Author’s Accepted Manuscript

Earnings management and annual report readability

Kin Lo, Felipe Ramos, Rafael Rogo

www.elsevier.com/locate/jae

PII: S0165-4101(16)30054-4
DOI: http://dx.doi.org/10.1016/j.jacceco.2016.09.002
Reference: JAE1115
To appear in: Journal of Accounting and Economics
Received date: 3 February 2015
Revised date: 2 August 2016
Accepted date: 1 September 2016
Cite this article as: Kin Lo, Felipe Ramos and Rafael Rogo, Earnings
management and annual report readability, Journal of Accounting and
Economics, http://dx.doi.org/10.1016/j.jacceco.2016.09.002
This is a PDF file of an unedited manuscript that has been accepted for
publication. As a service to our customers we are providing this early version of
the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting galley proof before it is published in its final citable form.
Please note that during the production process errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal pertain.
 Earnings management and annual report readability
Kin Loa*, Felipe Ramosb, Rafael Rogo
a
Sauder School of Business, The University of British Columbia, Vancouver, BC, Canada
b
FUCAPE Business School, Vitoria, ES, Brazil
c
Sauder School of Business, The University of British Columbia, Vancouver, BC, Canada
*
Contact author: Mail: 2053 Main Mall, Vancouver, BC, Canada, V6T 1Z2. Ph. 604-822-8430;
Email: [email protected]

Abstract

We explore how the readability of annual reports varies with earnings management. Using the
Fog Index to measure readability (Li 2008), and focusing on the management discussion and
analysis section of the annual report (MD&A), we predict and find that firms most likely to have
managed earnings to beat the prior year’s earnings have MD&As that are more complex. This
disruption of the overall pattern of readability increasing with the level of earnings found in Li
(2008) challenges the ontological explanation that good news is inherently easier to
communicate, and shows that obfuscation contributes to making disclosures more complex.

Key Words: Annual report readability, profitability, earnings management, computational

linguistics
JEL Classification: D82; G14; G18; K22; M41; M43; M45

“For more than forty years, I’ve studied the documents that public companies file. Too often,
I’ve been unable to decipher just what is being said or, worse yet, had to conclude that nothing
was being said.
[…]
Maybe we simply don’t have the technical knowledge to grasp what the writer wishes to convey.
Or perhaps the writer doesn’t understand what he or she is talking about. In some cases,
moreover, I suspect that a less-than scrupulous issuer doesn’t want us to understand a subject it
feels legally obligated to touch upon.”
Warren Buffett
Preface of “A Handbook of Plain English Handbook” – SEC


We appreciate the helpful comments from our colleagues at UBC, participants of the UBCOW conference, 2015
Conference on Convergence of Financial and Managerial Accounting Research, and ANPCONT 2014, and
workshop participants at Singapore Management University, Nanyang Technological University, and University of
Alberta. We thank the anonymous review and the Editor, Joanna Wu, for their careful attention to improving this
paper. Funding was provided in part by the Chartered Professional Accountants of BC and the Social Sciences and
Humanities Research Council.

1
1. INTRODUCTION

In a typical corporate report, the textual narrative represents the great majority of the

disclosure— an average of 80% of an annual report, for instance—versus the remainder that

consists of numbers and representations of quantitative data. The clarity of this large component

of mandatory disclosure is crucial to understanding and to interpreting the information contained

in the report. The U.S. Securities and Exchange Commission (SEC) has been very forthright

about the overly complex corporate reports. Christopher Cox, Chairman of the SEC, suggested

using direct measures of narrative clarity to enforce plain English communication. SEC seems to

believe that “the jargon of lawyers has taken over” and the trend towards hard-to-read disclos-

ures is due to the fact that “the main purpose of the drafting exercise has shifted from informing

investors to insuring the issuer and the underwriter against potential claims” (SEC, 2007). In this

paper, we examine whether managers use of complex disclosures goes beyond the presence of

“legalese”, but also whether they use complex disclosure to hide information from investors.

The seminal work of Li (2008) explored the relationship between the readability of

annual reports and financial performance. Borrowing the Fog Index from computation

linguistics, where a higher reading on the Fog Index indicates disclosures that are more difficult

to understand, Li finds a negative relationship between Fog and the level of earnings. It is

unclear, however, whether this result is due to managers providing complex disclosures to

obfuscate bad performance or that bad news is simply harder to be communicated (Bloomfield,

2008). To further explore these alternative explanations, obfuscation or ontology, and to better

understand managers’ use of complex disclosures, we look at instances in which firms are more

likely to have managed earnings upwards to meet or beat an earnings target (Burgstahler and

Dichev, 1997). In these cases, although firms are releasing good news about meeting a

benchmark, they have incentives to hide the tools used to achieve it, as suggested by Warren

2
Buffett. In other words, when reported performance differs from underlying fundamentals, we

expect managers to try to make it harder for investors to identify such earnings management

behavior and the underlying performance. Our results suggest that the readability level of

financial disclosures goes beyond the one derived from the ontological explanation of good vs.

bad news being disclosed. Instead, we find that managers strategically use corporate disclosure

to mislead or to influence investors’ understanding of firm’s value.

Our study is motivated by the importance and richness of the textual component of

financial disclosures. The SEC highlighted the importance of textual disclosures when it issued a

set of rules requiring plain English disclosures. Christopher Cox, Chairman of the SEC, went

further and suggested “just as the Black-Scholes model is commonplace when it comes to

compliance with the stock option compensation rules, we may soon be looking to the Gunning-

Fog and Flesch-Kincaid models to judge the level of compliance with the plain English rules.” If

readability is going to be used as a measure of compliance, then we should understand the factors

that affect how managers choose the level of readability.

Our analysis focuses on the readability of the management discussion and analysis

(MD&A) section of the annual report—a section that is required by law but also a medium in

which managers have discretion over how to present an explanation of the company’s business,

financial conditions, and results of operation. As opposed to conference calls and press releases,

the structure and content of MD&As are fixed; consequently managers are “legally obligated to

touch upon”(Warren Buffett) subjects they likely avoid in other disclosures.

The earnings benchmark we use is the prior year’s earnings (rather than earnings

forecasts or zero earnings) because anecdotal evidence suggests that management’s discussions

in the annual report are more likely to compare and contrast performance in the current fiscal

3
year with that in the prior year (or years). Forecasted earnings, whether by sell-side analysts or

by management, are seldom referenced in annual reports. Zero and small positive earnings

events are relatively infrequent, so we reserve this benchmark for supplemental analyses. 1

Moreover, small or zero earnings changes mean that performance this year was similar to that in

the previous year so little explanation is expected by readers and provided by management. This

idea is the basis for our null hypothesis–firms should provide disclosures that are easy to

understand when their performance does not change much from previous year.

The findings are consistent with our hypotheses. Controlling for the relationship between

Fog and the overall earnings level as well as other known factors, we find robust evidence that

Fog is higher for firms that meet or just beat prior year’s earnings (MBE). We further identified

firms that are more likely to have managed earnings to meet a benchmark. We use model-free

methods, and several accruals and real activities methods to identify firms that are more likely to

have managed earnings. We find that firms more likely to have managed earnings to meet the

benchmark by a couple of cents provide more complex disclosures than firms that either miss the

benchmark or were less likely to have engaged in earnings management to exceed the

benchmark; this finding holds both when the comparison group is the broad cross-section of

firms as well as when the comparison group is a matched group of firms with characteristics

similar to the firms with suspected earnings management.

Our conclusions remain the same when we add firm fixed effects that control for firm-

specific characteristics, such as the complexity of the firm and the industry, which may impact

1
Three examples showing that firms focus on the zero earnings change benchmark in their MD&A are as follows
(underlining added). Standard Pacific Corporation (31/12/2011): “For the year ended December 31, 2011, we
reported a net loss of $16.4 million, or $0.05 per diluted share, compared to a net loss of $11.7 million, or $0.05 per
diluted share, in 2010.” Human Genome Sciences (31/12/2005): “Net Income (Loss). We recorded a net loss of
$239.4 million, or $1.83 per share, for the year ended December 31, 2005, compared to a net loss of $242.9 million,
or $1.87 per share, for the year ended December 31, 2004.” TII Network Technologies (31/12/2009): “Net income
in 2009 was $73,000 or $0.01 per diluted share, compared to net income of $578,000 or $0.04 per diluted share in
2008.”

4
the complexity of the disclosure. In another specification check, we include last year’s

complexity (lagFOG) in our analysis. Both specifications account for the high persistence of

FOG, and our conclusions remain the same.

In two further tests, we use firm-years with earnings that were misstated or required

subsequent restatement as clear indicators of earnings management. In these smaller samples, we

find that Fog is higher for firm-years with earnings that were subsequently found to be misstated

or had to be restated.

We also investigate whether the higher FOG in the suspect firms is due to more

information being provided. In line with Li (2008), a longer MD&A section could suggest more

being explained, or more obfuscation. We find no significant change in the length of the MD&A

for the group of firms that are more likely to have managed earnings to meet the benchmark.

However, we find shorter MD&As for firms that meet the benchmark but with no earnings

management. This is consistent with our null hypothesis that when performance is similar to

previous year and no need for obfuscation, firms provide simpler disclosures. These findings are

consistent with and reinforce our Fog analysis: the suspect firms do not have longer disclosures,

but they do have more complex disclosures, so they are not communicating more information,

but saying about the same amount in more complex ways.

Although MD&A are required disclosures that the SEC views as essential to investor

understanding of firms’ operations and performance, one may also argue that these disclosures

are not relevant to investors because of their lack of timeliness, for instance relative to

conference calls. If true, this argument also suggests that managers will care more about beating

analysts’ expectations than the past year’s earnings number. To address this issue, we focus on

firms with no analyst following because for such firms, management is unconcerned with

5
meeting analysts’ expectations and therefore last year’s performance is likely to be a more

important target. We find that the incremental disclosure complexity for suspect firms is three

times larger for firms with no analyst following compared with firms that do have analyst

following. Secondly, we split our sample based on the date when the MD&A is made available

to investors. We group firm-years in which MD&As were made available in the same week as

earnings were first announced. We find that the incremental level of complexity found in our

previous tests is three times larger when MD&As are released within seven days of the public

announcement of earnings.

Our contribution to the literature is two-fold. First, we add to our understanding of the

determinants of readability. More specifically, we refine the overall relationship between

readability and financial performance. While the overall pattern documented by Li (2008) is one

where higher earnings associates with lower Fog, we provide evidence that this relationship is

discontinuous (or at least non-monotonic) around the benchmark of the prior year’s earnings,

particularly for firms that are likely to have managed accruals. Second, we show that earnings

management, that is, using accounting discretion with the aim of concealing underlying

performance, manifests itself as more complex disclosures. Our evidence is consistent with

managers strategically choosing higher reporting complexity in conjunction with earnings

management in an attempt to conceal the latter, which is a new finding in the literature. Overall,

we add to a more complete understanding of the relationship between financial report readability

and reported performance.

Aside from Li (2008), this paper is also closely related to Larcker and Zakolyukina

(2012, “LZ”). However, LZ and this paper differ in a number of important ways. First, LZ’s

primary interest is finding linguistic predictors for financial restatements (i.e., restatements are

6
the dependent variable) whereas this paper’s primary interest is in understanding the

determinants of readability (i.e., the Fog score is the dependent variable). Second, whereas LZ

examine voluntary conference calls, we examine MD&A disclosures that are required by law.

Third, LZ analyze verbal communication but we examine textual disclosures, which result in

differing levels of preparation, forethought, and spontaneity. This last difference is important

because our hypothesis is based upon management’s deliberate attempt to obscure the financial

picture to hide earnings management, whereas LZ’s hypothesized effect derives in large part

from inadvertent signals conveyed (e.g., use of different pronouns, hesitations, expression of

anxiety). In sum, this paper differs from LZ in research question, whether the disclosure is

mandatory, as well as the degree of preparation possible for the different avenues of disclosure.

The remainder of this paper is organized as follows. The next section describes our

hypotheses. Section 3 contains our analyses and Section 4 concludes.

2. HYPOTHESIS DEVELOPMENT

In computational linguistics, the Gunning Fog Index, or just Fog Index, is a function of the

number of words per sentence plus the percentage of words that are complex (i.e., having three

or more syllables). This sum is scaled by a constant (0.4) such that the Fog value approximates

the number of years of formal education required to understand the text.

The Fog Index was first brought into the accounting literature by Li (2008), who

examined how readability of annual reports varies with financial performance. Li found a

negative relationship between profitability and Fog (i.e., profitable firms have less complex

reports compared with firms with losses). He also found that firms with more persistent positive

earnings have lower Fog.

In the discussion of Li (2008), Bloomfield (2008) provides a number of potential

7
explanations for the observed relationships between readability and financial performance. Two

are particularly salient here. First is obfuscation—that managers try to hide bad news by writing

text that is more difficult to decipher. Second is ontology—that bad news is inherently more

difficult to communicate.

Bloomfield provides two other potential explanations that could also be considered

variations of ontology. He suggests that loss firms need to provide more explanation as a result

of “management by exception.” He also suggests that the nature of accounting conservatism—

recognizing bad news in a more timely fashion than good news—requires managers to provide

more explanation about the future when there are losses. In sum, the ontological explanations

suggest that readability is inherently a function of the circumstances. In contrast, the obfuscation

explanation requires management to intervene and affect the disclosure, manifesting in more

complex disclosures. As will be seen below, our analyses will have bearing on these two

explanations.

