial performance between firms. Prior studies mainly focus on the role of accounting standards in the production of comparability, but the role of economic agents has been largely overlooked. We find that a firm's audit committee size and financial expertise affect its financial statement comparability. Financial information tends to be more comparable among industry peers when audit committees are larger and more members have financial and accounting expertise. The effect of audit committee expertise on comparability is stronger for firms with less independent and smaller boards, for firms with non-Big 4 auditors and for firms with CEOs serving as the chairperson of the boards.
Article
We explore the relation between financial reporting uniformity and comparability. We also examine the effect of uniformity on analyst coverage, analyst forecast accuracy, and analyst forecast dispersion. We develop a Compustat‐based financial statement account uniformity measure based on the presentation of common financial statement line items. We define uniformity using a Jaccard similarity index where a firm's non‐missing Compustat data items are compared to a prototypical firm for that industry‐year. We emphasize the conceptual difference between uniformity and comparability, and our tests do not show a significant association between uniformity and output‐based accounting comparability. However, we find that the association between account uniformity and comparability becomes positive in high R&D firms, while it becomes negative in firms with higher managerial discretion. We explore the association between uniformity and information processing efficiencies, and we show that our uniformity measure is positively associated with analyst coverage. Furthermore, we partition our uniformity measure into separate income statement and balance sheet components. We find that income statement uniformity is associated with higher forecast accuracy and lower forecast dispersion, while balance sheet uniformity is associated with greater analyst coverage. Finally, we provide further support for these findings using an XBRL‐based uniformity measure. This article is protected by copyright. All rights reserved
Article
For many accounting research questions, empirical researchers cannot randomly assign observations to treatment conditions or identify a quasi-experimental setting. In these cases, entropy balancing (Hainmueller 2012) is an increasingly popular statistical method for identifying a control sample that is nearly identical to the treated sample with respect to observable covariates. In this paper, we compare entropy balancing's approach of reweighting control sample observations to ordinary least squares and propensity score matching. We demonstrate that researchers applying entropy balancing in empirical settings involving panel data with features common in accounting research may encounter implementation issues that render the resulting estimates sensitive to relatively minor changes in the control sample or the research design. Using the setting of estimating the Big-N audit fee premium, we empirically demonstrate these issues and propose solutions. Data Availability: Data are available from public sources cited in the text. JEL Classifications: C18; M4.
Article
We investigate CEOs who combine insider selling with stock splits, which is suspicious because dumping stocks is inconsistent with the positive stock-split signal. Our empirical results indicate that, compared with other splits, these mixed-signal splits perform poorly and are followed by much lower buy-and-hold-abnormal returns and much higher likelihoods of announcing an earnings restatement and CEO turnover in the post-split period. Our results are robust to entropy balancing and controlling for CEO characteristics, incentives, and corporate governance and highlight previously ignored agency issues around stock splits. Attention to insider trades is essential to properly interpret a stock-split signal. This article is protected by copyright. All rights reserved
Article
We examine the voluntary disclosure behavior of peer firms of hostile takeover targets. We find that peer firms under control threat use a disclosure strategy that emphasizes bad news: they provide more bad news forecasts, tend to bundle bad news forecasts with earnings announcements, use more negative tone in conference call presentations, and more evenly distribute negative tonal words throughout the presentation to heighten the visibility of bad news. This asymmetric disclosure of bad news is concentrated in firms whose managers have greater incentives to mitigate the control threat—firms with younger CEOs, CEOs with higher total compensation, and firms with weaker anti-takeover provisions. Further tests show that peer firms also manage accruals downward. We contribute to the sparse literature on the impact of corporate control contests on voluntary disclosure by demonstrating that peer firms under control threat emphasize bad news to preempt control threat. JEL Classifications: G14; G32; M4; M41.
Article
We hypothesize that if individual auditors possess unique audit styles that they consistently apply to different audit engagements, then client firms with a common signing auditor will exhibit higher earnings comparability. Using a large sample of Chinese firms, we find that client firms report more comparable earnings when they are audited by the same individual auditor than when they are audited by (1) different audit firms, (2) the same audit firm, but different audit offices, and (3) the same audit office, but different individual auditors. The individual auditor style effect is stronger for larger audit firms, senior signing auditors, and signing auditors with more stable teamwork experience. We also document that having a common signing auditor is associated with lower analyst earnings forecast error and dispersion for client firms. This study contributes to the literature by showing that individual auditors have a significant impact on client firms' earnings comparability.
Article
We show that propensity score matching (PSM), an enormously popular method of preprocessing data for causal inference, often accomplishes the opposite of its intended goal—thus increasing imbalance, inefficiency, model dependence, and bias. The weakness of PSM comes from its attempts to approximate a completely randomized experiment, rather than, as with other matching methods, a more efficient fully blocked randomized experiment. PSM is thus uniquely blind to the often large portion of imbalance that can be eliminated by approximating full blocking with other matching methods. Moreover, in data balanced enough to approximate complete randomization, either to begin with or after pruning some observations, PSM approximates random matching which, we show, increases imbalance even relative to the original data. Although these results suggest researchers replace PSM with one of the other available matching methods, propensity scores have other productive uses.
Article
Propensity score matching and weighting are popular methods when estimating causal effects in observational studies. Beyond the assumption of unconfoundedness, however, these methods also require the model for the propensity score to be correctly specified. The recently proposed covariate balancing propensity score (CBPS) methodology increases the robustness to model misspecification by directly optimizing sample covariate balance between the treatment and control groups. In this paper, we extend the CBPS to a continuous treatment. We propose the covariate balancing generalized propensity score (CBGPS) methodology, which minimizes the association between covariates and the treatment. We develop both parametric and nonparametric approaches and show their superior performance over the standard maximum likelihood estimation in a simulation study. The CBGPS methodology is applied to an observational study, whose goal is to estimate the causal effects of political advertisements on campaign contributions. We also provide open-source software that implements the proposed methods.