Guiding Through the Fog:
Financial Statement Complexity and Voluntary Disclosure
Wayne Guay
guay@wharton.upenn.edu
Delphine Samuels
dels@wharton.upenn.edu
Daniel Taylor
dtayl@wharton.upenn.edu
The Wharton School
University of Pennsylvania
Draft: September 25, 2015
Abstract:
A growing literature documents that complex financial statements negatively affect the
information environment. In this paper, we examine whether managers use voluntary disclosure
to mitigate these negative effects. Employing traditional cross-sectional and within-firm designs,
we find a robust positive relation between financial statement complexity and subsequent
voluntary disclosure. This relation is stronger when liquidity decreases around the filing of the
financial statements, is stronger when firms have more outside monitors, and is weaker when
firms have poor performance and greater earnings management. We also examine the relation
between financial statement complexity and voluntary disclosure using two quasi-natural
experiments. Employing a generalized difference-in-differences design, we find firms affected
by the adoption of complex accounting standards (e.g., SFAS 133 and SFAS 157) increase their
voluntary disclosure to a greater extent than unaffected firms. Collectively, these findings
suggest managers use voluntary disclosure to mitigate the negative effects of complex financial
statements on the information environment.
JEL classification: D82; D83; G14; M41; M45;
Keywords: financial statement complexity; voluntary disclosure; information environment
_______________
We thank The Wharton School for financial support. Additionally, we thank an anonymous referee, Brad
Badertscher, Mary Barth, Brian Bushee, Kimball Chapman, Luzi Hail, Michelle Hanlon (co-editor), Mirko Heinle,
Bradford Hepfer, John Kepler, Caleb Rawson, Katherine Schipper, Cathy Schrand, David Tsui, Andy Van Buskirk,
Ro Verrecchia, Joanna Wu (co-editor), and seminar participants at the 2015 AAA annual meeting, the 2015 Penn
State Accounting Research Conference, the 2015 University of Colorado Summer Accounting Research Conference,
the University of Rochester and The Wharton School for helpful comments.
“I am raising the question here and internally at the SEC as to whether investors need, and are
optimally served, by the detailed and lengthy disclosures about all of the topics that companies
currently provide in the reports they are required to prepare and file with us. […] In some cases,
lengthy and complex disclosure may indeed be a direct result of the Commission’s rules.”
– SEC Chair Mary Jo White, speech to the National Association of Corporate Directors, October
15, 2013
1. Introduction
Financial statements provide a structured medium for firms to disclose accounting
information and to supply discussion and analysis that explains financial results. However, the
growing complexity of accounting rules and explanatory language surrounding firms’ financial
statements has led to concerns about the effectiveness of these disclosures in communicating
information to shareholders. Many scholars and practitioners argue that the increasing
complexity of these disclosures makes it difficult for management to bring any given item to
investors’ attention (e.g., Bloomfield, 2002; Hirshleifer and Teoh, 2003; Miller, 2010; KPMG,
2011). Consistent with these arguments, a growing literature documents that complex financial
statements negatively affect the information environment. For example, experimental and
empirical evidence suggests that both professional and non-professional investors fail to
internalize information in complex financial statements, and that more complex financial
statements reduce price efficiency and increase uncertainty.1
With respect to managing their information environment, firms have at their disposal a
variety of disclosure mediums beyond financial statements that can be used to achieve an optimal
information environment (e.g., management forecasts, 8-K filings, press releases). While the
literature noted above documents the potential negative effects of complex financial statements,
1
For experimental evidence, see Hirst and Hopkins (1998), Maines and McDaniel (2000), and Hobson (2006); for a
review see Libby et al. (2002). For empirical evidence, see You and Zhang (2009), Miller (2010), Lehavy et al.
(2011), Lee (2012), Bonsall and Miller (2013), Callen et al. (2013), Lawrence (2013), Loughran and McDonald
(2014), Bozanic and Thevenot (2015), and Lang and Stice-Lawrence (2015).
1
it does not explore whether managers use alternative disclosure channels to mitigate these
effects. We analyze the relation between financial statement complexity and subsequent
voluntary disclosure using both traditional cross-sectional and within-firm designs as well as two
quasi-natural experiments. Across all tests, we find that managers release more frequent and
timely voluntary disclosures when their financial statements are more complex. Collectively, our
findings suggest that managers use voluntary disclosures to mitigate the negative effects of
complex financial statements on the information environment.
Theoretically, the relation between financial statement complexity and voluntary
disclosure depends on how the complexity arises. On the one hand, complex financial statements
might reflect an intentional choice by managers to obfuscate and hide information from
investors, for example poor performance (e.g., Li, 2008). In this case, managers choose to accept
the negative effects of financial statement complexity (as their personal benefits from a low
quality information environment exceed the costs) and are not expected to provide investors with
supplemental information outside of the required financial report. In this case, we expect no
relation––or perhaps even a negative relation––between financial statement complexity and
voluntary disclosure.
On the other hand, complex financial statements might reflect the complexity of the
firm’s business transactions (“business complexity”) and associated reporting standards
(“reporting complexity”).2 In this case, the negative effects of financial statement complexity do
not reflect an intentional choice by managers. Consequently, if the complexity of the financial
statements causes the quality of the information environment to decline, economic theory
suggests managers will use other disclosure channels to improve the information environment
(e.g., Jung and Kwon, 1988; Verrecchia, 1990). For example, when firms engage in complex
2
See for example, Bloomfield (2008) and Bushee et al. (2015).
2
transactions or apply complex reporting standards, investors may have difficulty understanding
the content of the 10-K. Observing these difficulties (or perhaps in some cases anticipating
them), managers may provide additional disclosures to help investors understand the
performance implications of these transactions. In this case, we expect a positive relation
between financial statement complexity and voluntary disclosure.
We examine the relation between financial statement complexity and voluntary
disclosure using multiple measures of these constructs and using three distinct sets of tests.
Following prior literature, we measure financial statement complexity using the readability and
length of the firm’s 10-K filing (e.g., Li, 2008), and we measure voluntary disclosure using the
frequency and immediacy of management forecasts (e.g., Balakrishnan et al., 2014a;
Balakrishnan et al., 2014b).3 In subsequent analyses, we confirm that our results are robust to
using alternative measures of financial statement complexity and to using the frequency and
immediacy of 8-K filings as alternative measures of voluntary disclosure (e.g., Leuz and
Schrand, 2009; Balakrishnan et al., 2014b).
In the first set of tests, we examine the relation between financial statement complexity
and subsequent voluntary disclosure using both traditional cross-sectional and within-firm
designs. A key advantage of using a within-firm design is that it helps alleviate concerns that our
measures of financial statement complexity capture omitted firm-specific characteristics (e.g.,
industry practices). We find robust evidence that complex financial statements are associated
with more frequent and more immediate management forecasts over periods ranging from one
month to twelve months after the filing of the 10-K.
3
With regard to financial statement complexity, an extensive prior literature uses both readability and length of the
firm’s 10-K to measure financial statement complexity, and finds these measures are positively associated with
information processing costs (e.g., You and Zhang, 2009; Miller, 2010; Lehavy et al., 2011; Lee, 2012; Bonsall and
Miller, 2013; Lawrence, 2013; Bozanic and Thevenot, 2015). With regard to voluntary disclosure, management
forecasts are generally considered to be among the most informative forms of voluntary disclosure (see Beyer et al.,
2010 for a review).
3
We assess the robustness of these results to controlling for voluntary disclosure issued
prior to the 10-K and to time trends in both financial statement complexity and voluntary
disclosure. We find (i) a positive relation between financial statement complexity and voluntary
disclosure immediately before the 10-K filing, and (ii) an incremental positive relation between
financial statement complexity and voluntary disclosure after the filing. These findings suggest
that managers appear to anticipate some––but not all––of the information problems related to
complex financial statements. We also find (iii) that our results are robust to estimating annual
cross-sectional regressions, and (iv) that the relation between financial statement complexity and
voluntary disclosure is strongest after RegFD. These findings suggest that our results are robust
to controlling for time-trends in financial statement complexity and voluntary disclosure, and are
consistent with the notion that prior to RegFD, managers could use an alternative channel
(“selective disclosure”) to clarify information in financial statements.
In our second set of tests, we examine cross-sectional variation in the relation between
financial statement complexity and voluntary disclosure. In particular, we examine how this
relation varies with: (1) the change in liquidity at the time the 10-K is filed, (2) the intensity of
external monitoring, and (3) firm performance and earnings management. Such tests are helpful
because, although our main tests examine voluntary disclosure locally around 10-K filings, one
might still harbor a concern that a correlated omitted variable (e.g., a merger) might explain both
complexity of the financial statements and the demand for voluntary disclosure. Testing multiple
predictions, and finding consistent evidence across these predictions, makes it less likely that our
collective results are attributable to alternative explanations.
We find that the relation between financial statement complexity and voluntary
disclosure is stronger when there is a greater reduction in liquidity around the filing of the 10-K,
and when firms have more outside monitors. We also find that the relation between financial
4
statement complexity and voluntary disclosure is weaker when firms have poor performance and
greater earnings management. Collectively, these findings suggest that managers give specific
consideration to the informational problems created by complex financial statements, and
suggest that the positive relation between financial statement complexity and voluntary
disclosure is strongest (weakest) in settings where managers have greater (lesser) incentives to
mitigate the informational problems created by complex financial statements.
In our third set of tests, we examine the relation between financial statement complexity
and voluntary disclosure using two quasi-natural experiments. Specifically, we use a generalized
difference-in-differences design to examine changes in voluntary disclosure around the adoption
of SFAS 133 (Accounting for Derivatives) and the adoption of SFAS 157 (Fair Value
Measurements). The advantages of this analysis are two-fold. First, because these are two of the
more complex accounting standards (KPMG, 2011), this analysis allows us to validate that our
text-based measures of financial statement complexity reflect, at least in part, the complexity of
the underlying accounting rules. Second, we can use the adoption of these accounting standards
to isolate the effect of reporting complexity on financial statement complexity, and in turn
voluntary disclosure. These tests provide additional direct evidence on whether managers give
specific consideration to the informational problems created by complex accounting standards.
We find that the adoptions of both SFAS 133 and SFAS 157 are associated with an
increase in financial statement complexity for affected firms, and in turn, an increase in the
firms’ voluntary disclosure. These results suggest managers give specific consideration to the
complexity of accounting standards, and help alleviate concerns that a correlated omitted
variable explains both complexity of the financial statements and the demand for voluntary
disclosure (e.g., that a merger may increase the complexity of the 10-K filing and simultaneously
create a demand for voluntary disclosure). To explain our results, an omitted variable would need
5
to be positively correlated with financial statement complexity, with voluntary disclosure, with
the timing of the rule changes, and with the likelihood of being affected by the new rules.
Taken together, our results suggest a more nuanced view of how complex financial
statements affect the mosaic of public information about the firm––while prior research
documents that complex financial statements negatively affect the information environment, our
results suggest that some firms attempt to mitigate these effects using voluntary disclosure. Our
study contributes to the literature that examines how managers use different disclosure mediums
to manage the information environment, and our findings are consistent with predictions from
extant theory that managers trade-off various disclosure mediums in attempting to achieve an
optimal information environment.
The remainder of the paper proceeds as follows. Section 2 reviews the related theoretical
and empirical literature on financial statement complexity and the relation between mandatory
and voluntary disclosure. Section 3 describes the research design and measurement choices.
Section 4 describes the sample. Section 5 presents the results and discusses our robustness tests.
Section 6 discusses several supplemental analyses, and Section 7 provides concluding remarks.
2. Literature review
2.1 Theory Literature
More complex financial statements require more time and effort to extract relevant
information, which makes them more costly to parse by investors (e.g., Bloomfield, 2002).
Regulators have long voiced concerns about excessively lengthy and complex financial
statements. In 1969, the SEC published the “Wheat Report,” which noted that the average
investor was unable to readily understand firms’ prospectuses, and recommended that companies
avoid unnecessarily complex, lengthy or verbose writing. More recently, the SEC adopted the
6
“plain English” rule in an effort to increase disclosure readability (SEC Rule 421(d)).4 Despite
these efforts, concerns remain that financial statement complexity adversely affects both
unsophisticated and sophisticated investors. For example, a recent study by KPMG finds that the
complexity of financial statements and associated information processing costs have continued to
grow––primarily due to changes in disclosure requirements related to fair value accounting,
derivatives, and hedging (KPMG, 2011).5
It is well known that an increase in information processing costs leads to a decrease in the
average precision of investors’ beliefs about future cash flow (e.g., Grossman and Stiglitz, 1980;
Kim and Verrecchia, 1991). In early theory papers on voluntary disclosure, Jung and Kwon
(1988) and Verrecchia (1990) show that voluntary disclosure depends on the precision of
investors’ beliefs about future cash flow—an increase in uncertainty about future cash flow (i.e.,
a reduction in precision) causes an increase in disclosure.
Our paper focuses on financial statement complexity as it relates to information
uncertainty. Taylor and Verrecchia (2015) discuss how uncertainty about future cash flow is
driven by the sum of two forces––the real volatility of future cash flow (fundamental
uncertainty) and common knowledge about that cash flow (information uncertainty). Either an
increase in fundamental uncertainty or an increase in information uncertainty works to increase
total uncertainty about future cash flow. Managers may not be able to reduce fundamental
uncertainty using voluntary disclosure (i.e., reduce the real volatility of cash flow), but may be
able to reduce information uncertainty using voluntary disclosure (i.e., increase common
4
Additional initiatives include the SECs “21st Century Disclosure Initiative” in 2008, the FASBs “Disclosure
Framework Project” in 2011, and the UK Financial Reporting Council’s “Cutting Clutter” initiative in 2011.
5
As an example of a practitioner-oriented perspective on financial statement complexity, see “The 109,894-Word
Annual Report,” Wall Street Journal, June 2, 2015.
7
knowledge).6 This motivates our focus on financial statement complexity, and on alleviating
“informational problems.”
We make two additional points regarding the content and timing of firms’ voluntary
disclosures. First, although economic theory predicts managers use voluntary disclosure to
alleviate the informational problems associated with complex financial statements, the theory
provides little guidance on the content or medium of the voluntary disclosure. For example,
suppose that in the year of adoption, SFAS 133 created some uncertainty about the hedging
aspects of a firm’s derivatives positions. Further, assume that observing this confusion,
management decides to provide voluntary disclosure to assist investors in sorting out the
implications of SFAS 133 for future cash flow. One possibility might be to issue a press release
or hold a conference call to provide additional detail on the derivatives positions. Another
possibility, however, might be to issue a management forecast of earnings, thereby helping
investors synthesize the information that was provided in the 10-K (i.e., effectively showing
investors how the derivatives disclosures map into future earnings). This example illustrates that
our empirical predictions do not require that the additional voluntary disclosure be verbiage that
is directly tied to the source of increased uncertainty.
Second, the theory provides little guidance on the timing of voluntary disclosures. When
managers anticipate informational problems, they could conceivably pre-empt these problems––
either by altering the content of the financial statements or increasing voluntary disclosure prior
to filing the financial statements. However, the fact that prior research finds complex financial
statements negatively affect the information environment suggests managers are unable (or
unwilling) to mitigate these effects by altering the content of the financial statements themselves.
This is perhaps not surprising given that predicting precisely where and when investors will
6
For example, several papers suggest that firms with more volatile fundamentals are less likely to provide voluntary
disclosure (e.g., Waymire, 1985; Chen et al., 2011).
8
struggle to understand a given set of financial statements is likely to prove difficult.7 Hence,
managers may need to observe analysts’ and investors’ interpretation of financial statement
information before understanding the source of uncertainty and how to resolve it. Consequently,
attempts to mitigate informational problems created by complex financial statements are likely
to, in large part, occur after the financial statements are filed. However, the expected timing of
any subsequent voluntary disclosure is not clear. We therefore tabulate our results over several
windows following the 10-K filing.
2.2 Empirical literature
Consistent with concerns in regulatory and practitioner circles that an increase in
financial statement complexity negatively affects the information environment, empirical
research shows that complex financial statements are associated with lower trading volume and
ownership among retail investors (e.g., Miller, 2010; Lawrence, 2013), increased analyst forecast
dispersion and reduced analyst forecast accuracy (e.g., Lehavy et al., 2011; Bozanic and
Thevenot, 2015), disagreement among credit rating agencies (e.g., Bonsall and Miller, 2013),
reductions in the extent to which prices impound information (e.g., You and Zhang, 2009; Lee,
2012; Callen et al., 2013), and increased idiosyncratic volatility (e.g., Loughran and McDonald,
2014). While the extant literature documents the potential negative effects of financial statement
complexity, it does not examine whether managers use alternative disclosure channels to
mitigate these effects.
One view of complex financial statements is that they reflect an information-based
agency problem. While the theoretical models discussed earlier are predicated on the notion that
managers seek to maximize firm value, this need not be the case. Managers may intentionally
7
Academic authors will appreciate the challenge in conveying complex information. Even when authors are using
their best efforts to be transparent, it may nevertheless be difficult to anticipate where a reader will struggle with the
material.