As suggested by Li (2012), we utilize the fraud triangle framework to further examine

managerial use of complex disclosures. The fraud triangle suggests that fraud is more likely to

occur when there is “incentive or pressure” to misreport earnings, when there is “opportunity” to

do so, and management has the “attitude or rationalization” for the fraud or misreporting.

Incentive or pressure to misreport: There is a substantial literature documenting the

frequency, motivations, and benefits that accrue to firms that are able to meet or beat

benchmarks. In recent decades, some two-third to three-quarters of firms will meet or beat

expectations in the capital market (as proxied by analyst forecasts). The rewards of doing so are

higher stock returns, lower information asymmetry, and lower cost of capital (Bartov et al., 2002;

Brown et al., 2009). These firm-level effects translate into personal benefits via executive

8
compensation directly through higher stock and option value, or indirectly through discretionary

bonuses. Bhojraj et al. (2009) show that firms that manage earnings via accruals or real activities

to meet a benchmark gain short-term benefits over firms that missed the benchmark and chose to

not manage earnings.

Opportunity: Management has the opportunity to misreport by managing accruals and

real activities, as well by other means such as balance sheet or cash flow management. In this

study, we focus on the first two because these activities are most directly related to meeting

earnings benchmarks. Cohen and Zarowin (2010) and Zhang (2012) show that firms use both

accruals and real activity earnings management to achieve benchmarks, and make tradeoff

between them.

Attitude or rationalization: As Li (2012, 398) suggests, management disclosures via

MD&A or other means reflects the attitude of management or provides the avenue for

management to rationalize its behavior. Together, our hypotheses and analyses combine the

elements of the fraud triangle, by analyzing whether the complexity of MD&A disclosures

(attitude / rationalization) relates to whether a firm meets or just beats an earnings benchmark

(incentive or pressure) and whether the firm likely used earnings management to meet or beat

that benchmark (opportunity).

Depending on the context, some benchmarks will be more salient than others. In the

capital market context, the expectations in the market is the most relevant—meeting or falling

short of the market’s expectations is what determines changes in stock prices. In other instances,

zero earnings is the relevant benchmark—maintaining a positive level of earnings is important

for reasons of contractual provisions and general loss aversion, for examples. A third benchmark

is the prior year’s performance, which is equivalent to a benchmark of zero change in earnings.

9
 We focus on the third benchmark for two reasons. First, our analysis of readability

focuses on annual reports, and the MD&A section in particular. MD&A is a disclosure required

by the Securities and Exchange Commission (SEC).2 As a regulated disclosure, it is reasonable

to expect management to discuss facts and figures that are already contained in the audited

financial statements and elsewhere in the annual report rather than information from the capital

markets, such as analyst forecasts, which can change frequently. Second, the requirement to

comment on trends suggests that management would rather have a zero or positive earnings

change rather than having to explain a decline in earnings, which could arguably be interpreted

as the beginning of a downward trend. Third, we focus on the zero earnings change benchmark

rather than the zero earnings benchmark to obtain more time-series variation in meeting/beating

vs. missing the benchmark. That is, some firms are persistently profitable while others are

persistently not; the number of firms at or close to zero profitability is relatively small.

(Nevertheless, we provide some supplementary tests of this alternative benchmark to augment

our main analyses.)

While the relationship between Fog and earnings levels documented by Li (2008) is

negative overall, we expect that firms at or just above the zero earnings change benchmark will

tend to show a different relationship under the obfuscation explanation. First, if the ontology

explanation (our null hypothesis) holds, earnings that met or just beat the prior year should have

disclosure that is less complex than for earnings that fell below the prior year’s. In addition,

reported results that are close to the prior year’s require little explanation and is likely to reduce

the complexity of the MD&A, and possibly even relative to earnings that beat the prior’s by a

larger amount, which would prompt more complex explanations.

2
SEC Regulation S-K, Item 303 specifies the MD&A requirement. Among other things, it requires registrants to
discuss financial condition, results of operation, and “currently known trends, events, and uncertainties …”
(Securities Act Release No. 6835, May 18, 1989).

10
 Some of the firms that meet or just beat prior year’s earnings accomplish this outcome by

engaging in upward earnings management, so it can be argued that the underlying performance

that they would have otherwise reported would have been lower, which would have a

commensurately higher Fog. That is, all else equal, the readability for the underlying

performance is lower than for the reported performance. However, management discussion is

unlikely to dwell on the underlying and unreported performance, particularly if management has

engaged in earnings management to meet/beat the benchmark. Rather, management will discuss

the earnings as presented, which is earnings that met or beat the past year’s level. In any case,

this ontological effect, if it exists in our research context, would be small because our analysis

focuses on a small range of earnings (i.e., earnings changes close to zero).

Second, we expect firms that are likely to have engaged in earnings management to meet

or beat prior year’s earnings to actively make disclosures more difficult to understand. Within

the accounting discretion available, management makes biased choices to increase earnings.

Management must try to hide the deception so as not to be discovered; i.e., earnings

management cannot be transparent for management to believe that it would have the intended

effect (Lo, 2008). Since analysis of financial information, whether numerical or qualitative, is

costly, we expect investors to analyze information only up to the point where marginal costs

equal marginal benefits of further analysis (Gross and Stiglitz, 1980; Bloomfield, 2002).3

Therefore, more complex disclosures increase the cost of analysis to investors, reducing the

depth of analysis, and decrease the likelihood that earnings management will be detected. We

3
A number of papers show evidence consistent with this theory that complexity reduces analysis. You and Zhang
(2009) show that investors underreact to information content in 10-K filings when the textual narrative is harder to
read. Lehavy et al., (2011) show analysts incur more effort when the clarity of annual reports is low. They also show
that investors demand more analyst services and information environment is of lower quality when 10-Ks are
complex. Lawrence (2013) finds that individual investors benefit from easy to read disclosures and the information
disadvantage to institutional investors is alleviated.

11
consider this to be the direct effect of obfuscation on readability.

Third, in addition to the direct effect on readability is the indirect effect that arises

subconsciously. Earnings management involves degrees of untruth. Deceptive communication is

linguistically more complex and also cognitively more demanding.4 This increase in disclosure

complexity is a subconscious result of cognitive dissonance between underlying and reported

performance (Li, 2012).

Potentially counteracting these effects is that liars will tend to tell simpler stories because

it is difficult to be untruthful. Simpler stories reduce the chance that fabricated details will

conflict with each other. This tendency to tell simpler stories can counteract the tendency of lies

to be linguistically more complex. Whether this effect dominates the direct and indirect effects

just discussed is an empirical question.

Based on the above discussion, our central hypothesis is as follows (in alternative form):

H1: Firms that have managed earnings in a particular year have annual report disclosures in
that year that are less readable, ceteris paribus.
Beyond this central hypothesis, we have three subsidiary hypotheses with increasing specificity:

H1A: Firm-years with zero or slightly positive earnings changes will have annual report
disclosures that are less readable, ceteris paribus.
H1B: Firm-years with (i) zero or slightly positive earnings changes and (ii) income-increasing
discretionary accruals or real activities earnings management will have annual report
disclosures that are less readable, ceteris paribus.
H1C: Firm-years with (i) zero or slightly positive earnings changes and (ii) high income
increasing discretionary accruals or real activities earnings management will have annual
report disclosures that are less readable, ceteris paribus.
These three predictions relate to firm-years that we suspect of containing managed earnings.

However, these classification schemes use purely quantitative data and there is likely a certain

amount of misidentification (Dechow et al. 1995). Therefore, we use the superior accuracy of

4
“… from a cognitive perspective, truth tellers should be able to discuss exactly what did and did not happen
because they were actually there to witness the event being discussed. Liars, on the other hand, would be forced to
keep track of what they have previously said to avoid contradicting themselves later” (Hancock et al., 2007).

12
human investigators to identify instances of earnings management. Therefore, we also make two

additional predictions using restatements and misstatements.

H1D: Firm-years with financial information that was subsequently restated will have annual
report disclosures that are less readable, ceteris paribus.
H1E: Firm-years with misstatements will have annual report disclosures that are less readable,
ceteris paribus.
Restatements are adjustments to financial statements as tracked by Audit Analytics; we do not

examine all restatements, but only those identified by Audit Analytics as relating to fraud or SEC

investigations. Misstatements are those reported in Accounting and Auditing Enforcement

Releases (AAER), which may or may not subsequently result in restatements. Using restatements

and misstatements is more accurate in identifying instances of earnings management, but also

potentially misses some earnings management activity that has not been the subject of regulatory

action. In this regard, this identification strategy is complementary to the earlier strategy using

small positive earnings changes, which likely classifies too many firms as having managed

earnings.

3. RESEARCH DESIGN

To test our hypotheses, we use a sample of firm-years with available data between 2000 and

2012. We require financial data from Computat as well as MD&A disclosures on the SEC’s

Edgar system. Details of required financial data items are provided below. We exclude firms in

the utilities and financial services industries (SIC 4400-5000 and 6000-6999) because of their

different operating and financial structures. Table 1 shows the results of the sample selection

procedure. The final sample consists of 26,967 firm-years and 4,855 unique firms.

Generally, our hypothesis concerns the influence of earnings management on readability.

Therefore, we require measures of readability, earnings management, and control variables that

are known to affect readability. Therefore, the general form of equation we use to test this

13
hypothesis is as follows:

∑ (1)

where EM refers to the earnings management proxy. The follow discussion provides additional

details for this equation.

3.1 Readability

We use the Gunning Fox Index to measure readability. As mentioned above, the Fog Index is

computed as follows:

( ). (2)

The number of words per sentence is computed as the ratio of the total number of words divided

by the number of sentences in the MD&A. Complex words are those having three or more

syllables. Longer sentences and a higher proportion of complex words increase Fog, meaning a

reduction in readability. The Fog Index has been used widely and has seen increasing usage in

the accounting literature (e.g., Miller, 2010; Lehavy et al., 2011; Rennekamp, 2012).

3.2 Earnings management

We use a number of different proxies for earnings management. Our first and simplest measure

uses the approach of Burgstahler and Dichev (1997): we identify firms having a higher

likelihood of managing earnings as those firms with earnings in the neighborhood of meeting or

just beating past year’s earnings. We conduct our main tests using earnings per share (EPS), but

we also present results for earnings deflated by total assets. In either case, we measure earnings

before extraordinary items. We define the variable MBE = 1 if ΔEPS falls in the neighborhood

from zero to a small positive number; otherwise MBE = 0. We use a range of values to define the

“small positive number” to ensure robustness of results.

The MBE measure is based on the outcome of earnings management, and misclassifies

14
firms that have earnings in the neighborhood just above the MBE benchmark even in the absence

of earnings management. Therefore, we also examine the process of earnings management,

namely using discretionary accruals or real activities.

For discretionary accruals, we use the Jones (1991) model in our main tests:

⁄ ( ⁄ )+ ( ⁄ ) ( ⁄ ). (3)

where TotAccrt are total operating accruals, ΔRevt is the change in revenues from year t-1 to t,

PPEt is gross property, plant, and equipment, and TAt-1 is total assets at the end of year t-1. We

estimate the model in cross-section by industry and year, and require at least 15 observations.

The residuals from this estimation form the discretionary accruals (DA). In supplementary tests,

we also modify the Jones model as suggested by Dechow et al. (1995) with similar results as

reported below. We rerun our analyses using performance-matched model of Kothari et al.

(2005). In particular, we calculate expected accruals based on 1-to-1 matching on industry (using

the Fama-French 48 industries), fiscal year, and closest previous-year return on assets (ROA,

measured as net income divided by lagged total assets). Results remain similar to those reported

below.

For real activities earnings management, we focus on discretionary expenses

(Roychowdhury 2006), specifically research and development (R&D) and advertising expenses.

We define real activities earnings management (RAM) as the negative sum of (ΔR&D expense +

ΔAdvertising expense), deflated by beginning total assets. Larger reductions in R&D or

advertising expenses result in more positive values of RAM.

Discretionary accrual models have large amounts of measurement error and suffer from

low power (Dechow et al., 1995), and similarly for measures of real activities earnings

management. We do not simply use these measures to proxy for earnings management. Rather,

we interact them with our first measure, MBE, to increase the power of detecting firms that have

15
managed earnings, because together the variables capture both the process and outcome of

earnings management. Furthermore, managing earnings to meet or beat past earnings presumably

involves upward (not downward) earnings management, so we identify firm-years with positive

earnings management using the indicator variables PosEM(DA) and PosEM(RA) for,

respectively, accrual and real activities management that increases income; the complement is

NegEM(). Therefore, our second measure to identify firms that are likely to have managed

earnings upwards is MBE × PosEM(DA) and MBE × PosEM(RA). We also construct a composite

variable to capture both types of earnings management: PosEM(Comb) = PosEM(DA) +

PosEM(RA).

Our third measure of earnings management refines the second one just described but uses

not only the sign of the earnings management, but also the magnitude. We separate firms with

positive discretionary accruals into high and low partitions using the median value, resulting in

HighPosEM(DA) and LowPosEM(DA); similary for positive real activities earnings

management. Our third measure of firms most likely to have managed earnings upwards is thus

MBE × HighPosEM(DA), MBE × HighPosEM(RA), and MBE × HighPosEM(Comb).

