You are on page 1of 17

Do Managers Really Guide Through the Fog?

On the Challenges in Assessing the Causes of Voluntary Disclosure

Travis Dyer
Mark Lang*
Lorien Stice-Lawrence

July 2016

Guay, Samuels, and Taylor (2016) document that firms with longer and more complex 10-Ks
provide relatively more voluntary disclosure, which they interpret as evidence that managers use
voluntary disclosure to mitigate negative effects of complex mandatory disclosure. We review the
results of Guay et al. and focus on two main challenges to inferring causality: (1) the coincidence
of upward over-time trends in annual report length, complexity, and voluntary disclosure, and (2)
the potential for omitted correlated variables, such as changes in firm economics, to drive changes
in 10-K textual characteristics and voluntary disclosure. While the results in Guay et al. are
extensive and robust, we argue that economic drivers of, and trends in, voluntary disclosure present
important avenues for future research.

The University of North Carolina at Chapel Hill * Corresponding author: Mark Lang, Kenan-Flagler
Business School, the University of North Carolina, Chapel Hill, NC 27516-3490, Mark_Lang@unc.edu.

Electronic copy available at: http://ssrn.com/abstract=2839274

I.

Introduction
The causes and consequences of complexity in financial disclosure are important and

recurring themes in the academic accounting and finance literatures. There is little disagreement
that financial reporting in general, and the annual report in particular, have become lengthier and
more complex. However, whether those trends reflect managerial discretion and are beneficial to
investors remain open and important questions. A major challenge in the literature is in
convincingly separating disclosure choices from underlying economics. In this discussion, we
explore these issues in the context of Guay, Samuels, and Taylor (2016) and use them to
highlight potential paths for future research.
Historically there have been two primary, non-mutually exclusive, views on the
complexity and length of disclosure which differ in terms of: (1) the desirability of lengthy and
complex disclosure for investors, and (2) the extent to which length and complexity are primarily
under the control of management (as opposed to reflecting the effects of mandated disclosure or
underlying economics).
The more traditional view is that if investors have unlimited processing capacity, then
lengthy and technically complex disclosure is, on balance, potentially beneficial. In particular,
investors can ignore disclosure that is irrelevant, so additional disclosure is not harmful.
Similarly, investors can efficiently process lengthy sentences and multisyllabic words, exploiting
outside resources as necessary to aid in interpreting complex technical disclosure, so that
complex disclosure is also not problematic. This view seems increasingly relevant given
advances in computing power and the increased prominence of large, sophisticated institutional
investors in the market. Under this perspective, lengthy and complex disclosure is consistent

Electronic copy available at: http://ssrn.com/abstract=2839274

with efforts by regulators and managers to increase the information available to investors and
enhance the transparency of capital markets.
While this perspective has conceptual appeal, more recently the view of lengthy and
complex disclosure has begun to change. Relying on research such as Grossman and Stiglitz
(1980) in which costly information acquisition plays an important role, papers like Li (2008)
argue that complex and lengthy disclosure may be costlier to process and, therefore, less likely to
be informative. This view is prominent in recent concerns about disclosure overload and in
empirical research such as Miller (2010), Lawrence (2013), and Loughran and McDonald
(2014). The underlying assumption is that many investors do not have unlimited processing
ability and disclosure that is complex and lengthy can obfuscate important information. Li (2008)
builds on that notion by arguing that annual report text is, to a substantial extent, under the firms
control and managers who have something to hide (e.g., poor performance) may therefore have
incentives to intentionally increase opacity by resorting to lengthy and complex disclosure.
Guay et al. (2016) introduce a third, related, perspective. As in Li (2008), they argue that
length and complexity can create opacity. However, consistent with research such as Cazier and
Pfeiffer (2015) and Dyer et al. (2016), they argue that the additional length and complexity in 10Ks likely reflect factors largely outside the control of the firm such as increased regulatory
disclosure or complex fundamentals. A primary innovation is in treating annual report disclosure
complexity and length as being largely exogenous from the standpoint of the firm and examining
managements response in terms of enhanced voluntary disclosure such as management forecasts
to mitigate the information processing costs associated with complex disclosure. This view takes
the notion from Li (2008) that length and complexity tend to reduce the informativeness of
disclosure but argues that in many cases, rather than exploiting complexity and length to
2

obfuscate, managements goal may be to guide through the Fog by increasing supplementary
voluntary disclosure such as management forecasts to clarify complex and lengthy mandatory
disclosure.
II.