9
choose to add (unnecessary) complexity to financial statements if the personal benefits from
doing so exceed the costs. For example, Li (2008) interprets the negative relation between firm
performance and financial statement complexity as evidence that managers intentionally increase
financial statement complexity to obfuscate poor performance. If high levels of financial
statement complexity reflect an intentional choice to obfuscate information, then it seems
unlikely that these managers would use alternative disclosure channels to increase the quality of
the information environment.
An alternative view of complex financial statements is that they reflect the complexity of
the firm’s business transactions and associated reporting and disclosure rules. That is, managers
attempt to avoid unnecessary complexity when preparing financial statements, but complex
transactions, complex reporting and mandatory disclosure rules (e.g., consolidation accounting,
hedge accounting, or accounting for financial assets) may necessitate complex financial
statements. In this case, as suggested by extant theory, voluntary disclosure may be a
supplemental medium that allows managers to communicate information to investors, and
thereby mitigate the negative effects of complex financial statements on the information
environment.
To be clear, the notion that financial statement complexity might not be the result of
intentional obfuscation is not new. Antecedent work that makes this point includes Bloomfield
(2008) and Bushee et al. (2015).8 The distinguishing feature of our study is that we examine how
financial statement complexity relates to voluntary disclosure, where this relation is predicted to
depend on the source of the complexity. While prior literature has examined the complexity of
8
Bloomfield (2008) suggests an alternative explanation for the negative relation between firm performance and
financial statement complexity is that poor performance requires managers to provide more detailed explanations.
Bushee et al. (2015) examines the complexity of language used in quarterly conference calls and compares the
complexity of language used by analysts on the call with that of managers. Conditional on analysts using complex
language on the call, Bushee et al. (2015) finds that managers use complex language to convey (rather than
obfuscate) information––when analysts ask complex questions, a complex response is more informative.
10
financial statements in isolation or voluntary disclosure in isolation, it has generally not sought to
link the two. In this regard, our paper is related to Lehavy et al. (2011) who examine how
analysts respond to the informational problems created by complex financial statements. In the
same vein, we examine how managers respond to the informational problems created by
complex financial statements.
Several related studies examine the interplay between firms’ mandatory and voluntary
disclosure. For the most part, these studies focus on measures of earnings quality and document a
positive relation between earnings quality and the incidence, frequency, and accuracy of
voluntary disclosure (e.g., Lennox and Park, 2006; Francis et al., 2008; Gong et al., 2009). One
notable exception is Ball et al. (2012), who find that more “verifiable” mandatory disclosures, as
measured by higher audit fees, are positively associated with the frequency and quality of
management forecasts.9 In contrast to these studies, we show that voluntary disclosure is
sometimes negatively related to the quality of mandatory disclosure. That is, lower information
accessibility in mandatory disclosure (in the form of complexity) is associated with more
information being released through voluntary disclosure. Similar to Balakrishnan et al. (2014a)
who find that managers respond to a reduction in analyst coverage by increasing voluntary
disclosure, our results suggest managers respond to an increase in financial statement complexity
by increasing voluntary disclosure.
3. Research design
3.1 Financial statement complexity and voluntary disclosure
We examine the relation between financial statement complexity and voluntary
disclosure by estimating regressions of the form:
9
Note that one interpretation of these results is that audit fees proxy for the complexity of the financial statements
(auditors charge a higher fee for more complex financial statements), in which case their results could be interpreted
as consistent with our evidence that financial statement complexity is positively associated with voluntary
disclosure.
11
VoluntaryDisct+1 = φ0 + φ1 FS_Complexityt + θ Controlst + εt,
(1)
where FS_Complexity is a measure of financial statement complexity (ReadIndex or Length),
VoluntaryDisc is a measure of voluntary disclosure (Frequency, Frequency90, Frequency60,
Frequency30, or Immediacy), and Controls is a vector of control variables.
Similar to prior research, we measure voluntary disclosure using the frequency and
immediacy of management forecasts. We define Frequency as the number of management
forecasts (including forecasts of EPS, cash flows, sales, etc.) issued during the 12-months
following the filing of the 10-K. To ensure that our results are not specific to a particular window
of disclosure, we also measure the frequency of forecasts over the 90, 60, and 30 days after the
filing of the 10-K, Frequency90, Frequency60, Frequency30, respectively. These narrower
windows allow us to sharpen identification that the disclosure serves as follow up to the 10-K,
but reduce the power of our tests.10
Regarding the immediacy of management forecasts, we define Immediacy as the number
of days between a firm’s 10-K filing and the first management forecast thereafter. We multiply
by negative one to obtain a measure that is increasing in immediacy. For example, an immediacy
of –118 implies the first forecast occurred 118 days after the 10-K was filed. Collectively, these
measures capture management’s use of voluntary disclosure to supply investors with additional
information after the 10-K, as well as how quickly such information is forthcoming.11
10
It is well established that management forecasts represent an important source of firm disclosure (Beyer et al.,
2010), and prior research suggests that managers who wish to improve the information environment issue more
frequent forecasts (see Hirst et al. (2008) for an overview of management guidance practices). We focus our
measures on frequency because such measures can be calculated for all firms, and because these measures capture
all forms of management guidance (e.g., EPS, cash flows, sales, etc.). Inferences are unchanged if we focus
specifically on earnings forecasts
11
From the incidence and timing of voluntary disclosure, it is difficult to infer when managers learned the private
information revealed in their forecast. As a result, we are not able to make statements about when managers obtain
their private information. Our tests rely only on the notion that disclosure reveals private information and improves
the information environment.
12
Following an extensive prior literature, we measure financial statement complexity using
the readability (ReadIndex) and length (Length) of the firm’s 10-K.12 While much of the prior
literature in accounting uses the Fog index to measure readability, to ensure that our results are
not specific to any single measure of readability, and to mitigate the influence of idiosyncratic
measurement error in any given measure, we construct a readability index that combines several
established measures of readability.13
ReadIndex is the first principal component of the following measures of readability:
Flesch-Kincaid readability, LIX readability, RIX readability, Gunning Fog readability, ARI
readability, and SMOG readability. Each of these measures is effectively a function of word
complexity and sentence length, and higher values correspond to less readable text.14 Appendix
B provides detailed definitions of these variables, present the results from our principal
component analysis, as well as the correlations among the six measures of readability and
ReadIndex. The analysis shows that only a single factor has an eigenvalue greater than one, that
this factor (ReadIndex) explains 94% of the variation in these measures, and that all measures of
readability are highly correlated with each other and with ReadIndex.
We measure length of the firm’s 10-K, Length, as the natural logarithm of the number of
words. Because the costs of processing longer documents are presumably higher, longer
documents can be more difficult to read. Consistent with the prior literature, we use the natural
logarithm rather than the raw number of words to reduce the impact of extreme values and
skewness in the number of words across firms.15
12
Several prior papers use readability and length interchangeably as proxies of financial statement complexity. See,
for example, Li (2008), Miller (2010), Lee (2012), Bonsall and Miller (2013), Lawrence (2013).
13
Results are robust to measuring readability using the Fog index, and in some cases stronger.
14
See Jones and Shoemaker (1994) and Moffit and Burns (2009) for reviews of studies using each of these measure
of readability.
15
In untabulated analyses, we find results are robust to using electronic file size of the 10-K filing to measure
financial statement complexity (e.g., Loughran and McDonald, 2014) and to including Length as an additional
variable in the construction of ReadIndex (see Section 6 for more details).
13
In all of our regression specifications, we include the following variables as controls. Size
is the natural logarithm of market value of equity as of the fiscal year-end, ROA is the industryyear adjusted return on assets, measured as income before extraordinary items scaled by total
assets, Loss is an indicator variable equal to one if net income is negative and zero otherwise,
Leverage is long-term debt plus short-term debt scaled by total assets, MTB is the market value
of equity plus book value of liabilities divided by book value of assets, SpecialItems is special
items scaled by total assets, Returns is the buy and hold return over the 12 months prior to the
10-K filing date, and σReturns is the standard deviation of monthly returns over the 12 months
prior to the filing date. See Appendix A for variable definitions. Note that all independent
variables are measured prior to our measures of voluntary disclosure (the dependent variable).
Throughout all of our analyses, we base inferences on standard errors clustered by firm
and filing date, and estimate regressions using the decile ranks of the independent variables
scaled to range from 0 to 1. We use the decile ranks of each independent variable to ensure that
all independent variables are of similar scale. As a result, each coefficient measures the change
in the respective measures of voluntary disclosure when moving from the bottom decile to the
top decile of the respective independent variable, ceteris paribus. This, in turn, allows us to
meaningfully compare the relative economic significance of each variable. The ranked
specification has the added advantage of being robust to both outliers and nonlinearities.16
The theoretical arguments for why financial statement complexity engenders subsequent
voluntary disclosure predict not only that variation in complexity across firms is related to
voluntary disclosure, but also that, within the firm, time-series variation in complexity should be
associated with time-series variation in voluntary disclosure (e.g., Verrecchia, 1990).
To
determine whether within-firm variation in financial statement complexity explains within-firm
16
Inferences are robust to using raw values.
14
variation in voluntary disclosure, we estimate an augmented version of equation (1) that includes
firm fixed effects. This research design eliminates the cross-sectional variation in financial
statement complexity and voluntary disclosure, and relies on within-firm time-series variation to
identify the coefficient on FS_Complexity. If the relation between financial statement complexity
and voluntary disclosure is driven exclusively by cross-sectional differences in firm
characteristics, then holding the firm constant, we expect to find no evidence of a relation
between financial statement complexity and voluntary disclosure. We present results separately,
with and without fixed effects, because the underlying theory we use to motivate our analysis
predicts both cross-sectional and temporal relations with voluntary disclosure, and we wish to
show that both results are observed in the data.17
3.2 Cross-sectional variation in the relation between financial statement complexity and
voluntary disclosure
In this section we describe the research design used to test our cross-sectional predictions.
We first examine how the relation between financial statement complexity and voluntary
disclosure varies with the change in liquidity at the time the financial statements are filed. More
complex financial statements require greater information processing which reduces liquidity, and
extant research suggests that managers use voluntary disclosure to achieve a target level of
liquidity.18 Consequently, we predict that the greater the reduction in liquidity when the financial
statements are filed, the more likely managers are to provide voluntary disclosure subsequent to
filing complex financial statements.
17
For example, the theory that underlies our analysis predicts: (i) industries with more complex financial statements
have more voluntary disclosure and (ii) as financial statement complexity increases over time, so too should
voluntary disclosure. We provide some evidence of these cross-sectional and time-series relations in our two quasinatural experiments. In this regard, controlling for firm fixed effects and temporal trends in complexity can be
viewed as over-controlling. Nevertheless, in untabulated analysis, we find inferences are robust to including industry
fixed effects and year fixed effects. Inferences are also robust to restricting our sample to those firms that have more
than one CEO during our sample period and including both firm and manager fixed effects (e.g., Bamber et al.,
2010; Brochet et al., 2011).
18
See for example, Coller and Yohn (1997), Figure 10 of Graham et al. (2005), and Balakrishnan et al. (2014a).
15
To test this prediction, we interact FS_Complexity in equation (1) with two measures of
the changes in illiquidity around the filing date of the 10-K, and predict positive coefficients on
the interaction terms. We measure the change in illiquidity using both the Amihud (2002)
measure of illiquidity:
Illiquidityt =
Rt
DVolumet
×106
(2)
where Rt is the daily return and DVolumet is the daily dollar volume (in millions), and the bid-ask
spread:
Spread t =
Ask t − Bid t
× 100
price t
(3)
where Askt (Bidt) is the quoted closing ask (bid) on day t, and pricet is the closing price on day t
from CRSP.
We calculate the Amihud (2002) measure of illiquidity and the bid-ask spread at the daily
level and then calculate the change around the filing date, ∆Illiquidity and ∆Spread, as the
difference between the average value on the day of and after the filing (t = 0, 1) and the average
value over the window beginning fifty days prior to the filing and ending five days prior to the
filing (t = –50, …, –5). To compute ∆Illiquidity and ∆Spread, we require non-missing data on
the daily CRSP file for at least 20 days prior to, the day of, and the day after the 10-K filing date.
Our second set of cross-sectional tests examines how the relation between financial
statement complexity and voluntary disclosure varies with scrutiny from external monitors. On
the one hand, managers under a high degree of scrutiny from external monitors may have
stronger incentives to mitigate the negative informational effects of complex financial statements
and maintain the quality of the information environment (e.g., Ajinkya et al., 2005; Armstrong et
al., 2010). On the other hand, in the absence of agency conflicts, a higher level of external
16
monitors might indicate a rich information environment which could decrease managers’
incentives to provide additional disclosure. To examine whether the relation between financial
statement complexity and voluntary disclosure varies with external monitoring, we interact
FS_Complexity in equation (1) with two measures of external monitoring. We measure external
monitoring using the number of analysts following the firm as of the filing date of the 10-K
(NAnalysts) and the number of institutional investors in the firm as of the latest calendar quarterend prior to the 10-K filing date (NInstitutions).19
Our third set of cross-sectional tests examines how the relation between financial
statement complexity and voluntary disclosure varies with firm performance and the level of
earnings management. Conditional on either poor firm performance or earnings management, the
benefits of a low quality information environment accrue to the manager (e.g., a reduced
likelihood of termination or reputational damage, inflated stock price, inflated compensation)
whereas the costs are borne by shareholders (e.g., reduction in liquidity, greater cost of capital).
Consequently, we predict that the positive relation between complexity and voluntary disclosure
is weaker when the firm underperforms its industry peers, posts a loss, and has large abnormal
accruals. To test this prediction, we interact FS_Complexity in equation (1) with two measures of
firm performance, ROA and Loss (as previously defined), and the firm’s performance-matched
abnormal accruals, AbAcc (e.g., Kothari et al., 2005).20
3.3 Two quasi-natural experiments
19
Results are robust to using percent of institutional ownership or the number of large blockholders (those with 5%
or more ownership) to measure external monitoring.
20
Firms are matched to a respective industry-year peer on Compustat by minimizing the absolute difference in ROA.
In untabulated analyses, we examine two additional measures of earnings management: (1) size-age-growth matched
accruals (e.g., Armstrong et al., 2015) and (2) an indicator for meeting-or-beating analyst earnings targets by
$0.01/share or less. While we find inferences are robust to (1), we find the interaction between financial statement
complexity and (2) is statistically insignificant at conventional levels. It is unclear how to interpret this finding in
light of the fact that analyst forecast accuracy is itself affected by both complexity and voluntary disclosure (see
discussion in Section 6.2).
17
In this section, we examine changes in financial statement complexity and voluntary
disclosure around the adoption of SFAS 133 (Accounting for Derivatives) and the adoption of
SFAS 157 (Fair Value Measurements) using a generalized difference-in-differences design.21
Both rules require extensive disclosures about fair value measurements, which we expect will
increase the complexity of affected firms’ financial statements––and if our hypotheses above are
correct––will also increase voluntary disclosure.22 The advantage of focusing on these two
changes in accounting standards are two-fold. First, because these are two of the more complex
accounting standards, the analysis allows us to validate that our text-based measures of financial
statement complexity reflect, at least in part, complexity of the underlying accounting and
mandatory disclosure rules. Second, we can use the adoption of these accounting standards to
isolate the effect of reporting complexity on financial statement complexity, and in turn
voluntary disclosure. These tests provide direct evidence on whether managers give specific
consideration to the complexity of accounting standards.
Regulators’ motives for the passage of fair value accounting standards such as SFAS 133
and SFAS 157 primarily relate to providing users with an understanding of the value of complex
assets and the sources of information used to estimate these values (see FASB, Appendix C of
SFAS 133 and Appendix C of SFAS 157). However, anecdotal evidence and academic studies
suggest that, to this day, the complexity associated with these standards impairs market
21
For a difference-in-differences design to provide meaningful estimates, the assignment variable needs to be
exogenous. The adoption of these accounting standards is “plausibly exogenous” with respect to the average firm’s
voluntary disclosure and financial statement complexity. So long as the change in standards is not the result of
actions on the part of the average firm, the change in standards can be treated as exogenous for the purposes of
econometric analysis (see for example, Larcker et al., 2011).
22
SFAS 133, Accounting for derivative instruments and hedging activities, was adopted for fiscal periods beginning
after June 15, 2000, and mandates that a firm record changes in the fair value of its cash flow hedges in accumulated
other comprehensive income (AOCI), a component of shareholders’ equity. As such, unrealized changes in fair
value appear on the balance sheet but do not appear on the income statement until they are realized. SFAS 157, Fair
value measurements, effective for fiscal periods beginning after November 15, 2007, provides a framework for the
measurement of financial assets and liabilities at fair value. Specifically, it distinguishes between three levels of
inputs used to derive fair value estimates: level 1 inputs consist of observable quoted prices in active markets for
identical assets or liabilities; level 2 inputs consist of observable sources other than quoted prices; and level 3 inputs
consist of unobservable, firm-supplied sources.