Our fourth proxy for earnings management is whether a firm-year’s financial statements

had been restated, as identified in Audit Analytics. Because some restatements result from

relatively benign reasons, we focus on restatements that Audit Analytics has flagged as (i)

resulting from fraud or (ii) being initiated by or resulted in an SEC investigation. Our fifth proxy

is whether a firm-year’s financial statements were misstated according to AAER.

3.3 Control variables

Our list of control variables derives from Li (2008). The most important of these in our context

are the earnings-related variables. The first is Earnings, defined as operating earnings deflated by

16
beginning total assets, which is expected to be negatively associated with Fog (i.e., firm-years

with high earnings have more readable MD&A). We also control for firm-years with losses with

the indicator Loss, which equal 1 when Earnings < 0, because losses require additional

explanations about the viability of the business, which is likely to make disclosures less readable

(Li 2008).

We include all of the 12 other control variables used in Li (2008). We provide details for

these variables in Appendix A.

3.4 Descriptive statistics

Table 2 shows the sample statistics. The mean and median value of Fog is around 18, meaning

that the typical MD&A requires two years beyond an undergraduate degree. This value is similar

to that reported in Li (2008), where the mean and median were 18.23 and 17.98, respectively. On

average, firms increase EPS by five to six cents year-over-year. Panel B and C show that, on a

univariate basis, firm-years with restatements or misstatements have higher Fog on average,

consistent with predictions.

Figure 1 provides a visual depiction of the level of Fog at various values of change in

EPS. The baseline value for comparison is a change in EPS exceeding $0.03 (i.e., good news),

which has an average Fog value of 17.99. Compared to this base value, large negative changes in

EPS (ΔEPS < -$0.03) have a negligible 0.01 higher readings of Fog. Firm-years with smaller

negative change in EPS have MD&A that is on average 0.34 higher in Fog. Most importantly,

MD&A for firms that have just met or slightly beat the past year’s EPS have the highest Fog,

specifically those observations where the firms are most likely to have managed earnings to

surpass the prior year’s EPS performance. The Fog value is 0.66 higher than the baseline of

17.99, and 0.32 higher than firms just missing the benchmark.

17
 Table 3 shows the correlation matrix for the variables. Fog negatively correlates with

firm size (Size) and number of geographic segments (NGSeg) and positively correlates with

market-to-book (MTB), with magnitude around 10%. The signs of these correlations are

consistent with Li (2008).5 While there are many correlations among the variables that are

statistically significant due to our large sample size, the magnitudes are mostly modest and

should not pose a problem with multicollinearity.

3.5 Results – Test of Hypothesis 1A

Table 4 presents our first test of Hypothesis 1. The definition of EM in Equation (1) in this

analysis is whether a firm met or just beat past year’s earnings. That is, we identify firms with

zero or small positive earnings changes as having a higher likelihood of having managed

earnings. We use three different definitions of “small positive,” being one, two, or three cents of

EPS before extraordinary items (Frankel et al, 2010).

The regression specification results in three groups in the analysis: i) firm-years with

earnings lower than in the previous year (identified by the indicator variable NegEarnChg), ii)

the group that met or just beat the past year’s earnings by a small amount (identified by the

indicator variable MBE), and iii) firm-years that beat the past year by a greater amount (more

than one cent, two cents, or three cents, depending on the regression specification). The structure

of the regression results in the third group forming the baseline for comparison.

Panel A presents the results from the full sample. The key variable is MBE, which has a

significantly positive coefficient (coefficient = 0.162 to 0.203, t = 2.39 to 3.74). Firms that met

or just beat prior year’s EPS had MD&A that were more complex to read by about a fifth or sixth

of a year of formal education as compared to firms that beat the benchmark by more than [one |

5
Li (2008) did not report a correlation table. We use the multivariate regression report in Li’s Table 2 to compare
without bivariate correlations, even though strictly they are not comparable.

18
two | three] cents, depending on the specification. When we compare MBE firm-years against

those with performance lower than prior year’s, the differences (β1 - β2) are statistically

significant at 5% level for all three cases (F(1, 38) = 7.11 to 15.48). This analysis controls for

other determinants of Fog identified in Li (2008, Table 2), including loss (Loss) and operating

earnings scaled by total assets (Earnings), as well as fixed effects for (Fama-French 48) industry

and year. Standard errors are clustered by industry as in Li (2008). As expected, the coefficient

for Loss is positive and for Earnings is negative, meaning that more profitable firms have less

complex MD&A.

While the significantly positive coefficient on MBE in Panel A indicates that firms that

just beat expectations have MD&A that are less readable compared to a broad cross-section of

firms, our analyses do not necessarily compare MBE firms against firms with similar

characteristics. To address this ambiguity, we repeated our analyses using smaller matched

samples, with results shown in Panels B and C of Table 4. For Panel B, we did 1-to-1 matching

based on the Fama-French 48 industries, fiscal year, market-to-book deciles, and the closest

market capitalization.

In Columns I to III, the matched sample, our baseline in this analysis, consists of firms

that missed the benchmark (i.e., ΔEPS < 0 or NegEarnChg=1). We find that even though MBE

firms are disclosing better news, they provide more complex disclosures as compared to similar

firms that missed the benchmark. In Columns IV to VI, the matched sample consists of firms that

beat the benchmark by more than the MBE observations (i.e., ΔEPS > [$0.01 | $0.02 | $0.03]). The

results are consistent with those in Panel A, with coefficients and statistical significance both of

comparable magnitudes.

For Panel C, we repeat the analysis with two changes in the matching algorithm: using a

19
matching based on industry, fiscal year, size deciles (vs. market-to-book decile above), and the

closest sales growth (vs. closest market capitalization above). Again, the results remain

consistent with the exception of Column I.

Overall, this first set of results—firms with relatively good news of a small positive

earnings change having MD&A sections that are on average less readable than both firms

disclosing worse news and firms disclosing better news—reveals a pattern different from the

general pattern found in Li (2008) and is inconsistent with a purely ontological explanation.

To address the scale issue related to using EPS, we show results of measuring

profitability as earnings before extraordinary items deflated by total assets. Table 5 Panel A

shows the results of this analysis, with MBE defined to equal 1 when deflated earnings is from

zero to 0.4%, 0.5% or 0.6% (Columns I to III). For brevity, we do not report the coefficients for

the control variables. Similar to Table 4, we find results as predicted. The coefficient on MBE

ranges from 0.120 to 0.132, somewhat smaller than in Table 4 but significant at the 1% level or

better (t = 2.63 to 3.26). When compared to firm with negative earnings changes (NegEarnChg),

the complexity of MD&A from MBE firms is also still statistically larger (F(1, 38) = 5.14 to

7.77).

We also conduct a placebo test to confirm that the results found are not spurious due to

persistent firm-specific factors. In particular, we want to reduce the possibility that some

unmodelled factors omitted from our analysis but correlated with the propensity to meet/beat

explain the variation in MD&A readability. To this end, we identify firms that just met or beat by

one, two, or three cents the benchmark in the prior year. We include MBEt-1 to see if meeting or

beating in the prior year is at all associated with current MD&A readability. Finding that MBEt-1

firms provide at time t disclosures more complex than others in the same group would suggest

20
the existence of persistent firm-specific factors related to readability. Table 5 Panel B shows that

the coefficient on MBEt-1 is not significantly different from zero (t = 0.542 to 1.364), while the

coefficient on MBEt continues to be significantly positive and with similar magnitudes as in

Table 4. This placebo test suggests that the result on MBE found in Table 4 is unlikely to be

spuriously caused by omitted variables.

A third test that we conduct to confirm our results in Table 4 is to increase the precision

of the MBE variable to distinguish firms that beat the benchmark with earnings management

from firms that do so without earnings management. (Later analyses in Tables 6 and 7 will also

do this using models of discretionary accruals and real activities; the test here uses the properties

of the earnings themselves.) We identify firms that just met or beat the past year’s EPS, with the

added condition of EPS in the first three fiscal quarters in the current year having fallen behind

the amount in the first three quarter of the preceding year. We define this variable as MBE_ByQ4

= (EPSt-1, Q1 + EPSt-1, Q2 + EPSt-1, Q3) - (EPSt, Q1 + EPSt, Q2 + EPSt, Q3 ), when part of the MBE

group, and 0 otherwise. The additional condition increases the likelihood of identifying firms

that managed earnings upwards in the final quarter of the fiscal year to meet or beat the annual

earnings of the prior year, since they were behind in the first three quarters, but managed to meet

or beat the annual figures after the fourth quarter. Panel C of Table 5 contains the results of this

analysis. Again, the variable of earnings management, MBE_ByQ4, is significantly positive

(coeff = 0.22 to 0.25, t = 2.19 to 2.40).

Overall, the adjusted R-squared values in both Tables 4 and 5 are around 13% (and

higher in the matched samples), which are similar but slightly higher than the 10% reported in Li

(2008). The results so far support Hypothesis 1A. We find consistent evidence that, while higher

earnings is associated with more readable MD&A, losses are associated with less readable

21
reports, but meeting or just beating is also associated with less readable reports. The ontology

explanation suggests that it is possible and even likely that good news is inherently easier and

bad news is inherently more difficult to explain. Our evidence suggests that the good news of

meeting or just beating past earnings is also harder to explain, which is contrary to the

ontological explanation. This result could arise if the good news is artificial and some amount of

obfuscation is required to make the underlying performance less transparent. We investigate this

possibility further in the next tests of H1B and H1C.

3.6 Results – tests of Hypotheses 1B and 1C

Considering only whether a firm met or just beat the prior year’s earnings as we did in the test of

H1A involves misclassifying firms that would have met or just beat prior year’s earnings without

any earnings management. To improve the identification of firms that are more likely to have

managed earnings, we consider the process of earnings management used to achieve that

earnings outcome. We estimate firms’ discretionary accruals and real activity earnings

management in addition to whether it met or beat prior earnings. Table 6 shows the results of

regressing Fog on MBE interacted with PosEM( ), an indicator variable identifying positive

earnings management, as well as the main effect for PosEM( ) and control variables. Columns I

to III show the results for the case of earnings management using discretional accruals (DA). The

results are as predicted: MBE × PosEM(DA) is significantly positive (coefficient = 0.29 to 0.33, t

= 3.72 to 5.28). Firms that are likely to have used earnings-increasing discretionary accruals to

meet or beat prior earnings have less readable MD&A than firms that beat the benchmark by

more than [one | two | three] cents (baseline), and also less readable than firms that miss the

benchmark (β2 – β4), (F(1, 38) = 15.45 to 27.76). The magnitudes of the coefficient are about 1.5

to 2 times as large as those estimated in Table 4 for MBE alone without considering accruals.

22
Consistent with this larger magnitude, we find the coefficient for MBE × NegEM(DA) to be

insignificant; firms that met or just beat prior earnings but have negative discretionary accruals

do not exhibit more complex MD&A than firms that beat the benchmark by more than [one | two

| three] cents (baseline). A comparison between the MBE firms with negative earnings

management against MBE firms with positive earnings management (β2 – β3; coefficient on MBE

× PosEM(DA) – coefficient on MBE × NegEM(DA)) shows statistical significance at the 10%

level in all three specifications (F(1, 38) = 2.85 to 5.49). Thus, the effect found in Table 4 is

concentrated in the subset with positive discretionary accruals. This evidence supports

Hypothesis 1B.

The results for real activities earnings management are similar, shown in Columns IV to

VI in Table 6. The coefficients on MBE × PosEM(RA) are similar to those for accruals

management. We find that MBE firms exhibit more complex MD&As than firms that beat by

more than the [one | two | three] cents, and more than firms with negative earnings change (β2 –

β4; coefficient on MBE × PosEM(DA) – coefficient on NegEarnChg) at a significance level of

1% (F(1, 38) = 11.63 to 13.56). The magnitudes of the coefficient of MBE x Pos(RA) are about

1.5 to 2 times as large as those of MBE x Neg(RA), however, the difference is statistically

significant only in Column IV (F(1, 38) = 2.91).

Combining the two sources of earnings management, Columns VII to IX show similar

results with significantly positive coefficients on MBE × PosEM(Comb). Note that the

coefficient value is somewhat smaller (0.22 to 0.24), but factoring in the variable’s range of 0 to

2 (instead of 0 to 1), the magnitude of the effect is actually higher than that shown in Columns I

to VI. Conclusions remain similar when we compare MBE firms against firms with negative

earnings changes (β2 – β4; coefficient on MBE × PosEM(DA) – coefficient on NegEarnChg)

23
(F(1, 38) = 21.03 to 31.39) and against MBE firms with no suspected upwards earnings

management (MBE × NegEM(Comb))(F(1, 38) = 3.77 to 8.95).