Summary of Guay et al. (2016) Results


Using management forecasts as their primary measure of voluntary disclosure, Guay et

al. (2016) provide consistent evidence across a range of specifications supporting the conclusion
that voluntary disclosure tends to be greater when annual reports are longer and more complex.
For example, in Table 2 there is a monotonic relation between both the frequency and immediacy
of management forecasts and quintiles of both annual report length and readability. That
conclusion is reinforced in the Table 3 regressions in which management forecast frequency is
strongly associated with annual report readability and length after including a range of controls
including firm fixed effects. In general, results for length tend to be stronger than those for
readability (and there is some evidence that length subsumes readability), consistent with
research such as Loughran and McDonald (2014).1
The authors then conduct a series of cross-sectional analyses using interactions to
document that the relation is, in fact, strongest where one would expect it to be strongest. The
results in Table 4 suggest that the increase in management forecast frequency is higher in cases
in which greater length and complexity are associated with greater illiquidity increases around
the 10-K release, suggesting that managers are more likely to provide additional disclosure if
lengthy and complex text in the annual report creates information asymmetry among market

The fact that length is the stronger result here is interesting in its own right. One might expect that it would be
relatively easy to ignore extraneous text while interpreting complex text would be more difficult. It would be
interesting to consider the interaction between length and complexity given that it might be particularly difficult for
investors to determine whether specific text is extraneous if the lengthy text also tends to be complex.

participants. In Table 5 results are stronger when there is greater institutional ownership and
analyst following, which the authors interpret as managers being more likely to use voluntary
disclosure to mitigate uncertainty when there is a greater level of monitoring. In Table 6 results
are weaker in cases in which firms are less profitable, reporting losses, or appear to be managing
earnings, suggesting that managers are less likely to provide additional disclosure in cases in
which there are greater incentives to obfuscate.
While the results of the interactions support the authors predictions, there are some
potentially surprising aspects that may merit further investigation. First, while the interactions
are generally consistent with expectations, in several cases the coefficients of the main effects of
the variables used in the interactions are counterintuitive. For example, in Table 4 the main effect
for the change in illiquidity on forecast frequency is negative. This suggests that managers do not
always respond to 10-Ks which increase illiquidity by increasing voluntary disclosure; they only
increase voluntary disclosure if the increase in illiquidity is accompanied by lengthy and
complex text. It is unclear why managers would be more inclined to provide guidance in cases in
which the source of information asymmetry resulted from complex text in the annual report
relative to other sources. Similarly, in Table 5 the main effect for number of analysts conditional
on institutional ownership is sometimes negative suggesting that the link between voluntary
disclosure and analysts may be subtler than a simple monitoring story would suggest. Finally, in
Table 6 the main effect for earnings management is significantly positive suggesting that
managers who have more to hide are more likely to provide forecasts; they are less likely to
provide management forecasts only if the annual report text is complex and redundant. As with
the main effects for information asymmetry and analyst following, the results suggest that there

may be something unique about complexity and length that provides particularly strong
incentives to supplement annual reports with additional disclosure.
Second, while the monitoring story underlying Table 5 seems plausible in the sense that
analysts and institutions may have greater influence over management, they also represent
relatively sophisticated users. Results in papers such as Miller (2010) and Lawrence (2013)
suggest that complex disclosure is more of an issue for small retail investors, consistent with
more sophisticated financial statement users such as analysts and institutions having a greater
capacity to deal with (and gain an information advantage from) length and complexity. Given
that sophisticated users are less likely to be disadvantaged by (and might gain an advantage
from) complex and lengthy disclosure, it would be worth exploring why their presence would
create particularly strong incentives to mitigate the effects of lengthy and complex disclosure.
Guay et al. (2016) conduct a number of additional analyses which provide supporting
evidence consistent with their overall conclusions, including a difference-in-difference analysis
around SFAS 133 and 157 and analyses using press releases and Forms 8-K as measures of
voluntary disclosure. Taken as a whole, the paper provides a thorough and cohesive set of results
supporting the general finding of a positive association between management forecast frequency
and length and complexity of 10-K disclosure. Our goal in the rest of this discussion is to focus
on two challenges in interpreting the primary results causally which provide opportunities for
future research: (1) the coincidence of upward trends in annual report length, complexity, and
voluntary disclosure over the sample period; and (2) the potential effects of omitted correlated
variables underpinning both 10-K characteristics and voluntary disclosure.