18
participants’ ability to understand and price this information (e.g., Riedl and Serafeim, 2011;
Campbell, 2015). For example, a recent KPMG survey seeking to identify sources of disclosure
overload and complexity concludes that “fair value, derivatives and hedging are the most
significant sources of disclosure complexity under specific GAAP requirements” (KPMG, 2011,
page 18).23
We identify the firms affected by the adoption of SFAS 133 and the adoption of SFAS
157 in our sample (treatment or affected group), and construct matched samples of unaffected
firms using propensity score matching (control or unaffected group). Specifically, we form oneto-one matched-pairs by estimating a propensity score in the year prior to the respective change
in accounting standards as a function of the control variables in equation (1). We then match
each affected firm to the corresponding unaffected firm that minimizes the squared difference in
propensity scores between the two firms. We repeat this procedure for each rule change,
resulting in separate treatment and control groups for SFAS 133 and SFAS 157. The test for
covariate balance between the two groups appears in Table C1 in Appendix C.
The advantage of using a matched sample of firms as a control group is two-fold. First, if
observations with certain characteristics (e.g., size) are more likely to be treated and to have a
differential trend in voluntary disclosure, controlling for these characteristics mitigates concerns
that the treatment is non-random. Second, using treatment and control groups that are similar
with respect to such characteristics can improve precision and better isolate the treatment effect
(Roberts and Whited, 2013).24
23
To provide an anecdote, in 2004 the SEC directed Fannie Mae to restate its earnings by $9 billion of losses on
derivatives due to improper application of SFAS 133. According to a case study by Risk Limited Corporation,
Fannie Mae failed to fully understand the SFAS 133 rules, which “illustrates the complexity and effort required for
ongoing FAS 133 compliance” (p. 16). See case study at: www.energyrisk.org/deutsch/fas-133-fannie-mae.ppt. Note
that our test rely only on the notion that the adoptions of SFAS 133 and 157 increased information processing costs
within three years of the respective adoptions.
24
Results are robust to using the entire sample of unaffected firms as the control group rather than the propensity
score matched sample.
19
We employ a generalized difference-in-differences design. Specifically, for each rule
change, we estimate the following regression:
VoluntaryDisct+1 = φ0 + φ1 Postt*Affected + f + δ + θ Controlst + εt,
(4)
where Post is an indicator variable equal to one for fiscal years after the rule change became
effective, and zero otherwise, Affected is an indicator variable equal to one for firms affected by
the respective rule change, and zero otherwise, f and δ are vectors of firm and year fixed effects,
respectively, and all other variables are as previously defined. We estimate equation (4) over a
period of six years around each regulatory change (i.e., three years prior to and three years after
the change). For SFAS 133, Affected firms are those reporting unrealized gains and losses on
derivatives in accumulated other comprehensive income (e.g., Campbell, 2015), and the preperiod (post-period) refers to three fiscal periods beginning before (after) June 15, 2000. For
SFAS 157, Affected firms are those reporting level 1, level 2 or level 3 assets or liabilities (e.g.,
Riedl and Serafiem, 2011), and the pre-period (post-period) refers to three fiscal periods
beginning before (after) November 15, 2007.
The distinguishing feature of the “generalized” difference-in-differences design is that it
includes both firm and year fixed effects, which controls for any fixed differences between
affected and unaffected firms (firm fixed effects absorb the Affected main effect) and any time
trends (year fixed effects absorb the Post main effect). Thus, the coefficient on the interaction
term, Post*Affected, represents the average treatment effect (or DiD ATE); see Bertrand et al.
(2004), Hansen (2007), and Angrist and Pischke (2009) for details on this design.
4. Sample and descriptive statistics
4.1 Sample construction
We construct our sample using data from Compustat, CRSP, I/B/E/S, Thomson-Reuters
database of 13-F filings, and the WRDS SEC Analytics Suite. The sample begins in 1995, when
20
management earnings forecasts first become available in the I/B/E/S Guidance database, and
ends in 2012. To be included in the sample, a firm must have a 10-K filing on EDGAR between
1995 and 2012, sufficient financial statement data on Compustat to calculate our control
variables, and monthly stock returns over the 12 months prior to the 10-K filing on CRSP. The
resulting sample consists of 72,366 firm-years.25
In two of our cross-sectional tests, we impose additional sample requirements. In the
cross-sectional tests that involve market-based measures of changes in illiquidity (∆Illiquidity
and ∆Spread), presented in Table 4, the sample is reduced to 69,066 observations due to data
requirements to calculate daily bid/ask spreads. In tests that involve earnings management,
presented in Table 6, the sample is reduced to 59,068 firm-years due to data requirements to
estimate performance-matched abnormal accruals (AbAcc).
4.2 Descriptive statistics
Table 1, Panel A presents descriptive statistics for our sample. Firms provide, on average,
2.62 forecasts per year, and for those firms providing a forecast, the first forecast occurs on
average 108.52 days after the 10-K filing. The average 10-K in our sample has a ReadIndex of –
0.01 and Length of 10.25, (or 33,896 words).26 Descriptive statistics for our control variables
show that our sample firms have a mean (median) industry-adjusted return-on-assets of 0.00
(0.02), a mean (median) leverage ratio of 0.22 (0.16), a mean (median) market-to-book ratio of
1.94 (1.03), and mean (median) special items of –0.02 (0.00). Approximately 32% of firms in
our sample report a loss (mean Loss is 0.32), and the average (median) annual buy and hold
return is about 14% (5%) over the twelve months prior to the 10-K. Similar to prior research
25
Without requiring control variables we have data for 86,236 10-Ks with CRSP identifiers over the 1995 to 2012
period.
26
The statistics for Length are similar to prior research (e.g., Li, 2008).
21
(e.g., Lehavy et al., 2011; Akins et al., 2012), average (median) analyst coverage is 4.61 (2.00)
analysts and the average (median) number of institutional investors is 97.97 (51.00).
Table 1, Panel B presents correlations among the variables used in our analysis, with
Spearman (Pearson) correlations above (below) the diagonal. The Spearman (Pearson)
correlation between our two measures of financial statement complexity is 0.47 (0.49), and the
correlation between our measures of voluntary disclosure over the various windows ranges
between 0.33 (Spearman correlation between Frequency and Frequency30) and 0.89 (Pearson
correlation between Frequency and Frequency90). Notably, all measures of voluntary disclosure
are positively correlated with all measures of financial statement complexity.
Table 2 presents the average values of our measures of voluntary disclosure by quintile of
financial statement complexity. Table 2 shows that the lowest quintile of ReadIndex (Length) has
on average 1.27 (0.61) forecasts over the 12 months following the 10-K filing, while the highest
quintile has on average 3.40 (4.17) forecasts, a difference of 2.13 (3.56) forecasts. Similarly, the
first forecast for the lowest quintile of ReadIndex (Length) comes on average 123.86 (145.78)
days after the filing of the 10-K, while the first forecast for the highest quintile comes on average
101.05 (96.27) days after the filing of the 10-K, a difference of 22.81 (49.51) days. All
differences are positive and statistically significant (two-tailed p-values <0.001).
Although effectively a univariate analysis and subject to the standard limitations, the
results in Table 2 are consistent with a positive relation between financial statement complexity
and voluntary disclosure. Figure 1 in Appendix C plots cumulative voluntary disclosure by
month after the 10-K filing for firms in the top and bottom quintiles of financial statement
complexity, where the initial value in month 0 is the number of management forecasts issued
between the fiscal year-end date and the 10-K filing date.
5. Results
22
5.1 Financial statement complexity and voluntary disclosure
Table 3 presents results from estimating equation (1). In Panel A, the dependent variable
is voluntary disclosure measured over a 12-month window following the filing of the 10-K
(Frequency), in Panel B a 90-day window following the filing of the 10-K (Frequency90), in
Panel C a 60-day window (Frequency60), in Panel D a 30-day window (Frequency30), and in
Panel E voluntary disclosure is measured as the number of days between the 10-K filing date and
the first management forecast thereafter, multiplied by minus one (Immediacy). In each panel,
financial statement complexity is measured by ReadIndex in columns (1) and (2) and Length in
columns (3) and (4).
Columns (1) and (3) of each panel present results from estimating pooled regressions. We
find positive and significant coefficients on both ReadIndex and Length in 7 out of 8 pooled
specifications across Panels A through D (two-tailed p-values < 0.01). The coefficients on
ReadIndex and Length become monotonically larger as the length of the window increases,
indicating that the bulk of the voluntary disclosures occur more than 30-days after the 10-K.
Despite that fact that managers do not often issue forecasts before the next earnings
announcement (average value of Frequency30 is 0.07; see Table 1), we do find results over
shorter windows, albeit weaker than over the longer windows. This suggests that the
informational problems created by complex financial statements are sufficiently large to
accelerate disclosure of forecasts, although not strong enough to result in instantaneous
disclosure.
As noted above, one explanation for the results being weaker in the 30-day window is
that managers may require time to sort out the extent to which the 10-K creates informational
problems and the issues that create such problems. That is, managers may need to observe
analysts’ and investors’ interpretation of financial statement information before understanding
23
the source of uncertainty and how to resolve it. Together with the fact that management rarely
issues forecasts within 30 days of the release of the 10-K, the 60-day or 90-day windows may be
the most convenient time to provide a forecast that aids investors in interpreting information
provided in a recent 10-K release.
We also find positive and significant coefficients on Immediacy in columns (1) and (3) of
Panel E, and the signs on the coefficients of our control variables are generally consistent with
prior literature (e.g., Ajinkya et al., 2005; Hirst et al., 2008). Combined, these results suggest that
firms issue both more frequent and more immediate management forecasts when their financial
statements are complex.
The marginal effect of financial statement complexity on voluntary disclosure also
appears to be economically significant. A firm in the top decile of ReadIndex issues about one
additional management forecast over the 12 months following the 10-K (φ1 = 0.98, t-stat = 7.40
in Panel A, column (1)), with the first forecast issued approximately 14 days earlier (φ1 = 14.49,
t-stat = 5.09 in Panel E, column (1)), relative to a firm in the bottom decile of ReadIndex, ceteris
paribus.27 Similarly, a firm in the top decile of Length issues about three additional management
forecasts over the 12 months following the 10-K filing (φ1 = 2.90, t-stat = 21.32 in Panel A,
column (3)), with the first forecast issued approximately 48 days earlier (φ1 = 48.40, t-stat =
11.15 in Panel E, column (3)) relative to a firm in the bottom decile of Length, ceteris paribus.
By comparison, the average firm issues 2.6 management forecasts.
Columns (2) and (4) of each panel present results after including firm fixed effects in the
respective regression specification. By including firm fixed effects, the regression coefficients
capture within-firm variation rather than cross-sectional variation, and provides an interesting
27
Recall that all independent variables are scaled decile ranks. As a result, the coefficient measures the change in the
dependent variable when moving from the bottom decile to the top decile of the respective independent variable,
ceteris paribus.
24
time-series perspective on how financial statement complexity influences firms’ voluntary
disclosure practices. The coefficients on both ReadIndex and Length are statistically significant
in all specifications across the panels (two-tailed p-values < 0.10) indicating that within-firm
changes in financial statement complexity are positively associated with within-firm changes in
management forecast frequency and immediacy. As in our pooled specification, the results
appear economically significant. For example, the results in column (4) of Panel E suggest that a
firm in the top decile of Length issues its first forecast approximately 78 days earlier (φ1 = 78.44,
t-stat = 13.22) than a firm in the bottom decile of Length.28
5.2 Cross-sectional variation in the relation between financial statement complexity and
voluntary disclosure
5.2.1 Changes in illiquidity around the filing of the 10-K
Table 4 presents results from estimating a model similar to Panel A of Table 3, except
that we include measures of the change in illiquidity around the 10-K filing date (∆Illiquidity and
∆Spread) and interact these measures with our measures of financial statement complexity.
Columns (1) through (4) of Table 4 present results using ReadIndex to measure financial
statement complexity, and columns (5) through (8) present results using Length to measure
financial statement complexity. We estimate regressions that interact each illiquidity measure
separately, followed by specifications that include both illiquidity variables in the same
regression specification, followed by specifications that include firm fixed effects. Finding
consistent evidence across multiple specifications where interactions are included separately and
simultaneously suggests our results are not influenced by multicollinearity.
28
Note that the signs on MTB and Leverage flip between our pooled and within-firm specifications which highlights
the differences between these two settings. Our results indicate that in the cross-section, firms with higher growth
options (lower leverage) tend to provide more voluntary disclosure. Over time within a given firm, however, the
results suggest that firms provide more voluntary disclosure as growth options decline and leverage increases. A
decline in growth options and an increase in leverage are likely to be issues of concern for both managers and
investors (or at least issues that can require some clarity).
25
Consistent with our predictions, across all specifications, we find that the coefficients on
the interaction between our measures of financial statement complexity and changes in illiquidity
are positive and statistically significant at conventional levels. These findings indicate that
managers appear to give specific consideration to the effect of the 10-K filing on liquidity when
deciding to provide voluntary disclosure.29
5.2.2
External monitors
Table 5 presents results from estimating a model similar to equation (1), except that we
include the measures of external monitoring (NAnalysts and NInstitutions), and interact these
measures with our measures of financial statement complexity. Columns (1) through (4) present
results using ReadIndex to measure financial statement complexity, and columns (5) through (8)
present results using Length to measure financial statement complexity. Consistent with the
prediction that managers under a high degree of scrutiny from external monitors have stronger
incentives to mitigate the negative informational effects of complex financial statements, across
all specifications, we find that the relation between financial statement complexity and voluntary
disclosure is significantly stronger when the firm has a greater number of analysts and
institutional investors.
Notably, our results also suggest that the sign of the relation between financial statement
complexity and voluntary disclosure depends on the presence of external monitors (i.e., the main
effect of FS_Complexity is negative and significant across all specifications). That is, while firms
in the top decile of analyst coverage or number of institutional investors provide more voluntary
29
We note that the main effect on ∆Illiquidity is often negative and significant across specifications (similarly, the
main effect on ∆Spread is negative and significant in one specification). We urge caution in interpreting this result.
Liquidity and bid-ask spread reflect both fundamental uncertainty driven by volatility of firms’ cash flows and
information uncertainty driven by the information environment (Taylor and Verrecchia, 2015). Consequently, if
firms with the least complex financial statements experience a large change in illiquidity around their 10-K filing,
that change is more likely attributable to fundamental uncertainty (as opposed to information uncertainty), and prior
literature suggests that firms with more volatile fundamentals are less likely to disclose (e.g., Waymire, 1985; Chen
et al., 2011).
26
disclosure after filing complex financial statements, firms in either the bottom decile of analyst
coverage (firms with zero analyst coverage) or in the bottom decile of institutional investor
coverage (firms with between 1 and 6 institutional investors) provide less voluntary disclosure
after filing complex financial statements. This is consistent with the notion that scrutiny from
external monitors disciplines managers’ disclosure decisions, and in the absence of that scrutiny,
managers can more easily engage in self-interested activity at the expense of shareholders (e.g.,
obfuscation).
5.2.3
Firm performance and earnings management
Table 6 presents results from estimating a model again similar to equation (1), except that
we include measures of firm performance (ROA and Loss) and earnings management (AbAcc) in
the model, and interact these measures with our measures of financial statement complexity.
Columns (1) through (5) present results using ReadIndex to measure financial statement
complexity, and columns (6) through (10) present results using Length to measure financial
statement complexity. Consistent with our predictions, across all specifications, we find: (i) that
the coefficient on the interaction between financial statement complexity and ROA is positive
and statistically significant, (ii) that the coefficient on the interaction between financial statement
complexity and Loss is negative and statistically significant, and (iii) that the coefficient on the
interaction between financial statement complexity and AbAcc is negative and statistically
significant. These findings are consistent with the notion that, conditional on either poor firm
performance or earnings management, the benefits of a low quality information environment
accrue to the manager whereas the costs are borne by shareholders.
Collectively, the results from examining cross-sectional variation in the relation between
financial statement complexity and voluntary disclosure suggest that managers are giving
specific consideration to the informational problems created by complex financial statements,
27
and suggest that the positive relation between financial statement complexity and voluntary
disclosure is strongest (weakest) in settings where managers have greater (lesser) incentives to
mitigate the informational problems created by complex financial statements. Testing multiple
predictions, and finding robust consistent evidence across these predictions, makes it less likely
that our collective results are attributable to alternative explanations.30
5.3 Robustness analyses
5.3.1 Pre-emptive voluntary disclosure
To examine whether managers anticipate some of the informational problems associated
with filing complex financial statements and pre-emptively issue voluntary disclosure, we reestimate our primary tests using voluntary disclosure issued in anticipation of the 10-K filing as
the dependent variable. We measure voluntary disclosure issued in anticipation of the 10-K filing
as the number of management forecasts issued between the fiscal year-end date and the 10-K
filing date, inclusive (PreFrequency).31 If managers anticipate financial statement complexity
and pre-empt complex financial statements with voluntary disclosure, we predict a positive
association between financial statement complexity and voluntary disclosure issued prior to the
10-K.