Next, we test if the level of complexity in the MD&A is increasing in the magnitude of

positive earnings management (H1C). To do so, we further refine the definition of earnings

management by identifying firms that have highly positive vs. less positive earnings

management by splitting at the median value of PosEM( ), resulting in HighPosEM( ) and

LowPosEM( ). In Table 7, coefficients for MBE × HighPosEM(DA) are significantly positive

(coeff = 0.39 to 0.42, t = 3.18 to 6.33), indicating higher complexity as compared to firms that

beat the benchmark by more than [one | two | three] cents, supporting H1C. Coefficients for MBE

× HighPosEM(DA) are also significantly larger than those of firms that missed the benchmark

(β2 – β5, F(1, 38) = 11.72 to 46.76), than those of firms that meet or just beat the benchmark with

no suspected accruals earnings management (β2 – β4, F(1, 38) = 5.95 to 6.62), and than those of

firms that meet or just beat the benchmark with low-level of accruals earnings management (β2 –

β3, F(1, 38) = 5.82 and 6.84), with the exception of the one in Column I (F(1, 38) = 1.73). The

magnitudes are larger than the corresponding coefficients from Table 6 by about one-third, and

about twice as large as that reported in Table 4, consistent with our expectations. The

coefficients for MBE × LowPosEM(DA) are also significantly positive in two of the three

specifications, but with magnitudes only about half as large as for the firms with high positive

discretionary accruals.

Columns IV to VI repeat the analysis with real activities earnings management. The

coefficients for MBE × HighPosEM(RA) are significantly positive (coefficient = 0.50 to 0.72, t =

2.82 to 3.95) and statistically larger than those of any other group. Columns VII to XI combine

the two sources of earnings management, with comparable results (coefficient = 0.36 to 0.41, t =

24
5.10 to 6.78). These results suggest that those firms most likely to have used income-increasing

discretionary accruals to meet or beat prior performance have MD&A reports that are more

complex by about 4/10th to 8/10th of a year of formal education when compared to firms that

either miss the benchmark, or beat the benchmark by more than [one | two | three] cents, which

is a material difference.

To obtain an alternative look at the differences in Fog as a function of earnings

management, we partition the sample based on the extent of accrual or real earnings

management. In Table 7, Panel B, the first group comprises all firm-years with no suspected

upward earnings management by accruals or real activities (Columns I to III). The second group

comprises firm-years with low levels of upward earnings management using accruals or real

activity (Columns IV to VI). The third group comprises firm-years with high level of upward

earnings management using either accruals or real activity (Columns VII to IX). When we

compare firms with no suspected upward earnings management (i.e., when DA < 0 and RAM <

0) in Column I to III, the MBE firms have a similar level of readability as non-MBE firms.

Moving to cases with low positive discretionary accruals or low real earnings management in

Columns IV to VI, the coefficients increase in magnitude and are significant in some

specifications. For cases with high positive earnings management in Columns VII to IX, the

coefficients on MBE are the highest and statistically most significant (coeff = 0.36 to 0.42, t =

3.2 to 5.34). We reach similar conclusions when we compare MBE firms against firms that

missed the benchmark (β1 – β2, (F(1, 37) = 17.32 to 35.63).

3.7 Supplemental robustness tests

We conduct six robustness checks to confirm our results above. For brevity, we present results

based on the most specific definition of earnings management that we just used to test H1C (i.e.,

25
using MBE × HighPosEM( ) ). Also, we present results only for MBE defined to be ΔEPS [$0,

$0.03]. Table 8 shows these results. Panel A shows regression estimates when we include firm-

level fixed effects (Column I) or when we include lag(Fog) as an independent variable (Column

II) to control for persistence in readability. These specifications are likely to be overly

conservative due to overfitting; indeed, the explanatory power is around 70% in R2.

Nevertheless, the coefficient on MBE × HighPosEM(Comb) remains significantly positive,

although the magnitude of the coefficient is lower compared to those in the previous tables.

When compared to firms that missed the benchmark, we find that firms that are more likely to

have managed earnings (MBE × HighPosEM(Comb)) exhibit more complex disclosures even

though they are disclosing good news. The coefficient on MBE × LowPosEM(Comp) is no longer

significant.

Panel B explores the results’ robustness to using other discretionary accrual models.

Column I uses the modified Jones model of Dechow et al. (1995) to measure discretionary

accruals, and the results remain essentially the same as those reported in Table 7, which uses the

simpler Jones (1991) model. Column II uses the performance-matched accrual model of Kothari

et al. (2005). Again, the results are very similar to those reported in Table 7 except that the

coefficient on MBE × LowPosDA is no longer significant, which is similar to what was found

with firm-level fixed effects in Panel A.

Panel C of Table 8 considers the zero-earnings benchmark, instead of the zero earnings

change benchmark we have examined in all the above analyses. We again observe very

consistent results: whether we look at discretional accruals, real activities management, or both

combined, the Fog score is higher for firms that meet or just beat the zero earnings benchmark

and have high amounts of income-increase earnings management (coefficient = 0.26 to 0.41; t =

26
2.31 to 4.43) as compared to firms that beat the benchmark by more than three cents. The

magnitudes of the coefficients are similar to that found previously (compare with Table 7A

Columns III, VI, and IX).

In Panel D of Table 8, we isolate the set of firms with earnings changes just above and

below zero for a more focused comparison. This is a small sample consisting of 1,501 firm-years

with EPS . When we compare firms with EPS vs EPS

)), we find significant positive coefficients on MBE × HighPosEM(), with

magnitude similar to that found above (coefficient = 0.259 to 0.435, t = 2.20 to 2.85)).

Furthermore, we find the coefficient values of the same magnitude as that shown in Figure 1,

which showed a difference in Fog of 0.32 between firms that MBE vs. firms that just fell short.

Panel E of Table 8 analyzes a subset of that used in Panel D with high positive earnings

management. Again, the results are similar.

Finally, in untabulated analysis, we also considered whether earnings management

systematically increased disclosure length. We repeat our analyses by substituting word count in

place of the Fog index as the dependent variable and did not observe significant results.

3.8 Additional tests – Evaluating the assumption that MD&A is important to investors and
management

A maintained assumption throughout our analysis is that investors do find useful information in

MD&A disclosures, and with rational expectations, management believes that the disclosures do

matter to investors. In other words, we assume management is not merely fulfilling a legal

obligation to provide the MD&A as part of its 10-K report filed with the SEC. This assumption

underpins (i) why management wants to meet or beat the previous year’s earnings, and (ii) our

hypothesis that management will try to obfuscate to hide any earnings management used to meet

or beat the previous year’s earnings. In this section, we provide evidence that this assumption is

27
justified.

First, we consider the timeliness of the MD&A. We posit that more timely disclosures

will be more useful to investors, and hence management will be more concerned about

establishing a pattern of stable or increasing earnings that is required as part of the MD&A. For

this analysis, we measure timeliness relative to the earnings announcement date. We identify

disclosures as timely if the company files its 10-K within 7 calendar days after the earnings

announcement, corresponding to the 5 trading days used in Li and Ramesh (2009). After

partitioning the data, we replicate the analysis reported in Panel A of Table 7.

In untabulated analysis, we find that the coefficient on MBE × HighPosEM() is larger in

magnitude when the MD&A is more timely compared with when it is more than 7 days after

earnings announcement date. This is the case when we look at discretionary accruals alone (2.6

times larger), real activities management alone (1.9 times larger), or both combined (1.4 times

larger),6 however the difference is only statistically significant for the case of earnings

management using accruals.

A second way we evaluate the maintained assumption of MD&A importance to investors

and management is to examine analyst following. Since analyst provide an alternate source of

information, we posit that MD&A disclosures are more important for firms without analyst

following. Furthermore, when there is no analyst following, management has one less

benchmark to beat, so the zero earnings change benchmark is likely to become more salient for

such firms.

We find that the coefficient on MBE × HighPosEM() is about twice as large for firms

without analyst following compared with those with analysts following, but again, only

significant for the case of accruals earnings management.

6
We find similar results with alternate cutoffs for timeliness of 0, 3, 5 days instead of 7 days.

28
 Overall, we find evidence supporting the validity of our maintained assumption that

MD&A disclosures are useful to investors, and it is rational for management to believe that

investors use those disclosures.

3.9 Restatement and Fraud Cases

Thus far, we have used quantitative techniques to identify firm-years that are more likely to

contain significant earnings management. Turning to the human-assisted identification of

earnings management, we now look at the results. Table 9 shows the results when earnings

management is identified by financial restatements that have been identified as fraudulent or

relating to SEC investigations. In the first specification (Column I), the indicator variable Restate

is significantly positive (coefficient = 0.25, t = 2.30), meaning that firm-years that were

subsequently restated have more complex MD&A. Now, since some restatements may be small

in magnitude, we incorporate the magnitude of restatement in the variable RestateSeverityDecile,

which ranges from 0 to 10, where 10 is the decile with the largest restatement and 1 the smallest,

and 0 for no restatement. Column II shows that the result using this variable is also significantly

positive (coeff = 0.036, t = 2.16). Column III looks at the AAER sample of misstatements.

Again, we find that firm-years with earnings management have less readable MD&A (Misstated

coefficient = 0.32, t = 2.08).

In sum, we find consistent and robust evidence supporting Hypothesis 1 with different

definitions of earnings management, different accrual models, controlling for different fixed

effects, using a placebo test, using mechanical as well as human detection of earnings

management.

4. DISCUSSION AND CONCLUSION

This paper extends the readability analysis of Li (2008). Beyond the overall negative relations

29
between the Fog Index and financial performance (i.e., the positive relationship between

readability and performance), we hypothesize and document a disruption to that overall pattern.

In the region where a firm meets or just beats the prior year’s earnings, the Fog score increases

and readability deteriorates. The effect is larger when we focus on subsets of firms within this

neighborhood of earnings performance that are more likely to have engaged in accruals or real

activities management to increase earnings. Overall, we find consistent and robust evidence that

firms that are likely to have managed earnings to meet or beat the benchmark of the prior year’s

earnings on average have more complex MD&A reports. We find similar results for firms that

meet or just beat the zero earnings benchmark.

If the ontological explanation proposed by Bloomfield (2008) were a sufficient

explanation (i.e., that good performance is inherently easier to communicate than bad

performance), then we should not observe the pattern we found. Our evidence suggests that, at

least for firms that are most suspected of having managed earnings, obfuscation is involved in

making the financial report more difficult to read.

Our results are also consistent with the commonly held belief, supported by empirical

evidence, that telling the truth is easier than telling lies (Hancock et al., 2007). Lying is difficult

if it is to be convincing because the communicator has to ensure the consistency of the purported

“facts.” While earnings management in many cases do not outright fall into the category of lying,

the activity does involve some active efforts on the part of management to bias the financial

statements through accruals or other means. Such actions create a discrepancy between

unmanaged performance and reported performance, creating cognitive dissonance, which can

make it mentally more taxing to explain reported performance when management knows the

underlying unmanaged performance to be different. Earnings management also challenges

30
managers’ ethical standards, which again can cause cognitive stress, which can be indirectly

connected to the readability of their writing.

While we believe that active obfuscation and cognitive dissonance contribute to

decreased readability, both causes are observationally equivalent in our analysis and we are

unable to distinguish between the two. Future research can seek to disentangle them.

Further research can also go beyond readability and explore how the specific content of

the MD&A relates to benchmark beating and earnings management. For instance, do firms

suspected of having managed earnings use different pronouns or a more passive writing style?

Another avenue is to investigate other earnings benchmarks and the corresponding disclosures

that would focus on those benchmarks (e.g., meeting or beating market expectations and

conference calls).

31
 Appendix A
Variable definitions
Readability variables

Fog = 0.4 × (words per sentence + percent of complex words).

Variables in test of Hypothesis 1

ΔEPS = change in EPS from year t-1 to t.

NegEarnChg = 1 if ΔEPS < 0.

MBE = 1 if ΔEPS falls in the neighborhood from zero to a small positive number; 0 otherwise.
(The small positive number is identified in each test.)

DA = discretionary accruals estimated using the Jones (1991) model; in robustness tests,
discretionary accruals are estimated using the modified Jones model of Dechow et al. (1995),
or the performance-matched model of Kothari et al. (2005).

PosEM(DA) = 1 if DA ≥ 0; otherwise 0. Indicates income increasing earnings management using

discretionary accruals.

HighPosEM(DA) = 1 if DA > median(DA | DA≥0); otherwise 0.

RAM = real activities earnings management = – (ΔR&D expense + ΔAdvertising expense) / total
assets.

PosEM(RA) = 1 if RAM 0; otherwise 0. Indicates income increasing earnings management

using real activities.

HighPosEM(RA) = 1 if RAM > median(RAM | RAM≥0); otherwise 0.

PosEM(Comb) = PosEM(DA) + PosEM(RA) = {0, 1, 2}. Identifies firm-years that have income
increasing earnings management using either discretional accruals or real activities, or both.

HighPosEM(Comb) = HighPosEM(DA) + HighPosEM(RA) = {0, 1, 2}.

Restated = 1 if restatement in Audit Analytics is identified as (i) relating to fraud or (ii) being
initiated by or resulted in an SEC investigation.

RestateSeverityDecile = decile rank of (total dollar change in Net Income due to the restatement
scaled by Total Assets)

Misstated = 1 if a firm-year was reported in an AAER (SEC Accounting and Auditing

Enforcement Release.)

Earnings = operating earnings deflated by total assets at the fiscal year-end.

32
Loss = 1 if Earnings < 0.

Size = log of market value of equity at fiscal year-end.

MTB = (market value of equity + book value of liabilities) / book value of total assets, measured
at the end of the fiscal year.

Age = number of years since a firm first appears in the CRSP monthly stock return file.

SpecItems = amount of special items divided by total assets.

EarnVol = standard deviation of operating earnings during the prior five years.

RetVol = standard deviation of monthly stock returns in the prior year.

NBSeg = natural log of the number of business segments.

NGSeg = natural log of the number of geographic segments.