III.

Disentangling Time Trends


A primary challenge for the literature is in convincingly separating disclosure choices

from the effects of underlying economics and other factors. This is particularly an issue for
documents such as the 10-K because it mixes the effects of mandatory disclosure (e.g., new
accounting standards) with the effects of underlying economics (e.g., mergers and other major
events that necessitate lengthy and complex disclosure) and voluntary disclosure (e.g., discretion
in how material events are discussed). This challenge is certainly not unique to Guay et al.
(2016). Perhaps the most notable example is Li (2008) and the accompanying discussion in
Bloomfield (2008). Li (2008) provides evidence that annual reports of firms reporting losses tend
to be longer and more complex, which he interprets as suggesting that managers of poorlyperforming firms have incentives to obfuscate by providing disclosure that is more difficult for
investors to interpret. However, as Bloomfield (2008) points out, a challenge to that
interpretation is the possibility that the economic events underpinning losses may themselves
necessitate lengthier and more complex disclosure.
Guay et al. (2016) face similar challenges in their setting. First is the potential effects of
time trends. One of the striking features of the Guay et al. (2016) analysis is the robustness to
inclusion of firm fixed effects. A typical concern with a pooled cross-sectional/time series design
is the possibility that omitted firm-level characteristics might drive the results. That is clearly not
the case here. As Table 3 in Guay et al. (2016) illustrates, results are generally stronger with firm
fixed effects than without, suggesting that within-firm time series variation drives the results.
The fact that results are stronger in the time series than the cross section raises the issue
of whether there might be important omitted time period factors that explain, at least in part, both
the within-firm changes in 10-K characteristics as well as the changes in voluntary disclosure.
6

For example, improvements in technology and data analysis could reduce the cost of preparing
and disseminating lengthy and complex disclosure as well as increasing the demand for
management forecasts by investors and the incentives and capacity of aggregators such as
I/B/E/S to gather the data. As a result, one might observe trends in the length and complexity of
10-Ks as well as trends in management forecast frequency without a direct causal link.
One way to get a sense for the potential importance of time period effects to the relation
between annual report characteristics and voluntary disclosure is to simply plot the trends over
time. In Figure 1 Panel A, we graph the average annual report length and the average number of
management forecasts over time. The overall pattern is striking in that both series trend markedly
upward over the sample period. Reflecting this shared trend, the two series are correlated at the
98% level. Similar results hold for management forecast immediacy in Panel B with both
immediacy and length trending upward strongly. Results are also similar for readability in Panel
C over the 2000-2011 period.2
Given the absence of controls for time period in the primary analysis in Guay et al.
(2016), these trends suggest that, for example, the splits across quintiles for the length variable in
Table 2 are largely comparisons within firm over time as opposed to across firms. Confirming
that conclusion, Tables 11 and 12 in Guay et al. (2016) illustrate that the association between
textual characteristics and voluntary disclosure becomes substantially weaker after controlling
for time period effects. For example, the coefficient on length drops from 2.90 in Table 3 to 0.26
in Table 11 when year fixed effects are added to the base specification and the t-statistic drops
from 21.32 to 1.98. Similarly, the coefficient on readability drops from 0.98 in Table 3 to 0.21 in

Readability increases significantly between 1998 and 2000, which Li (2008) attributes to the advent of the SECs
Plain English initiative in 1998.