We also examine whether the determinants of voluntary disclosure after filing a complex
10-K are distinct from the determinants of voluntary disclosure before the filing of a complex
10-K. If managers do not fully anticipate the informational problems associated with filing
30
For parsimony, we tabulate results from all our tests of cross-sectional variation using Frequency as the dependent
variable. Using Frequency90, Frequency60, or Frequency30 as the dependent variable yields similar inferences.
When Immediacy is defined for both forecasters and non-forecasters (i.e., set to –365 when the firm does not issue a
forecast), our cross-sectional tests on Immediacy also yields similar inferences. However, when Immediacy is
defined only for forecasters (set to missing when the firm does not issue a forecast) the sample shrinks to 26,509
firms (see Table 3, Panel E) and our cross-sectional results are mixed.
31
Note that any management forecasts on the day of the 10-K filing or during the earnings announcement will be
included in PreFrequency (and are excluded from our measures of voluntary disclosure after the filing). Results are
robust to extending the pre-emptive disclosure windows backward to include the 30, 60, or 90 days prior to the fiscal
year-end.
28
complex financial statements––and hence the determinants differ––we predict an incremental
relation between financial statement complexity (FS_Complexity) and voluntary disclosure
subsequent to the filing (Frequency) after controlling for voluntary disclosure prior to the filing
(PreFrequency). Otherwise, if the determinants of voluntary disclosure do not differ before and
after the filing of a complex 10-K, then controlling for voluntary disclosure before the 10-K, we
expect to find no evidence of a relation between financial statement complexity and voluntary
disclosure after the 10-K.
Panel A of Table 7 presents results from estimating a model similar to equation (1),
except that we use voluntary disclosure prior to the 10-K as our dependent variable
(PreFrequency). Panel A shows positive and statistically significant coefficients on both
measures of financial statement complexity (ReadIndex in columns (1) and (2), and Length in
columns (3) and (4)). These results suggest that firms do tend to provide more voluntary
disclosure in anticipation of filing complex financial statements. Panel B of Table 7 presents
results from repeating our primary tests in equation (1), but including PreFrequency as an
additional control variable. Panel B shows that voluntary disclosure prior to the filing of the 10K is very highly correlated with voluntary disclosure after the filing of the 10-K. Importantly, all
coefficients on our measures of financial statement complexity remain positive and statistically
significant. Collectively, the evidence in Table 7 suggests a positive relation between financial
statement complexity and voluntary disclosure immediately prior to the filing, and an
incremental association between financial statement complexity and voluntary disclosure
subsequent to the filing (after controlling for voluntary disclosure prior to the filing).
5.3.2 Alternative measures of voluntary disclosure: 8-K filings
We examine the robustness of our results to using the frequency and immediacy of firms’
8-K filings as alternative measures of voluntary disclosure. Prior research suggests there is a
29
significant discretionary component to 8-K filing decisions and that managers actively use 8-Ks
to alter their information environment (e.g., Leuz and Schrand, 2009; Balakrishnan et al., 2014b).
However, there is undoubtedly a mandatory aspect to these disclosures. In this regard, we
caution that management forecasts are perhaps a cleaner measure of voluntary disclosure than 8K filings, as 8-K filings can co-mingle both voluntary and mandatory disclosures.32
Nevertheless, we examine these additional measures of voluntary disclosure to ensure that our
results are not specific to management forecasts, but rather apply to a broader array of firms’
voluntary disclosures.
Mirroring our management forecast variables, we define Frequency8K, Frequency8K90,
Frequency8K60 and Frequency8K30 as the number of 8-Ks filed during the 12-month, 90-day,
60-day, and 30-day period following the 10-K, and Immediacy8K as the number of days between
a firm’s 10-K and its first 8-K filing thereafter. Panel A of Table 8 presents the average value for
each measure by quintile of financial statement complexity, and the difference between the
extreme quintiles. Panel A shows that the lowest quintile of ReadIndex (Length) has on average
4.93 (2.61) 8-Ks over the 12 months following the 10-K, while the highest quintile has on
average 9.39 (13.64) 8-Ks, a difference of 4.46 (11.03). Similarly, the first 8-K for the lowest
quintile of ReadIndex (Length) comes on average 84.25 (112.03) days after the filing of the 10K, while the first 8-K for the highest quintile comes on average 56.69 (33.96) days after the
filing of the 10-K, a difference of 27.56 (78.07) days. All differences are positive and statistically
significant (two-tailed p-values < 0.01).
32
As exploratory analysis in this regard, we examined characteristics of various types of 8-Ks. For example,
focusing on 8-Ks with material contracts that must be disclosed (e.g., Li, 2013), we find firms with longer 10-Ks
have significantly more 8-Ks related to material contracts/agreements. Firms in the bottom (top) quintile of length
have on average 0.1 (2.1) 8-Ks related to material contracts/agreements 12 months after the 10-K. We also find that
firms with less readable (longer) 10-Ks also have less readable (longer) 8-Ks. However, interpreting the readability
and length of 8-Ks is challenging, as the content is not standardized across 8-Ks. For example it is difficult to
compare the readability or length of a supply contract with, say, the readability or length of a director resignation.
30
Panel B presents results from repeating the tests in Panels A through E of Table 3, using
the each of the five measures of 8-K disclosure as the dependent variable. For each of the five
measures of 8-K disclosure, columns (1) through (5), we present results separately for our two
measures of financial statement complexity (ReadIndex and Length) and two regression designs
(pooled and within-firm), for a total of 20 specifications (5 x 2 x 2). Each column mirrors a Panel
from Table 3. For parsimony we do not tabulate coefficients on control variables. Regardless of
specification, we find the coefficients on our measures of financial statement complexity are
positive and both statistically and economically significant.
5.3.3 Time-series variation
To ensure that our results are not an artifact of common time-trends in financial statement
complexity and voluntary disclosure, we estimate separate annual cross-sectional regressions of
equation (1) from 1995 to 2012. This procedure allows both the intercept and coefficients to vary
by year and produces a time-series of 18 coefficients (e.g., Fama and MacBeth, 1973). We
estimate annual cross-sectional regressions for four separate regression specifications: one for
each measure of financial statement complexity and one for each of management forecasts and 8Ks (2 x 2). Panel A of Table 9 presents the average of the coefficients (FM coef.) and t-tests for
whether the average coefficient is different from zero (FM t-stat). For parsimony we do not
tabulate coefficients on control variables. In three out of the four specifications, we find that the
average coefficient on financial statement complexity is positive and statistically significant.
Figure 2 in Appendix C plots the annual cross-sectional coefficients over our sample period for
each of the four specifications.
Next, we examine various sub-periods. First, we analyze the period from 1995 to 1999,
before “Regulation Fair Disclosure” (“RegFD”) became effective. Prior to RegFD, managers
could use an alternative channel––private communication (or “selective disclosure”)––to clarify
31
information in financial statements. However, beginning in October 2000, RegFD prohibited
firms from disclosing material information to select groups of market participants and effectively
eliminated this channel. Second, we analyze the period from 2000 to 2004, after RegFD but
before the “Additional Form 8-K Disclosure Requirements and Acceleration of Filing Date” rule
(“Rule8K”) became effective. Rule8K, which became effective in August 2004, expanded the
mandatory disclosure requirements of Form 8-K filings (e.g., required firms to furnish earnings
announcements on Form 8-K) and shortened the filing deadlines for certain reportable events to
four business days after the occurrence of the event.33 Third, we analyze the sample period from
2005 to 2012, after both RegFD and Rule8K became effective.
Panel B presents results from estimating equation (1) over each of the three sub-periods.
As before, we estimate four specifications: one for each measure of financial statement
complexity and voluntary disclosure (2 x 2). We find that the association between financial
statement complexity and voluntary disclosure is relatively weak in the period before RegFD
became effective: the association between financial statement complexity and 8-Ks remains
positive and significant (columns (3) and (4)), but the association between financial statement
complexity and management forecasts is not statistically significant (columns (1) and (2)). In
both periods subsequent to RegFD however, we find that the association between financial
statement complexity and voluntary disclosure remains positive and significant in all our
specifications. In untabulated analyses, across all specifications, we find the difference in the
coefficients on financial statement complexity before and after RegFD is statistically significant
at conventional levels (two-tailed p-values < 0.10). These results are consistent with the notion
that RegFD increases the incentives for managers to use voluntary disclosure (as opposed to
33
Among other changes, Rule8K also implemented Section 409 of SOX that required firms to file an 8-K within
four business days of the consummation of a material contract (e.g., Li, 2013).
32
selective disclosure) to mitigate the informational problems created by complex financial
statements.
5.4 Two quasi-natural experiments
Panel A of Table 10 presents mean and median values for our measures of financial
statement complexity pre- and post-SFAS133, for affected firms and a matched sample of
unaffected firms. Panel B presents mean and median values for our measures of financial
statement complexity pre- and post-SFAS157, for affected and a matched sample of unaffected
firms. Consistent with our text-based measures of complexity reflecting the complexity of the
underlying accounting rules, we find significantly larger increases in both ReadIndex and Length
for firms affected by SFAS 133 and firms affected by SFAS 157.
Figure 3 in Appendix C plots the average values of each measure of voluntary disclosure
for the sample of affected and unaffected firms around the adoption of SFAS 133 and the
adoption of SFAS 157. These plots provide some visual evidence that both groups of firms had
similar levels and trends in voluntary disclosure prior to the rule changes, but widely divergent
levels of voluntary disclosure after the rule change.
Panel C of Table 10 presents results from estimating equation (4). Across all
specifications, we find the average treatment effect is highly statistically significant (DiD ATE tstats range from 2.56 to 4.77). Collectively, the results from our two quasi-natural experiments
suggest managers give specific consideration to the complexity of accounting standards, and help
alleviate concerns that a correlated omitted variable explains both complexity of the financial
statements and the demand for voluntary disclosure.
5.5 Alternative Interpretation: Voluntary disclosure as additional obfuscation
One alternative interpretation of the positive association between financial statement
complexity and voluntary disclosure is that complex financial statements reflect an intentional
33
choice by managers to obfuscate and hide information from investors, and such managers issue
more voluntary disclosure in an attempt to mislead investors and reinforce a “false narrative.”
While plausible, this alternative interpretation is generally inconsistent with the findings from
our cross-sectional tests on firm performance and earnings management (Table 6). Presumably,
if firms were issuing voluntary disclosures to mislead investors, the positive relation between
financial statement complexity and voluntary disclosure would be higher when managers have
greater incentives to obfuscate. However, we find the opposite––the positive relation between
financial statement complexity and voluntary disclosure is lower when managers have greater
incentives to obfuscate.
Nevertheless, in untabulated analyses we conduct two additional sets of tests in an effort
to further address this concern. First, we examine whether the accuracy of management forecasts
varies with financial statement complexity. Lehavy et al. (2011) argue that analysts are less
accurate when financial statements are complex because analysts have less information.
However, managers should not be similarly affected. Consistent with this––and counter to the
notion their forecasts are intended to mislead investors––we find no evidence that managers’
forecasts are less accurate when financial statements are complex, and some evidence that their
forecasts are actually more accurate when financial statements are complex.34 Second,
consistent with an extensive prior literature that finds management forecasts (and voluntary
disclosures more generally) improve liquidity and enhance transparency, we find that
management forecasts and liquidity are positively related in our sample. We argue that the
findings from our cross-sectional tests and these additional tests are inconsistent with the notion
that managers are issuing more voluntary disclosure in order to mislead investors.
34
We note that this result is already somewhat in the literature. While not their primary focus, Hutton et al. (2012)
examine when managers are more accurate than analysts, and find managers are more accurate than analysts when
the Fog index is high. They do not discuss the result further (p. 1233, Table 4).
34
6. Supplemental analyses
6.1 Alternative measures of financial statement complexity
6.1.1 Alternative text-based measures of complexity
Throughout all of our analyses, we treat ReadIndex and Length as two different empirical
measures of the same underlying construct—financial statement complexity. Accordingly, we
include them separately in our regressions. However, in untabulated analyses we include
ReadIndex and Length simultaneously in equation (1) and find that Length subsumes ReadIndex.
The fact that one of our two measures of financial statement complexity subsumes the other is
perhaps not surprising given that these measures are fairly highly, positively correlated (0.49).
The empirical challenge when two variables measure the same underlying construct is that it is
difficult to predict whether and when one measure will load incrementally to the other.
We use two common approaches to address the robustness of our analysis with respect to
the simultaneous inclusion of multiple measures of the same construct. The first approach
follows Barth et al. (1997). We include both measures simultaneously in equation (1) and use an
F-test to determine whether the coefficients on ReadIndex and Length are jointly different from
zero. We find that the coefficients are jointly different from zero (two-tailed p-value < 0.01).
The second approach follows Lang et al. (2012). We conduct a principal component
analysis of our six readability measures and length. We find that only a single factor has an
eigenvalue greater than one, that this factor explains 84% of the variation in all seven variables,
and that this factor loads positively on Length and each of the measures of readability (see factor
output in Table C2 of Appendix C). This provides a degree of confidence that both length and
readability are measuring the same economic construct.35 In untabulated analysis, we find
35
We note that this contrasts with Lang and Stice-Lawrence (2015) who find that––pooling across annual reports
from 42 different countries––Fog and Length are negatively correlated and load on two distinct principal
components.
35
inferences are similar if we consolidate Length and the six measures of readability into a single
text-based measure of financial statement complexity.
6.1.2 Alternative measures of financial statement complexity: Business complexity
In this section, we assess the robustness of our results to an alternative measure of
financial statement complexity based on the complexity of the firm’s business strategy.
Specifically, we regress ReadIndex and Length on two sets of predictors of the firm’s business
complexity. Our first set of predictors includes firm size, leverage, growth opportunities,
absolute returns, acquisitions, capital intensity, capital expenditures, research and development,
the amount raised from stock and debt issuances, cash flow volatility, and indicator variables for
whether the firm is audited by a Big 5 auditor, had a goodwill impairment, and had a
restructuring charge (e.g., Li, 2008; Bushee et al, 2015). Our second set of predictors adds ROA,
Loss and firm fixed effects to these variables (e.g., Lang and Stice-Lawrence, 2015).
The predicted values from these regressions are a linear combination of variables that
measure the complexity of the firm’s business strategy. As such, the predicted value can be
considered as an additional measure of financial statement complexity that combines
readability/length with more traditional measures of business complexity identified in the
literature (e.g., R&D, acquisitions). Variable definitions and results appear in Panel A of Table
C3 in Appendix C.36
Panel B of Table C3 in Appendix C presents results from re-estimating equation (1) using
the predicted and residual values from the respective first-stage model in Panel A. We find that
both the predicted and residual values are positively related to voluntary disclosure, and that the
relation is much stronger for the predicted values. We caveat, that to interpret the predicted
(residual) value as the non-discretionary (discretionary) component of financial statement
36
The computation of cash flow volatility requires five years of data for each firm-year and reduces our sample to
67,008 observations.
36
complexity, one needs to assume that discretion is uncorrelated with the determinants in the
model. Prior literature suggests that may not be the case. For example, Li (2008) interprets the
correlations between Loss and measures of readability and length as evidence of discretion.
Thus, this analysis should not be interpreted as separating non-discretionary and discretionary
components of readability and length. We advise readers that the results from this analysis
should be interpreted with caution.
6.2 Changes in analyst forecasts around the 10-K
We examine whether the relation between financial statement complexity and voluntary
disclosure varies with analysts’ reaction to the 10-K. Analyst forecast behavior is commonly
used to measure the quality of the firm’s information environment (e.g., Lang et al., 2012), so
one can think of these tests as analogous to cross-sectional tests regarding the changes in
liquidity around the 10-K. We construct two new variables that measure the effect of the 10-K on
analysts’ forecasts: (i) the magnitude of the revision in the consensus analyst forecast around the
10-K, and (ii) the ex post accuracy of the revised consensus forecast. Per the regression
specification used in our cross-sectional tests (see for example Table 4), we interact these analyst
variables with our measures of financial statement complexity and include them in equation (1).
Results appear in Table C4 of Appendix C. We find managers are less likely to provide
voluntary disclosure subsequent to filing complex financial statements when analysts revise their
forecasts in response to the 10-K and when those revisions are more ex post accurate. These
findings are consistent with the notion that managers use their forecasts to “guide” analysts’
expectations (e.g,, Ajinkya and Gift, 1984), and are related to prior work that documents
managers have a comparative advantage in forecasting earnings of firms with complex financial
statements––when financial statement complexity is high, analysts are less accurate (e.g., Lehavy
et al., 2011) but managers are more accurate (e.g., Hutton et al., 2012).