NItems = number of items in Compustat with non-missing values.

M&A = 1 for firm-years in which a company is an acquirer according to SDC Platinum M&A
database; 1 otherwise.

SEO =1 for firm-years in which a company has a seasoned equity offering according to SDC
Global New Issues database; 0 otherwise.

Delaware = 1 if the firm is incorporated in Delaware; 0 otherwise.

33
 References
Bartov, E., D. Givoly, C. Hayn, 2002. The rewards to meeting or beating earnings expectations.
Journal of Accounting and Economics 33, 173-204.

Bhojraj, S., P. Hribar, M. Picconi, and J. McInnis, 2009. Making sense of cents: An examination
of firms that marginally miss or beat analyst forecasts. Journal of Finance 65, 2361-2388.

Bloomfield, R.J., 2002. The incomplete revelation hypothesis and financial reporting.
Accounting Horizons 16, 233-243.

Bloomfield, R.J., 2008.Discussion of “Annual report readability, current earnings, and earnings
persistence”. Journal of Accounting and Economics 45, 248-252.

Brown, S., S.A. Hillegeist, K. Lo, 2009. The effect of earnings surprises on information
asymmetry. Journal of Accounting and Economics 47, 208-225.

Burgstahler, D., I. Dichev, I., 1997. Earnings management to avoid earnings decreases and
losses. Journal of Accounting and Economics 24, 99-126.

Cohen, D., P. Zarowin, 2010. Accrual-Based and Real Earnings Management Activities around
Seasoned Equity Offerings. Journal of Accounting and Economics 50(1), May 2010: 2-19

Dechow, P.M., W.A. Richardson, and I. Tuna, 2003. Why are earnings kinky? An examination
of the earnings management explanation. Review of Accounting Studies 8, 355-384.

Dechow, P., R.G. Sloan, A. Sweeney, 1995. Detecting Earnings Management. The Accounting
Review 70, 193-225.

Frankel, R., W.J. Mayew, and Y. Sun, 2010. Do pennies matter? Investor relations consequences
of small negative earnings surprises. Review of Accounting Studies 15, 220-242.

Grossman, S.J. and J.E. Stiglitz, 1980. On the impossibility of informationally efficient markets.
The American Economic Review 70, 393-408.

Hancock, J.T., L.E. Curry, S. Goorha, M. Woodworth, 2007. On lying and being lied to: A
linguistic analysis of deception in computer-mediated communication. Discourse Processes
45, 1-23.

Jones, J., 1991. Earnings management during import relief investigation. Journal of Accounting
Research 29, 193-228.

Kothari, S.P., A.J. Leone, C.E. Wasley, 2005. Performance matched discretionary accrual
measures. Journal of Accounting and Economics 39, 163-197.

Larcker, D.F., and A.A. Zakolyukina, 2012. Detecting deceptive discussions in conference calls.
Journal of Accounting Research 50, 495-540.

34
Lawrence, A., 2013. Individual investors and financial disclosure. Journal of Accounting and
Economics 56, 130-147.

Lehavy, R., F. Li, and K. Merkley, 2011. The effect of annual report readability on analyst
following and the properties of their earnings forecasts. The Accounting Review 86, 1087-
1115.

Li, E.X., and K. Ramesh, 2009. Market reaction surrounding the filing of periodic SEC reports.
The Accounting Review 84, 1171-1208.

Li, F., 2008. Annual report readability, current earnings, and persistence. Journal of Accounting
and Economics 45, 221-247.

Li, F., 2012. Discussion of Analyzing Speech to Detect Financial Misreporting. Journal of
Accounting Research 50, 393-400

Lo, K., 2008. Earnings management and earnings quality. Journal of Accounting and Economics
45, 350-357.

Miller, B.P., 2010. The effects of reporting complexity on small and large investor trading. The
Accounting Review 85, 2107-2143.

Rennekamp, K., 2012. Processing fluency and investors’ reactions to disclosure readability.
Journal of Accounting Research 50, 1319-1354.

Roychowdhury, S., 2006. Earnings management through real activities manipulation. Journal of
Accounting and Economics 42, 335-370.

SEC, 2007. Speech by SEC Chairman: Closing Remarks to the Second Annual Corporate
Governance Summit. https://www.sec.gov/news/speech/2007/spch032307cc.htm

You, H. and X. Zhang, 2009. Financial disclosure complexity and investor underreaction to 10-K
information. Review of Accounting Studies 14, 559-586.

Zang, A., 2012. Evidence on the Trade-Off between Real Activities Manipulation and Accrual-
Based Earnings Management. The Accounting Review 87, No. 2, pp. 675-703.

35
Figure 1 Level of Fog Index at various values of change in EPS. EPS = earnings per share before
extraordinary items, DA = discretionary accruals (see Appendix B for detailed definition)

Level of Fog Index at various values of change in EPS

18.8

18.6 Incremental Fog for ΔEPS in [0 to 0.03] and

DA > median(DA|DA>0)

18.4
Incremental Fog for ΔEPS in [0 to 0.03] and
0.58
0 < DA < median(DA|DA>0)
Fog

18.2
0.34 Incremental Fog for ΔEPS in [0 to 0.03] and
0 > DA
0.02
0.06
18 0.01

Incremental Fog on missing expectations

(Change in EPS <0)
17.8 17.99 17.99 17.99 17.99

Base Fog value for change EPS > $0.03

17.6
<-0.03 [-0.03 to 0) [0 to 0.03] >0.03
Change in EPS

36
Table 1: Sample selection
Observations
All Compustat firm-years 2000-2012 109,197
Less: observations in utilities or financial services (24,507)
Less: firm-years with insufficient data (57,723)
Number of firm-years used in tests of Hypothesis 1 26,967

Number of firms 4,855

Table 2: Descriptive statistics for variables used to test Hypothesis 1 (n = 26,967). Panel A – Full
sample for test of H1A, H1B, and H1C
Variable N Mean S.D. Q1 Median Q3
Fog 26,967 18.020 1.613 16.880 17.907 19.048
ΔEPS 26,967 0.010 1.654 -0.390 0.060 0.460
Earnings 26,967 -0.009 0.257 -0.029 0.059 0.114
Loss 26,967 0.371 0.483 0 0 1
Size 26,967 5.768 2.058 4.292 5.774 7.156
MTB 26,967 1.999 1.757 1.100 1.496 2.242
Age 26,967 15.576 11.853 6.000 12.000 22.000
SpecItems 26,967 -0.032 0.248 -0.016 -0.001 0.000
EarnVol 26,967 0.063 1.045 0.001 0.002 0.010
RetVol 26,967 0.158 0.099 0.090 0.131 0.194
NBSeg 26,967 1.032 0.531 0.693 0.693 1.386
NGSeg 26,967 1.057 0.647 0.693 1.099 1.609
NItems 26,967 278.878 29.838 255.000 283.000 301.000
M&A 26,967 0.398 0.489 0.000 0.000 1.000
SEO 26,967 0.061 0.239 0.000 0.000 0.000
Delaware 26,967 0.653 0.476 0.000 1.000 1.000

37
Panel B – Restatement sample for test of H1D
Compustat population (excl.
Restatement sample restatement obs.) Restate – Compustat
Diff. in
Variable N Mean Median N Mean Median means t-stat
FOG 659 18.238 18.118 26,308 18.015 17.902 0.224 3.515
Earnings 659 0.009 0.057 26,308 -0.021 0.060 0.030 2.225
Loss 659 0.399 0 26,308 0.371 0 0.028 1.470
Size 659 6.214 6.275 26,308 5.756 5.761 0.458 5.635
MTB 659 1.944 1.466 26,308 2.000 1.497 -0.056 -0.804
Age 659 14.200 11.000 26,308 15.600 12.000 -1.366 -2.921
SpecItems 659 -0.036 -0.004 26,308 -0.032 -0.001 -0.004 -0.415
EarnVol 659 0.032 0.002 26,308 0.063 0.002 -0.032 -0.766
RetVol 659 0.166 0.137 26,308 0.157 0.131 0.009 2.201
NBSeg 659 2.483 2.000 26,308 2.237 1.000 0.245 3.394
NGSeg 659 2.801 2.000 26,308 2.522 2.000 0.279 3.065
Nitems 659 272.400 274.000 26,308 279.000 284.000 -6.657 -5.660
M&A 659 0.476 0.000 26,308 0.396 0.000 0.081 4.194
SEO 659 0.050 0.000 26,308 0.061 0.000 -0.011 -1.157
Delaware 659 0.662 1.000 26,308 0.652 1.000 0.009 0.487

Panel C - Misstatement sample for test of H1E

Compustat population
Misstatement sample (excl. misstatement obs.) Fraud – Compustat
Diff. in
Variable N Mean Median N Mean Median means t-stat
FOG 114 18.286 18.314 26853 18.019 17.906 0.267 1.762
Earnings 114 0.033 0.058 26853 -0.020 0.060 0.053 1.662
Loss 114 0.421 0 26853 0.371 0 0.050 1.090
Size 114 7.157 6.952 26853 5.762 5.768 1.395 5.830
MTB 114 2.214 1.661 26853 1.998 1.496 0.216 1.309
Age 114 11.600 7.500 26853 15.600 12.000 -3.970 -3.570
SpecItems 114 -0.052 -0.010 26853 -0.032 -0.001 -0.020 -0.874
EarnVol 114 0.007 0.002 26853 0.063 0.002 -0.055 -0.565
RetVol 114 0.190 0.177 26853 0.157 0.131 0.033 3.497
NBSeg 114 2.947 3.000 26853 2.240 1.000 0.707 4.112
NGSeg 114 2.947 3.000 26853 2.527 2.000 0.420 1.940
Nitems 114 258.100 259.000 26853 279.000 284.000 -20.861 -7.457
M&A 114 0.605 1.000 26853 0.397 0.000 0.209 4.544
SEO 114 0.070 0.000 26853 0.061 0.000 0.010 0.424
Delaware 114 0.649 1.000 26853 0.653 1.000 -0.004 -0.080

Note: Refer to Appendix B for variable definitions.

38
Table 3: Correlation matrix with Pearson (Spearman) values above (below) the main diagonal (N =
26,967)
Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
(1) Fog 1 0.00 -0.23 0.17 -0.10 0.12 -0.02 -0.02 0.04 0.10 -0.02 -0.09 0.01 -0.03 -0.01 0.06
(2) ΔEPS 0.00 1 0.01 -0.01 0.01 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
(3) Earnings -0.22 0.20 1 -0.75 0.49 0.23 0.25 0.15 -0.40 -0.43 0.10 0.12 0.08 0.22 -0.06 -0.09
(4) Loss 0.17 -0.27 -0.59 1 -0.42 -0.12 -0.25 -0.32 0.38 0.44 -0.11 -0.11 -0.06 -0.2 0.05 0.09
(5) Size -0.11 0.10 0.40 -0.42 1 0.17 0.29 0.08 -0.05 -0.40 0.13 0.26 0.28 0.34 0.07 0.08
(6) MTB 0.09 0.15 -0.25 0.04 0.38 1 -0.12 -0.01 0.18 0.08 -0.11 -0.06 -0.01 -0.04 0.07 0.07
(7) Age -0.02 0.02 0.23 -0.24 0.24 -0.08 1 0.06 -0.03 -0.29 0.20 0.16 0.23 0.09 -0.09 -0.20
(8) SpecItems -0.02 0.27 0.18 -0.16 0.01 0.10 0.05 1 0.00 -0.11 0.01 0.01 0.01 0.00 0.01 -0.02
(9) EarnVol 0.19 0.04 -0.14 0.05 -0.38 0.20 -0.25 -0.02 1 0.06 -0.03 -0.03 -0.01 -0.01 0.01 0.02
(10) RetVol 0.10 -0.04 -0.40 0.41 -0.44 -0.08 -0.33 -0.13 0.43 1 -0.10 -0.09 -0.21 -0.18 0.07 0.06
(11) NBSeg -0.04 0.01 0.13 -0.11 0.13 -0.10 0.19 -0.04 -0.19 -0.12 1 0.14 0.02 0.11 -0.03 -0.04
(12) NGSeg -0.08 0.02 0.19 -0.11 0.27 0.03 0.15 -0.11 -0.09 -0.09 0.14 1 0.20 0.14 -0.04 0.02
(13) NItems 0.00 0.02 0.07 -0.06 0.27 0.05 0.23 -0.11 -0.13 -0.19 0.03 0.20 1 0.08 0.01 0.03
(14) M&A -0.03 -0.02 0.20 -0.20 0.34 0.05 0.07 -0.06 -0.19 -0.20 0.12 0.14 0.08 1 0.00 0.03
(15) SEO -0.01 0.04 -0.08 0.05 0.08 0.09 -0.10 0.02 0.05 0.07 -0.03 -0.04 0.01 0.00 1 0.06
(16) Delaware 0.07 0.02 -0.10 0.09 0.08 0.08 -0.21 -0.04 0.10 0.07 -0.05 0.03 0.04 0.03 0.06 1
Bolded coefficients are statistically significant at the 1% level. Refer to Appendix B for variable definitions.