Table 11 and the t-statistic drops from 7.40 to 1.91. Turning to the Fama-MacBeth specifications
in Table 12, the coefficient on length switches sign and becomes insignificant (-0.10, t-statistic 0.10) and the coefficient on readability drops to 0.29 (t-statistic 3.72).
This is not to question the authors overall conclusions. As they note, the results are
consistent with the notion that the trend in voluntary disclosure is a response to the trend in
annual report length and complexity, and it is beyond the scope of Guay et al. (2016) to
completely disentangle the two effects. That being said, it does raise interesting questions for
future research. While trends in 10-K disclosure have been investigated in research such as Li
(2008) and Dyer et al. (2016), trends in voluntary disclosure in general, and management
forecasts in particular, have received less attention yet appear to be striking and worthy of
investigation. While one possibility is that the trends in management forecasts reflect a
managerial response to the length and complexity of annual report disclosure, there could be
other potential explanations for the shared trends.
One way to get a sense for factors that might influence trends in management forecasts is
to consider changes in the types of forecasts included in the I/B/E/S database over time. Guay et
al. (2016) use the number of forecasts per firm as the primary unit of observation. As a result, a
given observation for the frequency measure takes on a value of 1 if only an EPS forecast was
provided and is incremented for each additional item forecasted (e.g., sales, cash flows, etc.)
during the window. In Figure 2, we graph the average number of I/B/E/S forecasts per firm per
year over time for EPS forecasts and for all forecasts (multiple types of forecasts per
observation). The top line which includes all forecasts mirrors Figure 1 Panel A, while the
bottom line plots the frequency of EPS forecasts (all forecast events in the dataset have at least
an EPS forecast). Up until 2000, I/B/E/S forecasts were largely limited to EPS forecasts, but after
8

2000 multiple forecasts became very common. While the number of total forecasts increased
steadily over time, the trend in overall forecast frequency is driven entirely by non-EPS forecasts
after 2003. In fact, the number of firms forecasting (as evidenced by the number of EPS
forecasts) peaks in 2002-2003 and, if anything, decreases after that point.
While that does not address the question of why firms became more likely to issue
multiple types of forecasts after 2000, it suggests that understanding the reason for multiple
forecasts is key to understanding the overall trend. One potential cause is changes in I/B/E/S data
gathering over time. In particular, it seems unlikely that firms were not providing any forecasts
other than EPS prior to 2000. Rather, it is possible that I/B/E/S expanded the range of forecasts
that it included in the database. For example, if improvements in technology since 1995 (or client
demand) made it more attractive to collect additional forecast information from the press releases
and corporate event transcripts that serve as the primary data sources for I/B/E/S, then one might
observe the apparent trends in forecasts over time because of changes in data gathering protocol.
Consistent with that conjecture, I/B/E/S notes a change in the number of forecasted items
collected around 2002, 2003, and again around 2007.3 As a result, the trends may be, at least in
part, an artifact of the I/B/E/S data collection protocol. This is also potentially important in
interpreting the sub period results in Table 12 Panel B. In particular, Guay et al. (2016) split the
sample to capture the effects of Regulation FD and Rule 8K in 2000 and 2005, which coincide
approximately with the changes in I/B/E/S forecast variables.

In particular, until the end of 2002 I/B/E/S provides only management forecasts of EPS and sales, with the addition
of operating cash flows in 2003 and capital expenditures, dividends, EBITDA, EPS including special items, gross
margin, operating profit, pretax income, ROA, and ROE beginning in 2007.

Although it is not clear what caused the increases in forecasts over time, it would be
interesting to know whether it was, in part, an artifact of the data collection process by I/B/E/S.
In particular, did the number of non-EPS forecasts made by firms actually increase so
substantially over time or did I/B/E/S simply begin to gather a broader set of forecasts? If the
number of forecasts did, in fact, increase so markedly, it would be worth understanding whether
such a substantial increase in voluntary disclosure reflected trends in annual report complexity or
other changes in the costs and benefits of disclosure during the period.
IV.

Timing of Management Forecasts


A second finding that may merit further analysis is the timing of the management

forecasts relative to the annual report release date. As noted earlier, a potential concern with
causally linking voluntary disclosure to annual report complexity is the possibility that economic
events (e.g., mergers, acquisitions, restructurings, or other disruptions) occurred which both
increased the length and complexity of annual report disclosure and also created an incentive to
provide management forecasts to mitigate the resulting uncertainty. As a consequence, the
positive correlation between annual report complexity and forecasts could be related to omitted
economic variables which are not entirely controlled for in the regression.4
An alternate approach to provide evidence on whether it is the text in the 10-K that
causes the increase in management forecasts is to consider the timing of the forecasts. As the
authors note, it is unclear ex ante when exactly to expect voluntary disclosure to mitigate
complex mandatory disclosure. For example, management might pre-empt complex mandatory
disclosure with enhanced voluntary disclosure or might wait until after the complex mandatory

Research such as Cazier and Pfeiffer (2015), for example, documents that annual report length is strongly
correlated with a variety of economic fundamentals including size, operating complexity, volatility, and firm events
such as acquisitions or the presence of special items.