37
7. Conclusion
A growing literature documents that complex financial statements negatively affect the
information environment. In this paper, we examine whether managers use voluntary disclosure
to mitigate these negative effects. We test our predictions by analyzing the relation between
financial statement complexity and subsequent voluntary disclosure using three distinct sets of
tests. First, using traditional cross-sectional and within-firm designs, we find that more complex
financial statements are followed by more frequent and immediate voluntary disclosure. Second,
examining cross-sectional variation in the relation between financial statement complexity and
voluntary disclosure, we find that the relation is stronger when liquidity decreases around the
filing of the financial statements and when firms have more outside monitors, and that the
relation is weaker when firms have poor performance and greater earnings management. Third,
employing two quasi-natural experiments, we find that firms affected by the adoption of complex
accounting standards (e.g., SFAS 133 and SFAS 157) increase voluntary disclosure to a greater
extent than unaffected firms. We conduct a battery of robustness tests and supplemental analyses
and find our results are robust to changes in research design and variable measurement.
By using three sets of tests—employing cross-sectional and within-firm designs,
examining cross-sectional variation, and using two quasi-natural experiments—we strive to
alleviate concerns that our collective results are attributable to alternative explanations.
However, we recognize that we cannot definitively rule out the possibility that our results are
driven by an omitted variable, and that more research is needed to further understand the effect
of financial statement complexity on the information environment.
Collectively, our results are consistent with the notion that managers trade off various
disclosure mediums in attempting to achieve an optimal information environment. While prior
research documents that complex financial statements negatively affect the information
38
environment, our results suggest that some firms attempt to mitigate these effects using voluntary
disclosure. Our results suggest a more nuanced view of how financial statement complexity
affects the mosaic of public information about the firm.
39
References
Akins, B., Ng., J., Verdi, R., 2012. Investor competition over information and the pricing of
information asymmetry. The Accounting Review 87, 35–58.
Amihud, Y., 2002. Illiquidity and stock returns: Cross-section and time-series effects. Journal of
Financial Markets 5, 31–56.
Angrist, J., Pischke, J., 2009. Mostly harmless econometrics: An empiricist’s companion.
Princeton: Princeton University Press.
Armstrong, C., Foster, G., Taylor, D., 2015. Abnormal accruals in newly public companies:
Opportunistic misreporting or economic activity? Management Science.
Armstrong, C., Guay, W., Weber, J., 2010. The role of information and financial reporting in
corporate governance and debt contracting. Journal of Accounting and Economics 50, 179–
234.
Ajinkya, B., Bhojraj, S., Sengupta, P., 2005. The association between outside directors,
institutional investors and the properties of management earnings forecasts. Journal of
Accounting Research 43, 343–376.
Ajinkya, B., Gift, M., 1984. Corporate managers’ earnings forecasts and symmetrical
adjustments of market expectations. Journal of Accounting Research, 425–444.
Balakrishnan, K., Billings, M., Kelly, B., Ljungqvist, A., 2014a. Shaping Liquidity: On the
Causal Effects of Voluntary Disclosure, Journal of Finance 69, 2237–2278.
Balakrishnan, K., Core, J., Verdi, R., 2014b. The relation between reporting quality and
financing and investment: evidence from changes in financing capacity. Journal of
Accounting Research 52, 1–36.
Ball, R., Jayaraman, S., Shivakumar, L., 2012. Audited financial reporting and voluntary
disclosure as complements: a test of the confirmation hypothesis. Journal of Accounting and
Economics 53, 136–166.
Bamber, L., Jiang, J., and Wang, I., 2010. What’s my style? The influence of top managers on
voluntary corporate financial disclosure. The Accounting Review 85, 1131–1162.
Barth, M., McNichols, M., Wilson, G., 1997. Factors influencing firms’ disclosures about
environmental liabilities. Review of Accounting Studies 2, 35–64.
Bertrand, M., Duflo, E., Mullinathan, S., 2004. How much should we trust difference in
differences estimates? Quarterly Journal of Economics 119, 249–275.
Beyer, A., Cohen, D., Lys T., Walther B., 2010. The financial reporting environment: Review of
the recent literature. Journal of Accounting and Economics 50, 296–343.
Bloomfield, R., 2002. The incomplete revelation hypothesis and financial reporting. Accounting
Horizons 16, 233–243.
Bloomfield, R, 2008. Discussion of annual report readability, current earnings, and earnings
persistence. Journal of Accounting and Economics 45, 248–252.
Bonsall, S., Miller, B., 2013. The impact of financial disclosure complexity on bond rating
agency disagreement and the cost of debt capital. Working Paper.
Bozanic, Z., Thevenot, M., 2015. Qualitative disclosure and changes in sell-side financial
analysts’ information environment. Contemporary Accounting Research.
Brochet, F., Faurel, L., McVay, S., 2011. Manager-specific effects on earnings guidance: An
analysis of top executive turnover. Journal of Accounting Research 49, 1123-1162.
Bushee, B., Gow, I., Taylor, D, 2015. Linguistic complexity in firm disclosures: Obfuscation or
information? Working Paper.
40
Callen, J., Khan, M., Lu, H., 2013. Accounting quality, stock price delay and future stock
returns. Contemporary Accounting Research 30, 269–295.
Campbell, J., 2015. The fair value of cash flow hedges, future profitability, and stock
returns. Contemporary Accounting Research 32, 243–279.
Chen, S., Matsumoto, D., Rajgopal, S., 2011. Is silence golden? An empirical analysis of firms
that stop giving quarterly earnings guidance. Journal of Accounting and Economics 51, 134–
150.
Coller, M., Yohn, T., 1997. Management forecasts and information asymmetry: An examination
of bid-ask spreads. Journal of Accounting Research 35, 181–191.
Fama, E., MacBeth, J., 1973. Risk, return, and equilibrium: empirical tests. Journal of Political
Economy 81, 607–636.
Francis, J., Nanda, D., Olsson, P., 2008. Voluntary disclosure, earnings quality, and cost of
capital. Journal of Accounting Research 46, 53–99.
Gong, G., Li, L., Xie, H., 2009. The Association between management earnings forecast errors
and accruals. The Accounting Review 84, 497–530.
Graham, J., Harvey, C., Rajgopal, S., 2005. The economic implications of corporate financial
reporting. Journal of Accounting and Economics 40, 3–73.
Grossman, S., Stiglitz, J., 1980. On the impossibility of informationally efficient markets. The
American economic review, 393-408.
Hansen, C., 2007. Generalized least squares inference in panel and multilevel models with serial
correlation and fixed effects. Journal of Econometrics 140, 670–694.
Hirshleifer, D., Teoh, S., 2003. Limited attention, information disclosure, and financial reporting.
Journal of Accounting and Economics 36, 337–386.
Hirst, E., Koonce, L., Venkataraman, S., 2008. Management earnings forecasts: A review and
framework. Accounting Horizons 22, 315–338.
Hutton, A., Lee, L., Shu, S., 2012. Do managers always know better? The relative accuracy of
management and analyst forecasts. Journal of Accounting Research 50, 1217–1244.
Jones, M., Shoemaker, P., 1994. Accounting narratives: A review of empirical studies of content
and readability. Journal of Accounting Literature 13, 142.
Jung, W., Kwon, Y., 1988. “Disclosure when the market is unsure of information endowment of
managers.” Journal of Accounting Research 26, 146–53.
Kim, O., Verrecchia, R., 1991. Market reaction to anticipated announcements. Journal of
Financial Economics 30, 273-309.
Kothari, S.P., Leone, A., Wasley, C., 2005. Performance matched discretionary accruals
measures. Journal of Accounting and Economics 39, 163‒197.
KPMG, 2011. Disclosure overload and complexity: Hidden in plain sight.
Lang, M., Lins, K., Maffett, M., 2012. Transparency, liquidity, and valuation: International
evidence on when transparency matters most. Journal of Accounting Research 50, 729-774.
Lang, M. Stice-Lawrence, L., 2015. Textual analysis and international financial reporting: Large
sample evidence. Working Paper.
Larcker, D., Ormazabal, G., Taylor, D., 2011. The market reaction to corporate governance
regulation, Journal of Financial Economics, 101, 431‒448.
Lawrence, A., 2013. Individual investors and financial disclosure. Journal of Accounting &
Economics 56, 130–147.
Lee, Y., 2012. The effect of quarterly report readability on information efficiency of stock prices.
Contemporary Accounting Research 29, 1137–1170.
41
Lehavy, R., Li, F., Merkley K., 2011. The effect of annual report readability on analyst following
and the properties of their earnings forecasts. The Accounting Review 86, 1087–1115.
Lennox, C., Park, C., 2006. The informativeness of earnings and management’s issuance of
earnings forecasts. Journal of Accounting and Economics 42, 439–458.
Leuz, C., Schrand, C., 2009. Disclosure and the cost of capital: Evidence from firms’ responses
to the Enron shock. Working paper.
Li, F., 2008. Annual report readability, current earnings, and earnings persistence. Journal of
Accounting and Economics 45, 221–247.
Li, E., 2013. Revealing future prospects without forecasts: The case of accelerating material
contract filings. The Accounting Review 88, 1769–1804.
Loughran, T., McDonald, B., 2014. Measuring readability in financial disclosures. Journal of
Finance 69, 1643–1671.
Miller, B., 2010. The effects of financial statement complexity on small and large investor
trading. The Accounting Review, 85, 2107–2143.
Moffitt, K., Burns, M., 2009. What does that mean? Investigating obfuscation and readability
cues as indicators of deception in fraudulent reports. Proceedings of the Fifteenth Americas
Conference on Information Systems, San Francisco, California.
Riedl, E., Serafeim, G., 2011. Information risk and fair values: An examination of equity
betas. Journal of Accounting Research 49, 1083–1122.
Roberts, M., Whited, T., 2013. Chapter 7—Endogeneity in empirical corporate
finance. Handbook of the Economics of Finance, 493-572.
Taylor, D., Verrecchia, R., 2015. Delegated trade and the pricing of public and private
information. Journal of Accounting and Economics, forthcoming.
Verrecchia, R., 1990. Information quality and discretionary disclosure. Journal of Accounting
and Economics 12, 365–80.
Waymire, G., 1985. Earnings volatility and voluntary management forecast disclosure. Journal of
Accounting Research, 268–295.
You, H. Zhang, X., 2009. Financial statement complexity and investor underreaction to 10-K
information. Review of Accounting Studies. 14, 559–586.
42
Appendix A. Variable definitions
Measures of financial statement complexity
ReadIndex
First principal component of the following six measures of readability: Flesch-Kincaid
readability, Gunning Fog readability index, RIX readability, ARI readability, SMOG
readability, and LIX readability. See Appendix B for details.
Length
Natural logarithm of the number of words in the firm’s 10-K
Measures of voluntary disclosure
Frequency
Number of management forecasts issued over the 12 months following the 10-K filing date.
Frequency90
Number of management forecasts issued over the 90 days following the 10-K filing date.
Frequency60
Number of management forecasts issued over the 60 days following the 10-K filing date.
Frequency30
Number of management forecasts issued over the 30 days following the 10-K filing date.
Immediacy
Number of days between the 10-K filing date and the firm's first management forecast
issued thereafter, multiplied by –1.
Number of management forecasts issued between the fiscal year-end date and the 10-K
filing date.
PreFrequency
Additional variables
Size
Natural logarithm of market value of equity, measured at the fiscal year-end.
ROA
Industry-year adjusted return on assets, measured as income before extraordinary items
scaled by total assets.
Loss
Indicator variable equal to one if net income is negative, and zero otherwise.
Leverage
Long term debt plus short term debt, scaled by total assets.
MTB
Market value of equity plus book value of liabilities divided by book value of assets,
measured at the fiscal year-end.
SpecialItems
Special items scaled by total assets.
Returns
Buy and hold return over the 12 months prior to the 10-K filing date.
σReturns
Standard deviation of monthly returns over the 12 months prior to the 10-K filing date.
∆Illiquidity
Average value of the Amihud (2002) measure of illiquidity on day t = 0 and day t = 1
relative to the 10-K filing date, less the average value from t = –5...–50, multiplied by 106.
∆Spread
Average value of the bid-ask spread on day t = 0 and day t = 1 relative to the 10-K filing
date, less the average value from t = –5...–50, multiplied by 100, where bid-ask spread is
calculated as (ask–bid)/price using data on closing prices and quotes from CRSP
NAnalysts
Number of analysts with one-year ahead earnings forecasts as of the 10-K filing date.
NInstitutions
Number of institutional investors as of the 10-K filing date.
AbAcc
Performance-matched abnormal accruals (e.g., Kothari et al., 2005).
43
Appendix B. Details on readability index
We construct our readability index (ReadIndex) as the first principal component of the six measures of readability
defined below. Higher values of ReadIndex correspond to less readable text. Each measure of readability is
normalized prior to conducting the factor analysis. Factor analysis is conducted on the full sample of 10-Ks between
1995 and 2012 prior to requiring data on control variables (sample of 86,236 firm-year observations).
Measures of readability
Flesch Kincaid
The Flesch-Kincaid Readability Index:
0.39 * (number of words / number of sentences)
+ 11.8 * (number of syllables / number of words) – 15.59
LIX
The LIX Readability Index is equal to:
(number of words / number of sentences)
+ (number of words over 6 letters * 100/ number of words)
RIX
The RIX Readability Index:
(number of words with 7 characters or more) / (number of sentences)
Fog
The Gunning Fog Index:
0.4 * (number of words / number of sentences)
+ 40 * (number of words with more than two syllables / number of words)
ARI
The Automated Readability Index (ARI):
4.71 * (number of characters / number of words)
+ 0.5 * (number of words / number of sentences) − 21.43
SMOG
The SMOG Index:
1.043 * sqrt(30 * number of words with more than two syllables / number of sentences) +
3.1291
Principal component output
Factor
1st
2nd
3rd
4th
5th
6th
Eigenvalue
5.62
0.21
0.11
0.05
0.01
0.01
Proportion of the
variation explained
93.62%
3.43%
1.85%
0.82%
0.19%
0.09%
Cumulative
Proportion of the
variation explained
93.62%
97.05%
98.91%
99.72%
99.91%
100.00%
Readability
Measures
FleschKincaid
LIX
RIX
Fog
ARI
SMOG
Correlation matrix
(1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
0.95
0.96
0.96
0.87
0.97
0.94
(2)
0.97
0.96
0.96
0.86
0.95
0.93
(3)
0.99
0.95
0.95
0.90
0.94
0.98
(4)
0.97
0.94
0.93
(5)
0.91
0.84
0.89
0.84
0.86
0.99
0.93
0.84
0.87
(6)
0.98
0.95
0.94
1.00
0.85
(7)
0.96
0.91
0.98
0.90
0.85
0.90
0.92
Legend: (1) ReadIndex, (2) Flesch Kincaid, (3) LIX, (4) RIX, (5) Fog, (6) ARI, (7) SMOG
44
First Principal
Component
Weights
0.17400
0.17584
0.17545
0.16306
0.17321
0.17253
Table 1. Descriptive statistics
Panel A presents descriptive statistics for the variables used in our analysis. Panel B presents correlations between
the variables used in our analysis. Spearman (Pearson) correlations appear above (below) the diagonal. All variables
are as defined in Appendix A. Sample of 72,366 firm-year observations from 1995 to 2012.