39
 Table 4: First test of Hypothesis 1 – earnings management identified by MBE.
Panel A: Full sample
The table reports determinant analyses of MD&A readability as a function of earnings
management. The dependent variable is the FOG index of the MD&A section of current annual
report. Firms that just meet or beat past year’s earnings by [one | two | three] cents are identified as
firms that likely managed reported earnings (MBE = 1), otherwise MBE = 0. The baseline is the
group of firm-years that beat past year’s earnings by more than [one | two | three] cents. Firm-years
with performance below prior year’s are identified by NegEarnChg=1. The control variables are
the Li (2008) controls listed in Appendix B. Predicted signs for variables from Li (2008) are as
listed in that paper’s Table 2, but if the predicted sign differs from the empirical association, then
we list it as “?” above. *, **, and *** indicate significance at the 10%, 5%, and 1% levels (two-
tailed test). The t-statistics are in parentheses.
MBE = 1 when ΔEPS
Independent variable Pred. [$0, $0.01] [$0, $0.02] [$0, $0.03]
Sign I II III
β1 MBE + 0.162** 0.187*** 0.203***
(2.393) (3.308) (3.744)
β2 NegEarnChg + -0.016 -0.012 -0.008
(-0.807) (-0.608) (-0.403)
Earnings – -0.375*** -0.375*** -0.376***
(-2.802) (-2.814) (-2.826)
Loss + 0.253*** 0.252*** 0.252***
(6.161) (6.159) (6.163)
Size ? -0.031 -0.031 -0.030
(-1.427) (-1.404) (-1.390)
MTB + 0.041*** 0.040*** 0.040***
(3.317) (3.288) (3.271)
Age ? 0.014*** 0.014*** 0.014***
(6.769) (6.752) (6.777)
SpecItems – 0.094*** 0.093*** 0.093***
(3.266) (3.263) (3.266)
EarnVol + -0.013 -0.013 -0.013
(-0.851) (-0.852) (-0.847)
RetVol + 0.442*** 0.444*** 0.451***
(2.958) (2.987) (3.030)
NBSeg ? 0.043 0.043 0.044
(1.153) (1.153) (1.157)
NGSeg ? -0.157*** -0.156*** -0.156***
(-3.402) (-3.388) (-3.382)
NItems ? -0.003** -0.003** -0.003**
(-2.243) (-2.241) (-2.238)
M&A ? 0.050* 0.050* 0.050*
(1.736) (1.739) (1.729)
SEO – -0.218*** -0.218*** -0.217***
(-4.831) (-4.818) (-4.798)
Delaware + 0.148*** 0.148*** 0.148***

40
 (3.126) (3.119) (3.121)
Constant 17.986*** 17.974*** 17.966***
(37.900) (38.138) (38.092)

Industry fixed effects Yes Yes Yes

Year fixed effects Yes Yes Yes
Obs with MBE = 1 438 679 898
Total observations 26,967 26,967 26,967
Adj. R-Squared 12.7% 12.8% 12.8%
Statistical test of
β1 - β2 = 0; F(1, 38)= 7.11** 12.35** 15.48***

41
Table 4 (continued)

Panel B: Matching Sample analysis based on Industry (Fama-French 48), Fiscal Year, MTB (deciles)
and closest market capitalization

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. Firms that just meet
or beat past year’s earnings by [one | two | three] cents are identified as firms that likely managed reported
earnings (MBE = 1), otherwise MBE = 0. In Columns I to III, firms that just meet or beat past year’s
earnings by [one | two | three] cents (MBE) are matched to firms that missed the benchmark, the baseline
group in the analysis. In Columns IV to VI, the MBE sample is matched to firms that beat the benchmark
by more than [one | two | three] cents, the baseline group in the analysis. The control variables are the Li
(2008) controls listed in Appendix B. Predicted signs for variables from Li (2008) are as listed in that
paper’s Table 2, but if the predicted sign differs from the empirical association, then we list it as “?”
above. *, **, and *** indicate significance at the 10%, 5%, and 1% levels (two-tailed test). The t-statistics
are in parentheses.

Compared to firms that MISSED benchmark Compared to firms that BEAT benchmark
MBE = 1 when ΔEPS MBE = 1 when ΔEPS
Independent Pred. [$0, $0.01] [$0, $0.02] [$0, $0.03] [$0, $0.01] [$0, $0.02] [$0, $0.03]
variable Sign I II III IV V VI
MBE + 0.144* 0.209** 0.225*** 0.175** 0.176** 0.242***
(1.740) (2.698) (2.885) (2.149) (2.276) (4.434)
Earnings – -0.068 0.003 -0.109 -0.394* -0.139 -0.157
(-0.178) (0.018) (-0.630) (-1.852) (-1.083) (-1.025)
Loss + 0.067 0.235* 0.339*** -0.050 0.232* 0.255**
(-0.519) (1.970) (3.467) (-0.421) (1.813) (2.226)

Controls Yes Yes Yes Yes Yes Yes

Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
Total observations 844 1,296 1,716 846 1,312 1,742
Adj. R-Squared 19.2% 19.4% 18.4% 21.5% 21.3% 20.5%

42
Table 4 (continued)
Panel C: Matching Sample analysis based on Industry (Fama-French 48), Fiscal Year, Size (deciles) and
closest sales growth

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. Firms that just meet
or beat past year’s earnings by [one | two | three] cents are identified as firms that likely managed reported
earnings (MBE = 1), otherwise MBE = 0. In Columns I to III, firms that just meet or beat past year’s
earnings by more than [one | two | three] cents (MBE) are matched to firms that missed the benchmark,
the baseline group in the analysis. In Columns IV to VI, the MBE sample is matched to firms that beat the
benchmark by [one | two | three] cents, the baseline group in the analysis. The control variables are the Li
(2008) controls listed in Appendix B. Predicted signs for variables from Li (2008) are as listed in that
paper’s Table 2, but if the predicted sign differs from the empirical association, then we list it as “?”
above. *, **, and *** indicate significance at the 10%, 5%, and 1% levels (two-tailed test). The t-statistics
are in parentheses.

Compared to firms that MISSED benchmark Compared to firms that BEAT benchmark
MBE = 1 when ΔEPS MBE = 1 when ΔEPS
Independent Pred. [$0, $0.01] [$0, $0.02] [$0, $0.03] [$0, $0.01] [$0, $0.02] [$0, $0.03]
variable Sign I II III IV V VI
MBE + 0.078 0.191** 0.167** 0.132 0.184** 0.176**
(0.737) (2.535) (2.148) (1.603) (2.768) (2.777)
Earnings – -0.340* -0.198** -0.252** -0.082 -0.060 -0.116
(-1.787) (-2.659) (-2.408) (-0.362) (-0.345) (-0.723)
Loss + -0.006 0.194 0.232** 0.226 0.302* 0.391**
(-0.042) (1.601) (2.445) (1.307) (1.763) (2.715)

Controls Yes Yes Yes Yes Yes Yes

Industry fixed effects Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes

Total observations 838 1,296 1,714 846 1,310 1,740

Adj. R-Squared 13.4% 15.6% 16.2% 16.5% 17.3% 18.0%

43
Table 5: First test of Hypothesis 1 – earnings management identified by MBE – supplemental
results

Panel A: MBE defined using earnings deflated by total assets instead of EPS

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. Firms that just meet
or beat past year’s earnings by less than [0.4% | 0.5% | 0.6%] of total assets are identified as firms that
likely managed reported earnings (MBE = 1), otherwise MBE = 0. The baseline is the group of firms that
beat past year’s earnings by more than [0.4% | 0.5% | 0.6%] of total assets. Firms with performance below
prior year’s are identified by NegEarnChg = 1. The control variables are the Li (2008) controls listed in
Appendix B. Predicted signs for variables from Li (2008) are as listed in that paper’s Table 2, but if the
predicted sign differs from the empirical association, then we list it as “?” above. *, **, and *** indicate
significance at the 10%, 5%, and 1% levels (two-tailed test). The t-statistics are in parentheses.
MBE = 1 when MBE = 1 when MBE = 1 when
ΔEarnings ΔEarnings ΔEarnings
Pred.
Independent variable Sign I II III
β1 MBE + 0.132** 0.130*** 0.120***
(2.634) (3.258) (2.923)
β2 NegEarnChg + -0.014 -0.014 -0.014
(-0.754) (-0.739) (-0.721)
Earnings – -0.746*** -0.745*** -0.745***
(-5.210) (-5.207) (-5.218)
Loss + 0.180*** 0.180*** 0.180***
(4.424) (4.431) (4.454)

Controls Yes Yes Yes

Industry fixed effects Yes Yes Yes
Year fixed effects Yes Yes Yes

Observations with MBE = 1 1040 1321 1595

Total observations 26,967 26,967 26,967
Adj. R-Squared 13.1% 13.1% 13.1%
Statistical test of
β1 - β2 = 0; F(1, 38) = 5.14** 7.77*** 7.66***

44
Table 5 (continued)
Panel B: Placebo test with MBE in prior year

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. Firms that just meet
or beat past year’s earnings by [one | two | three] cents are identified as firms that likely managed reported
earnings (MBE = 1), otherwise MBE = 0. The baseline is the group of firms that beat past year’s earnings
by more than [one | two | three] cents. Firms with performance below prior year’s are identified by
NegEarnChg = 1. Firms that just met or beat by [one | two | three] cents the benchmark the year before
are identified as MBEt-1=1. A significant coefficient for MBEt-1 could be interpreted as the existence of
unmodelled factors omitted from our analysis but correlated with the propensity to meet/beat explaining
the variation in MD&A readability. The control variables are the Li (2008) controls listed in Appendix B.
Predicted signs for variables from Li (2008) are as listed in that paper’s Table 2, but if the predicted sign
differs from the empirical association, then we list it as “?” above. Refer to Appendix B for variable
definitions. *, **, and *** indicate significance at the 10%, 5%, and 1% levels (two-tailed test). The t-
statistics are in parentheses.

MBE = 1 when MBE = 1 when MBE = 1 when

Pred. ΔEPS ΔEPS ΔEPS
Independent variable Sign I II III
β1 MBEt + 0.180** 0.196*** 0.211**
(2.698) (3.579) (2.748)
β2 MBEt-1 0 0.047 0.072 0.083
(0.542) (1.364) (1.123)
β3 NegEarnChg + -0.010 -0.006 -0.003
(-0.508) (-0.323) (-0.087)
Earnings – -0.392*** -0.392*** -0.392***
(-2.883) (-2.895) (-6.361)
Loss + 0.253*** 0.253*** 0.252***
(6.019) (6.023) (7.879)

Controls Yes Yes Yes

Industry fixed effects Yes Yes Yes
Year fixed effects Yes Yes Yes

Observations with MBE = 1 438 679 898

Total observations 26,967 26,967 26,967
Adj. R-Squared 12.8% 12.8% 12.8%

Statistical test of
β1 – β3 = 0; F(1, 38) = 8.34*** 13.60*** 6.48**

45
Table 5 (continued)
Panel C: Using performance in first 3 fiscal quarters to identify distance to MBE

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. MBE_ByQ4t
measures by how far a firm is underperforming in the first 3 fiscal quarters compared with the same
period in the prior year. Specifically, it is EPS in the first 3 quarter of year t-1 less EPS in the first 3
quarters of year t for firms that just meet or beat past year’s earnings by [one | two | three] cents,
otherwise MBE_ByQ4t = 0. The baseline is the group of firms that beat past year’s earnings by more than
[one | two | three] cents. Firms with performance below prior year’s are identified by NegEarnChg = 1.
The control variables are the Li (2008) controls listed in Appendix B. Predicted signs for variables from
Li (2008) are as listed in that paper’s Table 2, but if the predicted sign differs from the empirical
association, then we list it as “?” above. *, **, and *** indicate significance at the 10%, 5%, and 1%
levels (two-tailed test). The t-statistics are in parentheses.

MBE = 1 when MBE = 1 when MBE = 1 when

Pred. ΔEPS ΔEPS ΔEPS
Independent variable
Sign I II III
β1 MBE_ByQ4t + 0.223** 0.245** 0.216**
(2.271) (2.401) (2.187)
β2 NegEarnChg + -0.020 -0.020 -0.020
(-1.043) (-1.057) (-1.055)
Earnings – -0.745*** -0.745*** -0.744***
(-5.085) (-5.085) (-5.084)
Loss + 0.175*** 0.176*** 0.176***
(4.452) (4.471) (4.469)

Controls Yes Yes Yes

Industry fixed effects Yes Yes Yes
Year fixed effects Yes Yes Yes

Average MBE_sprint 0.029 0.020 0.010

Total observations 26,967 26,967 26,967
Adj. R-Squared 12.8% 12.8% 12.8%
Statistical test
β1 – β2 = 0; F(1, 41) = 5.13** 5.68** 4.96**

46
Table 6: Second test of Hypothesis 1 – earnings management identified by MBE and sign of
discretionary accruals or real earnings management

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. PosEM( ) identifies
positive earnings management. Column I to III use accruals earnings management, Column IV to VI use
real activity earnings management, and Column VII to IX use a combination of the two types of earnings
management. The coefficient on MBE × PosEM( ) identifies the incremental Fog for the group more
likely to managed reported earnings, which comprises firms that just met or beat past year’s earnings by
[one | two | three] cents (MBE = 1) and have positive earnings management. The coefficient on MBE ×
NegEM( ) identifies the incremental Fog for firms that just met or beat past year’s earnings by [one | two |
three] cents (MBE = 1) and have no suspected upwards earnings management. The baseline is the group
of firms that beat past year’s earnings by more than [one | two | three] cents. Firms with performance
below prior year’s are identified by NegEarnChg = 1. The control variables are the Li (2008) controls
listed in Appendix B. Predicted signs for variables from Li (2008) are as listed in that paper’s Table 2, but
if the predicted sign differs from the empirical association, then we list it as “?” above. Refer to Appendix
B for variable definitions. *, **, and *** indicate significance at the 10%, 5%, and 1% levels (two-tailed
test). The t-statistics are in parentheses.