10

disclosure is issued. However, if the voluntary disclosure is intended to mitigate complexity in


the text of the 10-K, it seems likely that the timing of the voluntary disclosure would at least be
correlated with the timing of the 10-K release.5 If, on the other hand, voluntary disclosure is a
response to forces that are not directly associated with the 10-K release (e.g., other events
affecting the firm), we might expect the frequency of disclosure during the year to be essentially
uncorrelated with the 10-K release date.
One way to get a sense for the spread of forecasts over time is based on the results for
alternate windows presented in Table 2 of Guay et al. (2016). The Diff Q5-Q1 column
documents the difference in forecast frequency between the most and least lengthy and complex
10-Ks based on alternate time windows. As the end of the window moves further away from the
10-K release date, the strength of the relation between complexity and voluntary disclosure
increases monotonically, suggesting that the forecasts are not particularly clustered around the
annual report release date. At some level, that is not surprising because the number of days in the
window is increasing. However, even dividing through by the number of days in the window to
focus on forecasts per day, the forecast frequency is higher further from the 10-K release date.
Further, Table 8 shows that the main results are similar in the pre-10-K period, suggesting that
disclosure changes are not limited to the post-10-K period.
To give a better sense for the pattern, in Figure 3 we plot voluntary disclosure over the
period 12 months prior to the 10-K to 12 months after for the highest and lowest complexity
observations. While the placebo and pre-/post- analyses in Guay et al. (2016) suggest that at least

While it is difficult to predict the timing of voluntary disclosure ex ante, it seems likely that there would be some
time clustering across firms. It is more difficult to envision a scenario in which there are equal incentives for the
typical firm to disclose in each month from twelve months prior to twelve months after the 10-K release if the
forecasts are intended to mitigate the effects of complex text in the 10-K.

11

a portion of the voluntary disclosure is clustered around the 10-K release, Figure 3 suggests that
the bulk of the effect seems uncorrelated with the 10-K timing.6 Again, this is not to dispute the
Guay et al. (2016) general conclusions. Rather, it is simply to highlight the difficulty in causally
tying management forecasts to characteristics of text in the 10-K. Given that the Guay et al.
(2016) results are stronger with firm fixed effects and do not simply reflect cross-sectional
variation, it seems likely that there would be trigger events that cause firms to begin (or increase)
forecasting. Research on the timing of forecasts would seem to have the potential to help
disentangle the causality behind management forecasts.
An impressive feature of Guay et al. (2016) is the wide range of tests and specifications
that yield similar conclusions and help mitigate causality concerns. A primary set of robustness
tests demonstrates that similar results hold for 8-K and press release frequency. These tests are
potentially useful in mitigating some concerns about causality and data issues (e.g., I/B/E/S
collection effects). However, it is possible that those measures are affected by some of the same
types of factors that influence forecasting behavior. For example, the 8-K is described by the
SEC as, the current report companies must file with the SEC to announce major events that
shareholders should know about.7 Similarly, press releases typically reflect discussion of major
events. While there is likely to be a voluntary element to 8-Ks and press releases, it also seems
likely that, in many cases, major events could result in an increase in management forecasts,
press releases, and 8-Ks, as well as longer and more complex disclosure in the annual report.

The fact that the pattern is similar with firm fixed effects suggests that the issue is not simply that more complex
firms always have more complex disclosure and therefore always disclose more.
7
https://www.sec.gov/answers/form8k.htm

12

V.