Panel A. Distribution of variables
Variable
Observations
Mean
Std
25th
Median
75th
Measures of voluntary disclosure
Frequency
Frequency90
Frequency60
Frequency30
Immediacy
72,366
72,366
72,366
72,366
26,509
2.62
0.59
0.37
0.07
–108.52
5.35
1.37
0.98
0.34
96.94
0.00
0.00
0.00
0.00
–153.00
0.00
0.00
0.00
0.00
–63.00
2.00
0.00
0.00
0.00
–44.00
–0.01
10.25
0.92
0.59
–0.59
9.89
–0.05
10.28
0.52
10.64
5.64
0.00
0.22
1.94
–0.02
0.32
0.14
0.15
–0.11
0.02
4.61
97.97
0.49
1.98
0.23
0.22
1.79
0.07
0.47
0.67
0.10
2.93
1.27
5.80
132.53
0.50
4.21
–0.02
0.02
1.03
–0.01
0.00
–0.25
0.08
–0.04
–0.26
1.00
16.00
0.00
5.55
0.02
0.16
1.29
0.00
0.00
0.05
0.12
0.00
–0.02
2.00
51.00
0.00
6.97
0.09
0.35
2.05
0.00
1.00
0.36
0.18
0.00
0.16
6.00
124.00
1.00
Measures of financial statement complexity
ReadIndex
Length
72,366
72,366
Additional variables
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
∆Illiquidity
∆Spread
NAnalysts
NInstitutions
AbAcc
72,366
72,366
72,366
72,366
72,366
72,366
72,366
72,366
69,066
69,066
72,366
72,366
59,068
45
Table 1. Descriptive statistics (cont’d)
Panel B. Correlation matrix
(1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
(16)
(17)
(18)
(19)
(20)
0.89
0.70
0.34
0.45
0.12
0.25
0.34
0.15
–0.01
0.04
0.01
–0.10
0.04
–0.10
0.02
–0.01
0.34
0.36
–0.05
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
(16)
(17)
(18)
(19)
(20)
0.81
0.66
0.81
0.33
0.43
0.52
0.51
0.75
0.82
0.60
0.14
0.14
0.09
0.04
0.06
0.27
0.27
0.19
0.07
0.14
0.47
0.40
0.35
0.24
0.14
0.11
0.22
0.36
0.30
0.26
0.22
0.10
0.13
0.07
0.09
0.12
0.01
0.01
–0.01
0.01
–0.02
0.02
0.15
0.11
–0.14
0.18
0.14
0.12
0.06
0.02
0.04
–0.05
0.33
0.26
–0.21
–0.13
–0.11
–0.09
–0.06
–0.05
–0.09
–0.11
–0.03
–0.06
–0.04
–0.02
–0.11
–0.10
–0.06
–0.04
–0.04
0.06
0.08
–0.33
–0.01
–0.02
0.01
–0.29
0.05
0.04
0.03
0.00
–0.01
–0.02
–0.03
0.25
0.02
–0.02
0.22
0.15
–0.29
–0.04
–0.07
–0.04
0.00
–0.07
–0.01
–0.06
–0.34
0.14
–0.09
0.13
–0.16
0.48
–0.18
0.13
0.11
0.09
0.05
0.05
0.08
0.13
0.29
0.05
0.03
0.13
–0.04
–0.03
0.02
–0.02
0.04
0.04
0.03
0.02
0.03
0.02
0.05
0.07
0.01
0.01
0.04
–0.01
0.01
–0.03
–0.01
0.20
0.47
0.39
0.29
0.18
0.11
0.19
0.26
0.69
0.13
0.02
0.26
–0.08
–0.18
0.07
–0.12
0.23
0.05
0.48
0.41
0.30
0.17
0.16
0.24
0.41
0.92
0.18
0.11
0.24
–0.08
–0.25
0.17
–0.26
0.29
0.06
0.71
–0.05
–0.05
–0.04
–0.02
–0.02
–0.03
–0.03
–0.08
0.00
0.01
–0.04
0.10
–0.01
–0.02
0.01
–0.02
0.00
–0.08
–0.07
0.80
0.45
0.58
0.12
0.23
0.30
0.13
–0.01
0.02
0.01
–0.09
0.02
–0.09
0.02
0.00
0.31
0.32
–0.05
0.49
0.52
0.08
0.16
0.21
0.11
–0.03
0.02
0.00
–0.06
0.02
–0.06
0.01
0.00
0.23
0.22
–0.04
0.36
0.04
0.07
0.15
0.06
–0.01
0.01
–0.01
–0.04
0.00
–0.01
0.01
0.00
0.17
0.18
–0.02
0.07
0.18
0.15
0.08
0.00
–0.05
0.01
–0.06
–0.03
–0.09
0.01
0.00
0.11
0.15
–0.02
0.49
0.22
0.06
0.04
0.01
–0.04
0.05
–0.02
–0.03
0.01
0.00
0.16
0.18
–0.03
0.35
0.12
0.15
–0.07
–0.04
0.08
–0.03
–0.03
0.02
0.01
0.23
0.32
–0.03
0.05
0.05
0.16
0.12
–0.33
0.16
–0.33
0.06
–0.01
0.70
0.80
–0.07
–0.04
0.04
0.00
0.04
–0.01
0.02
0.00
0.00
0.04
0.07
0.00
–0.17
–0.03
0.05
–0.04
–0.02
0.00
0.01
0.01
0.06
0.01
–0.03
0.12
0.23
0.21
0.02
0.01
0.09
0.05
–0.02
–0.34
0.14
–0.23
0.00
–0.02
0.04
0.06
0.11
–0.16
0.46
–0.01
0.04
–0.18
–0.23
–0.01
0.08
0.00
–0.04
0.03
0.06
–0.01
0.00
0.03
–0.17
–0.24
0.01
0.20
0.03
0.03
–0.01
–0.01
–0.01
0.01
0.73
–0.07
–0.05
Legend: (1) Frequency, (2) Frequency90 (3) Frequency60, (4) Frequency30, (5) Immediacy, (6) ReadIndex, (7) Length, (8) Size, (9) ROA, (10) Leverage, (11)
MTB, (12) SpecialItems, (13) Loss, (14) Returns, (15) σReturns, (16) ∆Illiquidity, (17) ∆Spread, (18) NAnalysts, (19) NInstitutions, (20) AbAcc
46
Table 2. Distribution of voluntary disclosure by quintiles of financial statement complexity
This table presents the average value for each measure of voluntary disclosure by quintile of financial statement
complexity. p-values (two-tailed) appear in brackets, test for a difference in voluntary disclosure between extreme
quintiles, and are based on standard errors clustered by firm and date. All variables are as defined in Appendix A.
Variable
Frequency
Frequency90
Frequency60
Frequency30
Immediacy
Variable
Frequency
Frequency90
Frequency60
Frequency30
Immediacy
Q1
1.27
0.27
0.18
0.05
–123.86
Q1
0.61
0.12
0.09
0.04
–145.78
Q2
2.26
0.49
0.33
0.07
–111.99
ReadIndex quintile
Q3
Q4
2.86
3.30
0.64
0.76
0.42
0.47
0.08
0.09
–104.72
–100.98
Q2
1.30
0.28
0.21
0.06
–112.38
Length quintile
Q3
Q4
2.71
4.3
0.60
0.99
0.43
0.61
0.08
0.09
–95.75
–92.43
47
Q5
3.40
0.78
0.44
0.09
–101.05
Diff
Q5–Q1
2.13
0.51
0.26
0.04
22.81
Diff
p-value
[<0.001]
[<0.001]
[<0.001]
[<0.001]
[<0.001]
Q5
4.17
0.96
0.5
0.11
–96.27
Diff
Q5–Q1
3.56
0.84
0.41
0.07
49.51
Diff
p-value
[<0.001]
[<0.001]
[<0.001]
[<0.001]
[<0.001]
Table 3. Financial statement complexity and voluntary disclosure
This table presents results from estimating the association between our two measures of financial statement
complexity and voluntary disclosure. In Panel A, voluntary disclosure is measured as the number of forecasts over
the 12 months following the filing of the 10-K (Frequency), in Panels B, C, and D voluntary disclosure is measured
as the number of forecasts over the 90, 60, and 30 days following the filing of the 10-K (Frequency90, Frequency60,
and Frequency30 respectively), and in Panel E, voluntary disclosure is measured using the number of days between
the filing date of the 10-K and the first management forecast thereafter, multiplied by minus one (Immediacy).
Columns (1) and (2) (Columns (3) and (4)) of each panel present results measuring financial statement complexity,
using ReadIndex (Length). All variables are as defined in Appendix A. Independent variables are transformed into
decile ranks and scaled to range from 0 to 1. t-statistics appear in parentheses and are based on standard errors
clustered by firm and date. ***, **, and * denote statistical significance at the 0.01, 0.05, and 0.10 levels (two-tail),
respectively.
Panel A. Voluntary disclosure 12 months subsequent to the 10-K filing
Dependent variable: Frequency
FS_Complexity = ReadIndex
Variable
FS_Complexity
Control variables
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Firm effects
Observations
R2 (%)
FS_Complexity = Length
(1)
0.98***
(7.40)
(2)
1.51***
(9.63)
(3)
2.90***
(21.32)
(4)
4.32***
(22.63)
4.62***
(20.30)
3.82***
(24.19)
–0.05
(–0.42)
0.35**
(2.21)
–1.73***
(–15.38)
–0.48***
(–5.18)
–0.29**
(–2.08)
–0.31*
(–1.90)
No
72,366
18.4
6.77***
(17.10)
2.30***
(15.20)
0.69***
(4.13)
–2.33***
(–10.78)
–0.59***
(–7.49)
–0.20***
(–2.72)
0.07
(0.59)
–1.30***
(–6.98)
Yes
72,366
61.2
3.47***
(16.33)
3.56***
(24.11)
–0.27**
(–2.20)
0.91***
(6.06)
–1.70***
(–15.31)
–0.84***
(–9.81)
–0.23**
(–1.98)
–0.32**
(–2.19)
No
72,366
20.4
4.92***
(15.20)
1.77***
(14.04)
0.33**
(2.12)
–1.47***
(–8.16)
–0.37***
(–5.16)
–0.46***
(–6.66)
0.12
(1.21)
–1.30***
(–8.40)
Yes
72,366
63.1
48
Table 3. Financial statement complexity and voluntary disclosure (cont’d)
Panel B. Voluntary disclosure 90 days subsequent to the 10-K filing
Dependent variable: Frequency90
FS_Complexity = ReadIndex
Variable
FS_Complexity
Control variables
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Firm effects
Observations
R2 (%)
(1)
0.27***
(8.22)
(2)
0.41***
(10.26)
(3)
0.71***
(20.84)
(4)
1.06***
(21.27)
1.04***
(18.71)
0.90***
(22.27)
–0.03
(–1.07)
0.05
(1.36)
–0.41***
(–14.47)
–0.12***
(–5.39)
–0.11***
(–3.00)
–0.08*
(–1.94)
No
72,366
15.0
1.62***
(16.40)
0.53***
(13.30)
0.19***
(4.49)
–0.65***
(–11.83)
–0.15***
(–7.07)
–0.08***
(–4.23)
–0.01
(–0.33)
–0.32***
(–6.72)
Yes
72,366
52.5
0.77***
(14.84)
0.83***
(22.14)
–0.09***
(–2.93)
0.19***
(5.19)
–0.40***
(–14.42)
–0.21***
(–9.73)
–0.09***
(–3.09)
–0.08**
(–2.20)
No
72,366
16.7
1.18***
(14.28)
0.40***
(11.78)
0.11***
(2.62)
–0.44***
(–9.44)
–0.10***
(–4.93)
–0.15***
(–7.76)
0.00
(0.02)
–0.32***
(–7.71)
Yes
72,366
54.2
49
FS_Complexity = Length
Table 3. Financial statement complexity and voluntary disclosure (cont’d)
Panel C. Voluntary disclosure 60 days subsequent to the 10-K filing
Dependent variable: Frequency60
FS_Complexity = ReadIndex
Variable
FS_Complexity
Control variables
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Firm effects
Observations
R2 (%)
(1)
0.12***
(5.76)
(2)
0.17***
(6.25)
(3)
0.37***
(15.78)
(4)
0.57***
(19.13)
0.51***
(18.21)
0.56***
(20.70)
–0.07***
(–3.63)
0.07***
(2.66)
–0.24***
(–13.10)
–0.06***
(–3.96)
–0.06**
(–2.43)
–0.05*
(–1.93)
No
72,366
8.9
0.81***
(13.57)
0.35***
(12.46)
0.11***
(3.59)
–0.30***
(–8.82)
–0.08***
(–5.19)
–0.03**
(–2.27)
–0.00
(–0.12)
–0.26***
(–8.45)
Yes
72,366
42.5
0.36***
(12.47)
0.53***
(20.38)
–0.10***
(–4.95)
0.14***
(5.78)
–0.23***
(–12.94)
–0.11***
(–7.32)
–0.05**
(–2.41)
–0.05**
(–2.16)
No
72,366
9.8
0.56***
(10.44)
0.28***
(11.22)
0.06**
(2.15)
–0.18***
(–6.05)
–0.05***
(–3.48)
–0.07***
(–4.95)
0.00
(0.20)
–0.26***
(–9.43)
Yes
72,366
43.5
50
FS_Complexity = Length
Table 3. Financial statement complexity and voluntary disclosure (cont’d)
Panel D. Voluntary disclosure 30 days subsequent to the 10-K filing
Dependent variable: Frequency30
FS_Complexity = ReadIndex
Variable
FS_Complexity
Control variables
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Firm effects
Observations
R2 (%)
(1)
0.00
(0.27)
(2)
0.01*
(1.65)
0.14***
(15.77)
0.08***
(11.74)
–0.00
(–0.21)
0.00
(0.34)
–0.06***
(–10.25)
–0.03***
(–7.06)
–0.03***
(–6.15)
0.04***
(4.94)
No
72,366
3.1
0.13***
(7.76)
0.07***
(7.13)
0.04***
(3.76)
–0.03***
(–2.87)
–0.03***
(–4.60)
–0.01***
(–3.29)
–0.02***
(–4.20)
0.01
(0.72)
Yes
72,366
23.5
51
FS_Complexity = Length
(3)
0.02***
(3.65)
(4)
0.07***
(7.33)
0.13***
(14.15)
0.08***
(11.52)
–0.00
(–0.49)
0.01
(1.09)
–0.06***
(–10.17)
–0.03***
(–7.96)
–0.03***
(–6.02)
0.04***
(4.94)
No
72,366
3.2
0.10***
(5.88)
0.06***
(6.33)
0.03***
(3.14)
–0.01
(–1.32)
–0.02***
(–3.97)
–0.02***
(–4.36)
–0.02***
(–4.09)
0.01
(0.75)
Yes
72,366
23.7
Table 3. Financial statement complexity and voluntary disclosure (cont’d)
Panel E. Voluntary disclosure immediacy subsequent to the 10-K filing
Dependent variable: Immediacy
FS_Complexity = ReadIndex
Variable
FS_Complexity
Control variables
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Firm effects
Observations
R2 (%)
(1)
14.49***
(5.09)
(2)
25.82***
(5.95)
(3)
48.40***
(11.15)
(4)
78.44***
(13.22)
49.58***
(11.72)
46.21***
(11.13)
–6.00**
(–2.28)
–15.81***
(–4.45)
–15.49***
(–7.33)
–5.31***
(–2.84)
–6.19*
(–1.87)
–12.20***
(–2.73)
No
26,509
5.1
116.82***
(11.70)
40.82***
(6.73)
3.74
(0.76)
–55.40***
(–9.43)
–9.32***
(–3.42)
–5.76**
(–2.48)
–3.56
(–1.00)
–20.93***
(–3.71)
Yes
26,509
36.1
30.64***
(6.68)
40.94***
(11.58)
–9.63***
(–3.68)
–5.96*
(–1.86)
–14.92***
(–7.05)
–10.89***
(–5.81)
–6.11**
(–2.00)
–10.63***
(–2.65)
No
26,509
6.6
86.52***
(8.70)
32.96***
(6.30)
–1.08
(–0.23)
–36.93***
(–6.81)
–6.10**
(–2.31)
–9.19***
(–4.10)
–3.50
(–1.06)
–15.37***
(–3.11)
Yes
26,509
37.6
52
FS_Complexity = Length
Table 4. Changes in illiquidity around the filing of the financial statements
This table presents results from examining whether the relation between financial statement complexity (FS_Complexity) and subsequent voluntary disclosure
(Frequency) varies with the change in illiquidity around the 10-K filing. Our model follows the specifications in Panel A of Table 3, except that we interact our
measures of financial statement complexity with our measures of the change in illiquidity around the 10-K filing (∆Illiquidity and ∆Spread). All variables are as
defined in Appendix A. For parsimony we do not tabulate coefficients on control variables. t-statistics appear in parentheses and are based on standard errors
clustered by firm and date. ***, **, and * denote statistical significance at the 0.01, 0.05, and 0.10 levels (two-tail), respectively.
Dependent variable: Frequency
FS_Complexity = ReadIndex
Variable
FS_Complexity
FS_Complexity x ∆Illiquidity
(1)
0.59***
(5.05)
0.85***
(4.79)
FS_Complexity x ∆Spread
∆Illiquidity
–0.27***
(–3.44)
∆Spread
Controls
Firm effects
Observations
R2 (%)
Yes
No
69,066
18.4
(2)
0.76***
(5.76)
0.50***
(3.70)
0.01
(0.20)
Yes
No
69,066
18.4
(3)
0.45***
(3.76)
0.78***
(4.42)
0.34***
(2.62)
–0.28***
(–3.63)
0.07
(1.08)
Yes
No
69,066
18.4
FS_Complexity = Length
(4)
1.06***
(6.59)
0.73***
(4.97)
0.24*
(1.90)
–0.32***
(–4.48)
0.11
(1.56)
Yes
Yes
69,066
61.8
53
(5)
2.15***
(16.43)
1.58***
(7.01)
–0.59***
(–7.53)
Yes
No
69,066
20.4
(6)
2.55***
(18.16)
0.79***
(4.97)
–0.13**
(–2.42)
Yes
No
69,066
20.4
(7)
1.95***
(13.94)
1.49***
(6.62)
0.50***
(3.20)
–0.59***
(–7.50)
–0.02
(–0.32)
Yes
No
69,066
20.4
(8)
3.54***
(19.08)
1.34***
(7.28)
0.30**
(2.07)
–0.57***
(–7.83)
0.04
(0.60)
Yes
Yes
69,066
63.7
Table 5. External monitors
This table presents results from examining whether the relation between financial statement complexity (FS_Complexity) and subsequent voluntary disclosure
(Frequency) varies with the intensity of external monitoring. Our model follows the specifications in Panel A of Table 3, except that we interact our measures of
financial statement complexity with our measures of external monitoring intensity (NAnalysts and NInstitutions). All variables are as defined in Appendix A. For
parsimony we do not tabulate coefficients on control variables. t-statistics appear in parentheses and are based on standard errors clustered by firm and date. ***,
**
, and * denote statistical significance at the 0.01, 0.05, and 0.10 levels (two-tail), respectively.