Accruals earnings Real activities

management management Combined
PosEM(DA) = 1 if DA > PosEM(RA) = 1 if RAM > PosEM(Comb)
0; 0; = PosEM(DA) +
0 otherwise 0 otherwise PosEM(RA)
MBE when ΔEPS MBE when ΔEPS MBE when ΔEPS
Pre
d. IV V VI VII VIII IX
Independent Sig
variable n I II III
β
1 PosEM( ) ? 0.037 0.038 0.037 0.003 0.003 0.003 0.021 0.021 0.020
(1.294) (1.289) (1.289) (0.072) (0.067) (0.090) (1.530) (1.520) (1.468)
β MBE × 0.329* 0.294* 0.294* 0.303* 0.315* 0.287* 0.237* 0.225* 0.220*
2 PosEM( ) + ** ** ** ** ** ** ** ** **
(3.715) (4.826) (5.279) (3.696) (3.260) (3.524) (4.479) (5.314) (6.188)
β MBE × 0.164* 0.190*
3 NegEM( ) 0 -0.038 0.065 0.097 0.144* * **
-0.130 -0.023 0.015
(- (- (-
0.298) (0.560) (0.960) (1.961) (2.543) (3.210) 1.136) 0.190) (0.142)
β
4 NegEarnChg + -0.009 -0.005 -0.001 -0.010 -0.006 -0.001 -0.008 -0.004 -0.001
(- (- (- (- (- (- (- (- (-
0.427) 0.231) 0.028) 0.518) 0.301) 0.072) 0.417) 0.228) 0.048)
- - - - - - - - -
0.393* 0.393* 0.393* 0.391* 0.390* 0.391* 0.390* 0.389* 0.389*
Earnings – ** ** ** ** ** ** ** ** **
(- (- (- (- (- (- (- (- (-

47
 2.899) 2.900) 2.912) 2.924) 2.936) 2.947) 2.902) 2.897) 2.908)
0.257* 0.257* 0.257* 0.252* 0.252* 0.251* 0.253* 0.253* 0.252*
Loss + ** ** ** ** ** ** ** ** **
(5.948) (5.931) (5.945) (5.792) (5.785) (5.787) (6.024) (5.998) (6.005)

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry fixed
effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes

Observations 26,143 26,143 26,143 26,143 26,143 26,143 26,143 26,143 26,143
Adj. R-Squared 12.8% 12.8% 12.8% 12.8% 12.8% 12.8% 12.8% 12.8% 12.8%
Statistical test of
8.95**
β2 – β3; F(1, 38) = 5.49** 2.85* 2.90* 2.91* 1.74 1.06 * 4.26** 3.77*
15.45* 22.13* 27.76* 13.56* 11.63* 12.09* 21.03* 26.49* 31.39*
β2 – β4; F(1, 38) = ** ** ** ** ** ** ** ** **

Table 7: Third test of Hypothesis 1 – earnings management identified by MBE and magnitude of
upward earnings announcement

Panel A:

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. The dependent
variable is the FOG index of the MD&A section of current annual report. PosEM( ) identifies positive
earnings management. Column I to III use accruals earnings management, Column IV to VI use real
activity earnings management, and Column VII to IX use a combination of the two processes of earnings
management. MBE × HighPosEM( ) identifies the group more likely to managed reported earnings, which
comprises firms that just met or beat past year’s earnings by [one | two | three] cents (MBE = 1) and have
above-median positive earnings management. MBE × LowPosEM( ) identifies firms that meet or beat past
year’s earnings by [one | two | three] cents (MBE = 1) and have below-median positive earnings
management. Specifically,
HighPosEM(DA) = 1 if DA > median(DA | DA≥0); LowPosEM(DA) = 1 if DA ≤ median(DA | DA≥0).
HighPosEM(RA) = 1 if DA > median(RAM | RAM≥0); LowPosEM(RA) = 1 if RAM ≤ median(RAM |
RAM≥0).

48
HighPosEM(Comb) = HighPosEM(DA) + HighPosEM(RA); LowPosEM(Comb) = 1 if LowPosEM(DA) =
1 and LowPosEM(RA) = 1.
MBE × NegEM( ) comprises firms that just meet or beat past year’s earnings by [one | two | three] cents
(MBE = 1) and have no suspected upwards earnings management. The baseline is the group of firms that
beat past year’s earnings by more than [one | two | three] cents. Firms with performance below prior
year’s are identified by NegEarnChg=1. The control variables are the Li (2008) controls listed in
Appendix B.
Predicted signs for variables from Li (2008) are as listed in that paper’s Table 2, but if the predicted sign
differs from the empirical association, then we list it as “?” above. *, **, and *** indicate significance at
the 10%, 5%, and 1% levels (two-tailed test). The t-statistics are in parentheses.
Accruals earnings Real activities
management management Combined
PosEM(DA) = 1 if DA > PosEM(RA) = 1 if RAM > PosEM(Comb)
0; 0; = PosEM(DA) +
0 otherwise 0 otherwise PosEM(RA)
MBE when ΔEPS MBE when ΔEPS MBE when ΔEPS
Pre
d. IV V VI VII VIII IX
Independent Sig
variable n I II III
β
1 PosEM() ? 0.038 0.038 0.037 0.003 0.003 0.004 0.021 0.021 0.021
(1.295) (1.292) (1.295) (0.089) (0.084) (0.108) (1.522) (1.510) (1.481)
β MBE × 0.418* 0.409* 0.391* 0.718* 0.600* 0.502* 0.409* 0.376* 0.359*
2 HighPosEM() + ** ** ** ** ** ** ** ** **
(3.175) (6.531) (6.330) (3.953) (2.815) (3.812) (5.140) (5.097) (6.777)
β MBE × 0.183* 0.158* 0.171*
3 LowPosEM( ) ? 0.220* 0.165 * -0.105 0.055 0.071 0.143 * *
(-
(1.927) (1.683) (2.199) 0.678) (0.396) (0.544) (1.429) (2.129) (2.355)
β MBE × 0.164* 0.191*
4 NegEM( ) 0 -0.038 0.065 0.097 0.144* * ** -0.129 -0.021 0.018
(- (- (-
0.298) (0.561) (0.961) (1.963) (2.548) (3.217) 1.128) 0.178) (0.164)
β
5 NegEarnChg + -0.009 -0.005 -0.001 -0.009 -0.005 -0.001 -0.007 -0.004 0.001
(- (- (- (- (- (- (- (-
0.429) 0.238) 0.034) 0.515) 0.298) 0.066) 0.383) 0.178) (0.030)
- - - - - - - - -
0.393* 0.394* 0.394* 0.389* 0.388* 0.389* 0.389* 0.388* 0.388*
Earnings – ** ** ** ** ** ** ** ** **
(- (- (- (- (- (- (- (- (-
2.908) 2.906) 2.918) 2.944) 2.962) 2.965) 2.919) 2.911) 2.923)
0.257* 0.257* 0.257* 0.252* 0.252* 0.251* 0.253* 0.253* 0.252*
Loss + ** ** ** ** ** ** ** ** **
(5.950) (5.931) (5.952) (5.843) (5.833) (5.817) (6.053) (6.025) (6.035)
Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry fixed
effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes

49
Observations 26,143 26,143 26,143 26,143 26,143 26,143 26,143 26,143 26,143
Adj. R-Squared 12.8% 12.8% 12.9% 12.8% 12.8% 12.8% 12.8% 12.9% 12.9%
Statistical test of
9.11**
β2 – β3; F(1, 38) = 1.73 5.82** 6.84** * 3.89*
5.22** 5.00** 4.26** 4.05*
12.80* 8.05**
β2 – β4; F(1, 38) = 6.62** 6.62** 5.95** 6.83** 3.59* 4.18** ** 7.21** *
11.72* 46.76* 45.62* 15.39* 8.29** 14.89* 30.41* 28.14* 49.95*
β2 – β5; F(1, 38) = ** ** ** ** * ** ** ** **

50
Table 7 (continued)
Panel B: Comparison across earnings management groups (no, low, or high earnings management)
The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. The sample is
partitioned into three groups: a group of firm-years with no suspected upward earnings management
(Columns I to III), a group of firm-years with low level of upward earnings management (Columns IV to
VI), and a group of firm-years with high upward earnings management (Columns VII to IX). The group
of interest (MBE) comprises firms that met or beat past year’s earnings by [one | two | three] cents. The
baseline is the group of firms that beat past year’s earnings by more than [one | two | three] cents. Firms
with performance below prior year’s are identified by NegEarnChg =1. All variables are as defined in
Appendix B. The control variables are the Li (2008) controls listed in Appendix B. Predicted signs for
variables from Li (2008) are as listed in that paper’s Table 2, but if the predicted sign differs from the
empirical association, then we list it as “?” above. *, **, and *** indicate significance at the 10%, 5%,
and 1% levels (two-tailed test). The t-statistics are in parentheses.
Low earnings High earnings
No earnings management management management
LowPosEM(DA) = 1 or
DA <0 and RAM < 0 LowPosEM(RA) = 1 HighPosEM(Comb) ≥ 1
MBE when ΔEPS MBE when ΔEPS MBE when ΔEPS
Pre
d. IV V VI VII VIII IX
Independent Sig
variable n I II III
β 0.167* 0.416* 0.390* 0.361*
1 MBE ?/+ -0.088 0.026 0.066 0.130 0.154* * ** ** **
(-
0.726) (0.207) (0.590) (1.321) (1.923) (2.160) (3.157) (5.195) (5.340)
β 0.066* 0.071* 0.074*
2 NegEarnChg + * * * -0.005 -0.001 0.002 -0.063 -0.059 -0.053
(- (- (- (- (-
(2.041) (2.115) (2.198) 0.168) 0.044) (0.053) 1.433) 1.376) 1.245)
- - - - - - - - -
0.376* 0.376* 0.376* 1.049* 1.045* 1.044* 0.259* 0.259* 0.260*
Earnings – * * * ** ** ** * * *
(- (- (- (- (- (- (- (- (-
2.698) 2.690) 2.693) 6.795) 6.764) 6.726) 2.035) 2.038) 2.058)
0.221* 0.221* 0.221* 0.230* 0.229* 0.229* 0.239* 0.239* 0.239*
Loss + ** ** ** ** ** ** ** ** **
(3.275) (3.258) (3.248) (3.138) (3.134) (3.133) (6.971) (6.977) (7.096)
Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry fixed
effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes

Observations 7,886 7,886 7,886 8,356 8,356 8,356 9,901 9,901 9,901
Adj. R-Squared 14.1% 14.1% 14.1% 11.1% 11.1% 11.1% 11.3% 11.3% 11.4%
Statistical test
17.32* 35.63* 35.27*
β1 – β2; F(1, 37) = 1.84 0.16 0.01 1.82 3.61* 4.47** ** ** **

51
Table 8: Third test of Hypothesis 1 – earnings management identified by MBE and magnitude of
upward earnings management – supplemental robustness tests

Panel A – Including firm-level (instead of industry) fixed effects or lag(Fog)

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. The dependent
variable is the FOG index of the MD&A section of current annual report. PosEM( ) identifies positive
earnings management, measured based on a combination of discretionary accruals and real activity
earnings management. MBE × HighPosEM( ) identifies the group more likely to managed reported
earnings, which comprises firms that just meet or beat past year’s earnings by [one | two | three] cents
(MBE = 1) and have above-median positive earnings management. MBE × LowPosEM( ) identifies firms
that meet or beat past year’s earnings by [one | two | three] cents (MBE = 1) and have below-median
positive earnings management. Specifically,
HighPosEM(DA) = 1 if DA > median(DA | DA≥0); LowPosEM(DA) = 1 if DA ≤ median(DA | DA≥0).
HighPosEM(RA) = 1 if DA > median(RAM | RAM≥0); LowPosEM(RA) = 1 if RAM ≤ median(RAM |
RAM≥0).
HighPosEM(Comb) = HighPosEM(DA) + HighPosEM(RA); LowPosEM(Comb) = 1 if LowPosEM(DA) =
1 and LowPosEM(RA) = 1.
The coefficient on MBE × NegEM( ) identifies the incremental Fog for firms that just met or beat past
year’s earnings by [one | two | three] cents (MBE = 1) and have no suspected upwards earnings
management. Column I uses firm-level fixed effects and clusters instead of industry fixed effects and
clusters. Column II uses lag(FOG) as an independent variable and industry-level fixed effects and
clusters.
The baseline is the group of firms that beat past year’s earnings by more than three cents. Firms with
performance below prior year’s are identified by NegEarnChg = 1. The control variables are the Li (2008)
controls listed in Appendix B.
Predicted signs for variables from Li (2008) are as listed in that paper’s Table 2, but if the predicted sign
differs from the empirical association, then we list it as “?” above. *, **, and *** indicate significance at
the 10%, 5%, and 1% levels (two-tailed test). The t-statistics are in parentheses.