Conclusion
Guay et al. (2016) provide novel and robust evidence that firms with longer and more

complex 10-Ks provide relatively more voluntary disclosure, which they interpret as evidence
that managers use voluntary disclosure to mitigate the negative effects of complex 10-K text.
Our goal is to highlight some of the challenges in causally linking annual report textual
characteristics to voluntary disclosure and to highlight potential directions for future research.
Going forward, there is substantial scope for additional research attempting to parse out
more nuanced causal connections and underlying mechanisms. For example, Bushee et al. (2015)
have begun to develop measures to separate informative complex disclosure from complex
disclosure which obfuscates. Topic modeling techniques such as Latent Dirichlet Allocation are
also likely to be useful in attempting to understand the content of complex disclosure.
Similarly, it would be interesting to investigate why textual length appears to be such an
important construct, in many cases subsuming other measures of complexity, given that it seems
relatively straightforward to ignore extraneous text. From a standard setters perspective, are
there better ways to organize text to permit investors to identify important information
(potentially similar to XBRL)? Potential synergies between archival and experimental research
could help researchers better understand human processing of textual information and identify
more useful ways to present lengthy and complex disclosure. Finally, there appear to be striking
trends in mandatory and voluntary disclosure over time, perhaps reflecting technological
advances, that would benefit from further study.

13

References
Bloomfield, R., 2008. Discussion of annual report readability, current earnings, and earnings
persistence. J. Account. Econ. 45, 248-252.
Bushee, B., Gow, I., Taylor, D., 2015. Linguistic complexity in firm disclosures: obfuscation or
information? Available at SSRN 2375424.
Cazier, R.A., Pfeiffer, R.J., 2015. Why are 10-K filings so long? Account. Horiz. 30, 1-21.
Dyer, T., Lang, M.H., Stice-Lawrence, L., 2016. The evolution of 10-K textual disclosure:
evidence from Latent Dirichlet Allocation. Available at SSRN 2741682.
Grossman, S.J., Stiglitz, J.E., 1980. On the impossibility of informationally efficient markets.
Am. Econ. Rev. 70, 393-408.
Guay, W., Samuels, D., Taylor, D., 2016. Guiding through the fog: financial statement
complexity and voluntary disclosure. J. Account. Econ., forthcoming.
Lawrence, A., 2013. Individual investors and financial disclosure. J. Account. Econ. 56, 130147.
Li, F., 2008. Annual report readability, current earnings, and earnings persistence. J. Account.
Econ. 45, 221-247.
Loughran, T., McDonald, B., 2014. Measuring readability in financial disclosures. J. Financ. 69,
1643-1671.
Miller, B.P., 2010. The effects of reporting complexity on small and large investor trading.
Account. Rev. 85, 2107-2143

14

Figure 1. Trends Over Time in Voluntary Disclosure and Financial Statement Complexity

Frequency

4
3

2
1

Frequency

21.6

21.5
21.4
21.3
21.2

21.1

21

Fog

Fog

Frequency

21.7

Frequency

-135

Immediacy

-115

-175

Figure 1 Panel C:
Average Frequency and Fog Over Time

-95

-155

decile(Length)

0.75
0.7
0.65
0.6
0.55
0.5
0.45
0.4
0.35
0.3

-75

Immediacy

0.8
0.75
0.7
0.65
0.6
0.55
0.5
0.45
0.4
0.35
0.3

decile(Length)

Figure 1 Panel B:
Average Immediacy and Decile Length Over Time

decile(Length)

Figure 1 Panel A:
Average Frequency and Decile Length Over Time

decile(Length)

Notes: Frequency is the number of all management forecasts (EPS, Sales,


CFO, etc.) issued over the 12 months following the firm's 10-K filing date.
Immediacy is the number of days between 10-K filing date and the firm's
first management forecast issued thereafter, multiplied by -1.
Decile(Length) is the decile rank of the number of words used in a firm's
10-K. Fog is measured as 0.4*(average number of words per sentence in a
firm's 10-K + percent of complex words in a firm's 10-K), where complex
words are words in excess of two syllables. Management forecast data
come from I/B/E/S and First Call. Length and Fog data come from 10-Ks
downloaded from EDGAR.

Notes: Frequency is the number of all management forecasts (EPS, Sales, CFO, etc.) issued over the 12
months following the firm's 10-K filing date. Frequency (EPS Forecasts) is the number of EPS forecasts
issued over the 12 months following the firm's 10-K filing date. Management forecast data come from
I/B/E/S and First Call.

Notes: Frequency is the number of all management forecasts (EPS, Sales, CFO, etc.) issued over the 12
months following the firm's 10-K filing date. Quintile(Length) is the quintile rank of the number of words
used in a firm's 10-K. Management forecast data come from I/B/E/S and First Call.