Dependent variable: Frequency
FS_Complexity = ReadIndex
Variable
FS_Complexity
FS_Complexity x NAnalysts
(1)
–0.87***
(–6.10)
3.40***
(8.44)
FS_Complexity x NInstitutions
NAnalysts
2.70***
(10.00)
NInstitutions
Controls
Firm effects
Observations
R2 (%)
Yes
No
72,366
22.1
(2)
–0.80***
(–6.20)
3.29***
(8.44)
5.84***
(17.50)
Yes
No
72,366
21.9
(3)
–1.37***
(–8.85)
2.61***
(5.78)
1.61***
(3.90)
2.28***
(9.04)
5.25***
(16.88)
Yes
No
72,366
24.1
54
FS_Complexity = Length
(4)
–1.03***
(–5.55)
2.47***
(5.75)
1.85***
(4.36)
–0.71**
(–2.57)
5.36***
(14.80)
Yes
Yes
72,366
62.5
(5)
–0.66***
(–3.75)
6.68***
(15.29)
0.87***
(3.25)
Yes
No
72,366
25.1
(6)
–0.45***
(–2.60)
5.67***
(13.20)
3.95***
(11.46)
Yes
No
72,366
23.9
(7)
–1.47***
(–7.38)
5.98***
(13.99)
1.62***
(3.94)
0.48**
(2.28)
4.67***
(14.93)
Yes
No
72,366
26.8
(8)
–1.66***
(–7.07)
6.04***
(12.72)
3.61***
(7.43)
–2.19***
(–7.46)
2.62***
(7.16)
Yes
Yes
72,366
64.7
Table 6. Firm performance and earning management
This table presents results from examining whether the relation between financial statement complexity (FS_Complexity) and subsequent voluntary disclosure
(Frequency) varies with firm performance and earnings management. Our model follows the specifications in Panel A of Table 3, except that we interact our
measures of financial statement complexity with our measures of firm performance and earnings management (ROA, Loss, and AbAcc). All variables are as
defined in Appendix A. For parsimony we do not tabulate coefficients on control variables. t-statistics appear in parentheses and are based on standard errors
clustered by firm and date. ***, **, and * denote statistical significance at the 0.01, 0.05, and 0.10 levels (two-tail), respectively.
Dependent variable: Frequency
FS_Complexity = ReadIndex
Variable
FS_Complexity
FS_Complexity x ROA
(1)
–0.93***
(–4.82)
4.17***
(10.66)
FS_Complexity x Loss
(2)
1.83***
(9.34)
–1.64***
(–8.04)
FS_Complexity x AbAcc
ROA
Loss
1.35***
(5.97)
–0.63***
(–6.15)
3.53***
(20.21)
0.22*
(1.85)
Yes
No
59,068
18.4
Yes
No
59,068
18.1
AbAcc
Controls
Firm effects
Observations
R2 (%)
(3)
1.88***
(8.17)
–1.20***
(–4.57)
3.52***
(20.09)
–0.61***
(–5.96)
0.21*
(1.73)
Yes
No
59,068
18.0
FS_Complexity = Length
(4)
0.29
(1.09)
3.94***
(10.24)
–1.40***
(–6.85)
–1.25***
(–4.79)
1.48***
(6.56)
0.07
(0.64)
0.25**
(2.07)
Yes
No
59,068
18.6
55
(5)
1.02***
(3.78)
2.40***
(6.95)
–0.63***
(–3.56)
–0.53***
(–2.60)
1.09***
(5.08)
0.04
(0.40)
0.42***
(3.64)
Yes
Yes
59,068
61.1
(6)
1.28***
(5.45)
4.31***
(9.16)
0.72***
(3.15)
–1.14***
(–12.51)
Yes
No
59,068
20.8
(7)
4.49***
(23.13)
–3.09***
(–14.31)
3.15***
(19.45)
0.50***
(4.83)
Yes
No
59,068
20.9
(8)
4.57***
(19.29)
–2.08***
(–7.74)
3.03***
(19.04)
–1.07***
(–11.29)
0.60***
(6.03)
Yes
No
59,068
20.5
(9)
3.51***
(11.62)
3.85***
(8.49)
–2.89***
(–13.67)
–2.17***
(–8.11)
1.08***
(4.80)
0.32***
(3.29)
0.61***
(6.32)
Yes
No
59,068
21.5
(10)
4.53***
(13.86)
3.07***
(7.52)
–1.80***
(–9.04)
–0.92***
(–4.29)
–0.04
(–0.20)
0.36***
(3.32)
0.53***
(5.04)
Yes
Yes
59,068
63.5
Table 7. Robustness—Financial statement complexity and pre-emptive voluntary
disclosure
This table presents results from estimating the association between financial statement complexity and voluntary
disclosure issued prior to the 10–K filing date. Panel A presents results from repeating the tests in Panel A of Table
3, replacing the dependent variable with voluntary disclosure issued between the fiscal year-end and the 10-K filing
date, inclusive (PreFrequency). Panel B presents results from repeating the tests in Panel A of Table 3 after
including PreFrequency as an additional control variable. All variables are as defined in Appendix A. t-statistics
appear in parentheses and are based on standard errors clustered by firm and date. ***, **, and * denote statistical
significance at the 0.01, 0.05, and 0.10 levels (two–tail), respectively.
Panel A. Financial statement complexity and voluntary disclosure prior to the 10-K
Dependent variable: PreFrequency
FS_Complexity = ReadIndex
Variable
FS_Complexity
Control variables
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Firm effects
Observations
R2 (%)
FS_Complexity = Length
(1)
0.26***
(8.23)
(2)
0.42***
(10.26)
(3)
0.69***
(20.99)
(4)
1.08***
(23.18)
0.95***
(18.34)
0.86***
(21.99)
–0.05
(–1.62)
0.08**
(2.05)
–0.40***
(–14.45)
–0.12***
(–5.04)
–0.15***
(–4.00)
–0.08**
(–2.11)
No
72,366
13.2
1.59***
(16.64)
0.48***
(11.50)
0.20***
(4.63)
–0.65***
(–12.09)
–0.14***
(–6.40)
–0.08***
(–3.75)
–0.05
(–1.58)
–0.36***
(–7.83)
Yes
72,366
50.7
0.69***
(14.28)
0.80***
(21.90)
–0.10***
(–3.42)
0.21***
(5.79)
–0.39***
(–14.34)
–0.20***
(–9.27)
–0.13***
(–4.27)
–0.08**
(–2.38)
No
72,366
14.9
1.14***
(14.26)
0.35***
(9.91)
0.12***
(2.79)
–0.43***
(–9.64)
–0.09***
(–4.26)
–0.14***
(–7.18)
–0.04
(–1.47)
–0.36***
(–8.97)
Yes
72,366
52.4
56
Table 7. Robustness—Financial statement complexity and pre-emptive voluntary
disclosure (cont’d)
Panel B. Controlling for voluntary disclosure prior to the 10-K
Dependent variable: Frequency
FS_Complexity = ReadIndex
Variable
FS_Complexity
PreFrequency
Control variables
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Firm effects
Observations
R2 (%)
(1)
0.51***
(6.03)
8.57***
(46.05)
(2)
0.91***
(7.86)
5.72***
(36.83)
(3)
1.35***
(14.45)
8.43***
(44.79)
(4)
2.65***
(18.87)
5.38***
(37.22)
1.74***
(17.55)
1.71***
(19.66)
–0.05
(–0.67)
0.34***
(3.55)
–0.74***
(–10.03)
–0.28***
(–4.50)
0.13
(1.59)
–0.29***
(–2.70)
No
72,366
49.0
3.97***
(15.25)
1.42***
(14.25)
0.38***
(2.86)
–1.32***
(–8.70)
–0.32***
(–4.97)
–0.14**
(–2.38)
0.21**
(2.48)
–0.86***
(–6.19)
Yes
72,366
69.4
1.27***
(12.29)
1.63***
(19.02)
–0.15**
(–2.06)
0.60***
(6.29)
–0.74***
(–10.16)
–0.44***
(–7.63)
0.15*
(1.94)
–0.29***
(–2.89)
No
72,366
49.4
3.01***
(12.63)
1.14***
(12.15)
0.18
(1.38)
–0.85***
(–6.06)
–0.20***
(–3.29)
–0.30***
(–5.26)
0.24***
(3.04)
–0.88***
(–7.15)
Yes
72,366
70.1
57
FS_Complexity = Length
Table 8: Robustness—Alternative measure of voluntary disclosure: 8-Ks
This table presents results from estimating the association between our two measures of financial statement
complexity and the frequency and immediacy of 8-K disclosures. Panel A presents the average value for each
measure of 8-K disclosure by quintile of financial statement complexity. Frequency8K is the number of 8-Ks issued
over the 12 months following the 10-K filing date. Frequency8K90, Frequency8K60, and Frequency8K30 are the
number of 8-Ks issued over the 90, 60, and 30 days following the 10-K filing date respectively. Immediacy8K is the
number of days between the 10-K filing date and the firm's first 8-K filing thereafter, multiplied by –1. Panel B
presents results from repeating the tests in Panels A through E of Table 3, using the each of the five measures of 8-K
disclosure. For each of the five measures, we present results separately for our two measures of financial statement
complexity and two regression designs––pooled and within-firm (5 x 2 x 2). For parsimony we do not tabulate
coefficients on control variables. All variables are as defined in Appendix A. t-statistics (two-tailed p-values) appear
in parentheses (brackets) and are based on standard errors clustered by firm and date. ***, **, and * denote statistical
significance at the 0.01, 0.05, and 0.10 levels (two-tail), respectively.
Panel A. Portfolio sorts
Variable
Frequency8K
Frequency8K90
Frequency8K60
Frequency8K30
Immediacy8K
Variable
Frequency8K
Frequency8K90
Frequency8K60
Frequency8K30
Immediacy8K
Q1
4.93
1.26
0.87
0.35
–84.25
Q1
2.61
0.63
0.44
0.18
–112.03
Q2
7.40
1.94
1.33
0.51
–64.34
Q2
4.90
1.24
0.88
0.33
–76.72
ReadIndex quintile
Q3
Q4
8.67
9.41
2.31
2.55
1.54
1.66
0.60
0.67
–56.58
–52.81
Length quintile
Q3
7.89
2.06
1.42
0.53
–52.21
58
Q4
10.75
2.90
1.91
0.74
–39.75
Q5
9.39
2.52
1.60
0.67
–56.69
Diff
Q5–Q1
4.46
1.26
0.73
0.32
27.56
Diff
p-value
[<0.001]
[<0.001]
[<0.001]
[<0.001]
[<0.001]
Q5
13.64
3.75
2.35
1.01
–33.96
Diff
Q5–Q1
11.03
3.12
1.91
0.83
78.07
Diff
p-value
[<0.001]
[<0.001]
[<0.001]
[<0.001]
[<0.001]
Table 8: Robustness—Financial statement complexity and 8-K disclosures (cont’d)
Panel B. Regressions of 8-K disclosures on financial statement complexity
Dependent variable:
FS_Complexity
ReadIndex
Controls
Within-Firm
R2
ReadIndex
Controls
Within-Firm
R2
Length
Controls
Within-Firm
R2
Length
Controls
Within-Firm
R2
Table 3
Panel A
Table 3
Panel B
Table 3
Panel C
Table 3
Panel D
Table 3
Panel E
Frequency8K
Frequency8K90
Frequency8K60
Frequency8K30
Immediacy8K
(1)
(2)
(3)
(4)
(5)
***
2.99
(12.28)
Yes
No
16.8
***
0.90
(12.44)
Yes
No
12.7
3.60***
(13.13)
Yes
Yes
54.5
***
0.54
(11.10)
Yes
No
7.7
***
0.23
(11.53)
Yes
No
5.2
21.58***
(10.51)
Yes
No
5.7
1.24***
(14.60)
Yes
Yes
45.1
0.72***
(12.23)
Yes
Yes
38.5
0.29***
(11.25)
Yes
Yes
30.3
30.98***
(9.19)
Yes
Yes
34.4
11.02***
(57.93)
Yes
No
33.8
3.21***
(53.27)
Yes
No
27.4
2.05***
(46.31)
Yes
No
19.6
0.83***
(42.53)
Yes
No
11.4
88.93***
(33.42)
Yes
No
15.0
11.33***
(48.21)
Yes
Yes
62.0
3.54***
(46.24)
Yes
Yes
52.3
2.25***
(39.81)
Yes
Yes
44.4
0.88***
(32.48)
Yes
Yes
33.2
107.13***
(28.22)
Yes
Yes
39.9
59
Table 9: Robustness—Time-period analysis
This table presents results from estimating the association between financial statement complexity and voluntary
disclosure subsequent to the 10-K filing over various time periods. Panel A presents results from estimating the
regression specification in Panel A of Table 3 using separate annual cross-sectional regressions. Columns (1) and (2)
present results using Frequency as the dependent variable. Columns (3) and (4) presents results using Frequency8K
as the dependent variable. We report the time-series average of the estimated annual slope coefficients (FM coef.),
their t-statistics (FM t-stat), and the average of the annual regression R2. Panel B presents results from estimating the
regression specification in Panel A of Table 3 over three sub-samples: 1995 to 1999, before Regulation Fair
Disclosure became effective (Pre-RegFD and Pre-Rule8K); 2000 to 2004, after RegFD but before the “Additional
Form 8-K Disclosure Requirements and Acceleration of Filing Date” rule became effective (Post-RegFD and PreRule8K); and 2005 to 2012, after both rules became effective (Post-RegFD and Post-Rule8K). Columns (1) and (2)
present results using Frequency as the dependent variable. Columns (3) and (4) presents results using Frequency8K
as the dependent variable. For parsimony we do not tabulate coefficients on control variables. All variables are as
defined in Appendix A. t-statistics appear in parentheses. ***, **, and * denote statistical significance at the 0.01, 0.05,
and 0.10 levels (two-tail), respectively.
Panel A. Annual cross-sectional regressions
Variable
Statistic
FS_Complexity
FM coef.
FM t-stat.
Controls
#Years
Avg R2
Dependent variable:
Frequency
Dependent variable:
Frequency8K
FS_Complexity =
FS_Complexity =
ReadIndex
(1)
Length
(2)
ReadIndex
(3)
Length
(4)
0.27***
(3.64)
–0.01
(–0.19)
1.04***
(10.20)
4.04***
(8.71)
Yes
18
17.0
Yes
18
16.8
Yes
18
10.6
Yes
18
26.0
Panel B. Time-period subsamples
Dependent variable:
Frequency
FS_Complexity =
Time Period
Variable
1995 to 1999
Pre-RegFD
Pre-Rule8k
FS_Complexity
2000 to 2004
Post-RegFD
Pre-Rule8K
2005 to 2012
Post-RegFD
Post-Rule8K
Controls
N
R2 (%)
FS_Complexity
Controls
N
R2 (%)
FS_Complexity
Controls
N
R2 (%)
ReadIndex
(1)
Length
(2)
ReadIndex
(3)
Length
(4)
–0.02
(–0.63)
Yes
22,102
8.7
–0.03
(–1.17)
Yes
22,102
8.7
0.66***
(7.24)
Yes
22,102
10.7
1.71***
(14.43)
Yes
22,102
12.0
0.41***
(3.21)
Yes
20,742
15.0
0.89***
(5.29)
Yes
20,742
15.2
2.08***
(9.49)
Yes
20,742
16.3
8.03***
(20.50)
Yes
20,742
25.2
0.79***
(3.07)
Yes
29,522
20.8
0.51*
(1.75)
Yes
29,522
20.7
1.32***
(5.85)
Yes
29,522
12.2
5.75***
(20.94)
Yes
29,522
15.5
60
Dependent variable:
Frequency8K
FS_Complexity =
Table 10: Quasi-natural experiments—Changes in accounting standards
This table presents results from estimating the effects of two changes in accounting standards, SFAS 133 and SFAS
157, on voluntary disclosure. For each change in accounting standards, we use a sample of firms affected by the
change (treatment group) and a propensity score matched sample of firms unaffected by the change (control group).