MBE when ΔEPS [$0, $0.03]

Pred. With firm fixed effects With lag(Fog)
Independent variable
Sign I II
β1 PosEM(Comb) ? -0.005 -0.009
(-0.413) (-0.860)
β2 MBE × HighPosEM(Comb) + 0.131** 0.101**
(2.280) (2.472)
β3 MBE × LowPosEM(Comb) ? -0.018 -0.035
(-0.302) (-0.574)
β4 MBE × NegEM(Comb) 0 0.016 0.024
(0.234) (0.350)
β5 NegEarnChg + 0.016 0.030***
(1.145) (3.267)
Earnings – -0.086* -0.117
(-1.694) (-3.691)
Loss + 0.078*** 0.086***

52
 (3.088) 5.345
lag(Fog) 0.804***
(107.753)

Controls Yes Yes

Industry fixed effects No Yes
Firm fixed effects Yes No
Year fixed effects Yes Yes

Observations 26,143 21,403

Adj. R-Squared 69.4% 70.5%

Statistical test F(1, 38) = F(1, 4739)=

β2 – β3 2.89* 2.97*
β2 – β4 0.87 1.28
β2 – β5 2.96* 3.29*

53
Table 8 (continued)
Panel B – Discretionary accruals estimated using modified Jones model of Dechow et al. (1995), Kothari
et al. (2005)

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. The dependent
variable is the FOG index of the MD&A section of current annual report. Pos EM( ) identifies positive
earnings management. Column I uses accruals earnings management, Column II uses real activity
earnings management, and Column III uses a combination of the two processes of earnings management.
MBE × HighPosEM( ) identifies the group more likely to managed reported earnings, which comprises
firms that just meet or beat past year’s earnings by [one | two | three] cents (MBE = 1) and have above-
median positive earnings management. MBE × LowPosEM( ) identifies firms that meet or beat past year’s
earnings by [one | two | three] cents (MBE = 1) and have below-median positive earnings management.
Specifically,
HighPosEM(DA) = 1 if DA > median(DA | DA≥0); LowPosEM(DA) = 1 if DA ≤ median(DA | DA≥0).
HighPosEM(RA) = 1 if DA > median(RAM | RAM≥0); LowPosEM(RA) = 1 if RAM ≤ median(RAM |
RAM≥0).
HighPosEM(Comb) = HighPosEM(DA) + HighPosEM(RA); LowPosEM(Comb) = 1 if LowPosEM(DA) =
1 and LowPosEM(RA) = 1.
The Coefficient on MBE × NegEM( ) identifies the incremental Fog for firms that just met or beat past
year’s earnings by [one | two | three] cents (MBE = 1) and have no suspected upwards earnings
management. Column I uses the modified Jones model of Dechow et al. (1995) to measure discretionary
accruals. Column II uses the performance-matched accrual model of Kothari et al. (2005) to measure
discretionary accruals: we calculate expected accruals based on 1-to-1 matching on industry (using the
Fama-French 48 industries), fiscal year, and closest previous-year return on assets (ROA, measured as net
income divided by lagged total assets). The baseline is the group of firms that beat past year’s earnings by
more than three cents. Firms with performance below prior year’s are identified by NegEarnChg = 1. The
control variables are the Li (2008) controls listed in Appendix B. Predicted signs for variables from Li
(2008) are as listed in that paper’s Table 2, but if the predicted sign differs from the empirical association,
then we list it as “?” above. *, **, and *** indicate significance at the 10%, 5%, and 1% levels (two-tailed
test). The t-statistics are in parentheses.

MBE when ΔEPS

Pred. Modified Jones Kothari et al.
Independent variable
Sign I II
β1 PosEM(DA) ? 0.022 0.012
(1.438) (0.681)
β2 MBE × HighPosEM(DA) + 0.353*** 0.376***
(6.659) (7.675)
β3 MBE × LowPosEM(DA) ? 0.155* 0.078
(1.958) (0.810)
β4 MBE × NegEM(DA) 0 0.031 0.116
(0.313) (1.361)
β5 NegEarnChg + 0.000 0.000
(0.016) (0.019)

54
 Earnings – -0.388*** -0.390***
(-2.925) (-2.921)
Loss + 0.253*** 0.252***
(6.039) (6.109)
Controls Yes Yes
Industry fixed effects Yes Yes
Year fixed effects Yes Yes
Observations 26,143 26,038
Adj. R-Squared 12.90% 12.90%
Statistical test
β2 – β3; F(1, 38) = 3.48* 6.18**
β2 – β4; F(1, 38) = 8.88*** 8.56***
β2 – β5; F(1, 38) = 43.29*** 58.58***

55
Table 8 (continued)

Panel C – Using zero-earnings benchmark (instead of zero earnings change)

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. The dependent
variable is the FOG index of the MD&A section of current annual report. Pos EM( ) identifies positive
earnings management. Column I uses accruals earnings management, Column II uses real activity
earnings management, and Column III uses a combination of the two processes of earnings management.
MBE × HighPosEM( ) identifies the group more likely to managed reported earnings, which comprises
firms that just meet or beat zero-earnings benchmark by less than three cents (MBE = 1) and have above-
median positive earnings management. MBE × LowPosEM( ) identifies firms that meet or beat zero-
earnings benchmark by less than three cents (MBE = 1) and have below-median positive earnings
management. Specifically,
HighPosEM(DA) = 1 if DA > median(DA | DA≥0); LowPosEM(DA) = 1 if DA ≤ median(DA | DA≥0).
HighPosEM(RA) = 1 if DA > median(RAM | RAM≥0); LowPosEM(RA) = 1 if RAM ≤ median(RAM |
RAM≥0).
HighPosEM(Comb) = HighPosEM(DA) + HighPosEM(RA); LowPosEM(Comb) = 1 if LowPosEM(DA) =
1 and LowPosEM(RA) = 1.
The coefficient on MBE × NegEM( ) identifies the incremental Fog for firms that just met or beat past
year’s earnings by [one | two | three] cents (MBE = 1) and have no suspected upwards earnings
management. The baseline is the group of firms that beat zero-earnings benchmark by more than three
cents. Firms with performance below zero-earnings benchmark are identified by NegEarnChg = 1. The
control variables are the Li (2008) controls listed in Appendix B. Predicted signs for variables from Li
(2008) are as listed in that paper’s Table 2, but if the predicted sign differs from the empirical association,
then we list it as “?” above. *, **, and *** indicate significance at the 10%, 5%, and 1% levels (two-tailed
test). The t-statistics are in parentheses.

MBE when EPS

Accruals earnings Real activities
Independent variable Pred. management management Combined
Sign I II III
β1 PosEM() ? 0.041 0.005 0.025
(1.415) (0.135) (1.678)
β2 MBE × HighPosEM() + 0.406*** 0.281** 0.258***
(4.432) (2.312) (3.919)
β3 MBE × LowPosEM( ) ? 0.260* 0.073 0.192
(2.001) (0.482) (1.650)
β4 MBE × NegEM( ) 0 0.101 0.258*** 0.133
(1.006) (3.521) (1.053)
β5 NegEarnChg + 0.266*** 0.260*** 0.261***
(6.407) (6.331) (6.536)
Earnings – -0.394*** -0.392*** -0.391***
(-2.910) (-2.932) (-2.900)
Controls Yes Yes Yes
Industry fixed effects Yes Yes Yes
Year fixed effects Yes Yes Yes
Observations 26,143 26,143 26,143
Adj. R-Squared 12.8% 12.8% 12.8%

56
 Statistical test
β2 – β3; F(1, 38) = 1.05 0.75 0.27
β2 – β4; F(1, 38) = 4.42** 0.07 1.11
β2 – β5; F(1, 38) = 4.17** 0.53 2.40

Table 8 (continued)
Panel D – MBE firms are compared to firms that missed the benchmark by less than 3 cents

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. The sample is
restricted to firm-years with . The dependent variable is the FOG index of the
MD&A section of current annual report. Pos EM( ) identifies positive earnings management. Column I
uses accruals earnings management, Column II uses real activity earnings management, and Column III
uses a combination of the two processes of earnings management. MBE × HighPosEM( ) identifies the
group more likely to have managed reported earnings, which comprises firms that just meet or beat past
year’s earnings by [one | two | three] cents (MBE = 1) and have above-median positive earnings
management. MBE × LowPosEM( ) identifies firms that meet or beat past year’s earnings by [one | two |
three] cents (MBE = 1) and have below-median positive earnings management. Specifically,
HighPosEM(DA) = 1 if DA > median(DA | DA≥0); LowPosEM(DA) = 1 if DA ≤ median(DA | DA≥0).
HighPosEM(RA) = 1 if DA > median(RAM | RAM≥0); LowPosEM(RA) = 1 if RAM ≤ median(RAM |
RAM≥0).
HighPosEM(Comb) = HighPosEM(DA) + HighPosEM(RA); LowPosEM(Comb) = 1 if LowPosEM(DA) =
1 and LowPosEM(RA) = 1.
The coefficient on MBE × NegEM( ) identifies the incremental Fog for firms that just met or beat past
year’s earnings by [one | two | three] cents (MBE = 1) and have no suspected upwards earnings
management. The baseline is the group of firms that missed past year’s earnings by less than three cents.
The control variables are the Li (2008) controls listed in Appendix B. Predicted signs for variables from
Li (2008) are as listed in that paper’s Table 2, but if the predicted sign differs from the empirical
association, then we list it as “?” above. *, **, and *** indicate significance at the 10%, 5%, and 1%
levels (two-tailed test). The t-statistics are in parentheses.

MBE when EPS

Accruals earnings Real activities
Independent variable Pred. management management Combined
Sign I II III
β1 PosEM() ? -0.092 -0.211 -0.159
(-0.491) (-1.253) (-1.438)
β2 MBE × HighPosEM() + 0.259** 0.435** 0.272***
(2.286) (2.201) (2.851)
β3 MBE × LowPosEM( ) ? 0.153 0.094 0.136
(1.299) (0.514) (1.425)
β4 MBE × NegEM( ) 0 -0.151 -0.008 -0.296**
(-0.928) (-0.093) (-2.032)
Earnings - -0.205* -0.198** -0.203**
(-1.866) (-2.136) (-2.030)
Loss + 0.370*** 0.366*** 0.371***
(3.883) (3.752) (3.843)

57
 Controls Yes Yes Yes
Industry fixed effects Yes Yes Yes
Year fixed effects Yes Yes Yes
Observations 1,501 1,501 1,501
Adj. R-Squared 19.0% 18.9% 19.4%
Statistical test
β2 – β3; F(1, 37) = 1.22 2.55 2.18
β2 – β4; F(1, 37) = 3.89* 4.23** 11.12***

Table 8 (continued)

Panel E – MBE firms with high earnings management compared to firms that missed benchmark by less
than 3 cents and had high earnings management

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. The sample is
restricted to firm-years with with high upward earnings management.
Specifically, and HighPosEM()=1, where
HighPosEM(DA) = 1 if DA > median(DA | DA≥0);
HighPosEM(RA) = 1 if DA > median(RAM | RAM≥0);
HighPosEM(Comb) = HighPosEM(DA) + HighPosEM(RA);
The baseline is the group of firms with high upward earnings management that missed past year’s
earnings by no more than three cents. The control variables are the Li (2008) controls listed in Appendix
B. Predicted signs for variables from Li (2008) are as listed in that paper’s Table 2, but if the predicted
sign differs from the empirical association, then we list it as “?” above. *, **, and *** indicate
significance at the 10%, 5%, and 1% levels (two-tailed test). The t-statistics are in parentheses.

MBE when EPS

Accruals earnings Real activities
Independent variable Pred. management management Combined
Sign I II III
MBE × HighPosEM() + 0.394*** 0.755*** 0.276**
(2.858) (6.782) (2.544)
Earnings – -0.091 -0.350*** 0.133
(-0.423) (-2.908) (1.026)
Loss + 0.502*** -0.243 0.506***
(3.349) (-0.785) (4.536)
Controls Yes Yes Yes
Industry fixed effects Yes Yes Yes
Year fixed effects Yes Yes Yes
Observations 475 180 531
Adj. R-Squared 14.4% 21.0% 16.4%

58
Table 9: Test of Hypothesis H1D and H1E – Restatement Sample

The table reports determinant analyses of MD&A readability as a function of earnings management. The
dependent variable is the FOG index of the MD&A section of current annual report. The group of interest
comprises firm-years with restatements (Column I), or the magnitude of restatement (Column II), or cases
of misstatement (Column III). The control variables are the Li (2008) controls listed in Appendix B.
Predicted signs for variables from Li (2008) are as listed in that paper’s Table 2, but if the predicted sign
differs from the empirical association, then we list it as “?” above. *, **, and *** indicate significance at
the 10%, 5%, and 1% levels (two-tailed test). The t-statistics are in parentheses.

Independent variable Pred. Sign I II II

Restated + 0.254**
(2.297)
RestateSeverityDecile + 0.036**
(2.156)
Misstated + 0.318**
(2.076)
Earnings – -0.377*** -0.377*** -0.377***
(-2.813) (-2.815) (-2.802)
Loss + 0.244*** 0.245*** 0.245***
(6.128) (6.1 (6.091)

Controls Yes Yes Yes

Industry fixed effects Yes Yes Yes
Year fixed effects Yes Yes Yes

Observations 26,967 26,967 26,967

Adj. R-Squared 12.80% 12.80% 12.70%
Restating/Fraud firm-years 659 659 114

59