Our analysis spans a window of three years prior to and three years after the change. Panel A presents the difference
in mean and median values of our measures of financial statement complexity before and after SFAS 133. Panel B
presents the difference in mean and median values of our measures of financial statement complexity before and
after SFAS 157. Panel C presents results from using a generalized difference-in-differences design to estimate the
effect of the change in accounting standards on voluntary disclosure. Columns (1) and (2) of Panel C show results
for SFAS 133. Columns (3) and (4) of Panel C show results for SFAS 157. In columns (1) and (2), Post is an
indicator variable equal to one for fiscal periods beginning after June 15, 2000 and zero otherwise, and Affected is an
indicator variable equal to one for firms with data on unrealized gains and losses on derivatives in accumulated other
comprehensive income and zero otherwise. In columns (3) and (4), Post is an indicator variable equal to one for
fiscal periods beginning after November 15, 2007 and zero otherwise, and Affected is an indicator variable equal to
one for firms with data on level 1, level 2 or level 3 assets or liabilities and zero otherwise. All other variables are as
defined in Appendix A. t-statistics (two-tailed p-values) appear in parentheses (brackets) and are based on standard
errors clustered by firm and date. ***, **, and * denote statistical significance at the 0.01, 0.05, and 0.10 levels (twotail), respectively.
Panel A. Changes in financial statement complexity around SFAS133
Affected
Measure of
FS_Complexity
ReadIndex
Length
Unaffected
Statistic
Mean
Median
PreFAS133
–0.04
–0.17
PostFAS133
0.00
–0.16
PreFAS133
–0.04
–0.12
PostFAS133
–0.20
–0.21
DiD
ATE
0.20
0.10
p–value
DiD = 0
[<0.001]
[0.003]
Mean
Median
9.96
9.96
10.19
10.19
9.98
9.99
10.11
10.11
0.10
0.11
[<0.001]
[<0.001]
Panel B. Changes in financial statement complexity around SFAS157
Affected
Measure of
FS_Complexity
ReadIndex
Length
Unaffected
Statistic
Mean
Median
PreFAS157
0.08
0.03
PostFAS157
0.20
0.15
PreFAS157
0.01
–0.02
PostFAS157
0.02
0.03
DiD
ATE
0.11
0.07
p–value
DiD = 0
[<0.001]
[<0.001]
Mean
Median
10.50
10.24
10.65
10.40
10.38
10.15
10.4
10.16
0.13
0.15
[<0.001]
[<0.001]
61
Table 10: Quasi-natural experiments—Changes in accounting standards (cont’d)
Panel C. Generalized difference-in-differences design
Variable
Post * Affected
DiD
ATE
FAS133
FAS157
Dependent variable:
Dependent variable:
Frequency
(1)
0.68***
(4.13)
Frequency8K
(2)
0.48**
(2.56)
2.96***
(5.50)
0.76***
(3.54)
0.00
(0.00)
0.04
(0.15)
–0.18
(–1.48)
–0.22*
(–1.91)
–0.53***
(–4.45)
–0.20
(–1.17)
Firm & Year
10,851
52.3
–2.30***
(–3.72)
–0.13
(–0.44)
–0.15
(–0.38)
2.38***
(5.93)
0.30**
(1.98)
–0.02
(–0.12)
–0.21
(–1.30)
0.24
(0.93)
Firm & Year
10,851
67.6
DiD
ATE
Frequency
(3)
1.17***
(4.77)
Frequency8K
(4)
0.77***
(2.77)
3.77***
(8.07)
0.31*
(1.66)
0.38
(1.16)
0.80**
(2.47)
0.11
(0.86)
–0.41***
(–3.50)
0.14
(0.86)
–0.48**
(–2.23)
Firm & Year
17,060
80.4
–1.46**
(–2.33)
0.20
(0.93)
0.08
(0.24)
1.34***
(3.69)
–0.07
(–0.44)
0.06
(0.46)
–0.51***
(–2.77)
–0.26
(–1.12)
Firm & Year
17,060
67.6
Control variables
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Fixed Effects
Observations
R2 (%)
62
Appendix C. Supplemental analyses
This appendix reports results for additional tests briefly described in the paper.
Contents:
Figure 1. Time-series of voluntary disclosure by quintile of financial statement complexity
Figure 2. Time-series of annual cross-sectional regression coefficients
Figure 3. Time-series of voluntary disclosures for treatment and control firms around SFAS
157 and SFAS 133
Table C1. Quasi-natural experiments: Covariate balance
Panel A. Covariate balance between observations affected and unaffected by SFAS 133
Panel B. Covariate balance between observations affected and unaffected by SFAS 157
Table C2. Principal component analysis including measures of readability and length
Table C3. Decomposing financial statement complexity
Panel A. Models of financial statement complexity
Panel B. Models of voluntary disclosure
Table C4. Changes in analyst forecasts around the filing of the financial statements
63
Figure 1. Cumulative voluntary disclosure by quintile of financial statement complexity
This figure plots cumulative number of management forecasts (Frequency) from the beginning of the fiscal year to
twelve months after the 10-K filing by intervals of one month. The initial value in month zero is the number of
management forecasts issued between the fiscal year-end date and the 10-K filing date. We plot the cumulative
number of forecasts separately for the lowest quintile of financial statement complexity (Quintile 1) and the highest
quintile of financial statement complexity (Quintile 5). Panel A presents results for ReadIndex, Panel B for Length.
Panel A. Cumulative voluntary disclosure by extreme quintiles of ReadIndex
4.5
Quintile 5
Quintile 1
4
3.5
Frequency
3
2.5
2
1.5
1
0.5
0
0
1
2
3
4
5
6
7
8
9
10
11
12
Months after 10-K filing
Panel B. Cumulative voluntary disclosure by extreme quintiles of Length
6
Quintile 5
Quintile 1
5
Frequency
4
3
2
1
0
0
1
2
3
4
5
6
7
8
Months after 10-K filing
64
9
10
11
12
Figure 2. Time-series of annual cross-sectional regression coefficients
This figure plots annual cross-sectional regression coefficients estimated in Panel A of Table 9. Cov(Freq,
ReadIndex) is the coefficient from a regression of Frequency on ReadIndex and control variables. Cov(Freq, Length)
is the coefficient from a regression of Frequency on Length and control variables. Cov(Freq8K, ReadIndex) is the
coefficient from a regression of Frequency8K on ReadIndex and control variables. Cov(Freq8K, Length) is the
coefficient from a regression of Frequency8K on Length and control variables.
7.00
Cov(Freq , ReadIndex)
Cov(Freq8K , ReadIndex)
6.00
Cov(Freq , Length)
Cov(Freq8K , Length)
Coefficient
5.00
4.00
3.00
2.00
1.00
0.00
Year
65
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
1999
1998
1997
1996
1995
-1.00
Figure 3. Time-series of voluntary disclosures for treatment and control firms around SFAS 157 and SFAS 133
This table presents average voluntary disclosure for affected firms and a matched sample of unaffected firms for the pre- and post- periods. Panel A presents results
for SFAS 133. Panel B presents results for SFAS 157.
Panel A. Voluntary disclosure around SFAS 133
Panel B. Voluntary disclosure around SFAS 157
66
Table C1. Quasi-natural experiments: Covariate balance
This table presents cross-sample differences in mean and median values of the control variables used to calculate the
propensity score in our quasi-natural experiment design. Panel A (Panel B) presents the difference in mean and
median values for the firms affected by SFAS 133 (SFAS 157) and their matched sample. p-values (two-tailed) test
for differences between means and medians and appear in brackets.
Panel A. Covariate balance between observations affected and unaffected by SFAS133
Affected
Variable
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Mean
6.44
0.28
0.31
1.89
–0.01
0.14
0.19
0.14
Median
6.36
0.16
0.29
1.19
0.00
0.00
–0.03
0.12
Unaffected
Mean
6.45
0.28
0.31
1.83
–0.01
0.17
0.23
0.14
Median
6.42
0.16
0.28
1.22
0.00
0.00
–0.03
0.12
Affected–Unaffected
Diff. in
means
–0.01
0.00
0.00
0.06
0.00
–0.03
–0.04
0.00
p–value
[0.893]
[0.945]
[0.553]
[0.354]
[0.404]
[0.051]
[0.220]
[0.303]
Diff. in
medians
–0.06
0.00
0.01
–0.03
0.00
0.00
0.00
0.00
p–value
[0.708]
[0.692]
[0.438]
[0.248]
[0.599]
NA
[0.716]
[0.474]
Panel B. Covariate balance between observations affected and unaffected by SFAS157
Affected
Variable
Size
ROA
Leverage
MTB
SpecialItems
Loss
Returns
σReturns
Mean
6.62
1.79
0.21
2.00
–0.01
0.20
0.14
0.09
Median
6.50
0.29
0.15
1.50
0.00
0.00
0.09
0.07
Unaffected
Mean
6.59
1.74
0.21
2.03
–0.01
0.20
0.15
0.09
Median
6.52
0.30
0.15
1.58
0.00
0.00
0.10
0.08
67
Affected–Unaffected
Diff. in
means
0.03
0.05
0.00
–0.03
0.00
0.00
–0.01
0.00
p–value
[0.531]
[0.750]
[0.608]
[0.458]
[0.878]
[0.808]
[0.753]
[0.212]
Diff. in
medians
–0.02
–0.01
0.00
–0.08
0.00
0.00
–0.01
–0.01
p–value
[0.829]
[0.132]
[0.235]
[0.021]
[0.415]
NA
[0.678]
[0.266]
Table C2. Principal component analysis including measures of readability and length
This table presents principal component output from consolidating the six readability measures defined in Appendix
B and Length into one index of financial statement complexity.
Factor
1st
2nd
3rd
4th
5th
6th
7th
Eigenvalue
5.88
0.76
0.20
0.11
0.04
0.01
0.01
Proportion of the
variation explained
83.94%
10.86%
2.80%
1.58%
0.59%
0.15%
0.07%
Cumulative
Proportion of the
variation explained
83.94%
94.81%
97.61%
99.19%
99.77%
99.93%
100.00%
68
Measures
FleschKincaid
LIX
RIX
Fog
ARI
SMOG
Length
First Principal
Component
Weights
0.16698
0.16785
0.16608
0.15409
0.16515
0.16415
0.09263
Table C3. Decomposing financial statement complexity
Panel A presents results from estimating financial statement complexity as a function of characteristics of the firm’s
business environment. Abs_Returns is the absolute value of the buy and hold return over the 12 months prior to the
annual report filing date. Acquisitions is acquisitions scaled by total assets. Capital is net plant, property, and
equipment scaled by total assets. Capex is the amount of capital expenditures scaled by total assets. R&D is the ratio
of research and development expense to sales. Financing is the amount raised from stock and debt issuances during
the year scaled by total assets. σCFO is the standard deviation of cash flows from operations over the prior five
years scaled by total assets. BigN is an indicator variable equal to one if the firm is audited by a Big 5 auditor and
zero otherwise. Goodwill is an indicator variable for whether the firm had a goodwill impairment charge that year.
Restructuring is an indicator variable for whether the firm had a restructuring charge that year. Panel B presents
results from estimating regressions of Frequency on the scaled decile ranks of the predicted and residual values of
the models estimated in Panel A and the control variables used in Table 3. For parsimony we do not tabulate
coefficients on control variables. All other variables are as defined in Appendix A. Sample of 67,008 observations. tstatistics appear in parentheses and are based on standard errors clustered by firm and date. p-values test for
differences between coefficients and appear in brackets. ***, **, and * denote statistical significance at the 0.01, 0.05,
and 0.10 levels (two–tail), respectively.
Panel A. Models of financial statement complexity
Size
Leverage
MTB
Abs_Returns
Acquisitions
Capital
Capex
R&D
Financing
σCFO
BigN
Goodwill
Restructuring
ROA
Loss
Firm effects
Observations
R2 (%)
ReadIndex
Coeff.
t-stat.
0.11***
(20.58)
0.23***
(6.49)
–0.04***
(–7.50)
0.07
(0.08)
0.23***
(2.91)
–0.57*** (–12.75)
–0.03
(–0.24)
0.04***
(7.34)
0.10***
(3.57)
0.34***
(5.86)
0.06***
(2.88)
0.05***
(3.27)
0.14***
(8.72)
No
67,008
8.8
ReadIndex
Coeff.
t-stat.
0.06***
(6.11)
0.15***
(3.55)
–0.03***
(–7.06)
–2.65***
(–3.69)
0.13**
(2.08)
–0.30***
(–3.91)
–0.07
(–0.62)
–0.01**
(–2.16)
–0.02
(–0.80)
–0.12*
(–1.74)
–0.13***
(–5.40)
0.08***
(4.82)
0.10***
(7.15)
0.00***
(2.72)
0.10***
(8.33)
Yes
67,008
60.2
Length
Coeff.
t-stat.
0.13***
(29.89)
0.31***
(14.97)
–0.06*** (–19.11)
4.79***
(6.89)
–0.50*** (–10.44)
–0.05*
(–1.72)
–0.53***
(–6.47)
0.03***
(13.88)
0.12***
(8.16)
0.49***
(14.67)
–0.23*** (–16.34)
0.07***
(7.42)
0.22***
(14.85)
No
67,008
23.2
Length
Coeff.
t-stat.
0.12***
(18.54)
0.25***
(9.21)
–0.05*** (–19.08)
1.17***
(2.82)
–0.18***
(–4.80)
–0.23***
(–4.88)
–0.53***
(–8.56)
–0.00
(–1.28)
–0.04***
(–3.38)
0.07*
(1.95)
–0.19*** (–15.41)
0.17***
(18.41)
0.23***
(18.59)
0.01***
(11.02)
0.14***
(21.03)
Yes
67,008
71.5
Panel B. Models of voluntary disclosure
PredictedValue
ResidualValue
1st Stage Model (1)
Coeff.
t-stat.
4.88***
(16.02)
1.25***
(7.80)
two-tailed p-value: predicted – residual = 0
[<0.001]
Controls
Yes
Firm effects
Yes
Observations
67,008
R2 (%)
61.6
Dependent variable: Frequency
1st Stage Model (2)
1st Stage Model (3)
Coeff.
t-stat.
Coeff.
t-stat.
5.54***
(21.12)
4.75***
(20.33)
0.70***
(7.30)
3.28***
(20.37)
[<0.001]
Yes
Yes
67,008
62.7
69
[<0.001]
Yes
Yes
67,008
63.0
1st Stage Model (4)
Coeff.
t-stat.
5.91***
(22.37)
1.81***
(18.77)
[<0.001]
Yes
Yes
67,008
63.5
Table C4. Changes in analyst forecasts around the filing of the financial statements
This table presents results from examining whether the relation between financial statement complexity
(FS_Complexity) and subsequent voluntary disclosure (Frequency) varies with the change in analyst forecasts
around the 10-K filing. Our model follows the specifications in Panel A of Table 3, except that we interact our
measures of financial statement complexity with our measures of the change in analyst forecasts around the 10-K
filing (∆Accuracy and Revision). ∆Accuracy is the difference in Accuracy between the first analyst consensus
forecast after the 10-K filing and the last analyst consensus forecast prior to the 10-K filing, where Accuracy is
defined as the absolute value of the difference between the mean analyst consensus forecast and realized EPS, scaled
by price at the beginning of the fiscal year and multiplied by negative one. Revision is the absolute value of the
difference between the first median analyst consensus forecast after the 10-K filing and the last median analyst
consensus forecast prior to the 10-K filing, scaled by price at the beginning of the fiscal year. All other variables are
as defined in Appendix A. For parsimony we do not tabulate coefficients on control variables. t-statistics appear in
parentheses and are based on standard errors clustered by firm and date. ***, **, and * denote statistical significance at
the 0.01, 0.05, and 0.10 levels (two-tail), respectively.
Dependent variable = Frequency
FS_Complexity = ReadIndex
Variable
FS_Complexity
FS_Complexity x ∆Accuracy
(1)
1.52***
(6.61)
–0.19
(–0.70)
FS_Complexity x Revision
∆Accuracy
–0.06
(–0.42)
Revision
Controls
Firm effects
Observations
R2 (%)
Yes
No
45,714
18.4
(2)
1.75***
(7.36)
–0.78**
(–2.45)
1.00***
(5.83)
Yes
No
45,714
18.5
70
(3)
1.64***
(6.54)
0.45
(1.46)
–1.01***
(–2.83)
–0.81***
(–4.44)
1.37***
(6.81)
Yes
No
45,714
18.6
FS_Complexity = Length
(4)
4.51***
(19.10)
–1.04***
(–4.08)
0.31**
(2.46)
Yes
No
45,714
21.3
(5)
4.81***
(19.75)
–1.83***
(–5.67)
1.03***
(7.01)
Yes
No
45,714
21.3
(6)
4.80***
(18.46)
0.08
(0.27)
–1.92***
(–5.43)
–0.48***
(–3.24)
1.28***
(7.49)
Yes
No
45,714
21.4